diff --git a/.clang-format b/.clang-format new file mode 100644 index 00000000000..96184e36d91 --- /dev/null +++ b/.clang-format @@ -0,0 +1,44 @@ +# +# Shogun clang-format configuration file +# +Language: Cpp +AccessModifierOffset: -4 +AlignAfterOpenBracket: AlwaysBreak +AllowAllParametersOfDeclarationOnNextLine: true +AllowShortBlocksOnASingleLine: false +AllowShortCaseLabelsOnASingleLine: false +AllowShortFunctionsOnASingleLine: None +AllowShortIfStatementsOnASingleLine: false +AllowShortLoopsOnASingleLine: false +AlwaysBreakTemplateDeclarations: true +BinPackArguments: true +BinPackParameters: true +BreakBeforeBraces: Custom +BraceWrapping: + AfterClass: true + AfterControlStatement: true + AfterEnum: true + AfterFunction: true + AfterNamespace: true + AfterObjCDeclaration: false + AfterStruct: true + AfterUnion: true + BeforeCatch: true + BeforeElse: true + IndentBraces: false +ColumnLimit: 80 +ContinuationIndentWidth: 4 +IndentWidth: 4 +NamespaceIndentation: All +PenaltyBreakComment: 10000 +PenaltyBreakBeforeFirstCallParameter: 19 +PenaltyBreakFirstLessLess: 120 +PenaltyBreakString: 1000 +PenaltyExcessCharacter: 10000000 +PenaltyReturnTypeOnItsOwnLine: 60 +PointerAlignment: Left +ReflowComments: true +SortIncludes: true +Standard: Cpp11 +TabWidth: 4 +UseTab: ForIndentation diff --git a/.gitignore b/.gitignore index aa34f8a780a..20415f123e3 100644 --- a/.gitignore +++ b/.gitignore @@ -52,53 +52,6 @@ cpplint.py /src/configure-*-*.c* /src/build-local -# modular interfaces -/src/interfaces/*_modular/*.doxy -/src/interfaces/*_modular/Evaluation.i -/src/interfaces/*_modular/Regression.i -/src/interfaces/*_modular/Library.i -/src/interfaces/*_modular/Distribution.i -/src/interfaces/*_modular/Structure.i -/src/interfaces/*_modular/Classifier.i -/src/interfaces/*_modular/Features.i -/src/interfaces/*_modular/Kernel.i -/src/interfaces/*_modular/Preprocessor.i -/src/interfaces/*_modular/Distance.i -/src/interfaces/*_modular/Clustering.i -/src/interfaces/*_modular/SGBase.i -/src/interfaces/*_modular/IO.i -/src/interfaces/*_modular/Mathematics.i -/src/interfaces/*_modular/ModelSelection.i -/src/interfaces/*_modular/modshogun.i -/src/interfaces/*_modular/modshogun_ignores.i -/src/interfaces/*_modular/*_includes.i -/src/interfaces/*_modular/Makefile -/src/interfaces/*_modular/Converter.i -/src/interfaces/*_modular/Multiclass.i -/src/interfaces/*_modular/Machine.i -/src/interfaces/*_modular/Transfer.i -/src/interfaces/*_modular/Loss.i -/src/interfaces/*_modular/Statistics.i -/src/interfaces/*_modular/Latent.i -/src/interfaces/*_modular/GaussianProcess.i - -# particular modular ones -/src/interfaces/csharp_modular/*.cs -/src/interfaces/csharp_modular/abstract_types_extension.i -/src/interfaces/csharp_modular/modshogun.dll -/src/interfaces/java_modular/*.java -/src/interfaces/java_modular/*.jar -/src/interfaces/java_modular/*.class -/src/interfaces/java_modular/org/* -/src/interfaces/java_modular/shogun/* -/src/interfaces/python_modular/*.py -/src/interfaces/python_modular/abstract_types_extension.i -/src/interfaces/r_modular/*.R -/src/interfaces/r_modular/*.RData -/src/interfaces/perl_modular/*.pm -/src/interfaces/octave_modular/abstract_types_extension.i -/.duped_py_pl.pb - # /examples/ *.log *.exe @@ -108,9 +61,9 @@ cpplint.py !/examples/undocumented/libshogun/*.cpp !/examples/undocumented/libshogun/CMakeLists.txt !/examples/undocumented/libshogun/tools/ -!examples/undocumented/python_modular/graphical/ -!examples/undocumented/python_modular/*.py -!examples/undocumented/python_modular/CMakeLists.txt +!examples/undocumented/python/graphical/ +!examples/undocumented/python/*.py +!examples/undocumented/python/CMakeLists.txt # /tests /tests/unit/shogun-unit-test @@ -120,12 +73,8 @@ cpplint.py /tests/unit/*.json /tests/unit/combined_kernel.weights Testing/ -examples/undocumented/python_modular/serialized_svm.bz2 -examples/undocumented/python_modular/tmp/blaah.asc -examples/undocumented/python_modular/tmp/blaah.h5 -examples/undocumented/python_modular/tmp/blaah.json -examples/undocumented/python_modular/tmp/blaah.xml -examples/undocumented/python_modular/vw_cache.dat.cache +examples/undocumented/python/serialized_svm.bz2 +examples/undocumented/python/vw_cache.dat.cache # cmake /CMakeFiles/ @@ -262,6 +211,7 @@ local.properties # CDT-specific (C/C++ Development Tooling) .cproject +.idea # JDT-specific (Eclipse Java Development Tools) .classpath diff --git a/.gitmodules b/.gitmodules index c5a3755999b..d67d6ee7dce 100644 --- a/.gitmodules +++ b/.gitmodules @@ -5,3 +5,6 @@ [submodule "debian"] path = debian url = git://github.com/shogun-toolbox/shogun-debian.git +[submodule "gpl"] + path = src/gpl + url = git://github.com/shogun-toolbox/shogun-gpl.git diff --git a/.travis.yml b/.travis.yml index bfcfc87275a..5b0a6e34839 100644 --- a/.travis.yml +++ b/.travis.yml @@ -1,5 +1,6 @@ sudo: required language: cpp +cache: ccache notifications: email: false irc: @@ -16,63 +17,74 @@ matrix: services: docker - compiler: clang services: docker + - compiler: gcc + services: docker + env: + - CMAKE_OPTIONS="-DCMAKE_BUILD_TYPE=Debug -DENABLE_COVERAGE=ON -DENABLE_CCACHE=OFF" + - CODE_COVERAGE=1 - compiler: clang services: docker - python: "2.7_with_system_site_packages" - language: python env: - - CMAKE_OPTIONS="-DPythonModular=ON -DTRAVIS_DISABLE_UNIT_TESTS=ON -DTRAVIS_DISABLE_LIBSHOGUN_TESTS=ON" + - CMAKE_OPTIONS="-DINTERFACE_PYTHON=ON -DTRAVIS_DISABLE_UNIT_TESTS=ON -DTRAVIS_DISABLE_LIBSHOGUN_TESTS=ON" - CC=clang - CXX=clang++ - compiler: clang services: docker - language: ruby env: - - CMAKE_OPTIONS="-DRubyModular=ON -DTRAVIS_DISABLE_UNIT_TESTS=ON -DTRAVIS_DISABLE_LIBSHOGUN_TESTS=ON" + - CMAKE_OPTIONS="-DINTERFACE_RUBY=ON -DTRAVIS_DISABLE_UNIT_TESTS=ON -DTRAVIS_DISABLE_LIBSHOGUN_TESTS=ON" - CC=clang - CXX=clang++ - compiler: clang services: docker - language: java env: - - CMAKE_OPTIONS="-DJavaModular=ON -DTRAVIS_DISABLE_UNIT_TESTS=ON -DTRAVIS_DISABLE_LIBSHOGUN_TESTS=ON" + - CMAKE_OPTIONS="-DINTERFACE_JAVA=ON -DTRAVIS_DISABLE_UNIT_TESTS=ON -DTRAVIS_DISABLE_LIBSHOGUN_TESTS=ON" - CC=clang - CXX=clang++ - compiler: clang services: docker env: - - CMAKE_OPTIONS="-DCSharpModular=ON -DTRAVIS_DISABLE_UNIT_TESTS=ON -DTRAVIS_DISABLE_LIBSHOGUN_TESTS=ON" + - CMAKE_OPTIONS="-DINTERFACE_CSHARP=ON -DTRAVIS_DISABLE_UNIT_TESTS=ON -DTRAVIS_DISABLE_LIBSHOGUN_TESTS=ON" - CC=clang - CXX=clang++ - compiler: clang services: docker env: - - CMAKE_OPTIONS="-DLuaModular=ON -DTRAVIS_DISABLE_UNIT_TESTS=ON -DTRAVIS_DISABLE_LIBSHOGUN_TESTS=ON" + - CMAKE_OPTIONS="-DINTERFACE_LUA=ON -DTRAVIS_DISABLE_UNIT_TESTS=ON -DTRAVIS_DISABLE_LIBSHOGUN_TESTS=ON" - CC=clang - CXX=clang++ - compiler: gcc services: docker env: - - CMAKE_OPTIONS="-DOctaveModular=ON -DTRAVIS_DISABLE_UNIT_TESTS=ON -DTRAVIS_DISABLE_LIBSHOGUN_TESTS=ON" - - compiler: clang - services: docker - env: - - CMAKE_OPTIONS="-DRModular=ON -DTRAVIS_DISABLE_UNIT_TESTS=ON -DTRAVIS_DISABLE_LIBSHOGUN_TESTS=ON" - - CC=clang - - CXX=clang++ + - CMAKE_OPTIONS="-DINTERFACE_OCTAVE=ON -DTRAVIS_DISABLE_UNIT_TESTS=ON -DTRAVIS_DISABLE_LIBSHOGUN_TESTS=ON" + - INTERFACE_OCTAVE=true - compiler: clang services: docker env: - - CMAKE_OPTIONS="-DScalaModular=ON -DTRAVIS_DISABLE_UNIT_TESTS=ON -DTRAVIS_DISABLE_LIBSHOGUN_TESTS=ON" + - CMAKE_OPTIONS="-DINTERFACE_R=ON -DTRAVIS_DISABLE_UNIT_TESTS=ON -DTRAVIS_DISABLE_LIBSHOGUN_TESTS=ON" - CC=clang - CXX=clang++ +# - compiler: clang +# services: docker +# env: +# - CMAKE_OPTIONS="-DScala=ON -DTRAVIS_DISABLE_UNIT_TESTS=ON -DTRAVIS_DISABLE_LIBSHOGUN_TESTS=ON" +# - CC=clang +# - CXX=clang++ before_install: - docker pull shogun/shogun-dev - - echo $CC; echo $CXX - - docker run -t -d -P -e "JAVA_HOME=/usr/lib/jvm/java-8-oracle" -e "CC=$CC" -e "CXX=$CXX" --name devenv -v $PWD:/opt/shogun shogun/shogun-dev /bin/sh -c "mkdir /opt/shogun/build;bash" + - perl -pe 's/\$(\w+)/$ENV{$1}/g' configs/shogun-sdk/travis.env.in > travis.env + - docker run -t -d -P --env-file travis.env --name devenv -v $HOME/.ccache:/root/.ccache -v $PWD:/opt/shogun shogun/shogun-dev /bin/sh -c "mkdir /opt/shogun/build;bash" before_script: - docker exec -t devenv /bin/sh -c "cd /opt/shogun/build; cmake -DCMAKE_INSTALL_PREFIX=$HOME/shogun-build -DENABLE_TESTING=ON $CMAKE_OPTIONS .." script: + - | + if [ $CC == "gcc" ] && [ -z ${INTERFACE_OCTAVE} ]; then + docker exec -t devenv /bin/sh -c "cd /opt/shogun/; if [ "$TRAVIS_PULL_REQUEST" != "false" ]; then ./scripts/check_format.sh "$TRAVIS_PULL_REQUEST_BRANCH" "$TRAVIS_BRANCH"; fi" + fi - docker exec -t devenv /bin/sh -c "cd /opt/shogun/build; make -j2" - docker exec -t devenv /bin/sh -c "cd /opt/shogun/build; make install" - docker exec -t devenv /bin/sh -c "cd /opt/shogun/build; ctest --output-on-failure -j 2" +after_success: + - | + if [ $CODE_COVERAGE ] ; then + docker exec -t devenv /bin/sh -c "cd /opt/shogun/build; lcov --directory src/shogun/ --capture --output-file all_coverage.info; lcov --remove all_coverage.info '/usr/*' > coverage.info; rm all_coverage.info; codecov -X gcov" + fi diff --git a/CMakeLists.txt b/CMakeLists.txt index 9f6901104c9..6c48bed88a5 100644 --- a/CMakeLists.txt +++ b/CMakeLists.txt @@ -35,12 +35,21 @@ SET(CMAKE_INCLUDE_DIRECTORIES_PROJECT_BEFORE ON) set(CMAKE_CXX_STANDARD 11) set(CMAKE_CXX_STANDARD_REQUIRED ON) set(CMAKE_CXX_EXTENSIONS OFF) +if (MSVC) + if (MSVC_VERSION VERSION_LESS 1900) + message(FATAL_ERROR "C++11 is required to use Shogun, but the version of Visual Studio you are using is too old and doesn't support C++11. You need Visual Studio 2015 or newer. ") + else() + include(CheckCXXCompilerFlag) + CHECK_CXX_COMPILER_FLAG("/std:c++14" _cpp_latest_flag_supported) + if (_cpp_latest_flag_supported) + add_compile_options("/std:c++14") + endif() + endif() +endif() ### FIXME: remove these flags when the codebase # is cleared up set(HAVE_CXX11 ON) -set(HAVE_CXX11_ATOMIC ON) -set(HAVE_STD_UNORDERED_MAP ON) ############# minimum library versions ################### SET(EIGEN_VERSION_MINIMUM 3.1.2) @@ -85,18 +94,19 @@ SET(EXT_SRC_TEST_TMP "${EXT_SRC_TEST}.jinja2") SET(THIRD_PARTY_DIR ${CMAKE_SOURCE_DIR}/third_party) SET(LIBSHOGUN_SRC_DIR ${CMAKE_SOURCE_DIR}/src/shogun) -SET(COMMON_MODULAR_SRC_DIR ${CMAKE_SOURCE_DIR}/src/interfaces/modular/) - -SET(AVAILABLE_INTERFACES PythonModular;OctaveModular;JavaModular;PerlModular;RubyModular;CSharpModular;RModular;LuaModular;ScalaModular) -SET(PythonModularDescription "Python") -SET(OctaveModularDescription "Octave") -SET(JavaModularDescription "Java") -SET(PerlModularDescription "Perl") -SET(RubyModularDescription "Ruby") -SET(CSharpModularDescription "C#") -SET(RModularDescription "R") -SET(LuaModularDescription "Lua") -SET(ScalaModularDescription "Scala") +SET(COMMON_INTERFACE_SRC_DIR ${CMAKE_SOURCE_DIR}/src/interfaces/swig/) + +SET(AVAILABLE_INTERFACES + INTERFACE_PYTHON;INTERFACE_OCTAVE;INTERFACE_JAVA;INTERFACE_PERL;INTERFACE_RUBY;INTERFACE_CSHARP;INTERFACE_R;INTERFACE_LUA;INTERFACE_SCALA) +SET(INTERFACE_PYTHON_DESCRIPTION "Python") +SET(INTERFACE_OCTAVE_DESCRIPTION "Octave") +SET(INTERFACE_JAVA_DESCRIPTION "Java") +SET(INTERFACE_PERL_DESCRIPTION "Perl") +SET(INTERFACE_RUBY_DESCRIPTION "Ruby") +SET(INTERFACE_CSHARP_DESCRIPTION "C#") +SET(INTERFACE_R_DESCRIPTION "R") +SET(INTERFACE_LUA_DESCRIPTION "Lua") +SET(INTERFACE_SCALA_DESCRIPTION "Scala") SET(LIBSHOGUN ON CACHE BOOL "Compile shogun library") IsAnyTrue("${AVAILABLE_INTERFACES}" ANY_INTERFACE_ENABLED) @@ -104,7 +114,7 @@ IF (${ANY_INTERFACE_ENABLED}) # SWIG3 is the minimum requirement because of C++11 support SET(SWIG_VERSION_MINIMUM 3.0.0) - IF(CSharpModular) + IF(INTERFACE_CSHARP) # We require SWIG 3.0.7 to support functions with a few SGVector or # SGMatrix arguments. The required SWIG feature is called # "Support for special variable expansion in typemap attributes." @@ -115,7 +125,7 @@ IF (${ANY_INTERFACE_ENABLED}) # typemapping created for earlier versions of SWIG. # see: http://www.swig.org/Doc3.0/CSharp.html#CSharp_introduction_swig2_compatibility LIST(APPEND CMAKE_SWIG_FLAGS "-DSWIG2_CSHARP") - ELSEIF(PythonModular) + ELSEIF(INTERFACE_PYTHON) # SWIG was broken for combining -builtin and -modernargs # from v3.0.0 and until 3.0.4. This bug was fixed in # v3.0.5. Make CMake emit an error and fail to configure. @@ -128,7 +138,7 @@ IF (${ANY_INTERFACE_ENABLED}) IF(ENABLE_CCACHE AND CCACHE_SWIG) SET(CCACHE_SWIG_EXECUTABLE ${CCACHE_SWIG}) ENDIF() - SET(COMPILE_MODULAR_INTERFACE 1) + SET(COMPILE_INTERFACE 1) ENDIF() # Detect OS @@ -137,18 +147,14 @@ DetectSystemName() # Get processor type, sets MACHINE macro SET(MACHINE ${CMAKE_SYSTEM_PROCESSOR}) -SET(EXT_LIB_SWIG_RUBY_MODULAR ".so") +SET(EXT_LIB_SWIG_RUBY ".so") if(DARWIN) - SET(EXT_LIB_SWIG_RUBY_MODULAR ".bundle") + SET(EXT_LIB_SWIG_RUBY ".bundle") ENDIF() ################ COMPILER ####################### # g++ version needs to be => 4.3 IF(CMAKE_COMPILER_IS_GNUCXX) - # in order to support cmake 2.8.7 and older - IF(NOT CMAKE_CXX_COMPILER_VERSION) - include(CheckCompiler) - ENDIF() IF("${CMAKE_CXX_COMPILER_VERSION}" VERSION_LESS "4.3.0") MESSAGE(FATAL_ERROR "g++ version is too old") ENDIF() @@ -184,6 +190,7 @@ IF(MSVC) SET(CMAKE_CXX_FLAGS_DISTRIBUTION "/Ot") SET(CMAKE_C_FLAGS_DEBUG "/DEBUG /Od /Zi") SET(CMAKE_CXX_FLAGS_DEBUG "/DEBUG /Od /Zi") + add_compile_options("/bigobj") ELSE() SET(CMAKE_C_FLAGS_RELEASE "-O3 ${RELEASE_COMPILER_FLAGS}") SET(CMAKE_CXX_FLAGS_RELEASE "-O3 ${RELEASE_COMPILER_FLAGS}") @@ -308,7 +315,7 @@ OPTION(BUILD_META_EXAMPLES "Generate API examples from meta-examples" ON) # note the examples dir is added below after tests have been defined ################# DATATYPES ################# -IF(COMPILE_MODULAR_INTERFACE) +IF(COMPILE_INTERFACE) OPTION(USE_CHAR "Support for char datatype" ON) OPTION(USE_BOOL "Support for bool datatype" ON) OPTION(USE_UINT8 "Support for uint8_t datatype" ON) @@ -323,10 +330,10 @@ IF(COMPILE_MODULAR_INTERFACE) OPTION(USE_FLOAT64 "Support for float64_t datatype" ON) OPTION(USE_COMPLEX128 "Support for complex128_t datatype" ON) OPTION(USE_FLOATMAX "Support for floatmax_t datatype" OFF) -ENDIF(COMPILE_MODULAR_INTERFACE) +ENDIF(COMPILE_INTERFACE) # detect word size -IF(CMAKE_SIZEOF_VOID_P EQUAL 8 AND NOT DARWIN) +IF(CMAKE_SIZEOF_VOID_P EQUAL 8 AND NOT (DARWIN OR WIN32)) SET(SWIGWORDSIZE64 TRUE) ENDIF() @@ -338,19 +345,6 @@ EndForEach(Interface) # Debugging Python-interface with CTest OPTION(ENABLE_PYTHON_DEBUG "Enable Python-interface-debugging with CTest" OFF) -# Allow to hide non-bsd compatible codes -OPTION(LICENSE_GPL_SHOGUN "Include GPL codes of Shogun (non-BSD compatible) in build" ON) -IF (LICENSE_GPL_SHOGUN) - SET(USE_GPL_SHOGUN 1) -ELSE() - SET(USE_GPL_SHOGUN 0) -ENDIF() - -# SVMLight -OPTION(USE_SVMLIGHT "SVMLight" ON) -IF(USE_SVMLIGHT AND NOT USE_GPL_SHOGUN) - MESSAGE(FATAL_ERROR "Can only use SVMLight when GPL codes are included") -ENDIF() SET(SVMLightWarning "\n!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!") SET(SVMLightWarning "${SVMLightWarning}\nWARNING: SHOGUN is built using SVMlight which was written") SET(SVMLightWarning "${SVMLightWarning}\nby Thorsten Joachims and uses a different non GPL compatible license.") @@ -379,9 +373,6 @@ OPTION(USE_PATHDEBUG "Viterbi path debugging" OFF) # big states OPTION(USE_BIGSTATES "Big (16bit) state" ON) -# Large file -OPTION(HAVE_LARGEFILE "Large file support" ON) - #kernelcache to use 4-byte-floating-point values instead of 8-byte-doubles OPTION(USE_SHORTREAL_KERNELCACHE "Kernelcache to use 4-byte-floating-point values instead of 8-byte-doubles" ON) @@ -390,9 +381,6 @@ SET(MEXP "19937" CACHE STRING "PRNG Mersenne exponent") SET(SFMT_MEXP ${MEXP}) SET(DSFMT_MEXP ${MEXP}) -# Reference counting -OPTION(USE_REFERENCE_COUNTING "Reference Counting" ON) - OPTION(USE_LOGCACHE "Use (1+exp(x)) log cache (is much faster but less accurate)" OFF) OPTION(USE_LOGSUMARRAY "Use sum array, supposed to be a bit more accurate" OFF) @@ -417,6 +405,9 @@ IF (LIBSHOGUN) MergeCFLAGS() add_subdirectory(${CMAKE_SOURCE_DIR}/src/shogun) add_library(shogun::shogun ALIAS shogun) + if(LIBSHOGUN_BUILD_STATIC) + add_library(shogun::shogun-static ALIAS shogun-static) + endif() set(shogun_INCLUDE_DIR ${CMAKE_BINARY_DIR}/src) ELSE() find_package(Shogun ${VERSION} CONFIG REQUIRED) @@ -427,9 +418,9 @@ ForEach(SwigFlag "-w473" "-w454" "-w312" "-w325" "-fvirtual") LIST(APPEND CMAKE_SWIG_FLAGS ${SwigFlag}) EndForEach() -OPTION(SWIG_SINGLE_THREADED "Build modular interfaces single-threaded to reduce memory usage" OFF) +OPTION(SWIG_SINGLE_THREADED "Build interfaces single-threaded to reduce memory usage" OFF) -OPTION(USE_SWIG_DIRECTORS "Enable SWIG director classes" OFF) +OPTION(USE_SWIG_DIRECTORS "Enable SWIG director classes" ON) # Respect system's or distribution's C[XX]FLAGS. OPTION(SWIG_WITH_SYSTEM_CFLAGS "Enable system's C[XX]FLAGS for compilation of swig-binaries" ON) @@ -448,71 +439,71 @@ IF(REDUCE_SWIG_DEBUG) SET(SWIG_CXX_COMPILER_FLAGS "${SWIG_CXX_COMPILER_FLAGS} -g1") ENDIF(REDUCE_SWIG_DEBUG) -# python modular -IF (PythonModular) - IF(EXISTS ${CMAKE_SOURCE_DIR}/src/interfaces/python_modular) - add_subdirectory(${CMAKE_SOURCE_DIR}/src/interfaces/python_modular) +# python +IF (INTERFACE_PYTHON) + IF(EXISTS ${CMAKE_SOURCE_DIR}/src/interfaces/python) + add_subdirectory(${CMAKE_SOURCE_DIR}/src/interfaces/python) ENDIF() ENDIF() -# lua modular -IF (LuaModular) - IF(EXISTS ${CMAKE_SOURCE_DIR}/src/interfaces/lua_modular) - add_subdirectory(${CMAKE_SOURCE_DIR}/src/interfaces/lua_modular) +# lua +IF (INTERFACE_LUA) + IF(EXISTS ${CMAKE_SOURCE_DIR}/src/interfaces/lua) + add_subdirectory(${CMAKE_SOURCE_DIR}/src/interfaces/lua) ENDIF() ENDIF() -# scala modular -IF (ScalaModular) +# scala +IF (INTERFACE_SCALA) # Java needed because Scala extends Java Classes and uses the executable generated from Java example for Integration testing - set(JavaModular "ON") + set(INTERFACE_JAVA "ON") FIND_PACKAGE(Scala REQUIRED) - IF(NOT JavaModular) - IF(EXISTS ${CMAKE_SOURCE_DIR}/src/interfaces/java_modular) - add_subdirectory(${CMAKE_SOURCE_DIR}/src/interfaces/java_modular) + IF(NOT INTERFACE_JAVA) + IF(EXISTS ${CMAKE_SOURCE_DIR}/src/interfaces/java) + add_subdirectory(${CMAKE_SOURCE_DIR}/src/interfaces/java) ENDIF() ENDIF() ENDIF() -# java modular -IF (JavaModular) - IF(EXISTS ${CMAKE_SOURCE_DIR}/src/interfaces/java_modular) - add_subdirectory(${CMAKE_SOURCE_DIR}/src/interfaces/java_modular) +# java +IF (INTERFACE_JAVA) + IF(EXISTS ${CMAKE_SOURCE_DIR}/src/interfaces/java) + add_subdirectory(${CMAKE_SOURCE_DIR}/src/interfaces/java) ENDIF() ENDIF() -# ruby modular -IF (RubyModular) - add_subdirectory(${CMAKE_SOURCE_DIR}/src/interfaces/ruby_modular) +# ruby +IF (INTERFACE_RUBY) + add_subdirectory(${CMAKE_SOURCE_DIR}/src/interfaces/ruby) ENDIF() -# octave modular -IF (OctaveModular) - IF(EXISTS ${CMAKE_SOURCE_DIR}/src/interfaces/octave_modular) - add_subdirectory(${CMAKE_SOURCE_DIR}/src/interfaces/octave_modular) +# octave +IF (INTERFACE_OCTAVE) + IF(EXISTS ${CMAKE_SOURCE_DIR}/src/interfaces/octave) + add_subdirectory(${CMAKE_SOURCE_DIR}/src/interfaces/octave) ENDIF() ENDIF() -# csharp modular -IF (CSharpModular) - IF(EXISTS ${CMAKE_SOURCE_DIR}/src/interfaces/csharp_modular) - add_subdirectory(${CMAKE_SOURCE_DIR}/src/interfaces/csharp_modular) +# csharp +IF (INTERFACE_CSHARP) + IF(EXISTS ${CMAKE_SOURCE_DIR}/src/interfaces/csharp) + add_subdirectory(${CMAKE_SOURCE_DIR}/src/interfaces/csharp) ENDIF() ENDIF() -# r modular -IF (RModular) - IF(EXISTS ${CMAKE_SOURCE_DIR}/src/interfaces/r_modular) - add_subdirectory(${CMAKE_SOURCE_DIR}/src/interfaces/r_modular) +# R +IF (INTERFACE_R) + IF(EXISTS ${CMAKE_SOURCE_DIR}/src/interfaces/r) + add_subdirectory(${CMAKE_SOURCE_DIR}/src/interfaces/r) ENDIF() ENDIF() -# perl modular -IF (PerlModular) +# perl +IF (INTERFACE_PERL) FIND_PACKAGE(FindPerlLibs REQUIRED) UNSET(TARGET_SWIGFLAGS) - IF(EXISTS ${CMAKE_SOURCE_DIR}/src/interfaces/perl_modular) - #add_subdirectory(${CMAKE_SOURCE_DIR}/src/interfaces/perl_modular) + IF(EXISTS ${CMAKE_SOURCE_DIR}/src/interfaces/perl) + #add_subdirectory(${CMAKE_SOURCE_DIR}/src/interfaces/perl) ENDIF() ENDIF() @@ -565,6 +556,11 @@ IF(EXISTS ${CMAKE_SOURCE_DIR}/examples) add_subdirectory(${CMAKE_SOURCE_DIR}/examples) ENDIF() + # always build minimal example(s) + IF(EXISTS ${CMAKE_SOURCE_DIR}/examples/minimal) + add_subdirectory(${CMAKE_SOURCE_DIR}/examples/minimal) + ENDIF() + IF(BUILD_META_EXAMPLES) # allow meta examples without adding examples dir itself add_subdirectory(${CMAKE_SOURCE_DIR}/examples/meta) @@ -583,7 +579,7 @@ include(FeatureSummary) feature_summary(WHAT ALL) PrintLine() -PrintStatus("Integration") +PrintStatus("Integrations") PrintInterfaceStatus("OpenCV Integration" OpenCV) @@ -591,7 +587,7 @@ PrintLine() PrintStatus("Interfaces") ForEach(Interface ${AVAILABLE_INTERFACES}) - PrintInterfaceStatus("${${Interface}Description} interface" ${Interface}) + PrintInterfaceStatus("${${Interface}_DESCRIPTION}" ${Interface}) EndForEach(Interface) PrintLine() diff --git a/NEWS b/NEWS index 3b1951d45c5..261ec7098bd 100644 --- a/NEWS +++ b/NEWS @@ -1,3 +1,34 @@ +2017-11-28 Viktor Gal + + * SHOGUN Release version 6.1.0 (libshogun 18.0, data 0.11, parameter 1) + + * This release is dedicated for Heiko's successful PhD defense + + * Add conda-forge packages, to get prebuilt binaries via the cross-platform conda package manager [Dougal Sutherland] + * Change interface cmake variables to INTERFACE_* + * Move GPL code to gpl submodule [Heiko Strathmann] + + * Features: + - Enable using BLAS/LAPACK from Eigen by default [Viktor Gal] + - Add iterators to SGVector and SGMatrix [Viktor Gal] + - Significantly lower the runtime of KernelPCA (GSoC '17) [Michele Mazzoni] + - Refactor FisherLDA and LDA solvers (GSoC '17) [Michele Mazzoni] + - Add automated test for trained model serialization (GSoC '17) [Michele Mazzoni] + - Enable SWIG director classes by default [Viktor Gal] + - Vectorize DotFeatures covariance/mean calculation [Michele Mazzoni] + - Support for premature stopping of model training (GSoC '17) [Giovanni De Toni] + - Add support for observable variables (GSoC '17) [Giovanni De Toni] + - Use TFLogger to serialize observed variables for TensorBoard (GSoC '17) [Giovanni De Toni] + - Drop CMath::dot and SGVector::dot and use linalg::dot [Viktor Gal] + - Added class probabilities for BaggingMachine (GSoC '17) [Olivier Nguyen] + * Bugfixes: + - Fix transpose bug in Ruby typemap for matrices [Elias Saalmann] + - Fix MKL detection and linking; use mkl_rt when available [Viktor Gal] + - Fix Windows static linking [Viktor Gal] + - Fix SWIG interface compilation on Windows [qcrist] + - Fix CircularBuffer bug that broke parsing of big CSV and LibSVM files #1991 [Viktor Gal] + - Fix R interface when using clang to compile the interface [Viktor Gal] + 2016-11-05 Viktor Gal * SHOGUN Release version 6.0.0 (libshogun 18.0, data 0.11, parameter 1) diff --git a/README.md b/README.md index c067c786439..8f70c468390 100644 --- a/README.md +++ b/README.md @@ -15,7 +15,12 @@ Develop branch build status: [![Build Status](https://travis-ci.org/shogun-toolbox/shogun.svg?branch=develop)](https://travis-ci.org/shogun-toolbox/shogun) [![Build status](https://ci.appveyor.com/api/projects/status/jx095rnr9qhg8dcv/branch/develop?svg=true)](https://ci.appveyor.com/project/vigsterkr/shogun/branch/develop) -[![Coverage Status](https://coveralls.io/repos/shogun-toolbox/shogun/badge.png?branch=develop)](https://coveralls.io/r/shogun-toolbox/shogun?branch=develop) +[![codecov](https://codecov.io/gh/shogun-toolbox/shogun/branch/develop/graph/badge.svg)](https://codecov.io/gh/shogun-toolbox/shogun) + +Donate to Shogun via NumFocus: + +[![Powered by NumFOCUS](https://img.shields.io/badge/powered%20by-NumFOCUS-orange.svg?style=flat&colorA=E1523D&colorB=007D8A)](http://numfocus.org) + Buildbot: http://buildbot.shogun-toolbox.org/waterfall. diff --git a/applications/arts/signal_sensor.py b/applications/arts/signal_sensor.py index a0a139b4800..0498570457a 100644 --- a/applications/arts/signal_sensor.py +++ b/applications/arts/signal_sensor.py @@ -12,11 +12,11 @@ from util import * -from shogun.Features import StringCharFeatures, StringWordFeatures, CombinedFeatures, DNA -from shogun.Kernel import CombinedKernel, WeightedDegreePositionStringKernel -from shogun.Kernel import K_COMMWORDSTRING, CommWordStringKernel, IdentityKernelNormalizer -from shogun.Preprocessor import SortWordString -from shogun.Classifier import KernelMachine +from shogun import StringCharFeatures, StringWordFeatures, CombinedFeatures, DNA +from shogun import CombinedKernel, WeightedDegreePositionStringKernel +from shogun import K_COMMWORDSTRING, CommWordStringKernel, IdentityKernelNormalizer +from shogun import SortWordString +from shogun import KernelMachine class Sensor(object): diff --git a/applications/asp/asp b/applications/asp/asp index 41f57f28201..e5f9d21eec4 100755 --- a/applications/asp/asp +++ b/applications/asp/asp @@ -24,9 +24,9 @@ try: import genomic import model import seqdict - import shogun.Kernel + import shogun - d=shogun.Kernel.WeightedDegreeStringKernel(1) + d=shogun.WeightedDegreeStringKernel(1) if (d.version.get_version_revision() < 2997): print print "ERROR: SHOGUN VERSION 0.6.2 or later required" diff --git a/applications/asp/signal_detectors.py b/applications/asp/signal_detectors.py index cd43df09f5b..fe9e084791e 100644 --- a/applications/asp/signal_detectors.py +++ b/applications/asp/signal_detectors.py @@ -13,10 +13,10 @@ import numpy import seqdict -from shogun.Classifier import LibSVM -from shogun.Features import StringCharFeatures,DNA -from shogun.Kernel import WeightedDegreeStringKernel -from shogun.Library import DynamicIntArray +from shogun import LibSVM +from shogun import StringCharFeatures,DNA +from shogun import WeightedDegreeStringKernel +from shogun import DynamicIntArray class svm_splice_model(object): def __init__(self, order, traindat, alphas, b, (window_left,offset,window_right), consensus): diff --git a/applications/classification/evaluate_multiclass_labels.py b/applications/classification/evaluate_multiclass_labels.py index 030f5361db1..f720bb687d0 100644 --- a/applications/classification/evaluate_multiclass_labels.py +++ b/applications/classification/evaluate_multiclass_labels.py @@ -32,7 +32,7 @@ import argparse import logging import numpy as np -from modshogun import (LibSVMFile, MulticlassLabels, MulticlassAccuracy) +from shogun import (LibSVMFile, MulticlassLabels, MulticlassAccuracy) from utils import get_features_and_labels LOGGER = logging.getLogger(__file__) diff --git a/applications/classification/predict_multiclass_svm.py b/applications/classification/predict_multiclass_svm.py index 88ce0d6b504..79585ec6784 100644 --- a/applications/classification/predict_multiclass_svm.py +++ b/applications/classification/predict_multiclass_svm.py @@ -32,7 +32,7 @@ import argparse import logging from contextlib import closing -from modshogun import (LibSVMFile, SparseRealFeatures, MulticlassLabels, +from shogun import (LibSVMFile, SparseRealFeatures, MulticlassLabels, MulticlassLibSVM, SerializableHdf5File, MulticlassAccuracy) from utils import get_features_and_labels diff --git a/applications/classification/random_fourier_classification.py b/applications/classification/random_fourier_classification.py index bf5468a893f..07350cde384 100644 --- a/applications/classification/random_fourier_classification.py +++ b/applications/classification/random_fourier_classification.py @@ -32,7 +32,7 @@ def parse_arguments(): return parser.parse_args() def evaluate(predicted_labels, labels, prefix="Results"): - from modshogun import PRCEvaluation, ROCEvaluation, AccuracyMeasure + from shogun import PRCEvaluation, ROCEvaluation, AccuracyMeasure prc_evaluator = PRCEvaluation() roc_evaluator = ROCEvaluation() @@ -58,9 +58,9 @@ def load_sparse_data(filename, dimension=None): return {'data':sparse_feats, 'labels':labels} if __name__=='__main__': - from modshogun import SparseRealFeatures, RandomFourierDotFeatures, GAUSSIAN - from modshogun import LibSVMFile, BinaryLabels, SVMOcas - from modshogun import Time + from shogun import SparseRealFeatures, RandomFourierDotFeatures, GAUSSIAN + from shogun import LibSVMFile, BinaryLabels, SVMOcas + from shogun import Time from numpy import array args = parse_arguments() diff --git a/applications/classification/train_multiclass_svm.py b/applications/classification/train_multiclass_svm.py index 5dfb1fa4c41..39e17d2619c 100644 --- a/applications/classification/train_multiclass_svm.py +++ b/applications/classification/train_multiclass_svm.py @@ -32,7 +32,7 @@ import argparse import logging from contextlib import contextmanager, closing -from modshogun import (LibSVMFile, GaussianKernel, MulticlassLibSVM, +from shogun import (LibSVMFile, GaussianKernel, MulticlassLibSVM, SerializableHdf5File, LinearKernel) from utils import get_features_and_labels, track_execution diff --git a/applications/classification/utils.py b/applications/classification/utils.py index 40da45d9ae0..e459c39e97e 100644 --- a/applications/classification/utils.py +++ b/applications/classification/utils.py @@ -30,7 +30,7 @@ import logging from contextlib import contextmanager -from modshogun import MulticlassLabels, SparseRealFeatures, Time +from shogun import MulticlassLabels, SparseRealFeatures, Time logging.basicConfig(level=logging.INFO, format='[%(asctime)-15s %(module)s] %(message)s') diff --git a/applications/easysvm/esvm/experiment.py b/applications/easysvm/esvm/experiment.py index 04595861ba6..ed4c3a79e46 100644 --- a/applications/easysvm/esvm/experiment.py +++ b/applications/easysvm/esvm/experiment.py @@ -28,28 +28,28 @@ from poim import compute_poims import shogun -from shogun.Kernel import GaussianKernel, WeightedDegreePositionStringKernel -from shogun.Kernel import WeightedDegreeStringKernel -from shogun.Kernel import LinearKernel, PolyKernel, LocalAlignmentStringKernel -from shogun.Kernel import LocalityImprovedStringKernel -from shogun.Kernel import CommWordStringKernel, WeightedCommWordStringKernel, CommUlongStringKernel -from shogun.Kernel import CombinedKernel -from shogun.Kernel import SLOWBUTMEMEFFICIENT -from shogun.Kernel import AvgDiagKernelNormalizer -from shogun.Features import RealFeatures, Labels, StringCharFeatures, DNA, StringWordFeatures, StringUlongFeatures, PROTEIN -from shogun.Features import CombinedFeatures -from shogun.Classifier import LibSVM,GPBTSVM +from shogun import GaussianKernel, WeightedDegreePositionStringKernel +from shogun import WeightedDegreeStringKernel +from shogun import LinearKernel, PolyKernel, LocalAlignmentStringKernel +from shogun import LocalityImprovedStringKernel +from shogun import CommWordStringKernel, WeightedCommWordStringKernel, CommUlongStringKernel +from shogun import CombinedKernel +from shogun import SLOWBUTMEMEFFICIENT +from shogun import AvgDiagKernelNormalizer +from shogun import RealFeatures, Labels, StringCharFeatures, DNA, StringWordFeatures, StringUlongFeatures, PROTEIN +from shogun import CombinedFeatures +from shogun import LibSVM,GPBTSVM DefaultSVM = LibSVM try: - from shogun.Classifier import SVMLight + from shogun import SVMLight LinAddSVM = SVMLight LinearSVM = SVMLight except: LinAddSVM = GPBTSVM LinearSVM = LibSVM -from shogun.Preprocessor import SortWordString, SortUlongString +from shogun import SortWordString, SortUlongString from utils import calcprc, calcroc, accuracy from utils import getPartitionedSet, getCurrentSplit diff --git a/applications/easysvm/esvm/plots.py b/applications/easysvm/esvm/plots.py index 1b0cd2f2948..98e7322e234 100644 --- a/applications/easysvm/esvm/plots.py +++ b/applications/easysvm/esvm/plots.py @@ -27,8 +27,8 @@ import warnings import shutil -from shogun.Features import Labels -from shogun.Evaluation import * +from shogun import Labels +from shogun import * def plotroc(output, LTE, draw_random=False, figure_fname="", roc_label='ROC'): """Plot the receiver operating characteristic curve""" diff --git a/applications/easysvm/esvm/utils.py b/applications/easysvm/esvm/utils.py index 3e6a0ad60fb..dfdefa24456 100644 --- a/applications/easysvm/esvm/utils.py +++ b/applications/easysvm/esvm/utils.py @@ -23,8 +23,8 @@ import warnings import shutil -from shogun.Features import Labels -from shogun.Evaluation import * +from shogun import Labels +from shogun import * ################################################################################ # evaluation functions diff --git a/applications/easysvm/galaxy/README b/applications/easysvm/galaxy/README index a45fe881a7b..1e7b2566082 100644 --- a/applications/easysvm/galaxy/README +++ b/applications/easysvm/galaxy/README @@ -1,3 +1,3 @@ The files in this directory are a copy of svn/projects/galaxy/tools/agr. If you edit them, make sure the -changes are also integrated into the main version. \ No newline at end of file +changes are also integrated into the main version. diff --git a/applications/easysvm/tutpaper/svm_params.py b/applications/easysvm/tutpaper/svm_params.py index 83a6f0b2206..a0aa7afd553 100644 --- a/applications/easysvm/tutpaper/svm_params.py +++ b/applications/easysvm/tutpaper/svm_params.py @@ -14,9 +14,9 @@ import numpy import shogun -from shogun.Kernel import GaussianKernel, LinearKernel, PolyKernel -from shogun.Features import RealFeatures, BinaryLabels -from shogun.Classifier import LibSVM +from shogun import GaussianKernel, LinearKernel, PolyKernel +from shogun import RealFeatures, BinaryLabels +from shogun import LibSVM from numpy import arange import matplotlib diff --git a/applications/msplicer/msplicer b/applications/msplicer/msplicer index 4e9417f3c7a..03101521cb3 100755 --- a/applications/msplicer/msplicer +++ b/applications/msplicer/msplicer @@ -22,9 +22,9 @@ try: import genomic import model import seqdict - import shogun.Structure + import shogun - d=shogun.Structure.DynProg() + d=shogun.DynProg() if (d.version.get_version_revision() < 2997): print print "ERROR: SHOGUN VERSION 0.6.2 or later required" @@ -152,7 +152,7 @@ class msplicer: def initialize_dynprog(self, seq): - dyn=shogun.Structure.DynProg() + dyn=shogun.DynProg() self.content.initialize_content(dyn) @@ -266,7 +266,7 @@ class msplicer: dyn.best_path_set_my_state_seq(my_states) dyn.best_path_set_my_pos_seq(my_pos) - dyn.io.set_loglevel(shogun.Structure.M_DEBUG) + dyn.io.set_loglevel(shogun.M_DEBUG) dyn.best_path_deriv_call() def print_version(): @@ -348,7 +348,7 @@ WS160, WS160gc, orfWS160gc if __name__ == '__main__': - dyn=shogun.Structure.DynProg() + dyn=shogun.DynProg() (startstop, fafname, modelfname, outfile ) = parse_options() p=msplicer() p.load_model(modelfname); diff --git a/applications/msplicer/plif.py b/applications/msplicer/plif.py index 059acd8a34e..b1abe8c6417 100644 --- a/applications/msplicer/plif.py +++ b/applications/msplicer/plif.py @@ -10,9 +10,9 @@ # from numpy import array -from shogun.Structure import Plif -from shogun.Structure import PlifArray -from shogun.Library import DynamicPlifArray +from shogun import Plif +from shogun import PlifArray +from shogun import DynamicPlifArray class plif: def __init__(self, model): diff --git a/applications/msplicer/signal_detectors.py b/applications/msplicer/signal_detectors.py index f10096d35cb..e12002c2d5f 100644 --- a/applications/msplicer/signal_detectors.py +++ b/applications/msplicer/signal_detectors.py @@ -13,7 +13,7 @@ import numpy import seqdict -from modshogun import KernelMachine,StringCharFeatures,DNA,WeightedDegreeStringKernel +from shogun import KernelMachine,StringCharFeatures,DNA,WeightedDegreeStringKernel class svm_splice_model(object): def __init__(self, order, traindat, alphas, b, (window_left,offset,window_right), consensus): diff --git a/applications/ocr/Ai.py b/applications/ocr/Ai.py index 73edfb261b0..f8a3284ae52 100644 --- a/applications/ocr/Ai.py +++ b/applications/ocr/Ai.py @@ -1,9 +1,9 @@ # File : $HeadURL$ # Version: $Id$ -from modshogun import RealFeatures, MulticlassLabels -from modshogun import GaussianKernel -from modshogun import GMNPSVM +from shogun import RealFeatures, MulticlassLabels +from shogun import GaussianKernel +from shogun import GMNPSVM import numpy as np import gzip as gz diff --git a/applications/tapkee/faces_embedding.py b/applications/tapkee/faces_embedding.py index 1931d9fad93..24cf327031d 100644 --- a/applications/tapkee/faces_embedding.py +++ b/applications/tapkee/faces_embedding.py @@ -8,7 +8,7 @@ # Written (W) 2011 Sergey Lisitsyn # Copyright (C) 2011 Sergey Lisitsyn -from modshogun import * +from shogun import * from numpy import * from matplotlib.offsetbox import TextArea, DrawingArea, OffsetImage, AnnotationBbox import re,os,time diff --git a/applications/tapkee/samples/dm.py b/applications/tapkee/samples/dm.py index 8c1d3b87efd..036028dfd44 100644 --- a/applications/tapkee/samples/dm.py +++ b/applications/tapkee/samples/dm.py @@ -1,4 +1,4 @@ -import modshogun as sg +import shogun as sg import data import numpy as np diff --git a/applications/tapkee/samples/hlle.py b/applications/tapkee/samples/hlle.py index 7360941b614..0dc860931c0 100644 --- a/applications/tapkee/samples/hlle.py +++ b/applications/tapkee/samples/hlle.py @@ -1,4 +1,4 @@ -import modshogun as sg +import shogun as sg import data # load data diff --git a/applications/tapkee/samples/isomap.py b/applications/tapkee/samples/isomap.py index 086bf2cbc2e..1d0569e1662 100644 --- a/applications/tapkee/samples/isomap.py +++ b/applications/tapkee/samples/isomap.py @@ -1,4 +1,4 @@ -import modshogun as sg +import shogun as sg import data import numpy as np diff --git a/applications/tapkee/samples/klle.py b/applications/tapkee/samples/klle.py index 1075a357c1e..77b9908cd68 100644 --- a/applications/tapkee/samples/klle.py +++ b/applications/tapkee/samples/klle.py @@ -1,4 +1,4 @@ -import modshogun as sg +import shogun as sg import data import numpy as np diff --git a/applications/tapkee/samples/la.py b/applications/tapkee/samples/la.py index 46e61451277..b125fcc67c8 100644 --- a/applications/tapkee/samples/la.py +++ b/applications/tapkee/samples/la.py @@ -1,4 +1,4 @@ -import modshogun as sg +import shogun as sg import data import numpy as np diff --git a/applications/tapkee/samples/lle.py b/applications/tapkee/samples/lle.py index 60740d3eb71..f0a05dd4599 100644 --- a/applications/tapkee/samples/lle.py +++ b/applications/tapkee/samples/lle.py @@ -1,4 +1,4 @@ -import modshogun as sg +import shogun as sg import data # load data diff --git a/applications/tapkee/samples/lltsa.py b/applications/tapkee/samples/lltsa.py index 12c19da02a4..f59a3fb2ea3 100644 --- a/applications/tapkee/samples/lltsa.py +++ b/applications/tapkee/samples/lltsa.py @@ -1,4 +1,4 @@ -import modshogun as sg +import shogun as sg import data # load data diff --git a/applications/tapkee/samples/lpp.py b/applications/tapkee/samples/lpp.py index f5d44662cbb..9c9c4e87f37 100644 --- a/applications/tapkee/samples/lpp.py +++ b/applications/tapkee/samples/lpp.py @@ -1,4 +1,4 @@ -import modshogun as sg +import shogun as sg import data # load data diff --git a/applications/tapkee/samples/ltsa.py b/applications/tapkee/samples/ltsa.py index 6d4f32734aa..2c2d760cd4a 100644 --- a/applications/tapkee/samples/ltsa.py +++ b/applications/tapkee/samples/ltsa.py @@ -1,4 +1,4 @@ -import modshogun as sg +import shogun as sg import data # load data diff --git a/applications/tapkee/samples/mds.py b/applications/tapkee/samples/mds.py index d758dce1ed4..4f0f8ffe633 100644 --- a/applications/tapkee/samples/mds.py +++ b/applications/tapkee/samples/mds.py @@ -1,4 +1,4 @@ -import modshogun as sg +import shogun as sg import data import numpy as np diff --git a/applications/tapkee/samples/npe.py b/applications/tapkee/samples/npe.py index a413a08856c..cefa4880bdf 100644 --- a/applications/tapkee/samples/npe.py +++ b/applications/tapkee/samples/npe.py @@ -1,4 +1,4 @@ -import modshogun as sg +import shogun as sg import data # load data diff --git a/applications/tapkee/swissroll_embedding.py b/applications/tapkee/swissroll_embedding.py index 057c902901c..e83d69d71a5 100644 --- a/applications/tapkee/swissroll_embedding.py +++ b/applications/tapkee/swissroll_embedding.py @@ -5,12 +5,12 @@ N = X.shape[1] converters = [] -from shogun.Converter import LocallyLinearEmbedding +from shogun import LocallyLinearEmbedding lle = LocallyLinearEmbedding() lle.set_k(9) converters.append((lle, "LLE with k=%d" % lle.get_k())) -from shogun.Converter import MultidimensionalScaling +from shogun import MultidimensionalScaling mds = MultidimensionalScaling() converters.append((mds, "Classic MDS")) @@ -19,29 +19,29 @@ lmds.set_landmark_number(20) converters.append((lmds,"Landmark MDS with %d landmarks" % lmds.get_landmark_number())) -from shogun.Converter import Isomap +from shogun import Isomap cisomap = Isomap() cisomap.set_k(9) converters.append((cisomap,"Isomap with k=%d" % cisomap.get_k())) -from shogun.Converter import DiffusionMaps -from shogun.Kernel import GaussianKernel +from shogun import DiffusionMaps +from shogun import GaussianKernel dm = DiffusionMaps() dm.set_t(2) dm.set_width(1000.0) converters.append((dm,"Diffusion Maps with t=%d, sigma=%.1f" % (dm.get_t(),dm.get_width()))) -from shogun.Converter import HessianLocallyLinearEmbedding +from shogun import HessianLocallyLinearEmbedding hlle = HessianLocallyLinearEmbedding() hlle.set_k(6) converters.append((hlle,"Hessian LLE with k=%d" % (hlle.get_k()))) -from shogun.Converter import LocalTangentSpaceAlignment +from shogun import LocalTangentSpaceAlignment ltsa = LocalTangentSpaceAlignment() ltsa.set_k(6) converters.append((ltsa,"LTSA with k=%d" % (ltsa.get_k()))) -from shogun.Converter import LaplacianEigenmaps +from shogun import LaplacianEigenmaps le = LaplacianEigenmaps() le.set_k(20) le.set_tau(100.0) @@ -67,7 +67,7 @@ plt.suptitle('Swissroll embedding',fontsize=9) plt.subplots_adjust(hspace=0.4) -from shogun.Features import RealFeatures +from shogun import RealFeatures for (i, (converter, label)) in enumerate(converters): X = numpy.genfromtxt('../../data/toy/swissroll.dat',unpack=True).T diff --git a/applications/tapkee/words_embedding.py b/applications/tapkee/words_embedding.py index faa1e8af613..d74deb215e6 100644 --- a/applications/tapkee/words_embedding.py +++ b/applications/tapkee/words_embedding.py @@ -10,7 +10,7 @@ from numpy import * from pylab import * -from modshogun import * +from shogun import * import random import difflib diff --git a/appveyor.yml b/appveyor.yml index 89f5f7f7e33..9adc097a25b 100644 --- a/appveyor.yml +++ b/appveyor.yml @@ -1,11 +1,13 @@ -version: 4.2.{build} +version: "{build}" environment: matrix: - APPVEYOR_BUILD_WORKER_IMAGE: Visual Studio 2015 VSVER: Visual Studio 14 2015 Win64 - - APPVEYOR_BUILD_WORKER_IMAGE: Visual Studio 2017 - VSVER: Visual Studio 15 2017 Win64 + PYTHON: "C:\\Python27" +# disable multiple builds until each build takes more than 30 minutes +# - APPVEYOR_BUILD_WORKER_IMAGE: Visual Studio 2017 +# VSVER: Visual Studio 15 2017 Win64 platform: - x64 @@ -20,14 +22,19 @@ before_build: - md %APPVEYOR_BUILD_FOLDER%\build - cd %APPVEYOR_BUILD_FOLDER%\build - cmake -G"%VSVER%" -DCMAKE_BUILD_TYPE=%CONFIGURATION% -DBUILD_META_EXAMPLES=OFF -DENABLE_TESTING=ON .. - - cd .. build: - parallel: true - project: build\shogun.sln verbosity: minimal +# TODO: fix cmake script in order that building examples +# triggers the build of shogun-static.vcxproj +build_script: + - msbuild src\shogun\libshogun.vcxproj + - msbuild src\shogun\shogun.vcxproj + - msbuild src\shogun\shogun-static.vcxproj + - msbuild shogun.sln + test_script: - cd %APPVEYOR_BUILD_FOLDER%\build - ctest --output-on-failure -C %CONFIGURATION% - - cd .. + diff --git a/benchmarks/elementwise_benchmark.cpp b/benchmarks/elementwise_benchmark.cpp deleted file mode 100644 index 3a1f2feaee8..00000000000 --- a/benchmarks/elementwise_benchmark.cpp +++ /dev/null @@ -1,138 +0,0 @@ -/* - * Copyright (c) The Shogun Machine Learning Toolbox - * Written (w) 2015 Soumyajit De - * All rights reserved. - * - * Redistribution and use in source and binary forms, with or without - * modification, are permitted provided that the following conditions are met: - * - * 1. Redistributions of source code must retain the above copyright notice, this - * list of conditions and the following disclaimer. - * 2. Redistributions in binary form must reproduce the above copyright notice, - * this list of conditions and the following disclaimer in the documentation - * and/or other materials provided with the distribution. - * - * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND - * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED - * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE - * DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR - * ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES - * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; - * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND - * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT - * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS - * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - * - * The views and conclusions contained in the software and documentation are those - * of the authors and should not be interpreted as representing official policies, - * either expressed or implied, of the Shogun Development Team. - */ - -#include -#include -#include -#include -#include -#include - -using namespace shogun; - -/** - * Instructions : - * 1. Install benchmarking toolkit "hayai" (https://github.com/nickbruun/hayai) - * 2. Compile against libhayai_main, e.g. - * g++ -O3 -std=c++11 elementwise_benchmark.cpp -I/usr/include/eigen3 -lshogun -lhayai_main -lOpenCL -o benchmark - * 3. ./benchmark - */ - -/** Generate data only once */ -struct Data -{ - Data() - { - init(); - } - - void init() - { - m_cpu=SGMatrix(num_rows, num_cols); - std::iota(m_cpu.data(), m_cpu.data()+m_cpu.size(), 1); - m_gpu=CGPUMatrix(m_cpu); - } - - SGMatrix m_cpu; - CGPUMatrix m_gpu; - - static constexpr index_t num_rows=1000; - static constexpr index_t num_cols=1000; -}; - -Data data; - -BENCHMARK(SGMatrix, elementwise, 10, 1000) -{ - float32_t weights=0.6; - float32_t std_dev=0.2; - float32_t mean=0.01; - - SGMatrix result=linalg::elementwise_compute(data.m_cpu, - [&weights, &std_dev, &mean](float32_t& sqr_dist) - { - float32_t outer_factor=-2*CMath::PI*CMath::sqrt(sqr_dist)*CMath::sq(weights); - float32_t exp_factor=CMath::exp(-2*CMath::sq(CMath::PI)*sqr_dist*CMath::sq(std_dev)); - float32_t sin_factor=CMath::sin(2*CMath::PI*CMath::sqrt(sqr_dist)*mean); - return outer_factor*exp_factor*sin_factor; - }); - -} - -BENCHMARK(SGMatrix, loop, 10, 1000) -{ - float32_t weights=0.6; - float32_t std_dev=0.2; - float32_t mean=0.01; - - SGMatrix result(data.m_cpu.num_rows, data.m_cpu.num_cols); - - for (index_t j=0; j(); - - std::string operation; - operation.append(linalg::implementation::ocl::format( - R"( - {type} {var1} = -2*{pi}*sqrt(element)*pow({weights}, 2); - {type} {var2} = exp(-2*pow({pi}, 2)*element*pow({stddev}, 2)); - {type} {var3} = sin(2*{pi}*sqrt(element)*{mean}); - return {var1}*{var2}*{var3}; - )", - { - linalg::ocl::Parameter("type")=data_type, - linalg::ocl::Parameter("var1")="outer_factor", - linalg::ocl::Parameter("var2")="exp_factor", - linalg::ocl::Parameter("var3")="sin_factor", - linalg::ocl::Parameter("pi")=CMath::PI, - linalg::ocl::Parameter("weights")=weights, - linalg::ocl::Parameter("stddev")=std_dev, - linalg::ocl::Parameter("mean")=mean - })); - - linalg::elementwise_compute_inplace(data.m_gpu, operation); -} diff --git a/cmake/CheckCompiler.cmake b/cmake/CheckCompiler.cmake deleted file mode 100644 index 5b2dcb43ce2..00000000000 --- a/cmake/CheckCompiler.cmake +++ /dev/null @@ -1,68 +0,0 @@ -# -*- mode: cmake; -*- -# -# Figure out the version of the used compiler -# Variables set by this module -# CMAKE_CXX_COMPILER_MAJOR major version of compiler -# CMAKE_CXX_COMPILER_MINR minor version of compiler -# CMAKE_CXX_COMPILER_PATCH patch level (e.g. gcc 4.1.0) -# - -#execute_process(COMMAND [args1...]] -# [COMMAND [args2...] [...]] -# [WORKING_DIRECTORY ] -# [TIMEOUT ] -# [RESULT_VARIABLE ] -# [OUTPUT_VARIABLE ] -# [ERROR_VARIABLE ] -# [INPUT_FILE ] -# [OUTPUT_FILE ] -# [ERROR_FILE ] -# [OUTPUT_QUIET] -# [ERROR_QUIET] -# [OUTPUT_STRIP_TRAILING_WHITESPACE] -# [ERROR_STRIP_TRAILING_WHITESPACE]) - -# check the version of the compiler -set(CMAKE_CXX_COMPILER_MAJOR "CMAKE_CXX_COMPILER_MAJOR-NOTFOUND") -set(CMAKE_CXX_COMPILER_MINOR "CMAKE_CXX_COMPILER_MINOR-NOTFOUND") -set(CMAKE_CXX_COMPILER_PATCH "CMAKE_CXX_COMPILER_PATCH-NOTFOUND") - -# extract the version of the compiler -if( ${CMAKE_CXX_COMPILER_ID} STREQUAL "Intel") - execute_process(COMMAND ${CMAKE_CXX_COMPILER} -dumpversion - OUTPUT_VARIABLE CMAKE_CXX_COMPILER_VERSION) - - string(REGEX REPLACE "^([0-9]+)\\.([0-9]+).*$" "\\1" - CMAKE_CXX_COMPILER_MAJOR ${CMAKE_CXX_COMPILER_VERSION}) - string(REGEX REPLACE "^([0-9]+)\\.([0-9]+).*" "\\2" - CMAKE_CXX_COMPILER_MINOR ${CMAKE_CXX_COMPILER_VERSION}) - set(CMAKE_CXX_COMPILER_PATCH "") -endif( ${CMAKE_CXX_COMPILER_ID} STREQUAL "Intel") - -if( ${CMAKE_CXX_COMPILER_ID} STREQUAL "GNU") - execute_process(COMMAND ${CMAKE_CXX_COMPILER} -dumpversion - OUTPUT_VARIABLE CMAKE_CXX_COMPILER_VERSION) - - string(STRIP ${CMAKE_CXX_COMPILER_VERSION} CMAKE_CXX_COMPILER_VERSION) - string(REGEX REPLACE "^([0-9]+).*$" "\\1" - CMAKE_CXX_COMPILER_MAJOR ${CMAKE_CXX_COMPILER_VERSION}) - string(REGEX REPLACE "^([0-9]+)\\.([0-9]+).*$" "\\2" - CMAKE_CXX_COMPILER_MINOR ${CMAKE_CXX_COMPILER_VERSION}) - string(REGEX REPLACE "^([0-9]+)\\.([0-9]+)\\.([0-9]+)$" "\\3" - CMAKE_CXX_COMPILER_PATCH ${CMAKE_CXX_COMPILER_VERSION}) -endif( ${CMAKE_CXX_COMPILER_ID} STREQUAL "GNU") - - -# just print the results if requested -function(info_compiler) - message(STATUS "CMAKE_FORCE_CXX_COMPILER = '${CMAKE_FORCE_CXX_COMPILER}'") - message(STATUS "CMAKE_CXX_COMPILER = '${CMAKE_CXX_COMPILER}'") - message(STATUS "CMAKE_CXX_COMPILER_ID = '${CMAKE_CXX_COMPILER_ID}'") - message(STATUS "CMAKE_CXX_COMPILER_INIT = '${CMAKE_CXX_COMPILER_INIT}'") - message(STATUS "CMAKE_GENERATOR_CXX = '${CMAKE_GENERATOR_CXX}'") - message(STATUS "CMAKE_GNULD_IMAGE_VERSION = '${CMAKE_GNULD_IMAGE_VERSION}'") - message(STATUS "CMAKE_CXX_COMPILER_VERSION= '${CMAKE_CXX_COMPILER_VERSION}'") - message(STATUS "CMAKE_CXX_COMPILER_MAJOR = '${CMAKE_CXX_COMPILER_MAJOR}'") - message(STATUS "CMAKE_CXX_COMPILER_MINOR = '${CMAKE_CXX_COMPILER_MINOR}'") - message(STATUS "CMAKE_CXX_COMPILER_PATCH = '${CMAKE_CXX_COMPILER_PATCH}'") -endfunction(info_compiler) diff --git a/cmake/CheckSVNRevision.cmake b/cmake/CheckSVNRevision.cmake deleted file mode 100644 index c4d6489b4d6..00000000000 --- a/cmake/CheckSVNRevision.cmake +++ /dev/null @@ -1,18 +0,0 @@ -EXECUTE_PROCESS( - COMMAND ${CMAKE_COMMAND} -E chdir ${SRC_DIR} ${SVN_EXEC} info - OUTPUT_VARIABLE SVN_INFO_OUTPUT - RESULT_VARIABLE SVN_INFO_RETURN - ERROR_VARIABLE SVN_INFO_ERROR - OUTPUT_STRIP_TRAILING_WHITESPACE - ) - -IF (NOT ${SVN_INFO_RETURN}) - STRING(REGEX MATCH "Revision: [0-9]+" CHECKED_OUT_REVISION ${SVN_INFO_OUTPUT}) - STRING(REGEX REPLACE "Revision: ([0-9]+)" "\\1" CHECKED_OUT_REVISION ${CHECKED_OUT_REVISION}) - - IF (${CHECKED_OUT_REVISION} VERSION_EQUAL ${REVISION}) - EXECUTE_PROCESS( - COMMAND ${CMAKE_COMMAND} -E touch ${STAMP_DIR}/${PROJECT_NAME}-download - ) - ENDIF() -ENDIF() \ No newline at end of file diff --git a/cmake/CommonModularInterface.cmake b/cmake/CommonModularInterface.cmake deleted file mode 100644 index 1e02e00cbaf..00000000000 --- a/cmake/CommonModularInterface.cmake +++ /dev/null @@ -1,89 +0,0 @@ -MACRO(GENERATE_MODULAR_TARGET MODULAR_NAME MODULAR_DIR MODULAR_LIBARIES) - -get_target_property(ShogunIncludes shogun::shogun INTERFACE_INCLUDE_DIRECTORIES) -INCLUDE_DIRECTORIES(${ShogunIncludes}) - -# set compiler SWIG generated cxx compiler flags -SET(CMAKE_CXX_FLAGS ${SWIG_CXX_COMPILER_FLAGS}) -# unset any release or distribution flags -# we don't want them when compiling SWIG generated source -SET(CMAKE_CXX_FLAGS_RELEASE "") -SET(CMAKE_CXX_FLAGS_DISTRIBUTION "") -SET(CMAKE_CXX_FLAGS_DEBUG "") - -if(${MODULAR_NAME} STREQUAL "python") - SET(PREPEND_TARGET "_") -endif() - -set(modular_files) -FILE(GLOB_RECURSE MODULAR_FILES ${COMMON_MODULAR_SRC_DIR}/*.i) -FILE(GLOB_RECURSE CUSTOM_MODULAR_FILES ${MODULAR_DIR}/*.i) -LIST(APPEND MODULAR_FILES ${CUSTOM_MODULAR_FILES}) -FOREACH(file ${MODULAR_FILES}) - get_filename_component(fname ${file} NAME) - list(APPEND modular_files ${fname}) - ADD_CUSTOM_COMMAND(OUTPUT ${fname} - DEPENDS ${file} - COMMAND "${CMAKE_COMMAND}" -E copy_if_different ${file} ${fname} - COMMENT "" - ) -ENDFOREACH() - -ADD_CUSTOM_TARGET(${MODULAR_NAME}_modular_src - DEPENDS shogun::shogun ${modular_files} - COMMENT "copying SWIG files") - -INCLUDE(${SWIG_USE_FILE}) -SET_SOURCE_FILES_PROPERTIES(modshogun.i PROPERTIES CPLUSPLUS ON) -IF(DEFINED TARGET_SWIGFLAGS) - SET_SOURCE_FILES_PROPERTIES(modshogun.i PROPERTIES SWIG_FLAGS ${TARGET_SWIGFLAGS}) -ENDIF() -SET(SWIG_MODULE_${MODULAR_NAME}_modular_EXTRA_DEPS ${modular_files}) -SWIG_ADD_MODULE(${MODULAR_NAME}_modular ${MODULAR_NAME} modshogun.i sg_print_functions.cpp) -SWIG_LINK_LIBRARIES(${MODULAR_NAME}_modular shogun::shogun ${MODULAR_LIBARIES}) -SET_TARGET_PROPERTIES(${SWIG_MODULE_${MODULAR_NAME}_modular_REAL_NAME} PROPERTIES OUTPUT_NAME ${PREPEND_TARGET}modshogun) -ADD_DEPENDENCIES(${SWIG_MODULE_${MODULAR_NAME}_modular_REAL_NAME} ${MODULAR_NAME}_modular_src) - -#ADD_CUSTOM_COMMAND(TARGETS ${PREPEND_TARGET}${MODULAR_NAME}_modular -# POST_BUILD -# COMMAND ${PYTHON_EXECUTABLE} -# ARGS ${CMAKE_SOURCE_DIR}/src/.scrub_docstrings.py ) - -IF(DOXYGEN_FOUND) - configure_file(${COMMON_MODULAR_SRC_DIR}/modshogun.doxy.in modshogun.doxy) - - ADD_CUSTOM_COMMAND( - OUTPUT modshogun - COMMAND ${DOXYGEN_EXECUTABLE} - ARGS modshogun.doxy - DEPENDS shogun::shogun - COMMENT "Generating doxygen doc" - ) - - ADD_CUSTOM_COMMAND( - OUTPUT modshogun_doxygen.i - COMMAND ${PYTHON_EXECUTABLE} ${CMAKE_SOURCE_DIR}/src/.doxy2swig.py - ARGS --quiet --no-function-definition modshogun/doxygen_xml/index.xml modshogun_doxygen.i - DEPENDS modshogun - ) - ADD_CUSTOM_TARGET(${MODULAR_NAME}_doxy2swig DEPENDS modshogun_doxygen.i) - ADD_DEPENDENCIES(${SWIG_MODULE_${MODULAR_NAME}_modular_REAL_NAME} ${MODULAR_NAME}_doxy2swig) -ELSE() - #TODO add scrubing -ENDIF() - -# Make sure all modular interfaces are build single-threaded to reduce -# excessive memory consumption during build. -IF(SWIG_SINGLE_THREADED) - FOREACH(SG_MODULAR_INTERFACE_TARGET ${SG_MODULAR_INTERFACE_TARGETS}) - ADD_DEPENDENCIES(${SWIG_MODULE_${MODULAR_NAME}_modular_REAL_NAME} - ${SG_MODULAR_INTERFACE_TARGET}) - ENDFOREACH(SG_MODULAR_INTERFACE_TARGET ${SG_MODULAR_INTERFACE_TARGETS}) - SET(SG_MODULAR_INTERFACE_TARGETS - "${SWIG_MODULE_${MODULAR_NAME}_modular_REAL_NAME};${SG_MODULAR_INTERFACE_TARGETS}" - CACHE STRING "List of modular-interfaces beeing build." FORCE) -ENDIF(SWIG_SINGLE_THREADED) - -CONFIGURE_FILE(${COMMON_MODULAR_SRC_DIR}/swig_config.h.in swig_config.h) - -ENDMACRO() diff --git a/cmake/FindAtlas.cmake b/cmake/FindAtlas.cmake index 60ed66320b0..d5cc159244d 100644 --- a/cmake/FindAtlas.cmake +++ b/cmake/FindAtlas.cmake @@ -7,10 +7,8 @@ # Atlas_FOUND # Atlas_INCLUDE_DIRS # Atlas_LIBRARIES -# Atlas_LIBRARYRARY_DIRS set(Atlas_INCLUDE_SEARCH_PATHS - /usr/include /usr/include/atlas /usr/include/atlas-base $ENV{Atlas_ROOT_DIR} @@ -20,36 +18,42 @@ set(Atlas_INCLUDE_SEARCH_PATHS set(Atlas_LIB_SEARCH_PATHS /usr/lib/atlas /usr/lib/atlas-base - /usr/lib64 /usr/lib64/atlas + /usr/lib64/atlas-base $ENV{Atlas_ROOT_DIR} $ENV{Atlas_ROOT_DIR}/lib ) - find_path(Atlas_CBLAS_INCLUDE_DIR NAMES cblas.h PATHS ${Atlas_INCLUDE_SEARCH_PATHS}) find_path(Atlas_CLAPACK_INCLUDE_DIR NAMES clapack.h PATHS ${Atlas_INCLUDE_SEARCH_PATHS}) -find_library(Atlas_CBLAS_LIBRARY NAMES ptcblas_r ptcblas cblas_r cblas blas PATHS ${Atlas_LIB_SEARCH_PATHS}) -find_library(Atlas_BLAS_LIBRARY NAMES atlas_r atlas tatlas satlas PATHS ${Atlas_LIB_SEARCH_PATHS}) -find_library(Atlas_LAPACK_LIBRARY NAMES alapack_r alapack lapack_atlas lapacke PATHS ${Atlas_LIB_SEARCH_PATHS}) - -set(LOOKED_FOR - Atlas_CBLAS_INCLUDE_DIR - Atlas_CLAPACK_INCLUDE_DIR - - Atlas_CBLAS_LIBRARY - Atlas_BLAS_LIBRARY - Atlas_LAPACK_LIBRARY -) +find_library(Atlas_BLAS_LIBRARY NAMES atlas_r atlas tatlas satlas PATHS ${Atlas_LIB_SEARCH_PATHS}) +set(ATLAS_LIBS_VAR Atlas_BLAS_LIBRARY) +if (Atlas_BLAS_LIBRARY) + include(CheckLibraryExists) + # atlas 3.10+ contains all the function in one shared lib so dont try to find other parts of atlas + check_library_exists("${Atlas_BLAS_LIBRARY}" cblas_dgemv "" FOUND_ATLAS_CBLAS_DGEMV) + if (NOT FOUND_ATLAS_CBLAS_DGEMV) + find_library(Atlas_CBLAS_LIBRARY NAMES ptcblas_r ptcblas cblas_r cblas PATHS ${Atlas_LIB_SEARCH_PATHS}) + list(APPEND ATLAS_LIBS_VAR Atlas_CBLAS_LIBRARY) + endif() + check_library_exists("${Atlas_BLAS_LIBRARY}" clapack_dpotrf "" FOUND_ATLAS_CLAPACK_DPOTRF) + if (NOT FOUND_ATLAS_CLAPACK_DPOTRF) + find_library(Atlas_LAPACK_LIBRARY NAMES alapack_r alapack lapack_atlas atllapack PATHS ${Atlas_LIB_SEARCH_PATHS}) + list(APPEND ATLAS_LIBS_VAR Atlas_LAPACK_LIBRARY) + endif() +endif() include(FindPackageHandleStandardArgs) -find_package_handle_standard_args(Atlas DEFAULT_MSG ${LOOKED_FOR}) +find_package_handle_standard_args(Atlas DEFAULT_MSG Atlas_CBLAS_INCLUDE_DIR Atlas_CLAPACK_INCLUDE_DIR ${ATLAS_LIBS_VAR}) if(ATLAS_FOUND) set(Atlas_INCLUDE_DIRS ${Atlas_CBLAS_INCLUDE_DIR} ${Atlas_CLAPACK_INCLUDE_DIR}) - set(Atlas_LIBRARIES ${Atlas_LAPACK_LIBRARY} ${Atlas_CBLAS_LIBRARY} ${Atlas_BLAS_LIBRARY}) - mark_as_advanced(${LOOKED_FOR}) + set(ATLAS_LIBS) + foreach(atlas_lib ${ATLAS_LIBS_VAR}) + list(APPEND ATLAS_LIBS ${${atlas_lib}}) + endforeach() + set(Atlas_LIBRARIES ${ATLAS_LIBS}) + mark_as_advanced(${Atlas_CBLAS_INCLUDE_DIR} ${Atlas_CLAPACK_INCLUDE_DIR} ${ATLAS_LIBS}) message(STATUS "Found Atlas (include: ${Atlas_CBLAS_INCLUDE_DIR}, library: ${Atlas_BLAS_LIBRARY})") endif(ATLAS_FOUND) - diff --git a/cmake/FindCBLAS.cmake b/cmake/FindCBLAS.cmake deleted file mode 100644 index 554d086d153..00000000000 --- a/cmake/FindCBLAS.cmake +++ /dev/null @@ -1,42 +0,0 @@ -# - Find CBLAS -# Find the native CBLAS headers and libraries. -# -# CBLAS_LIBRARIES - List of libraries when using cblas. -# CBLAS_FOUND - True if cblas found. -# -# Copyright 2009-2011 The VOTCA Development Team (http://www.votca.org) -# -# Licensed under the Apache License, Version 2.0 (the "License"); -# you may not use this file except in compliance with the License. -# You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -# Unless required by applicable law or agreed to in writing, software -# distributed under the License is distributed on an "AS IS" BASIS, -# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -# See the License for the specific language governing permissions and -# limitations under the License. -# - -find_file(CBLAS_LIBRARY NAMES libcblas.so.3 libcblas.so.3gf - PATHS /usr/lib /usr/local/lib /opt/local/lib) -find_library(CBLAS_LIBRARY NAMES cblas gslcblas HINTS $ENV{CBLASDIR}/lib $ENV{CBLASDIR}/lib64 ) - -set(CBLAS_LIBRARIES ${CBLAS_LIBRARY} ) - -include(FindPackageHandleStandardArgs) -# handle the QUIETLY and REQUIRED arguments and set FFTW3_FOUND to TRUE -# if all listed variables are TRUE - -find_package_handle_standard_args(CBLAS DEFAULT_MSG CBLAS_LIBRARY ) - -if (CBLAS_FOUND) - include(CheckLibraryExists) - check_library_exists("${CBLAS_LIBRARY}" cblas_dsyrk "" FOUND_DSYRK) - if(NOT FOUND_DSYRK) - message(FATAL_ERROR "Could not find cblas_dsyrk in ${CBLAS_LIBRARY}, take a look at the error message in ${CMAKE_BINARY_DIR}${CMAKE_FILES_DIRECTORY}/CMakeError.log to find out what was going wrong. If you don't have pkg-config installed you will most likely have to set CBLAS_LIBRARY by hand (i.e. -DCBLAS_LIBRARY='/path/to/libcblas.so') !") - endif(NOT FOUND_DSYRK) -endif (CBLAS_FOUND) - -mark_as_advanced( CBLAS_LIBRARY ) diff --git a/cmake/FindMetaExamples.cmake b/cmake/FindMetaExamples.cmake index 054f746a0b7..760d07ee53b 100644 --- a/cmake/FindMetaExamples.cmake +++ b/cmake/FindMetaExamples.cmake @@ -30,7 +30,10 @@ function(get_excluded_meta_examples) IF(NOT USE_GPL_SHOGUN) LIST(APPEND EXCLUDED_META_EXAMPLES gaussian_processes/gaussian_process_regression.sg + gaussian_processes/gaussian_process_classifier.sg multiclass_classifier/multiclass_logisticregression.sg + statistical_testing/linear_time_mmd.sg + statistical_testing/quadratic_time_mmd.sg ) ENDIF() diff --git a/cmake/FindPythonLibs.cmake b/cmake/FindPythonLibs.cmake deleted file mode 100644 index 45712a35b33..00000000000 --- a/cmake/FindPythonLibs.cmake +++ /dev/null @@ -1,279 +0,0 @@ -# - Find python libraries -# This module finds if Python is installed and determines where the -# include files and libraries are. It also determines what the name of -# the library is. This code sets the following variables: -# -# PYTHONLIBS_FOUND - have the Python libs been found -# PYTHON_LIBRARIES - path to the python library -# PYTHON_INCLUDE_PATH - path to where Python.h is found (deprecated) -# PYTHON_INCLUDE_DIRS - path to where Python.h is found -# PYTHON_DEBUG_LIBRARIES - path to the debug library (deprecated) -# PYTHONLIBS_VERSION_STRING - version of the Python libs found (since CMake 2.8.8) -# -# The Python_ADDITIONAL_VERSIONS variable can be used to specify a list of -# version numbers that should be taken into account when searching for Python. -# You need to set this variable before calling find_package(PythonLibs). -# -# If you'd like to specify the installation of Python to use, you should modify -# the following cache variables: -# PYTHON_LIBRARY - path to the python library -# PYTHON_INCLUDE_DIR - path to where Python.h is found - -#============================================================================= -# Copyright 2001-2009 Kitware, Inc. -# -# Distributed under the OSI-approved BSD License (the "License"); -# see accompanying file Copyright.txt for details. -# -# This software is distributed WITHOUT ANY WARRANTY; without even the -# implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. -# See the License for more information. -#============================================================================= -# (To distribute this file outside of CMake, substitute the full -# License text for the above reference.) - -include(CMakeFindFrameworks) -# Search for the python framework on Apple. -CMAKE_FIND_FRAMEWORKS(Python) - -set(_PYTHON1_VERSIONS 1.6 1.5) -set(_PYTHON2_VERSIONS 2.7 2.6 2.5 2.4 2.3 2.2 2.1 2.0) -set(_PYTHON3_VERSIONS 3.3 3.2 3.1 3.0) - -if(PythonLibs_FIND_VERSION) - if(PythonLibs_FIND_VERSION MATCHES "^[0-9]+\\.[0-9]+(\\.[0-9]+.*)?$") - string(REGEX REPLACE "^([0-9]+\\.[0-9]+).*" "\\1" _PYTHON_FIND_MAJ_MIN "${PythonLibs_FIND_VERSION}") - string(REGEX REPLACE "^([0-9]+).*" "\\1" _PYTHON_FIND_MAJ "${_PYTHON_FIND_MAJ_MIN}") - unset(_PYTHON_FIND_OTHER_VERSIONS) - if(PythonLibs_FIND_VERSION_EXACT) - if(_PYTHON_FIND_MAJ_MIN STREQUAL PythonLibs_FIND_VERSION) - set(_PYTHON_FIND_OTHER_VERSIONS "${PythonLibs_FIND_VERSION}") - else() - set(_PYTHON_FIND_OTHER_VERSIONS "${PythonLibs_FIND_VERSION}" "${_PYTHON_FIND_MAJ_MIN}") - endif() - else() - foreach(_PYTHON_V ${_PYTHON${_PYTHON_FIND_MAJ}_VERSIONS}) - if(NOT _PYTHON_V VERSION_LESS _PYTHON_FIND_MAJ_MIN) - list(APPEND _PYTHON_FIND_OTHER_VERSIONS ${_PYTHON_V}) - endif() - endforeach() - endif() - unset(_PYTHON_FIND_MAJ_MIN) - unset(_PYTHON_FIND_MAJ) - else() - set(_PYTHON_FIND_OTHER_VERSIONS ${_PYTHON${PythonLibs_FIND_VERSION}_VERSIONS}) - endif() -else() - set(_PYTHON_FIND_OTHER_VERSIONS ${_PYTHON3_VERSIONS} ${_PYTHON2_VERSIONS} ${_PYTHON1_VERSIONS}) -endif() - -# Set up the versions we know about, in the order we will search. Always add -# the user supplied additional versions to the front. -# If FindPythonInterp has already found the major and minor version, -# insert that version between the user supplied versions and the stock -# version list. -set(_Python_VERSIONS ${Python_ADDITIONAL_VERSIONS}) -if(DEFINED PYTHON_VERSION_MAJOR AND DEFINED PYTHON_VERSION_MINOR) - list(APPEND _Python_VERSIONS ${PYTHON_VERSION_MAJOR}.${PYTHON_VERSION_MINOR}) -endif() -list(APPEND _Python_VERSIONS ${_PYTHON_FIND_OTHER_VERSIONS}) - -unset(_PYTHON_FIND_OTHER_VERSIONS) -unset(_PYTHON1_VERSIONS) -unset(_PYTHON2_VERSIONS) -unset(_PYTHON3_VERSIONS) - -foreach(_CURRENT_VERSION ${_Python_VERSIONS}) - string(REPLACE "." "" _CURRENT_VERSION_NO_DOTS ${_CURRENT_VERSION}) - if(WIN32) - find_library(PYTHON_DEBUG_LIBRARY - NAMES python${_CURRENT_VERSION_NO_DOTS}_d python - PATHS - [HKEY_LOCAL_MACHINE\\SOFTWARE\\Python\\PythonCore\\${_CURRENT_VERSION}\\InstallPath]/libs/Debug - [HKEY_CURRENT_USER\\SOFTWARE\\Python\\PythonCore\\${_CURRENT_VERSION}\\InstallPath]/libs/Debug - [HKEY_LOCAL_MACHINE\\SOFTWARE\\Python\\PythonCore\\${_CURRENT_VERSION}\\InstallPath]/libs - [HKEY_CURRENT_USER\\SOFTWARE\\Python\\PythonCore\\${_CURRENT_VERSION}\\InstallPath]/libs - ) - endif() - - find_library(PYTHON_LIBRARY - NAMES - python${_CURRENT_VERSION_NO_DOTS} - python${_CURRENT_VERSION}mu - python${_CURRENT_VERSION}m - python${_CURRENT_VERSION}u - python${_CURRENT_VERSION} - PATHS - [HKEY_LOCAL_MACHINE\\SOFTWARE\\Python\\PythonCore\\${_CURRENT_VERSION}\\InstallPath]/libs - [HKEY_CURRENT_USER\\SOFTWARE\\Python\\PythonCore\\${_CURRENT_VERSION}\\InstallPath]/libs - # Avoid finding the .dll in the PATH. We want the .lib. - NO_SYSTEM_ENVIRONMENT_PATH - ) - # Look for the static library in the Python config directory - find_library(PYTHON_LIBRARY - NAMES python${_CURRENT_VERSION_NO_DOTS} python${_CURRENT_VERSION} - # Avoid finding the .dll in the PATH. We want the .lib. - NO_SYSTEM_ENVIRONMENT_PATH - # This is where the static library is usually located - PATH_SUFFIXES python${_CURRENT_VERSION}/config - ) - - # For backward compatibility, honour value of PYTHON_INCLUDE_PATH, if - # PYTHON_INCLUDE_DIR is not set. - if(DEFINED PYTHON_INCLUDE_PATH AND NOT DEFINED PYTHON_INCLUDE_DIR) - set(PYTHON_INCLUDE_DIR "${PYTHON_INCLUDE_PATH}" CACHE PATH - "Path to where Python.h is found" FORCE) - endif() - - set(PYTHON_FRAMEWORK_INCLUDES) - if(Python_FRAMEWORKS AND NOT PYTHON_INCLUDE_DIR) - foreach(dir ${Python_FRAMEWORKS}) - set(PYTHON_FRAMEWORK_INCLUDES ${PYTHON_FRAMEWORK_INCLUDES} - ${dir}/Versions/${_CURRENT_VERSION}/include/python${_CURRENT_VERSION}) - endforeach() - endif() - - find_path(PYTHON_INCLUDE_DIR - NAMES Python.h - PATHS - ${PYTHON_FRAMEWORK_INCLUDES} - [HKEY_LOCAL_MACHINE\\SOFTWARE\\Python\\PythonCore\\${_CURRENT_VERSION}\\InstallPath]/include - [HKEY_CURRENT_USER\\SOFTWARE\\Python\\PythonCore\\${_CURRENT_VERSION}\\InstallPath]/include - PATH_SUFFIXES - python${_CURRENT_VERSION}mu - python${_CURRENT_VERSION}m - python${_CURRENT_VERSION}u - python${_CURRENT_VERSION} - ) - - # For backward compatibility, set PYTHON_INCLUDE_PATH. - set(PYTHON_INCLUDE_PATH "${PYTHON_INCLUDE_DIR}") - - if(PYTHON_INCLUDE_DIR AND EXISTS "${PYTHON_INCLUDE_DIR}/patchlevel.h") - file(STRINGS "${PYTHON_INCLUDE_DIR}/patchlevel.h" python_version_str - REGEX "^#define[ \t]+PY_VERSION[ \t]+\"[^\"]+\"") - string(REGEX REPLACE "^#define[ \t]+PY_VERSION[ \t]+\"([^\"]+)\".*" "\\1" - PYTHONLIBS_VERSION_STRING "${python_version_str}") - unset(python_version_str) - endif() - - if(PYTHON_LIBRARY AND PYTHON_INCLUDE_DIR) - break() - endif() -endforeach() - -mark_as_advanced( - PYTHON_DEBUG_LIBRARY - PYTHON_LIBRARY - PYTHON_INCLUDE_DIR -) - -# We use PYTHON_INCLUDE_DIR, PYTHON_LIBRARY and PYTHON_DEBUG_LIBRARY for the -# cache entries because they are meant to specify the location of a single -# library. We now set the variables listed by the documentation for this -# module. -set(PYTHON_INCLUDE_DIRS "${PYTHON_INCLUDE_DIR}") -set(PYTHON_DEBUG_LIBRARIES "${PYTHON_DEBUG_LIBRARY}") - -# These variables have been historically named in this module different from -# what SELECT_LIBRARY_CONFIGURATIONS() expects. -set(PYTHON_LIBRARY_DEBUG "${PYTHON_DEBUG_LIBRARY}") -set(PYTHON_LIBRARY_RELEASE "${PYTHON_LIBRARY}") -include(SelectLibraryConfigurations) -SELECT_LIBRARY_CONFIGURATIONS(PYTHON) -# SELECT_LIBRARY_CONFIGURATIONS() sets ${PREFIX}_FOUND if it has a library. -# Unset this, this prefix doesn't match the module prefix, they are different -# for historical reasons. -unset(PYTHON_FOUND) - -include(FindPackageHandleStandardArgs) -FIND_PACKAGE_HANDLE_STANDARD_ARGS(PythonLibs - REQUIRED_VARS PYTHON_LIBRARIES PYTHON_INCLUDE_DIRS - VERSION_VAR PYTHONLIBS_VERSION_STRING) - -# PYTHON_ADD_MODULE( src1 src2 ... srcN) is used to build modules for python. -# PYTHON_WRITE_MODULES_HEADER() writes a header file you can include -# in your sources to initialize the static python modules -function(PYTHON_ADD_MODULE _NAME ) - get_property(_TARGET_SUPPORTS_SHARED_LIBS - GLOBAL PROPERTY TARGET_SUPPORTS_SHARED_LIBS) - option(PYTHON_ENABLE_MODULE_${_NAME} "Add module ${_NAME}" TRUE) - option(PYTHON_MODULE_${_NAME}_BUILD_SHARED - "Add module ${_NAME} shared" ${_TARGET_SUPPORTS_SHARED_LIBS}) - - # Mark these options as advanced - mark_as_advanced(PYTHON_ENABLE_MODULE_${_NAME} - PYTHON_MODULE_${_NAME}_BUILD_SHARED) - - if(PYTHON_ENABLE_MODULE_${_NAME}) - if(PYTHON_MODULE_${_NAME}_BUILD_SHARED) - set(PY_MODULE_TYPE MODULE) - else() - set(PY_MODULE_TYPE STATIC) - set_property(GLOBAL APPEND PROPERTY PY_STATIC_MODULES_LIST ${_NAME}) - endif() - - set_property(GLOBAL APPEND PROPERTY PY_MODULES_LIST ${_NAME}) - add_library(${_NAME} ${PY_MODULE_TYPE} ${ARGN}) -# target_link_libraries(${_NAME} ${PYTHON_LIBRARIES}) - - if(PYTHON_MODULE_${_NAME}_BUILD_SHARED) - set_target_properties(${_NAME} PROPERTIES PREFIX "${PYTHON_MODULE_PREFIX}") - if(WIN32 AND NOT CYGWIN) - set_target_properties(${_NAME} PROPERTIES SUFFIX ".pyd") - endif() - endif() - - endif() -endfunction() - -function(PYTHON_WRITE_MODULES_HEADER _filename) - - get_property(PY_STATIC_MODULES_LIST GLOBAL PROPERTY PY_STATIC_MODULES_LIST) - - get_filename_component(_name "${_filename}" NAME) - string(REPLACE "." "_" _name "${_name}") - string(TOUPPER ${_name} _nameUpper) - set(_filename ${CMAKE_CURRENT_BINARY_DIR}/${_filename}) - - set(_filenameTmp "${_filename}.in") - file(WRITE ${_filenameTmp} "/*Created by cmake, do not edit, changes will be lost*/\n") - file(APPEND ${_filenameTmp} -"#ifndef ${_nameUpper} -#define ${_nameUpper} - -#include - -#ifdef __cplusplus -extern \"C\" { -#endif /* __cplusplus */ - -") - - foreach(_currentModule ${PY_STATIC_MODULES_LIST}) - file(APPEND ${_filenameTmp} "extern void init${PYTHON_MODULE_PREFIX}${_currentModule}(void);\n\n") - endforeach() - - file(APPEND ${_filenameTmp} -"#ifdef __cplusplus -} -#endif /* __cplusplus */ - -") - - - foreach(_currentModule ${PY_STATIC_MODULES_LIST}) - file(APPEND ${_filenameTmp} "int ${_name}_${_currentModule}(void) \n{\n static char name[]=\"${PYTHON_MODULE_PREFIX}${_currentModule}\"; return PyImport_AppendInittab(name, init${PYTHON_MODULE_PREFIX}${_currentModule});\n}\n\n") - endforeach() - - file(APPEND ${_filenameTmp} "void ${_name}_LoadAllPythonModules(void)\n{\n") - foreach(_currentModule ${PY_STATIC_MODULES_LIST}) - file(APPEND ${_filenameTmp} " ${_name}_${_currentModule}();\n") - endforeach() - file(APPEND ${_filenameTmp} "}\n\n") - file(APPEND ${_filenameTmp} "#ifndef EXCLUDE_LOAD_ALL_FUNCTION\nvoid CMakeLoadAllPythonModules(void)\n{\n ${_name}_LoadAllPythonModules();\n}\n#endif\n\n#endif\n") - -# with configure_file() cmake complains that you may not use a file created using file(WRITE) as input file for configure_file() - execute_process(COMMAND ${CMAKE_COMMAND} -E copy_if_different "${_filenameTmp}" "${_filename}" OUTPUT_QUIET ERROR_QUIET) - -endfunction() diff --git a/cmake/FindScala.cmake b/cmake/FindScala.cmake index 46c683efe1b..b56b493b3ea 100644 --- a/cmake/FindScala.cmake +++ b/cmake/FindScala.cmake @@ -6,13 +6,15 @@ set(_SCALA_PATHS /sw /usr /usr/share/java + /usr/share/scala + /usr/share/scala/lib ) find_package(Java COMPONENTS Runtime) if(JAVA_FOUND) include(UseJava) else() - message(WARNING "JAVA count not be found!" "\nIt is required for Scala Modular Interface!!!") + message(WARNING "Java is not found." "\nIt is required for the Scala Interface.") endif() find_program(Scala_SCALA_EXECUTABLE @@ -25,29 +27,31 @@ find_program(Scala_SCALAC_EXECUTABLE PATHS ${_SCALA_PATHS} ) -find_jar(Scala_JAR_EXECUTABLE "scala-library") +find_jar(Scala_JAR_EXECUTABLE + NAMES "scala-library" + PATHS ${_SCALA_PATHS} + ) -if(Scala_SCALA_EXECUTABLE) - execute_process(COMMAND ${Scala_SCALA_EXECUTABLE} -version - RESULT_VARIABLE SCALA_SEARCH_SUCCESS - OUTPUT_VARIABLE SCALA_VERSION - ERROR_VARIABLE SCALA_VERSION - OUTPUT_STRIP_TRAILING_WHITESPACE - ERROR_STRIP_TRAILING_WHITESPACE) - if( SCALA_SEARCH_SUCCESS ) - message( FATAL_ERROR "Error executing scala -version" ) - else() - string(TOLOWER ${SCALA_VERSION} SCALA_VERSION) - string( REGEX REPLACE ".*([0-9]+\\.[0-9]+\\.[0-9_.]+.*)" "\\1" SCALA_VERSION "${SCALA_VERSION}" ) - string( REGEX REPLACE "([0-9]+\\.[0-9]+\\.[0-9_.]).*" "\\1" SCALA_VERSION ${SCALA_VERSION} ) - endif() +if( "${Scala_SCALA_EXECUTABLE}" MATCHES "Scala_SCALA_EXECUTABLE-NOTFOUND" ) + message( FATAL_ERROR "Error executing scala -version" ) +else() + execute_process(COMMAND ${Scala_SCALA_EXECUTABLE} -version + RESULT_VARIABLE SCALA_SEARCH_SUCCESS + OUTPUT_VARIABLE SCALA_VERSION + ERROR_VARIABLE SCALA_VERSION + OUTPUT_STRIP_TRAILING_WHITESPACE + ERROR_STRIP_TRAILING_WHITESPACE + ) + string(TOLOWER ${SCALA_VERSION} SCALA_VERSION) + string( REGEX REPLACE ".*([0-9]+\\.[0-9]+\\.[0-9_.]+.*)" "\\1" SCALA_VERSION "${SCALA_VERSION}" ) + string( REGEX REPLACE "([0-9]+\\.[0-9]+\\.[0-9_.]).*" "\\1" SCALA_VERSION ${SCALA_VERSION} ) endif() include(FindPackageHandleStandardArgs) if (CMAKE_VERSION LESS 2.8.3) find_package_handle_standard_args(SCALA DEFAULT_MSG Scala_SCALA_EXECUTABLE) else () - find_package_handle_standard_args(SCALA + find_package_handle_standard_args(SCALA REQUIRED_VARS Scala_SCALA_EXECUTABLE Scala_SCALAC_EXECUTABLE Scala_JAR_EXECUTABLE VERSION_VAR SCALA_VERSION) endif () diff --git a/cmake/FindSpinlock.cmake b/cmake/FindSpinlock.cmake deleted file mode 100644 index 8c590c0945c..00000000000 --- a/cmake/FindSpinlock.cmake +++ /dev/null @@ -1,23 +0,0 @@ -IF (CMAKE_USE_PTHREADS_INIT) - SET(_bindir "${CMAKE_BINARY_DIR}${CMAKE_FILES_DIRECTORY}") - IF (DARWIN) - TRY_COMPILE(HAVE_SPINLOCK "${_bindir}" "${CMAKE_MODULE_PATH}/spinlock-test-darwin.cpp") - ELSE () - TRY_COMPILE(HAVE_SPINLOCK "${_bindir}" - "${CMAKE_MODULE_PATH}/spinlock-test.cpp" - CMAKE_FLAGS "-DLINK_LIBRARIES:STRING=${CMAKE_THREAD_LIBS_INIT}" - ) - ENDIF () - - IF (HAVE_SPINLOCK) - MESSAGE(STATUS "Spinlock support found") - SET(SPINLOCK_FOUND TRUE) - ELSE (HAVE_SPINLOCK) - MESSAGE(STATUS "Spinlock support not found") - SET(SPINLOCK_FOUND FALSE) - ENDIF (HAVE_SPINLOCK) -ELSE () - MESSAGE(STATUS "Spinlock support not found due to no pthreads available") - SET(SPINLOCK_FOUND FALSE) -ENDIF () - diff --git a/cmake/Findrxcpp.cmake b/cmake/Findrxcpp.cmake new file mode 100644 index 00000000000..2577e3b6a35 --- /dev/null +++ b/cmake/Findrxcpp.cmake @@ -0,0 +1,30 @@ +# Copyright Gonzalo Brito Gadeschi 2015 +# Copyright Kirk Shoop 2016 +# Distributed under the Boost Software License, Version 1.0. +# (See accompanying file LICENSE.md or copy at http://boost.org/LICENSE_1_0.txt) +# +# Find the rxcpp include directory +# The following variables are set if rxcpp is found. +# rxcpp_FOUND - True when the rxcpp include directory is found. +# rxcpp_INCLUDE_DIR - The path to where the meta include files are. +# If rxcpp is not found, rxcpp_FOUND is set to false. + +# https://github.com/gnzlbg/ndtree/blob/master/cmake/Findrange-v3.cmake + +find_package(PkgConfig) + +if(NOT EXISTS "${rxcpp_INCLUDE_DIR}") + find_path(rxcpp_INCLUDE_DIR + NAMES rxcpp/rx.hpp + DOC "rxcpp library header files" + ) +endif() + +include(FindPackageHandleStandardArgs) +mark_as_advanced(rxcpp_INCLUDE_DIR) + +if(EXISTS "${rxcpp_INCLUDE_DIR}") + set(rxcpp_FOUND 1) +else() + set(rxcpp_FOUND 0) +endif() \ No newline at end of file diff --git a/cmake/ShogunFindLAPACK.cmake b/cmake/ShogunFindLAPACK.cmake new file mode 100644 index 00000000000..fe3d6db9c24 --- /dev/null +++ b/cmake/ShogunFindLAPACK.cmake @@ -0,0 +1,118 @@ +# Enable Eigen to use Lapack backend +OPTION(ENABLE_EIGEN_LAPACK "Enable Eigen to use detected BLAS and LAPACK backend" ON) + +FIND_PACKAGE(LAPACK QUIET) +IF (LAPACK_FOUND) + SET(HAVE_LAPACK 1) + + # find out the type of Lapack/BLAS implementation we are dealing with + IF("${LAPACK_LIBRARIES}" MATCHES ".*/Accelerate.framework$") + # Accelerate.framework we found for LaPack/BLAS + SET(HAVE_MVEC 1) + SET(HAVE_CATLAS 1) + MESSAGE(STATUS "Found Accelerate.framework using as BLAS/LAPACK backend.") + + if (ENABLE_EIGEN_LAPACK) + SET(EIGEN_USE_BLAS 1) + MESSAGE(STATUS "Enabling Accelerate.framework as BLAS backend for Eigen.") + find_library(LAPACKE_LIBRARY + NAMES lapacke + PATHS /usr/lib /usr/local/lib $ENV{LAPACKE_PATH}) + if (LAPACKE_LIBRARY) + MESSAGE(STATUS "Enabling Accelerate.framework as LAPACK backend for Eigen.") + SET(EIGEN_USE_LAPACKE_STRICT 1) + LIST(APPEND LAPACK_LIBRARIES ${LAPACKE_LIBRARY}) + endif() + endif() + ELSEIF("${LAPACK_LIBRARIES}" MATCHES ".*/.*mkl_.*") + # in case MKL is available enable Eigen to use it. + # for more fine grained control and details see: + # https://eigen.tuxfamily.org/dox/TopicUsingIntelMKL.html + # this is supported since Eigen version 3.1 and later + MESSAGE(STATUS "Found MKL using as BLAS/LAPACK backend.") + SET(HAVE_MKL 1) + + # trying to use the new Single Dynamic Library of MKL + # https://software.intel.com/en-us/articles/a-new-linking-model-single-dynamic-library-mkl_rt-since-intel-mkl-103 + IF (NOT "${LAPACK_LIBRARIES}" MATCHES ".*/.*mkl_rt.*") + # just use the mkl_rt and let the user specify/decide all the MKL specific + # optimisation runtime + SET(MKL_LIBRARIES ${LAPACK_LIBRARIES}) + LIST(FILTER MKL_LIBRARIES INCLUDE REGEX ".*/.*mkl_core.*") + get_filename_component(MKL_PATH ${MKL_LIBRARIES} DIRECTORY) + find_library(MKL_RT mkl_rt PATHS ${MKL_PATH}) + IF (MKL_RT) + IF (MSVC) + SET(LAPACK_LIBRARIES ${MKL_RT}) + ELSEIF(CMAKE_USE_PTHREADS_INIT) + SET(LAPACK_LIBRARIES ${MKL_RT}) + LIST(APPEND LAPACK_LIBRARIES ${CMAKE_THREAD_LIBS_INIT} -lm) + ENDIF() + ENDIF() + ENDIF() + + IF (ENABLE_EIGEN_LAPACK) + FIND_PATH(MKL_INCLUDE_DIR mkl.h) + IF(NOT MKL_INCLUDE_DIR) + MESSAGE(STATUS "Found MKL, but not mkl.h. Make sure that mkl headers are available in order to use MKL as BLAS/Lapack backend for Eigen.") + SET(ENABLE_EIGEN_LAPACK OFF CACHE BOOL "Use ${ENABLE_EIGEN_LAPACK}" FORCE) + ELSE() + MESSAGE(STATUS "Enabling MKL as BLAS/Lapack backend for Eigen.") + SET(EIGEN_USE_MKL_ALL 1) + target_include_directories(shogun PUBLIC ${MKL_INCLUDE_DIR}) + IF (LIBSHOGUN_BUILD_STATIC) + target_include_directories(shogun-static PUBLIC ${MKL_INCLUDE_DIR}) + ENDIF() + ENDIF() + ENDIF() + ELSE() + include(CheckLibraryExists) + # test whether we have cblas.h in the header paths and the detected + # LAPACK_LIBRARIES contains all the libraries to compile even with cblas_* functions + check_library_exists("${LAPACK_LIBRARIES}" cblas_dgemv "" FOUND_CBLAS_DGEMV) + + # detect if the detected Lapack is atlas + # clapack_* functions are implemented in atlas + check_library_exists("${LAPACK_LIBRARIES}" clapack_dpotrf "" FOUND_CLAPACK_DPOTRF) + IF (FOUND_CLAPACK_DPOTRF OR NOT FOUND_CBLAS_DGEMV) + FIND_PACKAGE(Atlas QUIET) + IF(Atlas_FOUND) + MESSAGE(STATUS "Found Atlas using as BLAS/LAPACK backend.") + SET(HAVE_ATLAS 1) + SHOGUN_INCLUDE_DIRS(SCOPE PUBLIC ${Atlas_INCLUDE_DIRS}) + IF (NOT FOUND_CBLAS_DGEMV) + # this usually happens on RHEL/CentOS; usually having Atlas is good + SET(LAPACK_LIBRARIES ${Atlas_LIBRARIES}) + ENDIF() + ELSEIF(NOT FOUND_CBLAS_DGEMV) + UNSET(LAPACK_FOUND CACHE) + UNSET(LAPACK_LIBRARIES) + UNSET(HAVE_LAPACK) + ENDIF() + ENDIF() + + # if LaPack is detected and Eigen is 3.3 or later + # use the lapack/blas backend in Eigen + IF(${EIGEN_VERSION} VERSION_GREATER 3.3.0 AND ENABLE_EIGEN_LAPACK AND HAVE_LAPACK) + SET(EIGEN_USE_BLAS 1) + MESSAGE(STATUS "Enabling detected BLAS library as backend for Eigen") + + find_library(LAPACKE_LIBRARY NAMES lapacke PATHS /usr/lib /usr/local/lib $ENV{LAPACKE_PATH}) + IF (LAPACKE_LIBRARY) + MESSAGE(STATUS "Enabling detected LAPACK backend for Eigen") + SET(EIGEN_USE_LAPACKE_STRICT 1) + LIST(APPEND LAPACK_LIBRARIES ${LAPACKE_LIBRARY}) + ENDIF() + ENDIF() + ENDIF() + + IF (ENABLE_EIGEN_LAPACK) + SET (LAPACK_SCOPE PUBLIC) + ELSE() + SET (LAPACK_SCOPE PRIVATE) + ENDIF() + target_link_libraries(shogun ${LAPACK_SCOPE} ${LAPACK_LIBRARIES}) + if (LIBSHOGUN_BUILD_STATIC) + target_link_libraries(shogun-static ${LAPACK_SCOPE} ${LAPACK_LIBRARIES}) + endif() +ENDIF() diff --git a/cmake/ShogunInterfaces.cmake b/cmake/ShogunInterfaces.cmake new file mode 100644 index 00000000000..3730a71088d --- /dev/null +++ b/cmake/ShogunInterfaces.cmake @@ -0,0 +1,105 @@ +MACRO(GENERATE_INTERFACE_TARGET INTERFACE_NAME INTERFACE_DIR INTERFACE_LIBRARIES) + +get_target_property(ShogunIncludes shogun::shogun INTERFACE_INCLUDE_DIRECTORIES) +INCLUDE_DIRECTORIES(${ShogunIncludes}) + +# set compiler SWIG generated cxx compiler flags +SET(CMAKE_CXX_FLAGS ${SWIG_CXX_COMPILER_FLAGS}) +# unset any release or distribution flags +# we don't want them when compiling SWIG generated source +SET(CMAKE_CXX_FLAGS_RELEASE "") +SET(CMAKE_CXX_FLAGS_DISTRIBUTION "") +SET(CMAKE_CXX_FLAGS_DEBUG "") + +if(${INTERFACE_NAME} STREQUAL "python") + SET(PREPEND_TARGET "_") +endif() + +SET(INTERFACE_FILES) +FILE(GLOB_RECURSE INTERFACE_FILES ${COMMON_INTERFACE_SRC_DIR}/*.i) +FILE(GLOB_RECURSE CUSTOM_INTERFACE_FILES ${INTERFACE_DIR}/*.i) +LIST(APPEND INTERFACE_FILES ${CUSTOM_INTERFACE_FILES}) +FOREACH(file ${INTERFACE_FILES}) + get_filename_component(fname ${file} NAME) + list(APPEND INTERFACE_FILES ${fname}) + ADD_CUSTOM_COMMAND(OUTPUT ${fname} + DEPENDS ${file} + COMMAND "${CMAKE_COMMAND}" -E copy_if_different ${file} ${fname} + COMMENT "" + ) +ENDFOREACH() + +SET(INTERFACE_TARGET interface_${INTERFACE_NAME}) +SET(INTERFACE_TARGET_SRC ${INTERFACE_TARGET}_src) + +ADD_CUSTOM_TARGET(${INTERFACE_TARGET_SRC} + DEPENDS shogun::shogun ${INTERFACE_FILES} + COMMENT "copying SWIG files") + +INCLUDE(${SWIG_USE_FILE}) +SET_SOURCE_FILES_PROPERTIES(shogun.i PROPERTIES CPLUSPLUS ON) +IF(DEFINED TARGET_SWIGFLAGS) + SET_SOURCE_FILES_PROPERTIES(shogun.i PROPERTIES SWIG_FLAGS ${TARGET_SWIGFLAGS}) +ENDIF() +SET(SWIG_MODULE_${INTERFACE_NAME}_EXTRA_DEPS ${INTERFACE_FILES}) +SWIG_ADD_MODULE(${INTERFACE_TARGET} ${INTERFACE_NAME} shogun.i sg_print_functions.cpp) +SWIG_LINK_LIBRARIES(${INTERFACE_TARGET} shogun::shogun ${INTERFACE_LIBRARIES}) + + +#get_cmake_property(_variableNames VARIABLES) +#foreach (_variableName ${_variableNames}) + # message(STATUS "${_variableName}=${${_variableName}}") + # endforeach() + +SET(INTERFACE_REAL_NAME ${SWIG_MODULE_interface_${INTERFACE_NAME}_REAL_NAME}) +SET_TARGET_PROPERTIES(${INTERFACE_REAL_NAME} PROPERTIES OUTPUT_NAME ${PREPEND_TARGET}shogun) +ADD_DEPENDENCIES(${INTERFACE_REAL_NAME} ${INTERFACE_TARGET_SRC}) + +#ADD_CUSTOM_COMMAND(TARGETS ${PREPEND_TARGET}interface_${INTERFACE_NAME} +# POST_BUILD +# COMMAND ${PYTHON_EXECUTABLE} +# ARGS ${CMAKE_SOURCE_DIR}/src/.scrub_docstrings.py ) + +IF(DOXYGEN_FOUND) + configure_file(${COMMON_INTERFACE_SRC_DIR}/shogun.doxy.in shogun.doxy) + + ADD_CUSTOM_COMMAND( + OUTPUT shogun_doxygen + COMMAND ${DOXYGEN_EXECUTABLE} + ARGS shogun.doxy + DEPENDS shogun::shogun + COMMENT "Generating doxygen doc" + ) + + ADD_CUSTOM_COMMAND( + OUTPUT shogun_doxygen.i + COMMAND ${PYTHON_EXECUTABLE} ${CMAKE_SOURCE_DIR}/src/.doxy2swig.py + ARGS --quiet --no-function-definition shogun_doxygen/xml/index.xml shogun_doxygen.i + DEPENDS shogun_doxygen + ) + ADD_CUSTOM_TARGET(${INTERFACE_NAME}_doxy2swig DEPENDS shogun_doxygen.i) + ADD_DEPENDENCIES(${INTERFACE_REAL_NAME} ${INTERFACE_NAME}_doxy2swig) +ELSE() + #TODO add scrubing +ENDIF() + +# Make sure all interfaces are build single-threaded to reduce +# excessive memory consumption during build. +IF(SWIG_SINGLE_THREADED) + FOREACH(SG_INTERFACE_TARGET ${SG_INTERFACE_TARGETS}) + ADD_DEPENDENCIES(${INTERFACE_REAL_NAME} + ${SG_INTERFACE_TARGET}) + ENDFOREACH(SG_INTERFACE_TARGET ${SG_INTERFACE_TARGETS}) + SET(SG_INTERFACE_TARGETS + "${INTERFACE_REAL_NAME};${SG_INTERFACE_TARGETS}" + CACHE STRING "List of modular-interfaces beeing build." FORCE) +ENDIF(SWIG_SINGLE_THREADED) + +CONFIGURE_FILE(${COMMON_INTERFACE_SRC_DIR}/swig_config.h.in swig_config.h) + +UNSET(INTERFACE_TARGET) +UNSET(INTERFACE_TARGET_SRC) +UNSET(INTERFACE_REAL_NAME) +UNSET(INTERFACE_FILES) + +ENDMACRO() diff --git a/cmake/ShogunUtils.cmake b/cmake/ShogunUtils.cmake index abc30353869..a133bd420ce 100644 --- a/cmake/ShogunUtils.cmake +++ b/cmake/ShogunUtils.cmake @@ -57,7 +57,7 @@ macro(DetectSystemName) list(FIND CMAKE_PLATFORM_IMPLICIT_LINK_DIRECTORIES "${CMAKE_INSTALL_PREFIX}/lib" isSystemDir) if(${isSystemDir} STREQUAL "-1") set(CMAKE_INSTALL_RPATH "${CMAKE_INSTALL_PREFIX}/lib") - endif(${isSystemDir} STREQUAL "-1") + endif(${isSystemDir} STREQUAL "-1") ELSEIF(${CMAKE_SYSTEM_NAME} MATCHES "Linux") SET(LINUX 1) ELSEIF(${CMAKE_SYSTEM_NAME} MATCHES "FreeBSD") @@ -70,11 +70,15 @@ MACRO(PrintInterfaceStatus INTERFACE_NAME INTERFACE_FLAG) message(STATUS " ${INTERFACE_NAME} is ON") ELSE() STRING(LENGTH ${INTERFACE_NAME} IFACE_NAME_LENGTH) - IF (IFACE_NAME_LENGTH LESS 10) - message(STATUS " ${INTERFACE_NAME} is OFF \t\t - enable with -D${INTERFACE_FLAG}=ON") + IF (IFACE_NAME_LENGTH LESS 3) + SET(INDENT "\t\t\t") + ELSEIF (IFACE_NAME_LENGTH LESS 10) + SET(INDENT "\t\t") ELSE () - message(STATUS " ${INTERFACE_NAME} is OFF \t - enable with -D${INTERFACE_FLAG}=ON") + SET(INDENT "\t") ENDIF () + message(STATUS " ${INTERFACE_NAME} is OFF ${INDENT} enable with -D${INTERFACE_FLAG}=ON") + UNSET(INDENT) ENDIF() ENDMACRO() @@ -107,6 +111,30 @@ MACRO(AddMetaIntegrationTest META_TARGET CONDITION) ENDIF() ENDMACRO() +MACRO(AddLibShogunExample EXAMPLE_CPP) + STRING(REGEX REPLACE ".cpp\$" "" EXAMPLE "${EXAMPLE_CPP}") + + add_executable(${EXAMPLE} ${CMAKE_CURRENT_SOURCE_DIR}/${EXAMPLE_CPP}) + if(WIN32) + target_link_libraries(${EXAMPLE} shogun::shogun-static ${SANITIZER_LIBRARY}) + else() + target_link_libraries(${EXAMPLE} shogun::shogun ${SANITIZER_LIBRARY}) + endif() + IF(SANITIZER_FLAGS) + set_target_properties(${EXAMPLE} PROPERTIES COMPILE_FLAGS ${SANITIZER_FLAGS}) + ENDIF() + add_test(libshogun-${EXAMPLE} ${CMAKE_CURRENT_BINARY_DIR}/${EXAMPLE}) + + # Add examples to the dependencies of modular interfaces to make sure + # nothing will infer with them being build single-threaded. + IF(SWIG_SINGLE_THREADED) + FOREACH(SG_INTERFACE_TARGET ${SG_INTERFACE_TARGETS}) + ADD_DEPENDENCIES(${SG_INTERFACE_TARGET} ${EXAMPLE}) + ENDFOREACH(SG_INTERFACE_TARGET ${SG_INTERFACE_TARGETS}) + ENDIF(SWIG_SINGLE_THREADED) +ENDMACRO() + + function(PrintLine) message(STATUS "===================================================================================================================") endfunction() @@ -151,7 +179,7 @@ macro(ADD_LIBRARY_DEPENDENCY) if (${TARGET_TYPE} STREQUAL INTERFACE_LIBRARY) target_include_directories(${element} INTERFACE ${LIBRARY_HEADER}) target_link_libraries(${element} INTERFACE ${${LIBRARY_PREFIX}_LIBRARIES}) - else() + else() if (NOT ${TARGET_TYPE} STREQUAL OBJECT_LIBRARY) target_link_libraries(${element} ${ADD_LIBRARY_DEPENDENCY_SCOPE} ${${LIBRARY_PREFIX}_LIBRARIES}) endif() @@ -245,9 +273,8 @@ function(GET_META_EXAMPLE_VARS META_EXAMPLE EX_NAME REL_DIR NAME_WITH_DIR) set(${NAME_WITH_DIR} ${EXAMPLE_NAME_WITH_DIR} PARENT_SCOPE) endfunction() - function(GET_INTERFACE_VARS INTERFACE DIRECTORY EXTENSION) - string(REGEX MATCH "([a-zA-Z]+)Modular" _dir ${INTERFACE}) + string(REGEX MATCH "INTERFACE_([a-zA-Z]+)" _dir ${INTERFACE}) STRING(TOLOWER "${CMAKE_MATCH_1}" _dir) SET(${DIRECTORY} ${_dir} PARENT_SCOPE) diff --git a/cmake/external/Eigen3.cmake b/cmake/external/Eigen3.cmake index fb9b698ab33..f947b0e7b1d 100644 --- a/cmake/external/Eigen3.cmake +++ b/cmake/external/Eigen3.cmake @@ -1,13 +1,17 @@ GetCompilers() set(EIGEN_INCLUDE_DIR ${THIRD_PARTY_INCLUDE_DIR}/eigen) +set(EIGEN_VERSION 3.2.8) include(ExternalProject) ExternalProject_Add( Eigen3 PREFIX ${CMAKE_BINARY_DIR}/Eigen3 DOWNLOAD_DIR ${THIRD_PARTY_DIR}/Eigen3 - URL https://bitbucket.org/eigen/eigen/get/3.2.8.tar.bz2 + URL https://bitbucket.org/eigen/eigen/get/${EIGEN_VERSION}.tar.bz2 URL_MD5 9e3bfaaab3db18253cfd87ea697b3ab1 +# FIXME: http://eigen.tuxfamily.org/bz/show_bug.cgi?id=1445 +# URL https://bitbucket.org/eigen/eigen/get/3.3.4.tar.bz2 +# URL_MD5 a7aab9f758249b86c93221ad417fbe18 CMAKE_ARGS -DEIGEN_INCLUDE_INSTALL_DIR:PATH=${EIGEN_INCLUDE_DIR} -DCMAKE_C_FLAGS:STRING=${CMAKE_C_FLAGS}${CMAKE_DEFINITIONS} -DCMAKE_CXX_FLAGS:STRING=${CMAKE_CXX_FLAGS}${CMAKE_DEFINITIONS} diff --git a/cmake/external/GoogleTestNMock.cmake b/cmake/external/GoogleTestNMock.cmake index 4f296b09028..05e093f2323 100644 --- a/cmake/external/GoogleTestNMock.cmake +++ b/cmake/external/GoogleTestNMock.cmake @@ -23,16 +23,27 @@ ELSE () ) ENDIF() -ExternalProject_Add( - GoogleMock - URL https://github.com/google/googletest/archive/release-1.8.0.tar.gz - URL_MD5 16877098823401d1bf2ed7891d7dce36 - TIMEOUT 10 - PREFIX ${CMAKE_BINARY_DIR}/GoogleMock - DOWNLOAD_DIR ${THIRD_PARTY_DIR}/GoogleMock - INSTALL_COMMAND "" - CMAKE_ARGS ${CUSTOM_CMAKE_ARGS} -) +IF(EXISTS /usr/src/googletest) + ExternalProject_Add( + GoogleMock + DOWNLOAD_COMMAND "" + SOURCE_DIR /usr/src/googletest + PREFIX ${CMAKE_BINARY_DIR}/GoogleMock + INSTALL_COMMAND "" + CMAKE_ARGS ${CUSTOM_CMAKE_ARGS} + ) +ELSE() + ExternalProject_Add( + GoogleMock + URL https://github.com/google/googletest/archive/release-1.8.0.tar.gz + URL_MD5 16877098823401d1bf2ed7891d7dce36 + TIMEOUT 10 + PREFIX ${CMAKE_BINARY_DIR}/GoogleMock + DOWNLOAD_DIR ${THIRD_PARTY_DIR}/GoogleMock + INSTALL_COMMAND "" + CMAKE_ARGS ${CUSTOM_CMAKE_ARGS} + ) +ENDIF() UNSET(C_COMPILER) UNSET(CXX_COMPILER) \ No newline at end of file diff --git a/cmake/external/TFLogger.cmake b/cmake/external/TFLogger.cmake new file mode 100644 index 00000000000..50e4daa3ebe --- /dev/null +++ b/cmake/external/TFLogger.cmake @@ -0,0 +1,21 @@ +GetCompilers() + +include(ExternalProject) +ExternalProject_Add( + rxcpp + PREFIX ${CMAKE_BINARY_DIR}/tflogger + DOWNLOAD_DIR ${THIRD_PARTY_DIR}/tflogger + URL https://github.com/shogun-toolbox/tflogger/archive/master.zip + CMAKE_ARGS + -DCMAKE_INSTALL_PREFIX:STRING=${CMAKE_BINARY_DIR}/src/shogun/lib/external + -DCMAKE_C_COMPILER:STRING=${C_COMPILER} + -DCMAKE_CXX_COMPILER:STRING=${CXX_COMPILER} + BUILD_COMMAND "" + ) + +add_dependencies(libshogun tflogger) + +set(TFLogger_INCLUDE_DIR ${THIRD_PARTY_INCLUDE_DIR}) + +UNSET(C_COMPILER) +UNSET(CXX_COMPILER) diff --git a/cmake/external/rxcpp.cmake b/cmake/external/rxcpp.cmake new file mode 100644 index 00000000000..8a25b2a5ca7 --- /dev/null +++ b/cmake/external/rxcpp.cmake @@ -0,0 +1,23 @@ +GetCompilers() + +include(ExternalProject) +ExternalProject_Add( + rxcpp + PREFIX ${CMAKE_BINARY_DIR}/rxcpp + DOWNLOAD_DIR ${THIRD_PARTY_DIR}/rxcpp + URL https://github.com/Reactive-Extensions/RxCpp/archive/v4.0.0.tar.gz + URL_MD5 feb89934f465bb5ac513c9adce8d3b1b + CMAKE_ARGS + -DCMAKE_BUILD_TYPE:STRING=RelWithDebInfo + -DCMAKE_INSTALL_PREFIX:STRING=${CMAKE_BINARY_DIR}/src/shogun/lib/external + -DCMAKE_C_COMPILER:STRING=${C_COMPILER} + -DCMAKE_CXX_COMPILER:STRING=${CXX_COMPILER} + BUILD_COMMAND "" + ) + +add_dependencies(libshogun rxcpp) + +set(rxcpp_INCLUDE_DIR ${THIRD_PARTY_INCLUDE_DIR}) + +UNSET(C_COMPILER) +UNSET(CXX_COMPILER) \ No newline at end of file diff --git a/cmake/spinlock-test-darwin.cpp b/cmake/spinlock-test-darwin.cpp deleted file mode 100644 index 7a042d14f5e..00000000000 --- a/cmake/spinlock-test-darwin.cpp +++ /dev/null @@ -1,8 +0,0 @@ -#include - -int main() -{ - volatile OSSpinLock lock; - OSSpinLockTry(&lock); - return 0; -} diff --git a/cmake/spinlock-test.cpp b/cmake/spinlock-test.cpp deleted file mode 100644 index e871211f13e..00000000000 --- a/cmake/spinlock-test.cpp +++ /dev/null @@ -1,8 +0,0 @@ -#include - -int main() -{ - volatile pthread_spinlock_t spinlock; - pthread_spin_init(&spinlock, 0); - return 0; -} diff --git a/configs/shogun-sdk/Dockerfile b/configs/shogun-sdk/Dockerfile index 8300927b1e2..17b2a580628 100644 --- a/configs/shogun-sdk/Dockerfile +++ b/configs/shogun-sdk/Dockerfile @@ -1,41 +1,45 @@ -FROM ubuntu:14.04 +FROM debian:stretch MAINTAINER shogun@shogun-toolbox.org -RUN apt-get update && apt-get install -qq software-properties-common lsb-release -RUN add-apt-repository "deb http://archive.ubuntu.com/ubuntu $(lsb_release -sc) multiverse" -RUN add-apt-repository -y ppa:rosmo/swig3.0.7 -RUN add-apt-repository -y ppa:webupd8team/java -RUN add-apt-repository -y ppa:george-edison55/cmake-3.x -RUN apt-get update -qq -RUN apt-get upgrade -y - -RUN echo debconf shared/accepted-oracle-license-v1-1 select true | debconf-set-selections -RUN echo debconf shared/accepted-oracle-license-v1-1 seen true | debconf-set-selections - -# install dependencies -RUN apt-get install -qq --force-yes --no-install-recommends make gcc g++ \ +RUN apt-get update -qq && apt-get upgrade -y && \ + apt-get install -qq --force-yes --no-install-recommends make gcc g++ \ libc6-dev libbz2-dev libjson-c-dev ccache libarpack2-dev libatlas-base-dev \ - libblas-dev libglpk-dev libhdf5-serial-dev zlib1g-dev liblapack-dev cmake \ + libblas-dev libglpk-dev libhdf5-serial-dev zlib1g-dev liblapacke-dev cmake \ libnlopt-dev liblpsolve55-dev libxml2-dev libsnappy-dev liblzo2-dev \ liblzma-dev libeigen3-dev swig3.0 python-dev python-numpy python-matplotlib python-scipy \ - python-jinja2 git-core wget jblas mono-devel mono-gmcs cli-common-dev \ + python-jinja2 python-setuptools git-core wget jblas mono-devel mono-dmcs cli-common-dev \ lua5.1 liblua5.1-0-dev octave liboctave-dev r-base-core clang \ - oracle-java8-installer ruby ruby-dev python-ply sphinx-doc python-pip \ - exuberant-ctags - -RUN pip install sphinx ply sphinxcontrib-bibtex sphinx_bootstrap_theme + openjdk-8-jdk ruby ruby-dev python-ply sphinx-doc python-pip \ + exuberant-ctags clang-format-3.8 libcereal-dev libcolpack-dev lcov \ + protobuf-compiler libprotobuf-dev scala googletest +RUN pip install sphinx ply sphinxcontrib-bibtex sphinx_bootstrap_theme codecov RUN gem install narray RUN cd /usr/bin && ln -s swig3.0 swig && ln -s ccache-swig3.0 ccache-swig -ADD http://crd.lbl.gov/~dhbailey/mpdist/arprec-2.2.18.tar.gz /tmp/ -RUN cd /tmp;\ - tar zxpf arprec-2.2.18.tar.gz;\ - cd arprec-2.2.18;\ - ./configure --enable-shared;\ - make install; ldconfig -ADD http://www.scala-lang.org/files/archive/scala-2.11.7.deb /tmp/scala.deb -RUN sudo dpkg -i /tmp/scala.deb +ADD http://crd.lbl.gov/~dhbailey/mpdist/arprec-2.2.19.tar.gz /tmp/ +RUN cd /tmp && \ + tar zxpf arprec-2.2.19.tar.gz && \ + cd arprec && ./configure --enable-shared && \ + make install && ldconfig ADD http://dl.bintray.com/sbt/debian/sbt-0.13.6.deb /tmp/sbt.deb -RUN sudo dpkg -i /tmp/sbt.deb +RUN dpkg -i /tmp/sbt.deb + +ADD https://github.com/Reactive-Extensions/RxCpp/archive/v4.0.0.tar.gz /tmp/ +RUN cd /tmp;\ + tar -xvf v4.0.0.tar.gz;\ + cd RxCpp-4.0.0/projects/;\ + mkdir build;\ + cd build;\ + cmake ../../;\ + make install; + +ADD https://github.com/shogun-toolbox/tflogger/archive/v0.1.0.tar.gz /tmp/ +RUN cd /tmp;\ + tar -xvf v0.1.0.tar.gz;\ + cd tflogger-0.1.0;\ + mkdir build;\ + cd build;\ + cmake ../;\ + make install; diff --git a/configs/shogun-sdk/travis.env.in b/configs/shogun-sdk/travis.env.in new file mode 100644 index 00000000000..39f20153848 --- /dev/null +++ b/configs/shogun-sdk/travis.env.in @@ -0,0 +1,15 @@ +TRAVIS=$TRAVIS +CI=$CI +TRAVIS_BRANCH=$TRAVIS_BRANCH +TRAVIS_COMMIT=$TRAVIS_COMMIT +TRAVIS_JOB_NUMBER=$TRAVIS_JOB_NUMBER +TRAVIS_PULL_REQUEST=$TRAVIS_PULL_REQUEST +TRAVIS_JOB_ID=$TRAVIS_JOB_ID +TRAVIS_REPO_SLUG=$TRAVIS_REPO_SLUG +TRAVIS_TAG=$TRAVIS_TAG +TRAVIS_OS_NAME=$TRAVIS_OS_NAME +TRAVIS_BUILD_DIR=/opt/shogun +TRAVIS_PULL_REQUEST_BRANCH=$TRAVIS_PULL_REQUEST_BRANCH +JAVA_HOME=/usr/lib/jvm/java-8-oracle +CC=$CC +CXX=$CXX diff --git a/configs/shogun/Dockerfile b/configs/shogun/Dockerfile index 66ab9fcd07d..ee0281c04cb 100644 --- a/configs/shogun/Dockerfile +++ b/configs/shogun/Dockerfile @@ -1,4 +1,4 @@ -FROM ubuntu:14.04 +FROM ubuntu:16.04 MAINTAINER shogun@shogun-toolbox.org RUN apt-get update && apt-get install -qq software-properties-common lsb-release @@ -8,4 +8,4 @@ RUN apt-get update -qq RUN apt-get upgrade -y # install shogun -RUN apt-get install -qq --force-yes --no-install-recommends libshogun17 +RUN apt-get install -qq --force-yes --no-install-recommends libshogun18 diff --git a/data b/data index 7854e3b08ab..4b85bd7faa0 160000 --- a/data +++ b/data @@ -1 +1 @@ -Subproject commit 7854e3b08ab99594dcba4c6e94d7cd48eb2d0556 +Subproject commit 4b85bd7faa0b2bbcbbc4458655e6f35f438f1e91 diff --git a/debian b/debian index add2d5b2eeb..e47abf2ac2c 160000 --- a/debian +++ b/debian @@ -1 +1 @@ -Subproject commit add2d5b2eebb7631023cb26a3ff5614c396befbe +Subproject commit e47abf2ac2c8741abf1352804d4726d471789717 diff --git a/doc/cookbook/source/examples/converter/ica_fast.rst b/doc/cookbook/source/examples/converter/ica_fast.rst new file mode 100644 index 00000000000..3683c5e4043 --- /dev/null +++ b/doc/cookbook/source/examples/converter/ica_fast.rst @@ -0,0 +1,38 @@ +============================== +Independent Component Analysis +============================== + +Independent component analysis (ICA) separates a multivariate signal into additive subcomponents that are maximally independent. +It is typically used for separating superimposed signals. + +The ICA algorithm presented here is fastICA, see :cite:`hyvarinen2000independent` for details. +There are many other ICA implementations, all based on :sgclass:`CICAConverter` + +------- +Example +------- + +Given a dataset which we assume consists of linearly mixed signals, we create CDenseFeatures +(RealFeatures, here 64 bit float values). + +.. sgexample:: ica_fast.sg:create_features + +We create the :sgclass:`CFastICA` instance, and set its parameter for the iterative optimization. + +.. sgexample:: ica_fast.sg:set_parameters + +Then we apply ICA, which gives the unmixed signals. + +.. sgexample:: ica_fast.sg:apply_convert + +We can also extract the estimated mixing matrix. + +.. sgexample:: ica_fast.sg:extract + +---------- +References +---------- +:wiki:`Independent_component_analysis` + +.. bibliography:: ../../references.bib + :filter: docname in docnames diff --git a/doc/cookbook/source/examples/evaluation/cross_validation_mkl_weights_storage.rst b/doc/cookbook/source/examples/evaluation/cross_validation_mkl_weights_storage.rst new file mode 100644 index 00000000000..7ca656b437a --- /dev/null +++ b/doc/cookbook/source/examples/evaluation/cross_validation_mkl_weights_storage.rst @@ -0,0 +1,53 @@ +============================================ +Cross Validation on Multiple Kernel Learning +============================================ + +Cross Validation is a model validation technique whose purpose is to give an insight +on how the model we are testing will generalize to an independent dataset. Essentially, +it is based on training and test the model many times on different complementary partitions of the original +training dataset and then to combine the validation results (e.g. average) to estimate a +the performance of the final predictive model. + +------- +Example +------- +We'll use as example a classification problem solvable by using :sgclass:`CMKLClassification`. +For the sake of brevity, we'll skip the initialization of features, kernels and so on +(see :doc:`../regression/multiple_kernel_learning` for a more complete example of MKL usage). + +.. sgexample:: cross_validation_mkl_weight_storage.sg:create_classifier + +Firstly, we initialize a splitting strategy :sgclass:`CStratifiedCrossValidationSplitting`, which is needed +to divide the dataset into folds, and an evaluation criterium :sgclass:`CAccuracyMeasure`, to evaluate the +performance of the trained models. Secondly, we create the :sgclass:`CCrossValidation` instance. +We set also the number of cross validation's runs. + +.. sgexample:: cross_validation_mkl_weight_storage.sg:create_cross_validation + +To observe also the partial folds' results, we create a cross validation's observer :sgclass:`CParameterObserverCV` +and then we register it into the :sgclass:`CCrossValidation` instance. + +.. sgexample:: cross_validation_mkl_weight_storage.sg:create_observer + +Finally, we evaluate the model and get the results (aka a :sgclass:`CCrossValidationResult` instance). + +.. sgexample:: cross_validation_mkl_weight_storage.sg:evaluate_and_get_result + +We get the :math:`mean` of all the evaluation results and its standard deviation :math:`stddev`. + +.. sgexample:: cross_validation_mkl_weight_storage.sg:get_results + +We can get more information about the single cross validation's runs and folds by using the observer we set before, like the kernels' weights. +We get the :sgclass:`CMKLClassification` machine used during the first run and trained on the first fold. + +.. sgexample:: cross_validation_mkl_weight_storage.sg:get_fold_machine + +Then, from the trained machine, we get the weights :math:`\mathbf{w}` of its kernels. + +.. sgexample:: cross_validation_mkl_weight_storage.sg:get_weights + +---------- +References +---------- + +:wiki:`Cross-validation_(statistics)` \ No newline at end of file diff --git a/doc/cookbook/source/index.rst b/doc/cookbook/source/index.rst index 1b979cb0399..4225acf2ec4 100644 --- a/doc/cookbook/source/index.rst +++ b/doc/cookbook/source/index.rst @@ -56,15 +56,15 @@ Statistical Testing examples/statistical_testing/** -Kernels -------- +Converter +--------- .. toctree:: :maxdepth: 1 :glob: - :caption: Kernels + :caption: Converter - examples/kernel/** + examples/converter/** Gaussian Processes ------------------ @@ -105,3 +105,13 @@ Neural Networks :caption: Neural Netwroks examples/neural_nets/** + +Evaluation +---------- + +.. toctree:: + :maxdepth: 1 + :glob: + :caption: Evaluation + + examples/evaluation/** \ No newline at end of file diff --git a/doc/cookbook/source/references.bib b/doc/cookbook/source/references.bib index b32bec37852..073d5a70290 100644 --- a/doc/cookbook/source/references.bib +++ b/doc/cookbook/source/references.bib @@ -153,3 +153,13 @@ @article{Quinonero-Candela2005 year = {2005}, pages = {1939--1959}, } + +@article{hyvarinen2000independent, + title={Independent component analysis: algorithms and applications}, + author={Hyv{\"a}rinen, A. and Oja, E.}, + journal={Neural networks}, + volume={13}, + number={4}, + pages={411--430}, + year={2000} +} diff --git a/doc/cookbook/source/static/shogun-style.css b/doc/cookbook/source/static/shogun-style.css index 194363f9ff3..2329d699def 100644 --- a/doc/cookbook/source/static/shogun-style.css +++ b/doc/cookbook/source/static/shogun-style.css @@ -1,5 +1,5 @@ .bs-sidenav { - background-color: #f5f5f5 !important; + background-color: #fff !important; padding-top: 0; margin-top: 20pt; } @@ -13,9 +13,11 @@ } #sidebar > ul > p.caption { - background-color: #cecece; padding: 10px; - border-radius: 5px; + text-shadow: none !important; + font-weight: 300; + font-size: 1.5em; + padding-left: 0px; } #sidebar li.toctree-l1:last-child { @@ -24,4 +26,34 @@ div.toctree-wrapper > p.caption { display: none; -} \ No newline at end of file +} + +#sidebar li.toctree-l1 > a { + padding-left: 0px; + font-size: 100%; +} + +#sidebar li.toctree-l1 { + text-shadow: none !important; + border: none; +} + +#sidebar li.toctree-l1 .current { + color: #000; +} + +#sidebar li.toctree-l1 :hover { + background-color: transparent; + border: none; + color: #000; +} + +.section > h1 { + font-weight: 300; + margin-bottom: 0.5em; +} + +.section > h2 { + font-weight: 300; + margin-bottom: 0.5em; +} diff --git a/doc/cookbook/source/templates/layout.html b/doc/cookbook/source/templates/layout.html index 94df908c4ab..1c16d0a1683 100644 --- a/doc/cookbook/source/templates/layout.html +++ b/doc/cookbook/source/templates/layout.html @@ -2,6 +2,11 @@ {% set script_files = script_files + ["_static/mathconf.js"] %} {% set bootswatch_css_custom = ['_static/shogun-style.css'] %} +{# + bootswatch_css_custom is removed in sphinx-bootstrap-theme v0.5.0 + css_files should be used instead +#} +{% set css_files = (css_files or []) + ['_static/shogun-style.css'] %} {% block rootrellink %}
  • Shogun »
    • diff --git a/doc/ipython-notebooks/classification/Classification.ipynb b/doc/ipython-notebooks/classification/Classification.ipynb index 8cf310c7280..e9364f1b4a5 100644 --- a/doc/ipython-notebooks/classification/Classification.ipynb +++ b/doc/ipython-notebooks/classification/Classification.ipynb @@ -48,7 +48,7 @@ "import numpy as np\n", "import matplotlib.pyplot as plt\n", "import os\nSHOGUN_DATA_DIR=os.getenv('SHOGUN_DATA_DIR', '../../../data')\n", - "from modshogun import *" + "from shogun import *" ] }, { diff --git a/doc/ipython-notebooks/classification/HashedDocDotFeatures.ipynb b/doc/ipython-notebooks/classification/HashedDocDotFeatures.ipynb index a413c70e7d1..34395685bbb 100644 --- a/doc/ipython-notebooks/classification/HashedDocDotFeatures.ipynb +++ b/doc/ipython-notebooks/classification/HashedDocDotFeatures.ipynb @@ -77,7 +77,7 @@ "The response to that is to read our collection as it is and compute the hash of every token only when it's required, on-the-fly.
    \n", "

    On-the-fly Hashing with Shogun

    \n", "We will now have a look at how the above idea is represented in the Shogun Toolbox. That is we will see how we can load our document collection in memory and consider a hashed document-term matrix with the hashing of every document (or token more specifically) happening on-the-fly, only when it's required to be computed. Altough it may sound a bit tricky, it's actually pretty straightforward and here is how.

    \n", - "First of all we import the required components from the modshogun library." + "First of all we import the required components from the shogun library." ] }, { @@ -90,7 +90,7 @@ "source": [ "%matplotlib inline\n", "import os\nSHOGUN_DATA_DIR=os.getenv('SHOGUN_DATA_DIR', '../../../data')\n", - "from modshogun import StringCharFeatures, RAWBYTE, HashedDocDotFeatures, NGramTokenizer" + "from shogun import StringCharFeatures, RAWBYTE, HashedDocDotFeatures, NGramTokenizer" ] }, { @@ -133,7 +133,7 @@ }, "outputs": [], "source": [ - "from modshogun import BinaryLabels\n", + "from shogun import BinaryLabels\n", "from numpy import array\n", "\n", "labels = BinaryLabels(array([-1, 1, 1]))" @@ -199,7 +199,7 @@ }, "outputs": [], "source": [ - "from modshogun import SVMOcas\n", + "from shogun import SVMOcas\n", "\n", "C = 0.1\n", "epsilon = 0.01\n", diff --git a/doc/ipython-notebooks/classification/MKL.ipynb b/doc/ipython-notebooks/classification/MKL.ipynb index f429c88fec4..d7d8bf5bffa 100644 --- a/doc/ipython-notebooks/classification/MKL.ipynb +++ b/doc/ipython-notebooks/classification/MKL.ipynb @@ -48,7 +48,7 @@ "%matplotlib inline\n", "import os\nSHOGUN_DATA_DIR=os.getenv('SHOGUN_DATA_DIR', '../../../data')\n", "# import all shogun classes\n", - "from modshogun import *" + "from shogun import *" ] }, { diff --git a/doc/ipython-notebooks/classification/SupportVectorMachines.ipynb b/doc/ipython-notebooks/classification/SupportVectorMachines.ipynb index 702a013bc53..a5b5bb51a2c 100644 --- a/doc/ipython-notebooks/classification/SupportVectorMachines.ipynb +++ b/doc/ipython-notebooks/classification/SupportVectorMachines.ipynb @@ -119,7 +119,7 @@ "import os\nSHOGUN_DATA_DIR=os.getenv('SHOGUN_DATA_DIR', '../../../data')\n", "import matplotlib.patches as patches\n", "#To import all shogun classes\n", - "import modshogun as sg\n", + "import shogun as sg\n", "import numpy as np\n", "\n", "#Generate some random data\n", diff --git a/doc/ipython-notebooks/clustering/GMM.ipynb b/doc/ipython-notebooks/clustering/GMM.ipynb index 884399a3b77..28b9f7cd373 100644 --- a/doc/ipython-notebooks/clustering/GMM.ipynb +++ b/doc/ipython-notebooks/clustering/GMM.ipynb @@ -114,7 +114,7 @@ "%matplotlib inline\n", "import os\nSHOGUN_DATA_DIR=os.getenv('SHOGUN_DATA_DIR', '../../../data')\n", "# import all Shogun classes\n", - "from modshogun import *" + "from shogun import *" ] }, { diff --git a/doc/ipython-notebooks/clustering/KMeans.ipynb b/doc/ipython-notebooks/clustering/KMeans.ipynb index eacb76706ff..83c070e0247 100644 --- a/doc/ipython-notebooks/clustering/KMeans.ipynb +++ b/doc/ipython-notebooks/clustering/KMeans.ipynb @@ -111,7 +111,7 @@ "cell_type": "code", "collapsed": false, "input": [ - "from modshogun import *\n", + "from shogun import *\n", "\n", "train_features = RealFeatures(rectangle)" ], diff --git a/doc/ipython-notebooks/computer_vision/Scene_classification.ipynb b/doc/ipython-notebooks/computer_vision/Scene_classification.ipynb index cb773bddc6d..cff09beb824 100644 --- a/doc/ipython-notebooks/computer_vision/Scene_classification.ipynb +++ b/doc/ipython-notebooks/computer_vision/Scene_classification.ipynb @@ -107,7 +107,7 @@ "import numpy as np\n", "import matplotlib.pyplot as plt\n", "%matplotlib inline\n", - "from modshogun import *\n", + "from shogun import *\n", "\n", "# get the list of all jpg images from the path provided\n", "import os\n", diff --git a/doc/ipython-notebooks/converter/Tapkee.ipynb b/doc/ipython-notebooks/converter/Tapkee.ipynb index ace962f0905..307c024eb54 100644 --- a/doc/ipython-notebooks/converter/Tapkee.ipynb +++ b/doc/ipython-notebooks/converter/Tapkee.ipynb @@ -130,7 +130,7 @@ "cell_type": "code", "collapsed": false, "input": [ - "from modshogun import Isomap, RealFeatures, MultidimensionalScaling\n", + "from shogun import Isomap, RealFeatures, MultidimensionalScaling\n", "\n", "# wrap data into Shogun features\n", "data, colors = generate_data('swissroll')\n", @@ -175,7 +175,7 @@ "cell_type": "code", "collapsed": false, "input": [ - "from modshogun import StochasticProximityEmbedding\n", + "from shogun import StochasticProximityEmbedding\n", "\n", "# wrap data into Shogun features\n", "data, colors = generate_data('helix')\n", diff --git a/doc/ipython-notebooks/distributions/KernelDensity.ipynb b/doc/ipython-notebooks/distributions/KernelDensity.ipynb index e28226916e3..f93c15563ab 100644 --- a/doc/ipython-notebooks/distributions/KernelDensity.ipynb +++ b/doc/ipython-notebooks/distributions/KernelDensity.ipynb @@ -123,7 +123,7 @@ "cell_type": "code", "collapsed": false, "input": [ - "from modshogun import KernelDensity, RealFeatures, K_GAUSSIAN, D_EUCLIDEAN, EM_KDTREE_SINGLE\n", + "from shogun import KernelDensity, RealFeatures, K_GAUSSIAN, D_EUCLIDEAN, EM_KDTREE_SINGLE\n", "\n", "def get_kde_result(bandwidth,samples):\n", " # set model parameters\n", @@ -352,7 +352,7 @@ "cell_type": "code", "collapsed": false, "input": [ - "from modshogun import KernelDensity, RealFeatures, K_GAUSSIAN, D_EUCLIDEAN, EM_BALLTREE_DUAL\n", + "from shogun import KernelDensity, RealFeatures, K_GAUSSIAN, D_EUCLIDEAN, EM_BALLTREE_DUAL\n", "import scipy.interpolate as interpolate\n", "\n", "def get_kde(samples):\n", diff --git a/doc/ipython-notebooks/evaluation/xval_modelselection.ipynb b/doc/ipython-notebooks/evaluation/xval_modelselection.ipynb index cb88db11067..9a786d8bdf5 100644 --- a/doc/ipython-notebooks/evaluation/xval_modelselection.ipynb +++ b/doc/ipython-notebooks/evaluation/xval_modelselection.ipynb @@ -69,7 +69,7 @@ "%matplotlib inline\n", "# include all Shogun classes\n", "import os\nSHOGUN_DATA_DIR=os.getenv('SHOGUN_DATA_DIR', '../../../data')\n", - "from modshogun import *\n", + "from shogun import *\n", "# generate some ultra easy training data\n", "gray()\n", "n=20\n", diff --git a/doc/ipython-notebooks/gaussian_process/gaussian_processes.ipynb b/doc/ipython-notebooks/gaussian_process/gaussian_processes.ipynb index a74e9d118fa..374b9a10093 100644 --- a/doc/ipython-notebooks/gaussian_process/gaussian_processes.ipynb +++ b/doc/ipython-notebooks/gaussian_process/gaussian_processes.ipynb @@ -31,7 +31,7 @@ "source": [ "%matplotlib inline\n", "# import all shogun classes\n", - "from modshogun import *\n", + "from shogun import *\n", "import random\n", "import numpy as np\n", "import matplotlib.pyplot as plt\n", diff --git a/doc/ipython-notebooks/gaussian_process/variational_classifier.ipynb b/doc/ipython-notebooks/gaussian_process/variational_classifier.ipynb index f8e5db47a6a..6031bf5ad52 100644 --- a/doc/ipython-notebooks/gaussian_process/variational_classifier.ipynb +++ b/doc/ipython-notebooks/gaussian_process/variational_classifier.ipynb @@ -90,7 +90,7 @@ "%matplotlib inline\n", "# import all shogun classes\n", "import os\nSHOGUN_DATA_DIR=os.getenv('SHOGUN_DATA_DIR', '../../../data')\n", - "from modshogun import *\n", + "from shogun import *\n", "\n", "# import all required libraries\n", "import scipy\n", diff --git a/doc/ipython-notebooks/ica/bss_image.ipynb b/doc/ipython-notebooks/ica/bss_image.ipynb index 7d92ceb1feb..a843d7816be 100644 --- a/doc/ipython-notebooks/ica/bss_image.ipynb +++ b/doc/ipython-notebooks/ica/bss_image.ipynb @@ -161,8 +161,8 @@ "cell_type": "code", "collapsed": false, "input": [ - "from modshogun import RealFeatures\n", - "from modshogun import Jade\n", + "from shogun import RealFeatures\n", + "from shogun import Jade\n", "\n", "mixed_signals = RealFeatures(X)\n", "\n", diff --git a/doc/ipython-notebooks/intro/Introduction.ipynb b/doc/ipython-notebooks/intro/Introduction.ipynb index f0a8e41fecf..f939e1b52ef 100644 --- a/doc/ipython-notebooks/intro/Introduction.ipynb +++ b/doc/ipython-notebooks/intro/Introduction.ipynb @@ -61,7 +61,7 @@ "%matplotlib inline\n", "import os\nSHOGUN_DATA_DIR=os.getenv('SHOGUN_DATA_DIR', '../../../data')\n", "#To import all Shogun classes\n", - "from modshogun import *" + "from shogun import *" ] }, { diff --git a/doc/ipython-notebooks/logdet/logdet.ipynb b/doc/ipython-notebooks/logdet/logdet.ipynb index 48d502246f7..c9e1e4d9391 100644 --- a/doc/ipython-notebooks/logdet/logdet.ipynb +++ b/doc/ipython-notebooks/logdet/logdet.ipynb @@ -93,7 +93,7 @@ }, "outputs": [], "source": [ - "from modshogun import RealSparseMatrixOperator, LanczosEigenSolver\n", + "from shogun import RealSparseMatrixOperator, LanczosEigenSolver\n", "\n", "op = RealSparseMatrixOperator(A.tocsc())\n", "\n", @@ -125,7 +125,7 @@ "source": [ "# We can specify the power of the sparse-matrix that is to be used for coloring, default values will apply a\n", "# 2-distance greedy graph coloring algorithm on the sparse-matrix itself. Matrix-power, if specified, is computed in O(lg p)\n", - "from modshogun import ProbingSampler\n", + "from shogun import ProbingSampler\n", "\n", "trace_sampler = ProbingSampler(op)\n", "# apply the graph coloring algorithm and generate the number of colors, i.e. number of trace samples\n", @@ -150,7 +150,7 @@ }, "outputs": [], "source": [ - "from modshogun import SerialComputationEngine, CGMShiftedFamilySolver, LogRationalApproximationCGM\n", + "from shogun import SerialComputationEngine, CGMShiftedFamilySolver, LogRationalApproximationCGM\n", "\n", "engine = SerialComputationEngine()\n", "cgm = CGMShiftedFamilySolver()\n", @@ -182,7 +182,7 @@ "outputs": [], "source": [ "import numpy as np\n", - "from modshogun import LogDetEstimator\n", + "from shogun import LogDetEstimator\n", "\n", "# number of log-det samples (use a higher number to get better estimates)\n", "# (this is 5 times number of colors estimate in practice, so usually 1 probing estimate is enough)\n", @@ -213,7 +213,7 @@ "# the following method requires massive amount of memory, for demonstration purpose\n", "# the following code is commented out and direct value obtained from running it once is used\n", "\n", - "# from modshogun import Statistics\n", + "# from shogun import Statistics\n", "# actual_logdet = Statistics.log_det(A)\n", "\n", "actual_logdet = 7120357.73878\n", @@ -282,7 +282,7 @@ "probing_estimates = log_det_estimator.sample(num_probing_estimates)\n", "\n", "# computing log-det estimates using Gaussian sampler\n", - "from modshogun import NormalSampler, Statistics\n", + "from shogun import NormalSampler, Statistics\n", "\n", "num_colors = probing_sampler.get_num_samples()\n", "normal_sampler = NormalSampler(op.get_dimension())\n", @@ -447,7 +447,7 @@ }, "outputs": [], "source": [ - "from modshogun import RealSparseMatrixOperator, ComplexDenseMatrixOperator\n", + "from shogun import RealSparseMatrixOperator, ComplexDenseMatrixOperator\n", "\n", "dim = 5\n", "np.random.seed(10)\n", @@ -487,7 +487,7 @@ "source": [ "from scipy.sparse import csc_matrix\n", "from scipy.sparse import identity\n", - "from modshogun import ConjugateGradientSolver\n", + "from shogun import ConjugateGradientSolver\n", "\n", "# creating a random spd matrix\n", "dim = 5\n", @@ -532,8 +532,8 @@ }, "outputs": [], "source": [ - "from modshogun import ComplexSparseMatrixOperator\n", - "from modshogun import ConjugateOrthogonalCGSolver\n", + "from shogun import ComplexSparseMatrixOperator\n", + "from shogun import ConjugateOrthogonalCGSolver\n", "\n", "# creating a random spd matrix\n", "dim = 5\n", @@ -580,7 +580,7 @@ }, "outputs": [], "source": [ - "from modshogun import CGMShiftedFamilySolver\n", + "from shogun import CGMShiftedFamilySolver\n", "\n", "cgm = CGMShiftedFamilySolver()\n", "\n", @@ -652,7 +652,7 @@ }, "outputs": [], "source": [ - "from modshogun import DirectSparseLinearSolver\n", + "from shogun import DirectSparseLinearSolver\n", "\n", "# creating a random spd matrix\n", "dim = 5\n", @@ -691,7 +691,7 @@ }, "outputs": [], "source": [ - "from modshogun import DirectLinearSolverComplex\n", + "from shogun import DirectLinearSolverComplex\n", "\n", "# creating a random spd matrix\n", "dim = 5\n", diff --git a/doc/ipython-notebooks/metric/LMNN.ipynb b/doc/ipython-notebooks/metric/LMNN.ipynb index 3ee07c04ba6..112ea1abc93 100644 --- a/doc/ipython-notebooks/metric/LMNN.ipynb +++ b/doc/ipython-notebooks/metric/LMNN.ipynb @@ -177,7 +177,7 @@ "cell_type": "code", "collapsed": false, "input": [ - "from modshogun import RealFeatures, MulticlassLabels\n", + "from shogun import RealFeatures, MulticlassLabels\n", "\n", "features = RealFeatures(x.T)\n", "labels = MulticlassLabels(y.astype(numpy.float64))" @@ -197,7 +197,7 @@ "cell_type": "code", "collapsed": false, "input": [ - "from modshogun import LMNN\n", + "from shogun import LMNN\n", "\n", "# number of target neighbours per example\n", "k = 1\n", @@ -384,7 +384,7 @@ "cell_type": "code", "collapsed": false, "input": [ - "from modshogun import KNN, EuclideanDistance, LMNN, RealFeatures, MulticlassLabels\n", + "from shogun import KNN, EuclideanDistance, LMNN, RealFeatures, MulticlassLabels\n", "\n", "def plot_neighborhood_graph(x, nn, axis=pyplot, cols=['r', 'b', 'g', 'm', 'k', 'y']):\n", "\tfor i in xrange(x.shape[0]):\n", @@ -480,7 +480,7 @@ "cell_type": "code", "collapsed": false, "input": [ - "from modshogun import CSVFile, RealFeatures, MulticlassLabels\n", + "from shogun import CSVFile, RealFeatures, MulticlassLabels\n", "\n", "ape_features = RealFeatures(CSVFile(os.path.join(SHOGUN_DATA_DIR, 'multiclass/fm_ape_gut.dat')))\n", "ape_labels = MulticlassLabels(CSVFile(os.path.join(SHOGUN_DATA_DIR, 'multiclass/label_ape_gut.dat')))" @@ -518,7 +518,7 @@ "collapsed": false, "input": [ "def visualize_tdsne(features, labels):\n", - " from modshogun import TDistributedStochasticNeighborEmbedding\n", + " from shogun import TDistributedStochasticNeighborEmbedding\n", " \n", " converter = TDistributedStochasticNeighborEmbedding()\n", " converter.set_target_dim(2)\n", @@ -561,9 +561,9 @@ "cell_type": "code", "collapsed": false, "input": [ - "from modshogun import KNN, EuclideanDistance\n", - "from modshogun import StratifiedCrossValidationSplitting, CrossValidation\n", - "from modshogun import CrossValidationResult, MulticlassAccuracy\n", + "from shogun import KNN, EuclideanDistance\n", + "from shogun import StratifiedCrossValidationSplitting, CrossValidation\n", + "from shogun import CrossValidationResult, MulticlassAccuracy\n", "\n", "# set up the classifier\n", "knn = KNN()\n", @@ -622,7 +622,7 @@ "cell_type": "code", "collapsed": false, "input": [ - "from modshogun import LMNN\n", + "from shogun import LMNN\n", "import numpy\n", "\n", "# to make training faster, use a portion of the features\n", @@ -712,7 +712,7 @@ "cell_type": "code", "collapsed": false, "input": [ - "from modshogun import CSVFile, RealFeatures, MulticlassLabels\n", + "from shogun import CSVFile, RealFeatures, MulticlassLabels\n", "\n", "wine_features = RealFeatures(CSVFile(os.path.join(SHOGUN_DATA_DIR, 'uci/wine/fm_wine.dat')))\n", "wine_labels = MulticlassLabels(CSVFile(os.path.join(SHOGUN_DATA_DIR, 'uci/wine/label_wine.dat')))\n", @@ -736,9 +736,9 @@ "cell_type": "code", "collapsed": false, "input": [ - "from modshogun import KNN, EuclideanDistance\n", - "from modshogun import StratifiedCrossValidationSplitting, CrossValidation\n", - "from modshogun import CrossValidationResult, MulticlassAccuracy\n", + "from shogun import KNN, EuclideanDistance\n", + "from shogun import StratifiedCrossValidationSplitting, CrossValidation\n", + "from shogun import CrossValidationResult, MulticlassAccuracy\n", "import numpy\n", "\n", "# kNN classifier\n", @@ -775,7 +775,7 @@ "cell_type": "code", "collapsed": false, "input": [ - "from modshogun import LMNN\n", + "from shogun import LMNN\n", "\n", "# train LMNN\n", "lmnn = LMNN(wine_features, wine_labels, k)\n", @@ -838,7 +838,7 @@ "cell_type": "code", "collapsed": false, "input": [ - "from modshogun import RescaleFeatures\n", + "from shogun import RescaleFeatures\n", "\n", "# preprocess features so that all of them vary within [0,1]\n", "preprocessor = RescaleFeatures()\n", diff --git a/doc/ipython-notebooks/multiclass/KNN.ipynb b/doc/ipython-notebooks/multiclass/KNN.ipynb index c5de58b52ab..68b2b174973 100644 --- a/doc/ipython-notebooks/multiclass/KNN.ipynb +++ b/doc/ipython-notebooks/multiclass/KNN.ipynb @@ -136,8 +136,8 @@ }, "outputs": [], "source": [ - "from modshogun import MulticlassLabels, RealFeatures\n", - "from modshogun import KNN, EuclideanDistance\n", + "from shogun import MulticlassLabels, RealFeatures\n", + "from shogun import KNN, EuclideanDistance\n", "\n", "labels = MulticlassLabels(Ytrain)\n", "feats = RealFeatures(Xtrain)\n", @@ -151,7 +151,7 @@ "print \"Predictions\", pred[:5]\n", "print \"Ground Truth\", Ytest[:5]\n", "\n", - "from modshogun import MulticlassAccuracy\n", + "from shogun import MulticlassAccuracy\n", "evaluator = MulticlassAccuracy()\n", "accuracy = evaluator.evaluate(pred, labels_test)\n", "\n", @@ -249,7 +249,7 @@ }, "outputs": [], "source": [ - "from modshogun import Time, KNN_COVER_TREE, KNN_BRUTE\n", + "from shogun import Time, KNN_COVER_TREE, KNN_BRUTE\n", "start = Time.get_curtime()\n", "knn.set_k(3)\n", "knn.set_knn_solver_type(KNN_BRUTE)\n", @@ -280,7 +280,7 @@ "outputs": [], "source": [ "def evaluate(labels, feats, use_cover_tree=False):\n", - " from modshogun import MulticlassAccuracy, CrossValidationSplitting\n", + " from shogun import MulticlassAccuracy, CrossValidationSplitting\n", " import time\n", " split = CrossValidationSplitting(labels, Nsplit)\n", " split.build_subsets()\n", @@ -422,7 +422,7 @@ }, "outputs": [], "source": [ - "from modshogun import GaussianKernel, GMNPSVM\n", + "from shogun import GaussianKernel, GMNPSVM\n", "\n", "width=80\n", "C=1\n", diff --git a/doc/ipython-notebooks/multiclass/Tree/DecisionTrees.ipynb b/doc/ipython-notebooks/multiclass/Tree/DecisionTrees.ipynb index 81ebc07ef7a..02f35db823f 100644 --- a/doc/ipython-notebooks/multiclass/Tree/DecisionTrees.ipynb +++ b/doc/ipython-notebooks/multiclass/Tree/DecisionTrees.ipynb @@ -199,7 +199,7 @@ }, "outputs": [], "source": [ - "from modshogun import ID3ClassifierTree, RealFeatures, MulticlassLabels\n", + "from shogun import ID3ClassifierTree, RealFeatures, MulticlassLabels\n", "from numpy import array, concatenate\n", "\n", "# encoding dictionary\n", @@ -561,7 +561,7 @@ }, "outputs": [], "source": [ - "from modshogun import MulticlassAccuracy\n", + "from shogun import MulticlassAccuracy\n", "\n", "# Shogun object for calculating multiclass accuracy\n", "accuracy = MulticlassAccuracy()\n", @@ -698,7 +698,7 @@ "source": [ "import matplotlib.pyplot as plt\n", "from numpy import ones, zeros, random, concatenate\n", - "from modshogun import RealFeatures, MulticlassLabels\n", + "from shogun import RealFeatures, MulticlassLabels\n", "% matplotlib inline\n", "\n", "def create_toy_classification_dataset(ncat,do_plot):\n", @@ -780,7 +780,7 @@ "outputs": [], "source": [ "from numpy import array\n", - "from modshogun import C45ClassifierTree\n", + "from shogun import C45ClassifierTree\n", "\n", "# steps in C4.5 Tree training bundled together in a python method\n", "def train_tree(feats,types,labels):\n", @@ -1028,7 +1028,7 @@ }, "outputs": [], "source": [ - "from modshogun import RealFeatures, MulticlassLabels\n", + "from shogun import RealFeatures, MulticlassLabels\n", "\n", "# training data\n", "feats_train = RealFeatures(feats_train)\n", @@ -1138,7 +1138,7 @@ }, "outputs": [], "source": [ - "from modshogun import MulticlassAccuracy\n", + "from shogun import MulticlassAccuracy\n", "\n", "# Shogun object for calculating multiclass accuracy\n", "accuracy = MulticlassAccuracy()\n", @@ -1258,7 +1258,7 @@ }, "outputs": [], "source": [ - "from modshogun import PT_MULTICLASS, CARTree\n", + "from shogun import PT_MULTICLASS, CARTree\n", "from numpy import array\n", "\n", "def train_carttree(feat_types,problem_type,num_folds,use_cv_pruning,labels,features):\n", @@ -1342,7 +1342,7 @@ }, "outputs": [], "source": [ - "from modshogun import RegressionLabels, RealFeatures\n", + "from shogun import RegressionLabels, RealFeatures\n", "from numpy import random, sin, linspace\n", "import matplotlib.pyplot as plt\n", "% matplotlib inline\n", @@ -1400,7 +1400,7 @@ }, "outputs": [], "source": [ - "from modshogun import PT_REGRESSION\n", + "from shogun import PT_REGRESSION\n", "from numpy import array\n", "\n", "# feature type - continuous\n", @@ -1539,7 +1539,7 @@ }, "outputs": [], "source": [ - "from modshogun import CARTree, PT_MULTICLASS\n", + "from shogun import CARTree, PT_MULTICLASS\n", "\n", "# set attribute types - all continuous\n", "feature_types = array([False, False, False, False])\n", @@ -1568,8 +1568,8 @@ }, "outputs": [], "source": [ - "from modshogun import RealFeatures, MulticlassLabels\n", - "from modshogun import CrossValidation, MulticlassAccuracy, CrossValidationSplitting, CrossValidationResult\n", + "from shogun import RealFeatures, MulticlassLabels\n", + "from shogun import CrossValidation, MulticlassAccuracy, CrossValidationSplitting, CrossValidationResult\n", "\n", "# training features\n", "feats_train = RealFeatures(feat)\n", @@ -1673,8 +1673,8 @@ }, "outputs": [], "source": [ - "from modshogun import CARTree, RegressionLabels, PT_REGRESSION, MeanSquaredError\n", - "from modshogun import CrossValidation, CrossValidationSplitting, CrossValidationResult\n", + "from shogun import CARTree, RegressionLabels, PT_REGRESSION, MeanSquaredError\n", + "from shogun import CrossValidation, CrossValidationSplitting, CrossValidationResult\n", "\n", "# form training features\n", "feats_train = RealFeatures(feat)\n", @@ -1818,7 +1818,7 @@ }, "outputs": [], "source": [ - "from modshogun import PT_MULTICLASS, CHAIDTree\n", + "from shogun import PT_MULTICLASS, CHAIDTree\n", "from numpy import array, dtype, int32\n", "\n", "def train_chaidtree(dependent_var_type,feature_types,num_bins,features,labels):\n", @@ -2002,7 +2002,7 @@ }, "outputs": [], "source": [ - "from modshogun import CSVFile, RealFeatures, MulticlassLabels\n", + "from shogun import CSVFile, RealFeatures, MulticlassLabels\n", "\n", "train_feats=RealFeatures(CSVFile( os.path.join(SHOGUN_DATA_DIR, 'uci/wine/fm_wine.dat')))\n", "train_labels=MulticlassLabels(CSVFile( os.path.join(SHOGUN_DATA_DIR, 'uci/wine/label_wine.dat')))" @@ -2028,7 +2028,7 @@ }, "outputs": [], "source": [ - "from modshogun import CHAIDTree, MulticlassLabels\n", + "from shogun import CHAIDTree, MulticlassLabels\n", "\n", "# set attribute types - all attributes are continuous(2)\n", "feature_types = array([2 for i in range(13)],dtype=int32) \n", @@ -2059,7 +2059,7 @@ "source": [ "# set up cross validation class\n", "\n", - "from modshogun import CrossValidation, CrossValidationSplitting, CrossValidationResult, MulticlassAccuracy\n", + "from shogun import CrossValidation, CrossValidationSplitting, CrossValidationResult, MulticlassAccuracy\n", "\n", "# set evaluation criteria - multiclass accuracy\n", "accuracy = MulticlassAccuracy()\n", @@ -2106,7 +2106,7 @@ }, "outputs": [], "source": [ - "from modshogun import CSVFile, RealFeatures, RegressionLabels\n", + "from shogun import CSVFile, RealFeatures, RegressionLabels\n", "from numpy import ptp\n", "\n", "train_feats=RealFeatures(CSVFile( os.path.join(SHOGUN_DATA_DIR, 'uci/housing/fm_housing.dat')))\n", @@ -2136,8 +2136,8 @@ }, "outputs": [], "source": [ - "from modshogun import CHAIDTree, MeanSquaredError\n", - "from modshogun import CrossValidation, CrossValidationSplitting, CrossValidationResult\n", + "from shogun import CHAIDTree, MeanSquaredError\n", + "from shogun import CrossValidation, CrossValidationSplitting, CrossValidationResult\n", "from numpy import array, dtype, int32\n", "\n", "def get_cv_error(max_depth):\n", diff --git a/doc/ipython-notebooks/multiclass/Tree/TreeEnsemble.ipynb b/doc/ipython-notebooks/multiclass/Tree/TreeEnsemble.ipynb index 5ee0e83915b..249e14b0ff1 100644 --- a/doc/ipython-notebooks/multiclass/Tree/TreeEnsemble.ipynb +++ b/doc/ipython-notebooks/multiclass/Tree/TreeEnsemble.ipynb @@ -66,7 +66,7 @@ "collapsed": false, "input": [ "import os\nSHOGUN_DATA_DIR=os.getenv('SHOGUN_DATA_DIR', '../../../../data')\n", - "from modshogun import CSVFile,RealFeatures,MulticlassLabels\n", + "from shogun import CSVFile,RealFeatures,MulticlassLabels\n", "\n", "def load_file(feat_file,label_file):\n", " feats=RealFeatures(CSVFile(feat_file))\n", @@ -92,7 +92,7 @@ "cell_type": "code", "collapsed": false, "input": [ - "from modshogun import RandomForest, MajorityVote\n", + "from shogun import RandomForest, MajorityVote\n", "from numpy import array\n", "\n", "def setup_random_forest(num_trees,rand_subset_size,combination_rule,feature_types):\n", @@ -149,7 +149,7 @@ "cell_type": "code", "collapsed": false, "input": [ - "from modshogun import CARTree, PT_MULTICLASS\n", + "from shogun import CARTree, PT_MULTICLASS\n", "\n", "def train_cart(train_feats,train_labels,feature_types,problem_type):\n", " c=CARTree(feature_types,problem_type,2,False)\n", @@ -180,7 +180,7 @@ "cell_type": "code", "collapsed": false, "input": [ - "from modshogun import MulticlassAccuracy\n", + "from shogun import MulticlassAccuracy\n", "\n", "accuracy=MulticlassAccuracy()\n", "\n", diff --git a/doc/ipython-notebooks/multiclass/multiclass_reduction.ipynb b/doc/ipython-notebooks/multiclass/multiclass_reduction.ipynb index 3f36e306e7f..22e3c21d651 100644 --- a/doc/ipython-notebooks/multiclass/multiclass_reduction.ipynb +++ b/doc/ipython-notebooks/multiclass/multiclass_reduction.ipynb @@ -185,10 +185,10 @@ "cell_type": "code", "collapsed": false, "input": [ - "from modshogun import RealFeatures, MulticlassLabels\n", - "from modshogun import LibLinear, L2R_L2LOSS_SVC, LinearMulticlassMachine\n", - "from modshogun import MulticlassOneVsOneStrategy, MulticlassOneVsRestStrategy\n", - "from modshogun import MulticlassAccuracy\n", + "from shogun import RealFeatures, MulticlassLabels\n", + "from shogun import LibLinear, L2R_L2LOSS_SVC, LinearMulticlassMachine\n", + "from shogun import MulticlassOneVsOneStrategy, MulticlassOneVsRestStrategy\n", + "from shogun import MulticlassAccuracy\n", "\n", "import time\n", "\n", @@ -273,7 +273,7 @@ "cell_type": "code", "collapsed": false, "input": [ - "from modshogun import MulticlassLibLinear\n", + "from shogun import MulticlassLibLinear\n", "mcsvm = MulticlassLibLinear(5.0, feats_train, lab_train)\n", "mcsvm.set_use_bias(True)\n", "\n", @@ -449,7 +449,7 @@ "cell_type": "code", "collapsed": false, "input": [ - "from modshogun import ECOCStrategy, ECOCRandomDenseEncoder, ECOCLLBDecoder\n", + "from shogun import ECOCStrategy, ECOCRandomDenseEncoder, ECOCLLBDecoder\n", "\n", "print \"\\nRandom Dense Encoder + Margin Loss based Decoder\"\n", "print \"=\"*60\n", @@ -482,7 +482,7 @@ "collapsed": false, "input": [ "def evaluate_multiclass_kernel(strategy):\n", - " from modshogun import KernelMulticlassMachine, LibSVM, GaussianKernel\n", + " from shogun import KernelMulticlassMachine, LibSVM, GaussianKernel\n", " width=2.1\n", " epsilon=1e-5\n", " \n", @@ -532,7 +532,7 @@ "cell_type": "code", "collapsed": false, "input": [ - "from modshogun import *\n", + "from shogun import *\n", "from numpy import *\n", "\n", "num=1000;\n", @@ -624,7 +624,7 @@ "cell_type": "code", "collapsed": false, "input": [ - "from modshogun import KernelMulticlassMachine, LibSVM, GaussianKernel\n", + "from shogun import KernelMulticlassMachine, LibSVM, GaussianKernel\n", "\n", "width=2.1\n", "epsilon=1e-5\n", diff --git a/doc/ipython-notebooks/multiclass/naive_bayes.ipynb b/doc/ipython-notebooks/multiclass/naive_bayes.ipynb index 9a833ef583f..661a4963ed1 100644 --- a/doc/ipython-notebooks/multiclass/naive_bayes.ipynb +++ b/doc/ipython-notebooks/multiclass/naive_bayes.ipynb @@ -140,9 +140,9 @@ "cell_type": "code", "collapsed": false, "input": [ - "from modshogun import GaussianNaiveBayes\n", - "from modshogun import RealFeatures\n", - "from modshogun import MulticlassLabels\n", + "from shogun import GaussianNaiveBayes\n", + "from shogun import RealFeatures\n", + "from shogun import MulticlassLabels\n", "\n", "X_train, Y_train = gen_samples(n_train)\n", "\n", diff --git a/doc/ipython-notebooks/neuralnets/autoencoders.ipynb b/doc/ipython-notebooks/neuralnets/autoencoders.ipynb index 64949fdab5d..8444fc9e03e 100644 --- a/doc/ipython-notebooks/neuralnets/autoencoders.ipynb +++ b/doc/ipython-notebooks/neuralnets/autoencoders.ipynb @@ -55,7 +55,7 @@ "%matplotlib inline\n", "import os\nSHOGUN_DATA_DIR=os.getenv('SHOGUN_DATA_DIR', '../../../data')\n", "from scipy.io import loadmat\n", - "from modshogun import RealFeatures, MulticlassLabels, Math\n", + "from shogun import RealFeatures, MulticlassLabels, Math\n", "\n", "# load the dataset\n", "dataset = loadmat(os.path.join(SHOGUN_DATA_DIR, 'multiclass/usps.mat'))\n", @@ -101,7 +101,7 @@ }, "outputs": [], "source": [ - "from modshogun import NeuralLayers, DeepAutoencoder\n", + "from shogun import NeuralLayers, DeepAutoencoder\n", "\n", "layers = NeuralLayers()\n", "layers = layers.input(256).rectified_linear(512).rectified_linear(128).rectified_linear(512).linear(256).done()\n", @@ -137,7 +137,7 @@ }, "outputs": [], "source": [ - "from modshogun import AENT_DROPOUT, NNOM_GRADIENT_DESCENT\n", + "from shogun import AENT_DROPOUT, NNOM_GRADIENT_DESCENT\n", "\n", "ae.pt_noise_type.set_const(AENT_DROPOUT) # use dropout noise\n", "ae.pt_noise_parameter.set_const(0.5) # each input has a 50% chance of being set to zero\n", @@ -150,7 +150,7 @@ "ae.pt_epsilon.set_const(0.0) # disable automatic convergence testing\n", "\n", "# uncomment this line to allow the training progress to be printed on the console\n", - "#from modshogun import MSG_INFO; ae.io.set_loglevel(MSG_INFO)\n", + "#from shogun import MSG_INFO; ae.io.set_loglevel(MSG_INFO)\n", "\n", "# start pre-training. this might take some time\n", "ae.pre_train(Xtrain)" @@ -287,7 +287,7 @@ }, "outputs": [], "source": [ - "from modshogun import NeuralSoftmaxLayer\n", + "from shogun import NeuralSoftmaxLayer\n", "\n", "nn = ae.convert_to_neural_network(NeuralSoftmaxLayer(10))\n", "\n", @@ -312,7 +312,7 @@ }, "outputs": [], "source": [ - "from modshogun import MulticlassAccuracy\n", + "from shogun import MulticlassAccuracy\n", "\n", "predictions = nn.apply_multiclass(Xtest)\n", "accuracy = MulticlassAccuracy().evaluate(predictions, Ytest) * 100\n", @@ -344,7 +344,7 @@ }, "outputs": [], "source": [ - "from modshogun import DynamicObjectArray, NeuralInputLayer, NeuralConvolutionalLayer, CMAF_RECTIFIED_LINEAR\n", + "from shogun import DynamicObjectArray, NeuralInputLayer, NeuralConvolutionalLayer, CMAF_RECTIFIED_LINEAR\n", "\n", "conv_layers = DynamicObjectArray()\n", "# 16x16 single channel images\n", diff --git a/doc/ipython-notebooks/neuralnets/neuralnets_digits.ipynb b/doc/ipython-notebooks/neuralnets/neuralnets_digits.ipynb index 98c0bc14061..02e056b18cd 100644 --- a/doc/ipython-notebooks/neuralnets/neuralnets_digits.ipynb +++ b/doc/ipython-notebooks/neuralnets/neuralnets_digits.ipynb @@ -51,7 +51,7 @@ "%matplotlib inline\n", "import os\nSHOGUN_DATA_DIR=os.getenv('SHOGUN_DATA_DIR', '../../../data')\n", "from scipy.io import loadmat\n", - "from modshogun import RealFeatures, MulticlassLabels, Math\n", + "from shogun import RealFeatures, MulticlassLabels, Math\n", "\n", "# load the dataset\n", "dataset = loadmat(os.path.join(SHOGUN_DATA_DIR, 'multiclass/usps.mat'))\n", @@ -109,8 +109,8 @@ }, "outputs": [], "source": [ - "from modshogun import NeuralNetwork, NeuralInputLayer, NeuralLogisticLayer, NeuralSoftmaxLayer\n", - "from modshogun import DynamicObjectArray\n", + "from shogun import NeuralNetwork, NeuralInputLayer, NeuralLogisticLayer, NeuralSoftmaxLayer\n", + "from shogun import DynamicObjectArray\n", "\n", "# setup the layers\n", "layers = DynamicObjectArray()\n", @@ -215,7 +215,7 @@ }, "outputs": [], "source": [ - "from modshogun import MulticlassAccuracy\n", + "from shogun import MulticlassAccuracy\n", "\n", "def compute_accuracy(net, X, Y):\n", " predictions = net.apply_multiclass(X)\n", @@ -246,7 +246,7 @@ "net_no_reg.set_max_num_epochs(600)\n", "\n", "# uncomment this line to allow the training progress to be printed on the console\n", - "#from modshogun import MSG_INFO; net_no_reg.io.set_loglevel(MSG_INFO)\n", + "#from shogun import MSG_INFO; net_no_reg.io.set_loglevel(MSG_INFO)\n", "\n", "net_no_reg.set_labels(Ytrain)\n", "net_no_reg.train(Xtrain) # this might take a while, depending on your machine\n", @@ -336,7 +336,7 @@ }, "outputs": [], "source": [ - "from modshogun import NNOM_GRADIENT_DESCENT\n", + "from shogun import NNOM_GRADIENT_DESCENT\n", "\n", "# set the dropout probabilty for neurons in the hidden layers\n", "net_dropout.set_dropout_hidden(0.5)\n", @@ -391,7 +391,7 @@ }, "outputs": [], "source": [ - "from modshogun import NeuralConvolutionalLayer, CMAF_RECTIFIED_LINEAR\n", + "from shogun import NeuralConvolutionalLayer, CMAF_RECTIFIED_LINEAR\n", "\n", "# prepere the layers\n", "layers_conv = DynamicObjectArray()\n", diff --git a/doc/ipython-notebooks/neuralnets/rbms_dbns.ipynb b/doc/ipython-notebooks/neuralnets/rbms_dbns.ipynb index 3d2e8c16589..0e7a05fe47b 100644 --- a/doc/ipython-notebooks/neuralnets/rbms_dbns.ipynb +++ b/doc/ipython-notebooks/neuralnets/rbms_dbns.ipynb @@ -174,7 +174,7 @@ "cell_type": "code", "collapsed": false, "input": [ - "from modshogun import RBM, RBMVUT_BINARY, Math\n", + "from shogun import RBM, RBMVUT_BINARY, Math\n", "\n", "# initialize the random number generator with a fixed seed, for repeatability\n", "Math.init_random(10)\n", @@ -223,10 +223,10 @@ "cell_type": "code", "collapsed": false, "input": [ - "from modshogun import RealFeatures, RBMMM_PSEUDO_LIKELIHOOD\n", + "from shogun import RealFeatures, RBMMM_PSEUDO_LIKELIHOOD\n", "\n", "# uncomment this line to allow the training progress to be printed on the console\n", - "#from modshogun import MSG_INFO; rbms[0].io.set_loglevel(MSG_INFO)\n", + "#from shogun import MSG_INFO; rbms[0].io.set_loglevel(MSG_INFO)\n", "\n", "for i in range(10):\n", " # obtain the data for digit i\n", @@ -305,7 +305,7 @@ "cell_type": "code", "collapsed": false, "input": [ - "from modshogun import DeepBeliefNetwork\n", + "from shogun import DeepBeliefNetwork\n", "\n", "dbn = DeepBeliefNetwork(256) # 256 visible units\n", "dbn.add_hidden_layer(200) # 200 units in the first hidden layer\n", @@ -386,7 +386,7 @@ "cell_type": "code", "collapsed": false, "input": [ - "from modshogun import NeuralSoftmaxLayer, MulticlassLabels\n", + "from shogun import NeuralSoftmaxLayer, MulticlassLabels\n", "\n", "# get the neural network\n", "nn = dbn.convert_to_neural_network(NeuralSoftmaxLayer(10))\n", @@ -413,7 +413,7 @@ "cell_type": "code", "collapsed": false, "input": [ - "from modshogun import MulticlassAccuracy\n", + "from shogun import MulticlassAccuracy\n", "\n", "predictions = nn.apply_multiclass(RealFeatures(Xtest))\n", "accuracy = MulticlassAccuracy().evaluate(predictions, MulticlassLabels(Ytest)) * 100\n", diff --git a/doc/ipython-notebooks/pca/pca_notebook.ipynb b/doc/ipython-notebooks/pca/pca_notebook.ipynb index 1e378b1b120..8d29e3f66a5 100644 --- a/doc/ipython-notebooks/pca/pca_notebook.ipynb +++ b/doc/ipython-notebooks/pca/pca_notebook.ipynb @@ -33,7 +33,7 @@ "%matplotlib inline\n", "import os\nSHOGUN_DATA_DIR=os.getenv('SHOGUN_DATA_DIR', '../../../data')\n", "# import all shogun classes\n", - "from modshogun import *" + "from shogun import *" ] }, { @@ -941,7 +941,7 @@ " imgplot.axes.get_xaxis().set_visible(False)\n", " imgplot.axes.get_yaxis().set_visible(False)\n", " \n", - "import Image\n", + "from PIL import Image\n", "from scipy import misc\n", "\n", "# to get a hang of the data, lets see some part of the dataset images.\n", diff --git a/doc/ipython-notebooks/regression/Regression.ipynb b/doc/ipython-notebooks/regression/Regression.ipynb index 8d69552897a..e7ebc689932 100644 --- a/doc/ipython-notebooks/regression/Regression.ipynb +++ b/doc/ipython-notebooks/regression/Regression.ipynb @@ -106,7 +106,7 @@ "import os\nSHOGUN_DATA_DIR=os.getenv('SHOGUN_DATA_DIR', '../../../data')\n", "from cycler import cycler\n", "# import all shogun classes\n", - "from modshogun import *\n", + "from shogun import *\n", "slope = 3\n", "\n", "X_train = rand(30)*10\n", diff --git a/doc/ipython-notebooks/statistical_testing/mmd_two_sample_testing.ipynb b/doc/ipython-notebooks/statistical_testing/mmd_two_sample_testing.ipynb index ebc165f1a19..2bedd3a02a9 100644 --- a/doc/ipython-notebooks/statistical_testing/mmd_two_sample_testing.ipynb +++ b/doc/ipython-notebooks/statistical_testing/mmd_two_sample_testing.ipynb @@ -41,7 +41,7 @@ "%pylab inline\n", "%matplotlib inline\n", "import os\nSHOGUN_DATA_DIR=os.getenv('SHOGUN_DATA_DIR', '../../../data')\n", - "import modshogun as sg\n", + "import shogun as sg\n", "import numpy as np" ] }, diff --git a/doc/ipython-notebooks/structure/Binary_Denoising.ipynb b/doc/ipython-notebooks/structure/Binary_Denoising.ipynb index 292b9245ba3..9b3d2039356 100644 --- a/doc/ipython-notebooks/structure/Binary_Denoising.ipynb +++ b/doc/ipython-notebooks/structure/Binary_Denoising.ipynb @@ -262,10 +262,10 @@ "cell_type": "code", "collapsed": false, "input": [ - "from modshogun import Factor, TableFactorType, FactorGraph\n", - "from modshogun import FactorGraphObservation, FactorGraphLabels, FactorGraphFeatures\n", - "from modshogun import FactorGraphModel, GRAPH_CUT, LP_RELAXATION\n", - "from modshogun import MAPInference" + "from shogun import Factor, TableFactorType, FactorGraph\n", + "from shogun import FactorGraphObservation, FactorGraphLabels, FactorGraphFeatures\n", + "from shogun import FactorGraphModel, GRAPH_CUT, LP_RELAXATION\n", + "from shogun import MAPInference" ], "language": "python", "metadata": {}, @@ -518,7 +518,7 @@ "cell_type": "code", "collapsed": false, "input": [ - "from modshogun import StochasticSOSVM\n", + "from shogun import StochasticSOSVM\n", "import time\n", "\n", "# Training with Stocastic Gradient Descent\n", diff --git a/doc/ipython-notebooks/structure/FGM.ipynb b/doc/ipython-notebooks/structure/FGM.ipynb index a89cae5d57f..8d724e7ba6e 100644 --- a/doc/ipython-notebooks/structure/FGM.ipynb +++ b/doc/ipython-notebooks/structure/FGM.ipynb @@ -248,7 +248,7 @@ }, "outputs": [], "source": [ - "from modshogun import TableFactorType\n", + "from shogun import TableFactorType\n", "\n", "# unary, type_id = 0\n", "cards_u = np.array([n_stats], np.int32)\n", @@ -294,8 +294,8 @@ "source": [ "def prepare_data(x, y, ftype, num_samples):\n", " \"\"\"prepare FactorGraphFeatures and FactorGraphLabels \"\"\"\n", - " from modshogun import Factor, TableFactorType, FactorGraph\n", - " from modshogun import FactorGraphObservation, FactorGraphLabels, FactorGraphFeatures\n", + " from shogun import Factor, TableFactorType, FactorGraph\n", + " from shogun import FactorGraphObservation, FactorGraphLabels, FactorGraphFeatures\n", "\n", " samples = FactorGraphFeatures(num_samples)\n", " labels = FactorGraphLabels(num_samples)\n", @@ -443,7 +443,7 @@ }, "outputs": [], "source": [ - "from modshogun import FactorGraphModel, TREE_MAX_PROD\n", + "from shogun import FactorGraphModel, TREE_MAX_PROD\n", "\n", "# create model and register factor types\n", "model = FactorGraphModel(samples, labels, TREE_MAX_PROD)\n", @@ -468,8 +468,8 @@ }, "outputs": [], "source": [ - "from modshogun import DualLibQPBMSOSVM\n", - "from modshogun import BmrmStatistics\n", + "from shogun import DualLibQPBMSOSVM\n", + "from shogun import BmrmStatistics\n", "import pickle\n", "import time\n", "\n", @@ -577,7 +577,7 @@ }, "outputs": [], "source": [ - "from modshogun import StochasticSOSVM\n", + "from shogun import StochasticSOSVM\n", "\n", "# the 3rd parameter is do_weighted_averaging, by turning this on, \n", "# a possibly faster convergence rate may be achieved.\n", @@ -760,7 +760,7 @@ }, "outputs": [], "source": [ - "from modshogun import FactorGraphFeatures, FactorGraphObservation, TREE_MAX_PROD, MAPInference\n", + "from shogun import FactorGraphFeatures, FactorGraphObservation, TREE_MAX_PROD, MAPInference\n", "\n", "# get a factor graph instance from test data\n", "fg0 = samples_ts.get_sample(100)\n", @@ -800,7 +800,7 @@ }, "outputs": [], "source": [ - "from modshogun import LabelsFactory, SOSVMHelper\n", + "from shogun import LabelsFactory, SOSVMHelper\n", "\n", "# training error of BMRM method\n", "bmrm.set_w(w_bmrm)\n", diff --git a/doc/ipython-notebooks/structure/multilabel_structured_prediction.ipynb b/doc/ipython-notebooks/structure/multilabel_structured_prediction.ipynb index 6e73a76082f..7eb42989a03 100644 --- a/doc/ipython-notebooks/structure/multilabel_structured_prediction.ipynb +++ b/doc/ipython-notebooks/structure/multilabel_structured_prediction.ipynb @@ -165,14 +165,14 @@ "cell_type": "code", "collapsed": false, "input": [ - "from modshogun import RealFeatures, MultilabelSOLabels, MultilabelModel\n", + "from shogun import RealFeatures, MultilabelSOLabels, MultilabelModel\n", "\n", "def create_features(X, constant):\n", " features = RealFeatures(\n", " np.c_[X, constant * np.ones(X.shape[0])].T)\n", " \n", " return features\n", - "from modshogun import MultilabelSOLabels\n", + "from shogun import MultilabelSOLabels\n", "\n", "def create_labels(Y, n_classes):\n", " try:\n", @@ -236,7 +236,7 @@ "cell_type": "code", "collapsed": false, "input": [ - "from modshogun import StochasticSOSVM, DualLibQPBMSOSVM, StructuredAccuracy, LabelsFactory\n", + "from shogun import StochasticSOSVM, DualLibQPBMSOSVM, StructuredAccuracy, LabelsFactory\n", "from time import time\n", "\n", "sgd = StochasticSOSVM(model, labels)\n", @@ -386,7 +386,7 @@ "cell_type": "code", "collapsed": false, "input": [ - "from modshogun import SparseMultilabel_obtain_from_generic\n", + "from shogun import SparseMultilabel_obtain_from_generic\n", "\n", "def plot_decision_plane(machine,\n", " title,\n", diff --git a/doc/ipython-notebooks/template.ipynb b/doc/ipython-notebooks/template.ipynb index 678beeca981..8a3c7846fbe 100644 --- a/doc/ipython-notebooks/template.ipynb +++ b/doc/ipython-notebooks/template.ipynb @@ -108,4 +108,4 @@ "metadata": {} } ] -} \ No newline at end of file +} diff --git a/doc/license/LICENSE.md b/doc/license/LICENSE.md index 94a9ed024d3..8a19f3af801 100644 --- a/doc/license/LICENSE.md +++ b/doc/license/LICENSE.md @@ -1,674 +1,28 @@ - GNU GENERAL PUBLIC LICENSE - Version 3, 29 June 2007 - - Copyright (C) 2007 Free Software Foundation, Inc. - Everyone is permitted to copy and distribute verbatim copies - of this license document, but changing it is not allowed. - - Preamble - - The GNU General Public License is a free, copyleft license for -software and other kinds of works. - - The licenses for most software and other practical works are designed -to take away your freedom to share and change the works. By contrast, -the GNU General Public License is intended to guarantee your freedom to -share and change all versions of a program--to make sure it remains free -software for all its users. We, the Free Software Foundation, use the -GNU General Public License for most of our software; it applies also to -any other work released this way by its authors. You can apply it to -your programs, too. - - When we speak of free software, we are referring to freedom, not -price. Our General Public Licenses are designed to make sure that you -have the freedom to distribute copies of free software (and charge for -them if you wish), that you receive source code or can get it if you -want it, that you can change the software or use pieces of it in new -free programs, and that you know you can do these things. - - To protect your rights, we need to prevent others from denying you -these rights or asking you to surrender the rights. Therefore, you have -certain responsibilities if you distribute copies of the software, or if -you modify it: responsibilities to respect the freedom of others. - - For example, if you distribute copies of such a program, whether -gratis or for a fee, you must pass on to the recipients the same -freedoms that you received. You must make sure that they, too, receive -or can get the source code. And you must show them these terms so they -know their rights. - - Developers that use the GNU GPL protect your rights with two steps: -(1) assert copyright on the software, and (2) offer you this License -giving you legal permission to copy, distribute and/or modify it. - - For the developers' and authors' protection, the GPL clearly explains -that there is no warranty for this free software. For both users' and -authors' sake, the GPL requires that modified versions be marked as -changed, so that their problems will not be attributed erroneously to -authors of previous versions. - - Some devices are designed to deny users access to install or run -modified versions of the software inside them, although the manufacturer -can do so. This is fundamentally incompatible with the aim of -protecting users' freedom to change the software. The systematic -pattern of such abuse occurs in the area of products for individuals to -use, which is precisely where it is most unacceptable. Therefore, we -have designed this version of the GPL to prohibit the practice for those -products. If such problems arise substantially in other domains, we -stand ready to extend this provision to those domains in future versions -of the GPL, as needed to protect the freedom of users. - - Finally, every program is threatened constantly by software patents. -States should not allow patents to restrict development and use of -software on general-purpose computers, but in those that do, we wish to -avoid the special danger that patents applied to a free program could -make it effectively proprietary. To prevent this, the GPL assures that -patents cannot be used to render the program non-free. - - The precise terms and conditions for copying, distribution and -modification follow. - - TERMS AND CONDITIONS - - 0. Definitions. - - "This License" refers to version 3 of the GNU General Public License. - - "Copyright" also means copyright-like laws that apply to other kinds of -works, such as semiconductor masks. - - "The Program" refers to any copyrightable work licensed under this -License. Each licensee is addressed as "you". "Licensees" and -"recipients" may be individuals or organizations. - - To "modify" a work means to copy from or adapt all or part of the work -in a fashion requiring copyright permission, other than the making of an -exact copy. The resulting work is called a "modified version" of the -earlier work or a work "based on" the earlier work. - - A "covered work" means either the unmodified Program or a work based -on the Program. - - To "propagate" a work means to do anything with it that, without -permission, would make you directly or secondarily liable for -infringement under applicable copyright law, except executing it on a -computer or modifying a private copy. Propagation includes copying, -distribution (with or without modification), making available to the -public, and in some countries other activities as well. - - To "convey" a work means any kind of propagation that enables other -parties to make or receive copies. Mere interaction with a user through -a computer network, with no transfer of a copy, is not conveying. - - An interactive user interface displays "Appropriate Legal Notices" -to the extent that it includes a convenient and prominently visible -feature that (1) displays an appropriate copyright notice, and (2) -tells the user that there is no warranty for the work (except to the -extent that warranties are provided), that licensees may convey the -work under this License, and how to view a copy of this License. If -the interface presents a list of user commands or options, such as a -menu, a prominent item in the list meets this criterion. - - 1. Source Code. - - The "source code" for a work means the preferred form of the work -for making modifications to it. "Object code" means any non-source -form of a work. - - A "Standard Interface" means an interface that either is an official -standard defined by a recognized standards body, or, in the case of -interfaces specified for a particular programming language, one that -is widely used among developers working in that language. - - The "System Libraries" of an executable work include anything, other -than the work as a whole, that (a) is included in the normal form of -packaging a Major Component, but which is not part of that Major -Component, and (b) serves only to enable use of the work with that -Major Component, or to implement a Standard Interface for which an -implementation is available to the public in source code form. A -"Major Component", in this context, means a major essential component -(kernel, window system, and so on) of the specific operating system -(if any) on which the executable work runs, or a compiler used to -produce the work, or an object code interpreter used to run it. - - The "Corresponding Source" for a work in object code form means all -the source code needed to generate, install, and (for an executable -work) run the object code and to modify the work, including scripts to -control those activities. However, it does not include the work's -System Libraries, or general-purpose tools or generally available free -programs which are used unmodified in performing those activities but -which are not part of the work. For example, Corresponding Source -includes interface definition files associated with source files for -the work, and the source code for shared libraries and dynamically -linked subprograms that the work is specifically designed to require, -such as by intimate data communication or control flow between those -subprograms and other parts of the work. - - The Corresponding Source need not include anything that users -can regenerate automatically from other parts of the Corresponding -Source. - - The Corresponding Source for a work in source code form is that -same work. - - 2. Basic Permissions. - - All rights granted under this License are granted for the term of -copyright on the Program, and are irrevocable provided the stated -conditions are met. This License explicitly affirms your unlimited -permission to run the unmodified Program. The output from running a -covered work is covered by this License only if the output, given its -content, constitutes a covered work. This License acknowledges your -rights of fair use or other equivalent, as provided by copyright law. - - You may make, run and propagate covered works that you do not -convey, without conditions so long as your license otherwise remains -in force. You may convey covered works to others for the sole purpose -of having them make modifications exclusively for you, or provide you -with facilities for running those works, provided that you comply with -the terms of this License in conveying all material for which you do -not control copyright. Those thus making or running the covered works -for you must do so exclusively on your behalf, under your direction -and control, on terms that prohibit them from making any copies of -your copyrighted material outside their relationship with you. - - Conveying under any other circumstances is permitted solely under -the conditions stated below. Sublicensing is not allowed; section 10 -makes it unnecessary. - - 3. Protecting Users' Legal Rights From Anti-Circumvention Law. - - No covered work shall be deemed part of an effective technological -measure under any applicable law fulfilling obligations under article -11 of the WIPO copyright treaty adopted on 20 December 1996, or -similar laws prohibiting or restricting circumvention of such -measures. - - When you convey a covered work, you waive any legal power to forbid -circumvention of technological measures to the extent such circumvention -is effected by exercising rights under this License with respect to -the covered work, and you disclaim any intention to limit operation or -modification of the work as a means of enforcing, against the work's -users, your or third parties' legal rights to forbid circumvention of -technological measures. - - 4. Conveying Verbatim Copies. - - You may convey verbatim copies of the Program's source code as you -receive it, in any medium, provided that you conspicuously and -appropriately publish on each copy an appropriate copyright notice; -keep intact all notices stating that this License and any -non-permissive terms added in accord with section 7 apply to the code; -keep intact all notices of the absence of any warranty; and give all -recipients a copy of this License along with the Program. - - You may charge any price or no price for each copy that you convey, -and you may offer support or warranty protection for a fee. - - 5. Conveying Modified Source Versions. - - You may convey a work based on the Program, or the modifications to -produce it from the Program, in the form of source code under the -terms of section 4, provided that you also meet all of these conditions: - - a) The work must carry prominent notices stating that you modified - it, and giving a relevant date. - - b) The work must carry prominent notices stating that it is - released under this License and any conditions added under section - 7. This requirement modifies the requirement in section 4 to - "keep intact all notices". - - c) You must license the entire work, as a whole, under this - License to anyone who comes into possession of a copy. This - License will therefore apply, along with any applicable section 7 - additional terms, to the whole of the work, and all its parts, - regardless of how they are packaged. This License gives no - permission to license the work in any other way, but it does not - invalidate such permission if you have separately received it. - - d) If the work has interactive user interfaces, each must display - Appropriate Legal Notices; however, if the Program has interactive - interfaces that do not display Appropriate Legal Notices, your - work need not make them do so. - - A compilation of a covered work with other separate and independent -works, which are not by their nature extensions of the covered work, -and which are not combined with it such as to form a larger program, -in or on a volume of a storage or distribution medium, is called an -"aggregate" if the compilation and its resulting copyright are not -used to limit the access or legal rights of the compilation's users -beyond what the individual works permit. Inclusion of a covered work -in an aggregate does not cause this License to apply to the other -parts of the aggregate. - - 6. Conveying Non-Source Forms. - - You may convey a covered work in object code form under the terms -of sections 4 and 5, provided that you also convey the -machine-readable Corresponding Source under the terms of this License, -in one of these ways: - - a) Convey the object code in, or embodied in, a physical product - (including a physical distribution medium), accompanied by the - Corresponding Source fixed on a durable physical medium - customarily used for software interchange. - - b) Convey the object code in, or embodied in, a physical product - (including a physical distribution medium), accompanied by a - written offer, valid for at least three years and valid for as - long as you offer spare parts or customer support for that product - model, to give anyone who possesses the object code either (1) a - copy of the Corresponding Source for all the software in the - product that is covered by this License, on a durable physical - medium customarily used for software interchange, for a price no - more than your reasonable cost of physically performing this - conveying of source, or (2) access to copy the - Corresponding Source from a network server at no charge. - - c) Convey individual copies of the object code with a copy of the - written offer to provide the Corresponding Source. This - alternative is allowed only occasionally and noncommercially, and - only if you received the object code with such an offer, in accord - with subsection 6b. - - d) Convey the object code by offering access from a designated - place (gratis or for a charge), and offer equivalent access to the - Corresponding Source in the same way through the same place at no - further charge. You need not require recipients to copy the - Corresponding Source along with the object code. If the place to - copy the object code is a network server, the Corresponding Source - may be on a different server (operated by you or a third party) - that supports equivalent copying facilities, provided you maintain - clear directions next to the object code saying where to find the - Corresponding Source. Regardless of what server hosts the - Corresponding Source, you remain obligated to ensure that it is - available for as long as needed to satisfy these requirements. - - e) Convey the object code using peer-to-peer transmission, provided - you inform other peers where the object code and Corresponding - Source of the work are being offered to the general public at no - charge under subsection 6d. - - A separable portion of the object code, whose source code is excluded -from the Corresponding Source as a System Library, need not be -included in conveying the object code work. - - A "User Product" is either (1) a "consumer product", which means any -tangible personal property which is normally used for personal, family, -or household purposes, or (2) anything designed or sold for incorporation -into a dwelling. In determining whether a product is a consumer product, -doubtful cases shall be resolved in favor of coverage. For a particular -product received by a particular user, "normally used" refers to a -typical or common use of that class of product, regardless of the status -of the particular user or of the way in which the particular user -actually uses, or expects or is expected to use, the product. A product -is a consumer product regardless of whether the product has substantial -commercial, industrial or non-consumer uses, unless such uses represent -the only significant mode of use of the product. - - "Installation Information" for a User Product means any methods, -procedures, authorization keys, or other information required to install -and execute modified versions of a covered work in that User Product from -a modified version of its Corresponding Source. The information must -suffice to ensure that the continued functioning of the modified object -code is in no case prevented or interfered with solely because -modification has been made. - - If you convey an object code work under this section in, or with, or -specifically for use in, a User Product, and the conveying occurs as -part of a transaction in which the right of possession and use of the -User Product is transferred to the recipient in perpetuity or for a -fixed term (regardless of how the transaction is characterized), the -Corresponding Source conveyed under this section must be accompanied -by the Installation Information. But this requirement does not apply -if neither you nor any third party retains the ability to install -modified object code on the User Product (for example, the work has -been installed in ROM). - - The requirement to provide Installation Information does not include a -requirement to continue to provide support service, warranty, or updates -for a work that has been modified or installed by the recipient, or for -the User Product in which it has been modified or installed. Access to a -network may be denied when the modification itself materially and -adversely affects the operation of the network or violates the rules and -protocols for communication across the network. - - Corresponding Source conveyed, and Installation Information provided, -in accord with this section must be in a format that is publicly -documented (and with an implementation available to the public in -source code form), and must require no special password or key for -unpacking, reading or copying. - - 7. Additional Terms. - - "Additional permissions" are terms that supplement the terms of this -License by making exceptions from one or more of its conditions. -Additional permissions that are applicable to the entire Program shall -be treated as though they were included in this License, to the extent -that they are valid under applicable law. If additional permissions -apply only to part of the Program, that part may be used separately -under those permissions, but the entire Program remains governed by -this License without regard to the additional permissions. - - When you convey a copy of a covered work, you may at your option -remove any additional permissions from that copy, or from any part of -it. (Additional permissions may be written to require their own -removal in certain cases when you modify the work.) You may place -additional permissions on material, added by you to a covered work, -for which you have or can give appropriate copyright permission. - - Notwithstanding any other provision of this License, for material you -add to a covered work, you may (if authorized by the copyright holders of -that material) supplement the terms of this License with terms: - - a) Disclaiming warranty or limiting liability differently from the - terms of sections 15 and 16 of this License; or - - b) Requiring preservation of specified reasonable legal notices or - author attributions in that material or in the Appropriate Legal - Notices displayed by works containing it; or - - c) Prohibiting misrepresentation of the origin of that material, or - requiring that modified versions of such material be marked in - reasonable ways as different from the original version; or - - d) Limiting the use for publicity purposes of names of licensors or - authors of the material; or - - e) Declining to grant rights under trademark law for use of some - trade names, trademarks, or service marks; or - - f) Requiring indemnification of licensors and authors of that - material by anyone who conveys the material (or modified versions of - it) with contractual assumptions of liability to the recipient, for - any liability that these contractual assumptions directly impose on - those licensors and authors. - - All other non-permissive additional terms are considered "further -restrictions" within the meaning of section 10. If the Program as you -received it, or any part of it, contains a notice stating that it is -governed by this License along with a term that is a further -restriction, you may remove that term. If a license document contains -a further restriction but permits relicensing or conveying under this -License, you may add to a covered work material governed by the terms -of that license document, provided that the further restriction does -not survive such relicensing or conveying. - - If you add terms to a covered work in accord with this section, you -must place, in the relevant source files, a statement of the -additional terms that apply to those files, or a notice indicating -where to find the applicable terms. - - Additional terms, permissive or non-permissive, may be stated in the -form of a separately written license, or stated as exceptions; -the above requirements apply either way. - - 8. Termination. - - You may not propagate or modify a covered work except as expressly -provided under this License. Any attempt otherwise to propagate or -modify it is void, and will automatically terminate your rights under -this License (including any patent licenses granted under the third -paragraph of section 11). - - However, if you cease all violation of this License, then your -license from a particular copyright holder is reinstated (a) -provisionally, unless and until the copyright holder explicitly and -finally terminates your license, and (b) permanently, if the copyright -holder fails to notify you of the violation by some reasonable means -prior to 60 days after the cessation. - - Moreover, your license from a particular copyright holder is -reinstated permanently if the copyright holder notifies you of the -violation by some reasonable means, this is the first time you have -received notice of violation of this License (for any work) from that -copyright holder, and you cure the violation prior to 30 days after -your receipt of the notice. - - Termination of your rights under this section does not terminate the -licenses of parties who have received copies or rights from you under -this License. If your rights have been terminated and not permanently -reinstated, you do not qualify to receive new licenses for the same -material under section 10. - - 9. Acceptance Not Required for Having Copies. - - You are not required to accept this License in order to receive or -run a copy of the Program. Ancillary propagation of a covered work -occurring solely as a consequence of using peer-to-peer transmission -to receive a copy likewise does not require acceptance. However, -nothing other than this License grants you permission to propagate or -modify any covered work. These actions infringe copyright if you do -not accept this License. Therefore, by modifying or propagating a -covered work, you indicate your acceptance of this License to do so. - - 10. Automatic Licensing of Downstream Recipients. - - Each time you convey a covered work, the recipient automatically -receives a license from the original licensors, to run, modify and -propagate that work, subject to this License. You are not responsible -for enforcing compliance by third parties with this License. - - An "entity transaction" is a transaction transferring control of an -organization, or substantially all assets of one, or subdividing an -organization, or merging organizations. If propagation of a covered -work results from an entity transaction, each party to that -transaction who receives a copy of the work also receives whatever -licenses to the work the party's predecessor in interest had or could -give under the previous paragraph, plus a right to possession of the -Corresponding Source of the work from the predecessor in interest, if -the predecessor has it or can get it with reasonable efforts. - - You may not impose any further restrictions on the exercise of the -rights granted or affirmed under this License. For example, you may -not impose a license fee, royalty, or other charge for exercise of -rights granted under this License, and you may not initiate litigation -(including a cross-claim or counterclaim in a lawsuit) alleging that -any patent claim is infringed by making, using, selling, offering for -sale, or importing the Program or any portion of it. - - 11. Patents. - - A "contributor" is a copyright holder who authorizes use under this -License of the Program or a work on which the Program is based. The -work thus licensed is called the contributor's "contributor version". - - A contributor's "essential patent claims" are all patent claims -owned or controlled by the contributor, whether already acquired or -hereafter acquired, that would be infringed by some manner, permitted -by this License, of making, using, or selling its contributor version, -but do not include claims that would be infringed only as a -consequence of further modification of the contributor version. For -purposes of this definition, "control" includes the right to grant -patent sublicenses in a manner consistent with the requirements of -this License. - - Each contributor grants you a non-exclusive, worldwide, royalty-free -patent license under the contributor's essential patent claims, to -make, use, sell, offer for sale, import and otherwise run, modify and -propagate the contents of its contributor version. - - In the following three paragraphs, a "patent license" is any express -agreement or commitment, however denominated, not to enforce a patent -(such as an express permission to practice a patent or covenant not to -sue for patent infringement). To "grant" such a patent license to a -party means to make such an agreement or commitment not to enforce a -patent against the party. - - If you convey a covered work, knowingly relying on a patent license, -and the Corresponding Source of the work is not available for anyone -to copy, free of charge and under the terms of this License, through a -publicly available network server or other readily accessible means, -then you must either (1) cause the Corresponding Source to be so -available, or (2) arrange to deprive yourself of the benefit of the -patent license for this particular work, or (3) arrange, in a manner -consistent with the requirements of this License, to extend the patent -license to downstream recipients. "Knowingly relying" means you have -actual knowledge that, but for the patent license, your conveying the -covered work in a country, or your recipient's use of the covered work -in a country, would infringe one or more identifiable patents in that -country that you have reason to believe are valid. - - If, pursuant to or in connection with a single transaction or -arrangement, you convey, or propagate by procuring conveyance of, a -covered work, and grant a patent license to some of the parties -receiving the covered work authorizing them to use, propagate, modify -or convey a specific copy of the covered work, then the patent license -you grant is automatically extended to all recipients of the covered -work and works based on it. - - A patent license is "discriminatory" if it does not include within -the scope of its coverage, prohibits the exercise of, or is -conditioned on the non-exercise of one or more of the rights that are -specifically granted under this License. You may not convey a covered -work if you are a party to an arrangement with a third party that is -in the business of distributing software, under which you make payment -to the third party based on the extent of your activity of conveying -the work, and under which the third party grants, to any of the -parties who would receive the covered work from you, a discriminatory -patent license (a) in connection with copies of the covered work -conveyed by you (or copies made from those copies), or (b) primarily -for and in connection with specific products or compilations that -contain the covered work, unless you entered into that arrangement, -or that patent license was granted, prior to 28 March 2007. - - Nothing in this License shall be construed as excluding or limiting -any implied license or other defenses to infringement that may -otherwise be available to you under applicable patent law. - - 12. No Surrender of Others' Freedom. - - If conditions are imposed on you (whether by court order, agreement or -otherwise) that contradict the conditions of this License, they do not -excuse you from the conditions of this License. If you cannot convey a -covered work so as to satisfy simultaneously your obligations under this -License and any other pertinent obligations, then as a consequence you may -not convey it at all. For example, if you agree to terms that obligate you -to collect a royalty for further conveying from those to whom you convey -the Program, the only way you could satisfy both those terms and this -License would be to refrain entirely from conveying the Program. - - 13. Use with the GNU Affero General Public License. - - Notwithstanding any other provision of this License, you have -permission to link or combine any covered work with a work licensed -under version 3 of the GNU Affero General Public License into a single -combined work, and to convey the resulting work. The terms of this -License will continue to apply to the part which is the covered work, -but the special requirements of the GNU Affero General Public License, -section 13, concerning interaction through a network will apply to the -combination as such. - - 14. Revised Versions of this License. - - The Free Software Foundation may publish revised and/or new versions of -the GNU General Public License from time to time. Such new versions will -be similar in spirit to the present version, but may differ in detail to -address new problems or concerns. - - Each version is given a distinguishing version number. If the -Program specifies that a certain numbered version of the GNU General -Public License "or any later version" applies to it, you have the -option of following the terms and conditions either of that numbered -version or of any later version published by the Free Software -Foundation. If the Program does not specify a version number of the -GNU General Public License, you may choose any version ever published -by the Free Software Foundation. - - If the Program specifies that a proxy can decide which future -versions of the GNU General Public License can be used, that proxy's -public statement of acceptance of a version permanently authorizes you -to choose that version for the Program. - - Later license versions may give you additional or different -permissions. However, no additional obligations are imposed on any -author or copyright holder as a result of your choosing to follow a -later version. - - 15. Disclaimer of Warranty. - - THERE IS NO WARRANTY FOR THE PROGRAM, TO THE EXTENT PERMITTED BY -APPLICABLE LAW. EXCEPT WHEN OTHERWISE STATED IN WRITING THE COPYRIGHT -HOLDERS AND/OR OTHER PARTIES PROVIDE THE PROGRAM "AS IS" WITHOUT WARRANTY -OF ANY KIND, EITHER EXPRESSED OR IMPLIED, INCLUDING, BUT NOT LIMITED TO, -THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR -PURPOSE. THE ENTIRE RISK AS TO THE QUALITY AND PERFORMANCE OF THE PROGRAM -IS WITH YOU. SHOULD THE PROGRAM PROVE DEFECTIVE, YOU ASSUME THE COST OF -ALL NECESSARY SERVICING, REPAIR OR CORRECTION. - - 16. Limitation of Liability. - - IN NO EVENT UNLESS REQUIRED BY APPLICABLE LAW OR AGREED TO IN WRITING -WILL ANY COPYRIGHT HOLDER, OR ANY OTHER PARTY WHO MODIFIES AND/OR CONVEYS -THE PROGRAM AS PERMITTED ABOVE, BE LIABLE TO YOU FOR DAMAGES, INCLUDING ANY -GENERAL, SPECIAL, INCIDENTAL OR CONSEQUENTIAL DAMAGES ARISING OUT OF THE -USE OR INABILITY TO USE THE PROGRAM (INCLUDING BUT NOT LIMITED TO LOSS OF -DATA OR DATA BEING RENDERED INACCURATE OR LOSSES SUSTAINED BY YOU OR THIRD -PARTIES OR A FAILURE OF THE PROGRAM TO OPERATE WITH ANY OTHER PROGRAMS), -EVEN IF SUCH HOLDER OR OTHER PARTY HAS BEEN ADVISED OF THE POSSIBILITY OF -SUCH DAMAGES. - - 17. Interpretation of Sections 15 and 16. - - If the disclaimer of warranty and limitation of liability provided -above cannot be given local legal effect according to their terms, -reviewing courts shall apply local law that most closely approximates -an absolute waiver of all civil liability in connection with the -Program, unless a warranty or assumption of liability accompanies a -copy of the Program in return for a fee. - - END OF TERMS AND CONDITIONS - - How to Apply These Terms to Your New Programs - - If you develop a new program, and you want it to be of the greatest -possible use to the public, the best way to achieve this is to make it -free software which everyone can redistribute and change under these terms. - - To do so, attach the following notices to the program. It is safest -to attach them to the start of each source file to most effectively -state the exclusion of warranty; and each file should have at least -the "copyright" line and a pointer to where the full notice is found. - - - Copyright (C) - - This program is free software: you can redistribute it and/or modify - it under the terms of the GNU General Public License as published by - the Free Software Foundation, either version 3 of the License, or - (at your option) any later version. - - This program is distributed in the hope that it will be useful, - but WITHOUT ANY WARRANTY; without even the implied warranty of - MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the - GNU General Public License for more details. - - You should have received a copy of the GNU General Public License - along with this program. If not, see . - -Also add information on how to contact you by electronic and paper mail. - - If the program does terminal interaction, make it output a short -notice like this when it starts in an interactive mode: - - Copyright (C) - This program comes with ABSOLUTELY NO WARRANTY; for details type `show w'. - This is free software, and you are welcome to redistribute it - under certain conditions; type `show c' for details. - -The hypothetical commands `show w' and `show c' should show the appropriate -parts of the General Public License. Of course, your program's commands -might be different; for a GUI interface, you would use an "about box". - - You should also get your employer (if you work as a programmer) or school, -if any, to sign a "copyright disclaimer" for the program, if necessary. -For more information on this, and how to apply and follow the GNU GPL, see -. - - The GNU General Public License does not permit incorporating your program -into proprietary programs. If your program is a subroutine library, you -may consider it more useful to permit linking proprietary applications with -the library. If this is what you want to do, use the GNU Lesser General -Public License instead of this License. But first, please read -. +Copyright (c) 2017, Shogun Machine Learning Toolbox developers +All rights reserved. + +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions are met: + + 1. Redistributions of source code must retain the above copyright notice, + this list of conditions and the following disclaimer. + + 2. Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + 3. Neither the name of the copyright holder nor the names of its + contributors may be used to endorse or promote products derived from + this software without specific prior written permission. + +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. diff --git a/doc/license/README_gpl.md b/doc/license/README_gpl.md new file mode 100644 index 00000000000..c5b323d423b --- /dev/null +++ b/doc/license/README_gpl.md @@ -0,0 +1,5 @@ +Shogun can be built (optionally) with GPL codes enabled. +In order to allow for a less restrictive BSD compatible distribution of the +main repository, these have been moved to a separate repository, shogun-gpl. +The build automatically integrates these codes if it is checked out. +Please refer to the submodule for exact licensing text and code details. diff --git a/doc/readme/ABOUT.md b/doc/readme/ABOUT.md index 327eee2f9d0..269f764d5cf 100644 --- a/doc/readme/ABOUT.md +++ b/doc/readme/ABOUT.md @@ -1,12 +1,10 @@ -The Shogun Machine learning toolbox offers a wide range of efficient and unified Machine Learning methods. - +Shogun is and open-source machine learning library that offers a wide range of efficient and unified machine learning methods. #### Shogun is accessible * Supports many languages (Python, Octave, R, Java/Scala, Lua, C#, Ruby, etc) and platforms (Linux/Unix, MacOS and Windows) and integrates with their scientific computing environments. * Try Shogun in the [cloud](https://cloud.shogun.ml) from your browser. - #### Shogun is state-of-the-art * Efficient implementation (from standard to cutting edge algorithms), modern software architecture in C++. diff --git a/doc/readme/DEVELOPING.md b/doc/readme/DEVELOPING.md index d9b684f8de7..5e5368b290f 100644 --- a/doc/readme/DEVELOPING.md +++ b/doc/readme/DEVELOPING.md @@ -49,7 +49,6 @@ The steps are Read error messages and use the internet to find solutions. Compile errors are the easiest to fix! If all that does not help, ask us. - 7. Commit locally, using neat and informative commit messages, grouping commits, potentially iterate over more changes to the code, git commit FILENAME(S) -m "Fix issue #1234" @@ -63,22 +62,29 @@ The steps are git reset --soft HEAD~3 git commit -m 'Clear commit message' +8. Check if the code was written in conformity with the [Shogun Code Style Guidelines](https://github.com/shogun-toolbox/shogun/wiki/Code-style), or the Continuous Integration will fail. +Shogun has a custom script called `check_format.sh` which can be used to verify the code formatting. + + ./scripts/check_format.sh "feature/BRANCH_NAME" "develop" -8. [Rebase](https://git-scm.com/book/en/v2/Git-Branching-Rebasing) against shogun's develop branch. + The script will provide you with the necessary information to fix potential style errors. All of this is done by + using [clang-format](https://clang.llvm.org/docs/ClangFormat.html). Make sure to have it installed + on your local machine, or the above script won't work. Update the commit once you have fixed the errors. +9. [Rebase](https://git-scm.com/book/en/v2/Git-Branching-Rebasing) against shogun's develop branch. This might cause rebase errors, which you need to [solve](https://help.github.com/articles/resolving-merge-conflicts-after-a-git-rebase/) git pull --rebase upstream develop -9. Push your commits to your fork +10. Push your commits to your fork git push origin feature/BRANCH_NAME If you squashed or amended commits after you had pushed already, you might be required to force push via using the `git push -f` option **with care**. -10. Send a [pull request](https://help.github.com/articles/about-pull-requests/) (PR) via GitHub. +11. Send a [pull request](https://help.github.com/articles/about-pull-requests/) (PR) via GitHub. As described above, you can always **update** a pull request using the the `git push -f` option. Please **do not** close and send new ones instead, always update. -11. Once the PR is merged, keep an eye on the [buildfarm](#buildfarm) to see whether your patch broke something. +12. Once the PR is merged, keep an eye on the [buildfarm](#buildfarm) to see whether your patch broke something. ## Requirements for merging your PR * Read some [tips](http://blog.ploeh.dk/2015/01/15/10-tips-for-better-pull-requests/) on how to write good pull requests. @@ -127,7 +133,7 @@ You can execute single tests via `ctest`, or via directly executing the unit tes Note that wildcards are allowed. Running single sub-tests is sometimes useful (i.e. for bug hunting) - ./bin/shogun-unit-test --gtest_filter=GaussianProcessRegression.apply_apply_regression + ./bin/shogun-unit-test --gtest_filter=GaussianProcessRegression.apply_regression ### Debugging and Memory leaks **All your C++ code and unit tests must be checked to not leak memory!** @@ -136,8 +142,8 @@ If you do that, you might want to compile with debugging symbols and without com Then - valgrind ./shogun-unit-test --gtest_filter=GaussianProcessRegression.apply_apply_regression - gdb ./shogun-unit-test --gtest_filter=GaussianProcessRegression.apply_apply_regression + valgrind ./bin/shogun-unit-test --gtest_filter=GaussianProcessRegression.apply_regression + gdb --args ./bin/shogun-unit-test --gtest_filter=GaussianProcessRegression.apply_regression The option `--leak-check=full` for valgrind might be useful. In addition to manually running valgrind on your tests, you can use `ctest` to check multiple tests. diff --git a/doc/readme/INSTALL.md b/doc/readme/INSTALL.md index be284527e43..b66fa038505 100644 --- a/doc/readme/INSTALL.md +++ b/doc/readme/INSTALL.md @@ -7,6 +7,7 @@ For other cases, we describe how to build Shogun from source code. # Quicklinks * [Ready-to-install packages](#binaries) + - [Anaconda](#anaconda) - [Ubuntu](#ubuntu) - [Debian](#debian) - [Fedora](#fedora) @@ -23,10 +24,22 @@ For other cases, we describe how to build Shogun from source code. - [Problems](#manual-problems) - [CMake tips](#manual-cmake) - [Customized Python](#manual-python) - - [Winows](#manual-windows) + - [Windows](#manual-windows) ## Ready-to-install packages +### Anaconda packages +The base shogun library and its Python interface are available through the conda package manager, via conda-forge. +To install both: + + conda install -c conda-forge shogun + +or to get just the library: + + conda install -c conda-forge shogun-cpp + +These packages include most of the optional dependencies and are currently available for Linux and MacOS; we're [working on a Windows build](https://github.com/conda-forge/shogun-cpp-feedstock/issues/1). + ### Ubuntu ppa We are working on integrating Shogun with Debian/Ubuntu. In the meantime, we offer a [prepackaged ppa](https://launchpad.net/~shogun-toolbox/+archive/ubuntu/stable). @@ -38,7 +51,7 @@ Add this to your system as Then, install as - sudo apt-get install libshogun17 + sudo apt-get install libshogun18 The Python (2) bindings can be installed as @@ -47,19 +60,19 @@ The Python (2) bindings can be installed as In addition to the latest stable release, we offer [nightly builds](https://launchpad.net/~shogun-toolbox/+archive/ubuntu/nightly) of our development branch. ### Debian -Latest packages for Debian jessie are available in our own repository at [http://apt.shogun.ml](http://apt.shogun.ml). -We provide both the stable and nightly packages, currenlty only for amd64 architecture. +Latest packages for Debian `jessie` and `stretch` are available in our own repository at [http://apt.shogun.ml](http://apt.shogun.ml). +We provide both the stable and nightly packages, currently only for amd64 architecture. In order to add the stable packages to your system, simply run the following commands sudo apt-key adv --keyserver keyserver.ubuntu.com --recv-keys 3DD2174ACAB30365 - echo "deb http://apt.shogun.ml/ jessie main" | sudo tee /etc/apt/sources.list.d/shogun-toolbox.list > /dev/null + echo "deb http://apt.shogun.ml/ stretch main" | sudo tee /etc/apt/sources.list.d/shogun-toolbox.list > /dev/null sudo apt-get update After this just simply install the shogun library - sudo apt-get install libshogun17 + sudo apt-get install libshogun18 -The nightly packages are available in the `nightly` component, i.e. `deb http://apt.shogun.ml/ jessie nightly` +The nightly packages are available in the `nightly` component, i.e. `deb http://apt.shogun.ml/ stretch nightly` ### Fedora Shogun is part of [Fedora 25](https://admin.fedoraproject.org/pkgdb/package/rpms/shogun/). @@ -72,10 +85,7 @@ Install as Shogun is part of [homebrew-science](https://github.com/Homebrew/homebrew-science). Install the latest stable version as - sudo brew install shogun - -Note: Shogun in homebrew is outdated. -Contact us if this changed or if you want to help changing it. + sudo brew install homebrew/science/shogun ### Windows Shogun natively compiles under Windows using MSVC, see the [AppVeyor CI build](https://ci.appveyor.com/project/vigsterkr/shogun) and the [Windows section](#manual-windows) @@ -89,11 +99,9 @@ You can run Shogun in [our own cloud](cloud.shogun.ml) or set up your own using sudo docker pull shogun/shogun:master sudo docker run -it shogun/shogun:master bash -We offer images for both the latest release and nightly development builds. - -For the [developer version](https://hub.docker.com/r/shogun/shogun-dev/), replace `shogun/shogun:master` with `shogun/shogun-dev`. +The docker image follows both the `master` and the `develop` branch of the repository, just specify the desired branch name as tag for the image. For example in order to use the develop version of shogun simply pull the `shogun/shogun:develop` docker image. -Check the "details" tab before downloading to check if the latest build was successful (otherwise you might run into errors when running the docker image)." +There's an [SDK docker image](https://hub.docker.com/r/shogun/shogun-dev/) for shogun development as well, which we use to run our [Travis CI](https://travis-ci.org/shogun-toolbox/shogun/) jobs. Sometimes mounting a local folder into the docker image is useful. You can do this via passing an additional option @@ -178,10 +186,10 @@ In both cases, it is necessary to set a number of system libraries for using Sho ## Interfaces The native C++ interface is always included. -The cmake options for building interfaces are `-DPythonModular -DOctaveModular -DRModular -DJavaModular -DRubyModular -DLuaModular -DCSharpModular` etc. +The cmake options for building interfaces are `-DINTERFACE_PYTHON=ON -DINTERFACE_R ..` etc. For example, replace the cmake step above by ``` -cmake -DPythonModular=ON [potentially more options] .. +cmake -DINTERFACE_PYTHON=ON [potentially more options] .. ``` The required packages (here debian/Ubuntu package names) for each interface are @@ -228,10 +236,6 @@ Make sure to read the [docs](https://cmake.org/documentation/) and [CMake_Useful Make sure to understand the concept of [out of source builds](https://cmake.org/Wiki/CMake_FAQ#Out-of-source_build_trees). Here are some tips on common options that are useful -Getting a list of possible interfaces to enable: - - grep -E "OPTION.*(Modular)" CMakeLists.txt - Specify a different swig executable: cmake -DSWIG_EXECUTABLE=/usr/bin/swig_custom @@ -256,16 +260,16 @@ For that, you need to do something similar to For example, for `brew` installed Python under MacOS, use something like: - cmake -DPYTHON_INCLUDE_DIR=/usr/local/Cellar/python/2.7.5/Frameworks/Python.framework/Headers -DPYTHON_LIBRARY=/usr/local/Cellar/python/2.7.5/Frameworks/Python.framework/Versions/2.7/lib/libpython2.7.dylib -DPythonModular=ON .. + cmake -DPYTHON_INCLUDE_DIR=/usr/local/Cellar/python/2.7.5/Frameworks/Python.framework/Headers -DPYTHON_LIBRARY=/usr/local/Cellar/python/2.7.5/Frameworks/Python.framework/Versions/2.7/lib/libpython2.7.dylib -DINTERFACE_PYTHON=ON .. Under Linux, where you want to use Python 3, which is not the system's default: - cmake -DPYTHON_INCLUDE_DIR=/usr/include/python3.3 -DPYTHON_EXECUTABLE:FILEPATH=/usr/bin/python3 -DPYTHON_PACKAGES_PATH=/usr/local/lib/python3.3/dist-packages -DPythonModular=ON .. + cmake -DPYTHON_INCLUDE_DIR=/usr/include/python3.3 -DPYTHON_EXECUTABLE:FILEPATH=/usr/bin/python3 -DPYTHON_PACKAGES_PATH=/usr/local/lib/python3.3/dist-packages -DINTERFACE_PYTHON=ON .. On a Linux cluster without root access, using [Anaconda](https://www.continuum.io/downloads) (note you will need to activate your environment everytime you want to run Shogun): source path/to/anaconda/bin/activate - cmake -DCMAKE_INSTALL_PREFIX=path/to/shogun/install/dir -DPYTHON_INCLUDE_DIR=path/to/anaconda/include/python2.7/ -DPYTHON_LIBRARY=path/to/anaconda/lib/libpython2.7.so -DPYTHON_EXECUTABLE=path/to/anaconda/bin/python -DPythonModular=On .. + cmake -DCMAKE_INSTALL_PREFIX=path/to/shogun/install/dir -DPYTHON_INCLUDE_DIR=path/to/anaconda/include/python2.7/ -DPYTHON_LIBRARY=path/to/anaconda/lib/libpython2.7.so -DPYTHON_EXECUTABLE=path/to/anaconda/bin/python -DINTERFACE_PYTHON=ON .. ## Windows build @@ -273,7 +277,7 @@ Please see our [AppVeyor](https://ci.appveyor.com/project/vigsterkr/shogun) buil It is recommended to use "Visual Studio 14 2015" or "MSBuild". You will need to adjust all path names to the Windows style, e.g. - git clone https://github.com/shogun-toolbox/shogun.git C:\projects\shogun + git clone https://github.com/shogun-toolbox/shogun.git C:\projects\shogun git submodule -q update --init cd C:\projects\shogun md build && cd build diff --git a/doc/readme/INTERFACES.md b/doc/readme/INTERFACES.md index b348a5b3217..b5e2aef34ca 100644 --- a/doc/readme/INTERFACES.md +++ b/doc/readme/INTERFACES.md @@ -39,53 +39,53 @@ Running it: ./native_example ### Python -This needs `modshogun.py` to be visible, which is either in `path/to/build/src/interfaces/python_modular/` or in something similar to `path/to/shogun-install/lib/python2.7/dist-packages/` +This needs `shogun.py` to be visible, which is either in `path/to/build/src/interfaces/python_modular/` or in something similar to `path/to/shogun-install/lib/python2.7/dist-packages/` - export PYTHONPATH="path/to/modshogun.py:$PYTHONPATH" + export PYTHONPATH="path/to/shogun.py:$PYTHONPATH" Running an example: python path/to/python_example.py ### Octave -This needs `modshogun.oct` to be visible, which is either in `path/to/build/src/interfaces/octave_modular/` or in something similar to `path/to/shogun-install/lib/x86_64-linux-gnu/octave/site/oct/api-v50+/x86_64-pc-linux-gnu/shogun/` +This needs `shogun.oct` to be visible, which is either in `path/to/build/src/interfaces/octave_modular/` or in something similar to `path/to/shogun-install/lib/x86_64-linux-gnu/octave/site/oct/api-v50+/x86_64-pc-linux-gnu/shogun/` - export OCTAVE_PATH="path/to/modshogun.oct:$OCTAVE_PATH" + export OCTAVE_PATH="path/to/shogun.oct:$OCTAVE_PATH" Running an example: python path/to/octave_example.py ### Ruby -This needs `modshogun.rb` to be visible, which is either in `path/to/build/src/interfaces/ruby_modular/` or in something similar to `path/to/shogun-install/lib/x86_64-linux-gnu/site_ruby` - export RUBYLIB="path/to/modshogun.rb:$RUBYLIB" +This needs `shogun.rb` to be visible, which is either in `path/to/build/src/interfaces/ruby_modular/` or in something similar to `path/to/shogun-install/lib/x86_64-linux-gnu/site_ruby` + export RUBYLIB="path/to/shogun.rb:$RUBYLIB" Running an example: ruby path/to/ruby_example.rb ### R -This needs `modshogun.R` to be visible, which is either in `path/to/build/src/interfaces/r_modular/` or in something similar to `path/to/shogun-install/lib/R/site-library` - export R_LIBS_USER="path/to/modshogun.R:$R_LIBS_USER" +This needs `shogun.R` to be visible, which is either in `path/to/build/src/interfaces/r_modular/` or in something similar to `path/to/shogun-install/lib/R/site-library` + export R_LIBS_USER="path/to/shogun.R:$R_LIBS_USER" Running an example: R --no-restore --no-save --no-readline --slave -f path/to/r_example.rb ### Lua -This needs `libmodshogun.so` (this is the interface file, not the shared library file `libshogun.so`) to be visible, which is either in `path/to/build/src/interfaces/lua_modular/` or in something similar to `path/to/shogun-install/lib/lua/5.1/` +This needs `libshogun.so` (this is the interface file, not the shared library file `libshogun.so`) to be visible, which is either in `path/to/build/src/interfaces/lua_modular/` or in something similar to `path/to/shogun-install/lib/lua/5.1/` - export LUA_CPATH="path/to/libmodshogun.so:$LUA_CPATH" + export LUA_CPATH="path/to/libshogun.so:$LUA_CPATH" Running an example: R --no-restore --no-save --no-readline --slave -f path/to/r_example.R ### CSharp -This needs `modshogun.dll` to be visible, which is either in `path/to/build/src/interfaces/csharp_modular` or in something similar to `path/to/shogun-install/lib/cli/shogun/` +This needs `shogun.dll` to be visible, which is either in `path/to/build/src/interfaces/csharp_modular` or in something similar to `path/to/shogun-install/lib/cli/shogun/` Compiling code works with the mono C# compiler and passing location of the above file - mcs path/to/csharp_example.cs /lib:path/to/modshogun.dll/r:modshogun -out:csharp_example.exe + mcs path/to/csharp_example.cs /lib:path/to/shogun.dll/r:shogun -out:csharp_example.exe Running requires setting the mono path @@ -103,7 +103,7 @@ usually in `/usr/share/java/`. Compiling code works with the java compiler and passing location of `shogun.jar`, `jblas.jar`, and the example itself in the class path - javac -cp /path/to/jblas.jar:/path/to/modshogun.jar:path/to/java_example.java -d /path/to/output/ /path/to/java_example.java + javac -cp /path/to/jblas.jar:/path/to/shogun.jar:path/to/java_example.java -d /path/to/output/ /path/to/java_example.java Running it again requires the above class path and some more options diff --git a/examples/CMakeLists.txt b/examples/CMakeLists.txt index a82e3a06382..0b3d5780f52 100644 --- a/examples/CMakeLists.txt +++ b/examples/CMakeLists.txt @@ -2,8 +2,8 @@ IF (NOT TRAVIS_DISABLE_LIBSHOGUN_TESTS AND EXISTS ${CMAKE_CURRENT_SOURCE_DIR}/un add_subdirectory(${CMAKE_CURRENT_SOURCE_DIR}/undocumented/libshogun) ENDIF() -IF (PythonModular AND EXISTS ${CMAKE_CURRENT_SOURCE_DIR}/undocumented/python_modular) - add_subdirectory(${CMAKE_CURRENT_SOURCE_DIR}/undocumented/python_modular) +IF (INTERFACE_PYTHON AND EXISTS ${CMAKE_CURRENT_SOURCE_DIR}/undocumented/python) + add_subdirectory(${CMAKE_CURRENT_SOURCE_DIR}/undocumented/python) ENDIF() IF(BUILD_META_EXAMPLES) diff --git a/examples/meta/CMakeLists.txt b/examples/meta/CMakeLists.txt index 2ffca780f55..9fd5374bdc6 100644 --- a/examples/meta/CMakeLists.txt +++ b/examples/meta/CMakeLists.txt @@ -27,7 +27,7 @@ LIST(APPEND GENERATOR_DEPENDENCIES ${TARGET_LANGUAGES}) file(MAKE_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}/parser_files) # list of interfaces for which we dont generate meta examples -SET(DISABLED_INTERFACES PerlModular) +SET(DISABLED_INTERFACES INTERFACE_PERL) FOREACH(META_EXAMPLE ${META_EXAMPLES}) # assume a structure //listing.sg @@ -40,7 +40,7 @@ FOREACH(META_EXAMPLE ${META_EXAMPLES}) # FIXME # Hack to avoid generation of scala meta examples dependencies that will # not be generated by generate.py script (since there is no scala.json) - LIST(REMOVE_ITEM AVAILABLE_INTERFACES "ScalaModular") + LIST(REMOVE_ITEM AVAILABLE_INTERFACES INTERFACE_SCALA) FOREACH(interface ${AVAILABLE_INTERFACES}) list(FIND DISABLED_INTERFACES ${interface} disabled) if (${disabled} STREQUAL "-1") @@ -48,7 +48,7 @@ FOREACH(META_EXAMPLE ${META_EXAMPLES}) LIST(APPEND EXAMPLE_LISTINGS ${CMAKE_CURRENT_BINARY_DIR}/${DIRECTORY}/${BASENAME}.${EXTENSION}) endif() ENDFOREACH() - LIST(APPEND AVAILABLE_INTERFACES "ScalaModular") + LIST(APPEND AVAILABLE_INTERFACES INTERFACE_SCALA) # Set generate.py flags SET(GENERATOR_FLAGS @@ -75,7 +75,7 @@ FOREACH(META_EXAMPLE ${META_EXAMPLES}) ENDFOREACH() SET_SOURCE_FILES_PROPERTIES(${TRANSLATED_META_EXAMPLES} PROPERTIES GENERATED TRUE) -add_custom_target(meta_examples +add_custom_target(meta_examples ALL DEPENDS ${TRANSLATED_META_EXAMPLES}) INSTALL(DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}/cpp @@ -104,11 +104,11 @@ ENDIF() # FIXME: these are the interfaces which has various problems # hence need to disable their testing # temporarily disabled as R and static calls has to be fixed -LIST(APPEND DISABLED_INTERFACES RModular) +LIST(APPEND DISABLED_INTERFACES INTERFACE_R) # temporarily removed since lua modular currently # does not support overloaded c++ methods # see https://github.com/shogun-toolbox/shogun/issues/3018 -LIST(APPEND DISABLED_INTERFACES LuaModular) +LIST(APPEND DISABLED_INTERFACES INTERFACE_LUA) FOREACH(interface ${AVAILABLE_INTERFACES}) diff --git a/examples/meta/cpp/CMakeLists.txt b/examples/meta/cpp/CMakeLists.txt index a6a9e154928..f79be391918 100644 --- a/examples/meta/cpp/CMakeLists.txt +++ b/examples/meta/cpp/CMakeLists.txt @@ -32,9 +32,9 @@ FOREACH(META_EXAMPLE ${META_EXAMPLES}) # Add examples to the dependencies of modular interfaces to make sure # nothing will infer with them being build single-threaded. IF(SWIG_SINGLE_THREADED) - FOREACH(SG_MODULAR_INTERFACE_TARGET ${SG_MODULAR_INTERFACE_TARGETS}) - ADD_DEPENDENCIES(${SG_MODULAR_INTERFACE_TARGET} ${GENERATED_CPP_BINARY}) - ENDFOREACH(SG_MODULAR_INTERFACE_TARGET ${SG_MODULAR_INTERFACE_TARGETS}) + FOREACH(SG_INTERFACE_TARGET ${SG_INTERFACE_TARGETS}) + ADD_DEPENDENCIES(${SG_INTERFACE_TARGET} ${GENERATED_CPP_BINARY}) + ENDFOREACH(SG_INTERFACE_TARGET ${SG_INTERFACE_TARGETS}) ENDIF(SWIG_SINGLE_THREADED) LIST(APPEND GENERATED_CPP_EXAMPLES ${GENERATED_CPP_BINARY}) diff --git a/examples/meta/csharp/CMakeLists.txt b/examples/meta/csharp/CMakeLists.txt index 38a7b28c4f1..7e0d86bfc81 100644 --- a/examples/meta/csharp/CMakeLists.txt +++ b/examples/meta/csharp/CMakeLists.txt @@ -1,4 +1,4 @@ -SET(CSHARP_FLAGS "/lib:${CSHARP_MODULAR_BUILD_DIR};/r:modshogun") +SET(CSHARP_FLAGS "/lib:${INTERFACE_CSHARP_BUILD_DIR};/r:shogun") # add test case for each generated example # (not generated yet so have to fake filenames from META_EXAMPLES list) @@ -11,7 +11,7 @@ FOREACH(META_EXAMPLE ${META_EXAMPLES}) ADD_CUSTOM_COMMAND(OUTPUT ${CMAKE_CURRENT_BINARY_DIR}/${EXAMPLE_REL_DIR}/${EXAMPLE_NAME}.exe COMMAND ${CSHARP_COMPILER} ${EXAMPLE_NAME}.cs ${CSHARP_FLAGS} -out:${EXAMPLE_NAME}.exe - DEPENDS csharp_modular shogun meta_examples ${CSHARP_META_EXAMPLE_SOURCE} + DEPENDS interface_csharp shogun meta_examples ${CSHARP_META_EXAMPLE_SOURCE} WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}/${EXAMPLE_REL_DIR}) ADD_CUSTOM_TARGET(csharp_${EXAMPLE_NAME} ALL @@ -23,7 +23,7 @@ FOREACH(META_EXAMPLE ${META_EXAMPLES}) WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}/${EXAMPLE_REL_DIR} COMMAND ${CSHARP_INTERPRETER} ${CMAKE_CURRENT_BINARY_DIR}/${EXAMPLE_REL_DIR}/${EXAMPLE_NAME}.exe) set_property(TEST generated_csharp-${EXAMPLE_NAME_WITH_DIR} PROPERTY - ENVIRONMENT "MONO_PATH=${CSHARP_MODULAR_BUILD_DIR}") + ENVIRONMENT "MONO_PATH=${INTERFACE_CSHARP_BUILD_DIR}") ENDFOREACH() add_custom_target(build_csharp_meta_examples ALL diff --git a/examples/meta/generator/generate.py b/examples/meta/generator/generate.py index 908667be078..ba8d1e8d355 100755 --- a/examples/meta/generator/generate.py +++ b/examples/meta/generator/generate.py @@ -88,8 +88,13 @@ def translateExamples(inputDir, outputDir, targetsDir, ctagsFile, extension = target["FileExtension"] # Create directory if it does not exist - if not os.path.exists(directory): - os.makedirs(directory) + try: + os.makedirs(name=directory, exist_ok=True) + except TypeError: + try: + os.makedirs(directory) + except OSError: + pass # Write translation outputFile = os.path.join(directory, @@ -98,9 +103,12 @@ def translateExamples(inputDir, outputDir, targetsDir, ctagsFile, # create subdirectories if they don't exist yet try: - os.makedirs(os.path.dirname(outputFile)) - except OSError: - pass + os.makedirs(name=os.path.dirname(outputFile), exist_ok=True) + except TypeError: + try: + os.makedirs(os.path.dirname(outputFile)) + except OSError: + pass with open(outputFile, "w") as nf: nf.write(translation) diff --git a/examples/meta/generator/parse.py b/examples/meta/generator/parse.py index 2d68ed33a5f..e84247630dc 100644 --- a/examples/meta/generator/parse.py +++ b/examples/meta/generator/parse.py @@ -213,6 +213,10 @@ def p_staticCall(self, p): "staticCall : type COLON identifier LPAREN argumentList RPAREN" p[0] = {"StaticCall": [p[1], p[3], p[5]]} + def p_globalCall(self, p): + "globalCall : identifier LPAREN argumentList RPAREN" + p[0] = {"GlobalCall": [p[1], p[3]]} + def p_indexList(self, p): """ indexList : int @@ -262,6 +266,7 @@ def p_expr(self, p): expr : enum | methodCall | staticCall + | globalCall | elementAccess | string | char diff --git a/examples/meta/generator/targets/cpp.json b/examples/meta/generator/targets/cpp.json index 73a224161e0..2dbcbe2dd03 100644 --- a/examples/meta/generator/targets/cpp.json +++ b/examples/meta/generator/targets/cpp.json @@ -3,6 +3,7 @@ "Dependencies": { "IncludeAllClasses": true, "IncludeEnums": true, + "IncludeGlobalFunctions": true, "DependencyListElement": "#include <$includePath>", "DependencyListSeparator": "\n" }, @@ -85,6 +86,7 @@ "FloatLiteral": "${number}f", "MethodCall": "$object->$method($arguments)", "StaticCall": "C$typeName::$method($arguments)", + "GlobalCall": "$method($arguments)", "Identifier": "$identifier", "Enum":"$typeName::$value" }, diff --git a/examples/meta/generator/targets/csharp.json b/examples/meta/generator/targets/csharp.json index 50d40a6baf0..5325014e41c 100644 --- a/examples/meta/generator/targets/csharp.json +++ b/examples/meta/generator/targets/csharp.json @@ -1,5 +1,5 @@ { - "Program": "using System;\n\npublic class classifier_knn_modular {\npublic static void Main() {\nmodshogun.init_shogun_with_defaults();\n\n$program\n}\n}\n", + "Program": "using System;\n\npublic class Application {\npublic static void Main() {\nshogun.init_shogun_with_defaults();\n\n$program\n}\n}\n", "Statement": "$statement;\n", "Comment": "//$comment\n", "Init": { @@ -70,6 +70,7 @@ "FloatLiteral": "${number}f", "MethodCall": "$object.$method($arguments)", "StaticCall": "$typeName.$method($arguments)", + "GlobalCall": "shogun.$method($arguments)", "Identifier": "$identifier", "Enum":"$typeName.$value" }, diff --git a/examples/meta/generator/targets/java.json b/examples/meta/generator/targets/java.json index 59ec7da993c..bf46a84921f 100644 --- a/examples/meta/generator/targets/java.json +++ b/examples/meta/generator/targets/java.json @@ -1,8 +1,9 @@ { - "Program": "import org.jblas.DoubleMatrix;\nimport org.jblas.FloatMatrix;\n\nimport org.shogun.modshogun;\n$dependencies\n\npublic class $programName {\nstatic {\nSystem.loadLibrary(\"modshogun\");\n}\n\npublic static void main(String argv[]) {\nmodshogun.init_shogun_with_defaults();\n\n$program\n}\n}\n", + "Program": "import org.jblas.DoubleMatrix;\nimport org.jblas.FloatMatrix;\n\nimport org.shogun.shogun;\n$dependencies\n\npublic class $programName {\nstatic {\nSystem.loadLibrary(\"shogun\");\n}\n\npublic static void main(String argv[]) {\nshogun.init_shogun_with_defaults();\n\n$program\n}\n}\n", "Dependencies": { "IncludeAllClasses": true, "IncludeEnums": true, + "IncludeGlobalFunctions": false, "DependencyListElement": "import org.shogun.$typeName;", "DependencyListSeparator": "\n" }, @@ -75,6 +76,7 @@ "FloatLiteral": "${number}f", "MethodCall": "$object.$method($arguments)", "StaticCall": "$typeName.$method($arguments)", + "GlobalCall": "shogun.$method($arguments)", "Identifier": "$identifier", "Enum":"$typeName.$value" }, diff --git a/examples/meta/generator/targets/lua.json b/examples/meta/generator/targets/lua.json index a5a5171f525..9cef4e54ef1 100644 --- a/examples/meta/generator/targets/lua.json +++ b/examples/meta/generator/targets/lua.json @@ -1,9 +1,9 @@ { - "Program": "require 'modshogun'\n\n$program", + "Program": "require 'shogun'\n\n$program", "Statement": "$statement\n", "Comment": "--$comment\n", "Init": { - "Construct": "$name = modshogun.$typeName($arguments)", + "Construct": "$name = shogun.$typeName($arguments)", "Copy": "$name = $expr" }, "Assign": "$identifier = $expr", @@ -22,8 +22,9 @@ "FloatLiteral": "$number", "MethodCall": "$object:$method($arguments)", "StaticCall": "$typeName:$method($arguments)", + "GlobalCall": "$method($arguments)", "Identifier": "$identifier", - "Enum":"modshogun.$value" + "Enum":"shogun.$value" }, "Element": { "Access": { diff --git a/examples/meta/generator/targets/octave.json b/examples/meta/generator/targets/octave.json index 010aa47f040..07bb58c247a 100644 --- a/examples/meta/generator/targets/octave.json +++ b/examples/meta/generator/targets/octave.json @@ -1,5 +1,5 @@ { - "Program": "modshogun\n\n$program", + "Program": "shogun\n\n$program", "Statement": "$statement;\n", "Comment": "%$comment\n", "Init": { @@ -40,6 +40,7 @@ "FloatLiteral": "$number", "MethodCall": "$object.$method($arguments)", "StaticCall": "$typeName.$method($arguments)", + "GlobalCall": "$method($arguments)", "Identifier": "$identifier", "Enum":"$value" }, diff --git a/examples/meta/generator/targets/python.json b/examples/meta/generator/targets/python.json index b83e113b50b..228b2e75bcf 100644 --- a/examples/meta/generator/targets/python.json +++ b/examples/meta/generator/targets/python.json @@ -4,8 +4,9 @@ "IncludeAllClasses": false, "IncludeInterfacedClasses": true, "IncludeEnums": true, - "DependencyListElement": "from modshogun import $typeName", - "DependencyListElementEnum": "from modshogun import $value", + "IncludeGlobalFunctions": true, + "DependencyListElement": "from shogun import $typeName", + "DependencyListElementEnum": "from shogun import $value", "DependencyListSeparator": "\n" }, "Statement": "$statement\n", @@ -50,6 +51,7 @@ "FloatLiteral": "$number", "MethodCall": "$object.$method($arguments)", "StaticCall": "$typeName.$method($arguments)", + "GlobalCall": "$method($arguments)", "Identifier": "$identifier", "Enum":"$value" }, diff --git a/examples/meta/generator/targets/r.json b/examples/meta/generator/targets/r.json index 0c79eed971e..c270184ea6f 100644 --- a/examples/meta/generator/targets/r.json +++ b/examples/meta/generator/targets/r.json @@ -22,6 +22,7 @@ "FloatLiteral": "$number", "MethodCall": "$object$$$method($arguments)", "StaticCall": "$typeName$$$method($arguments)", + "GlobalCall": "$method($arguments)", "Identifier": "$identifier", "Enum":"\"$value\"" }, diff --git a/examples/meta/generator/targets/ruby.json b/examples/meta/generator/targets/ruby.json index d6dd8c9441a..c9d889c6856 100644 --- a/examples/meta/generator/targets/ruby.json +++ b/examples/meta/generator/targets/ruby.json @@ -1,9 +1,9 @@ { - "Program": "require 'modshogun'\n\n$dependencies$program", + "Program": "require 'shogun'\n\n$dependencies$program", "Statement": "$statement\n", "Comment": "#$comment\n", "Init": { - "Construct": "$name = Modshogun::$typeName.new $arguments", + "Construct": "$name = Shogun::$typeName.new $arguments", "Copy": "$name = $expr", "CharVector": "$name = NArray.byte($arguments)", "ByteVector": "$name = NArray.byte($arguments)", @@ -41,9 +41,10 @@ "RealLiteral": "$number", "FloatLiteral": "$number", "MethodCall": "$object.$method $arguments", - "StaticCall": "Modshogun::$typeName.$method $arguments", + "StaticCall": "Shogun::$typeName.$method $arguments", + "GlobalCall": "Shogun::$method $arguments", "Identifier": "$identifier", - "Enum":"Modshogun::$value" + "Enum":"Shogun::$value" }, "Element": { "Access": { diff --git a/examples/meta/generator/translate.py b/examples/meta/generator/translate.py index 25477a85241..d6e73551ad2 100644 --- a/examples/meta/generator/translate.py +++ b/examples/meta/generator/translate.py @@ -33,6 +33,7 @@ def getDependencies(program): allClasses = set() interfaceClasses = set() enums = set() + globalFunctions = set() # All classes used for objectType in find("ObjectType", program): @@ -55,13 +56,18 @@ def getDependencies(program): objectKey = list(typeDict.keys())[0] interfaceClasses.add(typeDict[objectKey]) + # All global function calls + for globalCall in find("GlobalCall", program): + identifier = list(globalCall[0].values())[0] + globalFunctions.add(identifier) + # All enums used for enum in find("Enum", program): enumType = enum[0]["Identifier"] enumValue = enum[1]["Identifier"] enums.add((enumType, enumValue)) - return allClasses, interfaceClasses, enums + return allClasses, interfaceClasses, enums, globalFunctions def getBasicTypesToStore(): @@ -160,11 +166,12 @@ def translateProgram(self, program, programName=None, storeVars=False): pass raise e - allClasses, interfacedClasses, enums = getDependencies(program) + allClasses, interfacedClasses, enums, globalFunctions = getDependencies(program) try: dependenciesString = self.dependenciesString(allClasses, interfacedClasses, - enums) + enums, + globalFunctions) except Exception as e: print("Translation of dependencies failed!") raise @@ -239,9 +246,10 @@ def injectVarsStoring(self, statementList, programName, varsToStore): } statementList.append({"Statement": storageSerialize}) - def dependenciesString(self, allClasses, interfacedClasses, enums): + def dependenciesString(self, allClasses, interfacedClasses, enums, + globalFunctions): """ Returns dependency import string - e.g. for python: "from modshogun import RealFeatures\n\n" + e.g. for python: "from shogun import RealFeatures\n\n" """ if "Dependencies" not in self.targetDict: @@ -256,13 +264,19 @@ def dependenciesString(self, allClasses, interfacedClasses, enums): dependencies = dependencies.union(interfacedClasses) if self.targetDict["Dependencies"].get("IncludeEnums"): dependencies = dependencies.union(enums) + if self.targetDict["Dependencies"].get("IncludeGlobalFunctions"): + dependencies = dependencies.union(globalFunctions) + + dependencies = list(dependencies) - translations = set(map(self.translateDependencyElement, dependencies)) + translations = list(map(self.translateDependencyElement, dependencies)) + translations.sort() separator = self.targetDict["Dependencies"]["DependencyListSeparator"] - return reduce(lambda l, r: r if l == "" else l+separator+r, + result = reduce(lambda l, r: r if l == "" else l+separator+r, translations, "") + return result def translateDependencyElement(self, dependencyElement): """ Translates a dependency element @@ -288,7 +302,6 @@ def translateDependencyElement(self, dependencyElement): elif "DependencyListElementClass" in dependencyRules: elementTemplate = Template(dependencyRules["DependencyListElementClass"]) - if "$includePath" in elementTemplate.template: includePath = self.getIncludePathForClass(typeName) @@ -465,6 +478,19 @@ def translateExpr(self, expr): method=method, arguments=translatedArgsList) + elif key == "GlobalCall": + template = Template(self.targetDict["Expr"]["GlobalCall"]) + method = expr[key][0]["Identifier"] + argsList = None + try: + argsList = expr[key][2] + except IndexError: + pass + translatedArgsList = self.translateArgumentList(argsList) + + return template.substitute(typeName=type,method=method, + arguments=translatedArgsList) + elif key == "ElementAccess": return self.translateElementAccess(expr[key]) diff --git a/examples/meta/java/CMakeLists.txt b/examples/meta/java/CMakeLists.txt index e97abcfb3b5..dee81196e12 100644 --- a/examples/meta/java/CMakeLists.txt +++ b/examples/meta/java/CMakeLists.txt @@ -1,6 +1,6 @@ SET(CLASSPATH "${JBLAS}:${SHOGUN_JAR}:${CMAKE_CURRENT_BINARY_DIR}") SET(JAVAOPTS "-Xmx1024m") -SET(JAVA_LIB_PATH "${JAVA_MODULAR_BUILD_DIR}") +SET(JAVA_LIB_PATH "${INTERFACE_JAVA_BUILD_DIR}") # add test case for each generated example # (not generated yet so have to fake filenames from META_EXAMPLES list) @@ -15,7 +15,7 @@ FOREACH(META_EXAMPLE ${META_EXAMPLES}) COMMAND ${Java_JAVAC_EXECUTABLE} -cp ${CLASSPATH} -d ${CMAKE_CURRENT_BINARY_DIR}/${EXAMPLE_REL_DIR} ${JAVA_META_EXAMPLE_SOURCE} - DEPENDS java_modular meta_examples shogun ${JAVA_META_EXAMPLE_SOURCE}) + DEPENDS interface_java meta_examples shogun ${JAVA_META_EXAMPLE_SOURCE}) ADD_CUSTOM_TARGET(java_${EXAMPLE_NAME} ALL DEPENDS ${COMPILED_JAVA_EXAMPLE} diff --git a/examples/meta/lua/CMakeLists.txt b/examples/meta/lua/CMakeLists.txt index 88bb1fead64..fa79808184f 100644 --- a/examples/meta/lua/CMakeLists.txt +++ b/examples/meta/lua/CMakeLists.txt @@ -8,5 +8,5 @@ FOREACH(META_EXAMPLE ${META_EXAMPLES}) WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}/{EXAMPLE_REL_DIR} COMMAND ${LUA_EXECUTABLE} ${CMAKE_CURRENT_BINARY_DIR}/${EXAMPLE_REL_DIR}/${EXAMPLE_NAME}.lua) set_property(TEST generated_lua-${EXAMPLE_NAME_WITH_DIR} PROPERTY - ENVIRONMENT "LUA_CPATH=${LUA_MODULAR_BUILD_DIR}/libmodshogun.so") + ENVIRONMENT "LUA_CPATH=${INTERFACE_LUA_BUILD_DIR}/libshogun.so") ENDFOREACH() diff --git a/examples/meta/octave/CMakeLists.txt b/examples/meta/octave/CMakeLists.txt index 571414ec640..355ae6b5a72 100644 --- a/examples/meta/octave/CMakeLists.txt +++ b/examples/meta/octave/CMakeLists.txt @@ -8,6 +8,6 @@ FOREACH(META_EXAMPLE ${META_EXAMPLES}) WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}/${EXAMPLE_REL_DIR} COMMAND ${OCTAVE_EXECUTABLE} ${CMAKE_CURRENT_BINARY_DIR}/${EXAMPLE_REL_DIR}/${EXAMPLE_NAME}.m) set_property(TEST generated_octave-${EXAMPLE_NAME_WITH_DIR} PROPERTY - ENVIRONMENT "OCTAVE_PATH=${OCTAVE_MODULAR_BINARY_DIR}") + ENVIRONMENT "OCTAVE_PATH=${INTERFACE_OCTAVE_BINARY_DIR}") ENDFOREACH() diff --git a/examples/meta/ruby/CMakeLists.txt b/examples/meta/ruby/CMakeLists.txt index fd24f5a41fa..ef4d3ef72f3 100644 --- a/examples/meta/ruby/CMakeLists.txt +++ b/examples/meta/ruby/CMakeLists.txt @@ -10,5 +10,5 @@ FOREACH(META_EXAMPLE ${META_EXAMPLES}) WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}/${EXAMPLE_REL_DIR} COMMAND ${RUBY_EXECUTABLE} ${CMAKE_CURRENT_BINARY_DIR}/${EXAMPLE_REL_DIR}/${EXAMPLE_NAME}.rb) set_property(TEST generated_ruby-${EXAMPLE_NAME_WITH_DIR} PROPERTY - ENVIRONMENT "RUBYLIB=${RUBY_MODULAR_SOURCE_DIR}:${RUBY_MODULAR_BINARY_DIR}:.") + ENVIRONMENT "RUBYLIB=${INTERFACE_RUBY_SOURCE_DIR}:${INTERFACE_RUBY_BINARY_DIR}:.") ENDFOREACH() diff --git a/examples/meta/scala/CMakeLists.txt b/examples/meta/scala/CMakeLists.txt index c7e3a0f3771..f3907d75f7f 100644 --- a/examples/meta/scala/CMakeLists.txt +++ b/examples/meta/scala/CMakeLists.txt @@ -1,6 +1,6 @@ string( REGEX REPLACE "/scala$" "/java" CMAKE_JAVA_DIR "${CMAKE_CURRENT_BINARY_DIR}" ) SET(CLASSPATH "${JBLAS}:${SHOGUN_JAR}:${Scala_JAR_EXECUTABLE}:${CMAKE_CURRENT_BINARY_DIR}") -SET(JAVA_LIB_PATH "${JAVA_MODULAR_BUILD_DIR}") +SET(JAVA_LIB_PATH "${INTERFACE_JAVA_BUILD_DIR}") # add test case for each generated example # (not generated yet so have to fake filenames from META_EXAMPLES list) diff --git a/examples/meta/src/converter/ica_fast.sg b/examples/meta/src/converter/ica_fast.sg new file mode 100644 index 00000000000..616f6767275 --- /dev/null +++ b/examples/meta/src/converter/ica_fast.sg @@ -0,0 +1,24 @@ +CSVFile f_feats("../../data/ica_2_sources.dat") + +Math:init_random(1) + +#![create_features] +RealFeatures features(f_feats) +#![create_features] + +#![set_parameters] +FastICA ica() +ica.set_max_iter(200) +ica.set_tol(0.00001) +#![set_parameters] + +#![apply_convert] +Features converted = ica.apply(features) +#![apply_convert] + +#![extract] +RealMatrix mixing_matrix = ica.get_mixing_matrix() +RealFeatures casted = RealFeatures:obtain_from_generic(converted) +RealMatrix unmixed_signal = casted.get_feature_matrix() +#![extract] + diff --git a/examples/meta/src/converter/ica_ff_sep.sg b/examples/meta/src/converter/ica_ff_sep.sg new file mode 100644 index 00000000000..c8d6b47af0d --- /dev/null +++ b/examples/meta/src/converter/ica_ff_sep.sg @@ -0,0 +1,24 @@ +CSVFile f_feats("../../data/ica_2_sources.dat") + +Math:init_random(1) + +#![create_features] +RealFeatures features(f_feats) +#![create_features] + +#![set_parameters] +FFSep ica() +ica.set_max_iter(200) +ica.set_tol(0.00001) +#![set_parameters] + +#![apply_convert] +Features converted = ica.apply(features) +#![apply_convert] + +#![extract] +RealMatrix mixing_matrix = ica.get_mixing_matrix() +RealFeatures casted = RealFeatures:obtain_from_generic(converted) +RealMatrix unmixed_signal = casted.get_feature_matrix() +#![extract] + diff --git a/examples/meta/src/converter/ica_jade.sg b/examples/meta/src/converter/ica_jade.sg new file mode 100644 index 00000000000..633e0eedf45 --- /dev/null +++ b/examples/meta/src/converter/ica_jade.sg @@ -0,0 +1,24 @@ +CSVFile f_feats("../../data/ica_2_sources.dat") + +Math:init_random(1) + +#![create_features] +RealFeatures features(f_feats) +#![create_features] + +#![set_parameters] +Jade ica() +ica.set_max_iter(200) +ica.set_tol(0.00001) +#![set_parameters] + +#![apply_convert] +Features converted = ica.apply(features) +#![apply_convert] + +#![extract] +RealMatrix mixing_matrix = ica.get_mixing_matrix() +RealFeatures casted = RealFeatures:obtain_from_generic(converted) +RealMatrix unmixed_signal = casted.get_feature_matrix() +#![extract] + diff --git a/examples/meta/src/converter/ica_jedi_sep.sg b/examples/meta/src/converter/ica_jedi_sep.sg new file mode 100644 index 00000000000..68501bc9ab5 --- /dev/null +++ b/examples/meta/src/converter/ica_jedi_sep.sg @@ -0,0 +1,24 @@ +CSVFile f_feats("../../data/ica_2_sources.dat") + +Math:init_random(1) + +#![create_features] +RealFeatures features(f_feats) +#![create_features] + +#![set_parameters] +JediSep ica() +ica.set_max_iter(200) +ica.set_tol(0.00001) +#![set_parameters] + +#![apply_convert] +Features converted = ica.apply(features) +#![apply_convert] + +#![extract] +RealMatrix mixing_matrix = ica.get_mixing_matrix() +RealFeatures casted = RealFeatures:obtain_from_generic(converted) +RealMatrix unmixed_signal = casted.get_feature_matrix() +#![extract] + diff --git a/examples/meta/src/converter/ica_sobi.sg b/examples/meta/src/converter/ica_sobi.sg new file mode 100644 index 00000000000..51ddd07c713 --- /dev/null +++ b/examples/meta/src/converter/ica_sobi.sg @@ -0,0 +1,24 @@ +CSVFile f_feats("../../data/ica_2_sources.dat") + +Math:init_random(1) + +#![create_features] +RealFeatures features(f_feats) +#![create_features] + +#![set_parameters] +SOBI ica() +ica.set_max_iter(200) +ica.set_tol(0.00001) +#![set_parameters] + +#![apply_convert] +Features converted = ica.apply(features) +#![apply_convert] + +#![extract] +RealMatrix mixing_matrix = ica.get_mixing_matrix() +RealFeatures casted = RealFeatures:obtain_from_generic(converted) +RealMatrix unmixed_signal = casted.get_feature_matrix() +#![extract] + diff --git a/examples/meta/src/evaluation/cross_validation_mkl_weight_storage.sg b/examples/meta/src/evaluation/cross_validation_mkl_weight_storage.sg new file mode 100644 index 00000000000..fc78600bd8d --- /dev/null +++ b/examples/meta/src/evaluation/cross_validation_mkl_weight_storage.sg @@ -0,0 +1,62 @@ +CSVFile f_feats("../../data/fm_train_real.dat") +CSVFile f_labels("../../data/label_train_twoclass.dat") + +#![create_features] +RealFeatures features(f_feats) +CombinedFeatures combined_features() +combined_features.append_feature_obj(features) +combined_features.append_feature_obj(features) +combined_features.append_feature_obj(features) +BinaryLabels labels(f_labels) +#![create_features] + +#![create_kernels] +CombinedKernel kernel() +GaussianKernel k_1(10, 0.1) +GaussianKernel k_2(10, 1) +GaussianKernel k_3(10, 2) +kernel.append_kernel(k_1) +kernel.append_kernel(k_2) +kernel.append_kernel(k_3) +#![create_kernels] + +#![create_classifier] +LibSVM libsvm() +MKLClassification svm(libsvm) +svm.set_interleaved_optimization_enabled(False) +svm.set_kernel(kernel) +#![create_classifier] + +#![create_cross_validation] +StratifiedCrossValidationSplitting splitting_strategy(labels, 5) +AccuracyMeasure evaluation_criterium() +CrossValidation cross(svm, combined_features, labels, splitting_strategy, evaluation_criterium) +cross.set_autolock(False) +cross.set_num_runs(2) +#![create_cross_validation] + +#![create_observer] +ParameterObserverCV mkl_obs(True) +cross.subscribe_to_parameters(mkl_obs) +#[!create_observer] + +#![evaluate_and_get_result] +CrossValidationResult result() +result = CrossValidationResult:obtain_from_generic(cross.evaluate()) +#![evaluate_and_get_result] + +#![get_results] +real mean = result.get_mean() +real stddev = result.get_std_dev() +#![get_results] + +#![get_fold_machine] +CrossValidationStorage obs = mkl_obs.get_observation(0) +CrossValidationFoldStorage fold = obs.get_fold(0) +MKLClassification machine = MKLClassification:obtain_from_generic(fold.get_trained_machine()) +#![get_fold_machine] + +#![get_weights] +CombinedKernel k = CombinedKernel:obtain_from_generic(machine.get_kernel()) +RealVector w = k.get_subkernel_weights() +#![get_weights] \ No newline at end of file diff --git a/examples/meta/src/meta_api/calls.sg b/examples/meta/src/meta_api/calls.sg new file mode 100644 index 00000000000..14b0ef034e6 --- /dev/null +++ b/examples/meta/src/meta_api/calls.sg @@ -0,0 +1,9 @@ +# static call +Math:init_random(1) + +# global function call +get_global_io() + +# member function call +GaussianKernel k() +k.set_width(1) diff --git a/examples/meta/src/multiclass_classifier/multiclass_ecoc_random.sg b/examples/meta/src/multiclass_classifier/multiclass_ecoc_random.sg index 79e79d4a10d..8d9f459ae19 100644 --- a/examples/meta/src/multiclass_classifier/multiclass_ecoc_random.sg +++ b/examples/meta/src/multiclass_classifier/multiclass_ecoc_random.sg @@ -1,8 +1,9 @@ +Math:init_random(1) + CSVFile f_feats_train("../../data/classifier_4class_2d_linear_features_train.dat") CSVFile f_feats_test("../../data/classifier_4class_2d_linear_features_test.dat") CSVFile f_labels_train("../../data/classifier_4class_2d_linear_labels_train.dat") CSVFile f_labels_test("../../data/classifier_4class_2d_linear_labels_test.dat") -Math:init_random(1) #![create_features] RealFeatures features_train(f_feats_train) diff --git a/examples/meta/src/regression/krr_nystrom.sg b/examples/meta/src/regression/krr_nystrom.sg index de84de05f40..8e651753fd0 100644 --- a/examples/meta/src/regression/krr_nystrom.sg +++ b/examples/meta/src/regression/krr_nystrom.sg @@ -27,12 +27,12 @@ RegressionLabels labels_predict = nystrom.apply_regression(features_test) #![train_and_apply] #![extract_alpha] -RealVector alpha = nystrom.get_alphas() +nystrom.get_alphas() #![extract_alpha] #![evaluate_error] MeanSquaredError eval() -real mse = eval.evaluate(labels_predict, labels_test) +eval.evaluate(labels_predict, labels_test) #![evaluate_error] # integration testing variables diff --git a/examples/minimal/CMakeLists.txt b/examples/minimal/CMakeLists.txt new file mode 100644 index 00000000000..c5eb8fa789e --- /dev/null +++ b/examples/minimal/CMakeLists.txt @@ -0,0 +1,11 @@ +FILE(GLOB EXAMPLES_CPP RELATIVE "${CMAKE_CURRENT_SOURCE_DIR}" "${CMAKE_CURRENT_SOURCE_DIR}/*.cpp") + +FOREACH(EXAMPLE_CPP ${EXAMPLES_CPP}) + STRING(REGEX REPLACE ".cpp\$" "" EXAMPLE "${EXAMPLE_CPP}") + AddLibShogunExample(${EXAMPLE_CPP}) + LIST(APPEND INSTALL_EXAMPLES ${CMAKE_CURRENT_BINARY_DIR}/${EXAMPLE}) +ENDFOREACH() + +INSTALL(PROGRAMS ${INSTALL_EXAMPLES} + DESTINATION share/shogun/examples/libshogun/ + COMPONENT minimal-example) diff --git a/examples/undocumented/libshogun/basic_minimal.cpp b/examples/minimal/minimal.cpp similarity index 100% rename from examples/undocumented/libshogun/basic_minimal.cpp rename to examples/minimal/minimal.cpp diff --git a/examples/undocumented/libshogun/CMakeLists.txt b/examples/undocumented/libshogun/CMakeLists.txt index fe114912e0c..92bedd0631c 100644 --- a/examples/undocumented/libshogun/CMakeLists.txt +++ b/examples/undocumented/libshogun/CMakeLists.txt @@ -4,6 +4,16 @@ IF (NOT USE_MOSEK) LIST(REMOVE_ITEM EXAMPLES_CPP "so_fg_model.cpp" "so_multiclass.cpp" "so_multilabel.cpp") ENDIF () +IF (NOT USE_GPL_SHOGUN) + LIST(REMOVE_ITEM EXAMPLES_CPP + "modelselection_model_selection_parameters_test.cpp" + "modelselection_parameter_tree.cpp" + "modelselection_grid_search_string_kernel.cpp" + "parameter_modsel_parameters.cpp" + "classifier_svmlight_string_features_precomputed_kernel.cpp" + ) +ENDIF() + LIST(REMOVE_ITEM EXAMPLES_CPP "classifier_bagging_liblinear.cpp" "classifier_larank.cpp" @@ -37,23 +47,8 @@ LIST(REMOVE_ITEM EXAMPLES_CPP FOREACH(EXAMPLE_CPP ${EXAMPLES_CPP}) STRING(REGEX REPLACE ".cpp\$" "" EXAMPLE "${EXAMPLE_CPP}") - - add_executable(${EXAMPLE} ${CMAKE_CURRENT_SOURCE_DIR}/${EXAMPLE_CPP}) - target_link_libraries(${EXAMPLE} shogun::shogun ${SANITIZER_LIBRARY}) - IF(SANITIZER_FLAGS) - set_target_properties(${EXAMPLE} PROPERTIES COMPILE_FLAGS ${SANITIZER_FLAGS}) - ENDIF() - add_test(libshogun-${EXAMPLE} ${CMAKE_CURRENT_BINARY_DIR}/${EXAMPLE}) + AddLibShogunExample(${EXAMPLE_CPP}) LIST(APPEND INSTALL_EXAMPLES ${CMAKE_CURRENT_BINARY_DIR}/${EXAMPLE}) - - # Add examples to the dependencies of modular interfaces to make sure - # nothing will infer with them being build single-threaded. - IF(SWIG_SINGLE_THREADED) - FOREACH(SG_MODULAR_INTERFACE_TARGET ${SG_MODULAR_INTERFACE_TARGETS}) - ADD_DEPENDENCIES(${SG_MODULAR_INTERFACE_TARGET} ${EXAMPLE}) - ENDFOREACH(SG_MODULAR_INTERFACE_TARGET ${SG_MODULAR_INTERFACE_TARGETS}) - ENDIF(SWIG_SINGLE_THREADED) - ENDFOREACH() INSTALL(PROGRAMS ${INSTALL_EXAMPLES} diff --git a/examples/undocumented/libshogun/classifier_latent_svm.cpp b/examples/undocumented/libshogun/classifier_latent_svm.cpp index 626d7f1e587..18471b8d506 100644 --- a/examples/undocumented/libshogun/classifier_latent_svm.cpp +++ b/examples/undocumented/libshogun/classifier_latent_svm.cpp @@ -1,10 +1,11 @@ -#include +#include +#include +#include #include +#include +#include #include -#include -#include #include -#include #include #include @@ -113,6 +114,7 @@ static void read_dataset(char* fname, CLatentFeatures*& feats, CLatentLabels*& l feats = new CLatentFeatures(num_examples); SG_REF(feats); + auto pb = progress(range(num_examples)); CMath::init_random(); for (int i = 0; (!feof(fd)) && (i < num_examples); ++i) { @@ -150,7 +152,7 @@ static void read_dataset(char* fname, CLatentFeatures*& feats, CLatentLabels*& l CBoundingBox* bb = new CBoundingBox(x,y); labels->add_latent_label(bb); - SG_SPROGRESS(i, 0, num_examples); + pb.print_progress(); CHOGFeatures* hog = new CHOGFeatures(width, height); hog->hog = SG_CALLOC(float64_t**, hog->width); for (int j = 0; j < width; ++j) @@ -176,7 +178,7 @@ static void read_dataset(char* fname, CLatentFeatures*& feats, CLatentLabels*& l labels->set_labels(ys); - SG_SDONE(); + pb.complete(); } int main(int argc, char** argv) diff --git a/examples/undocumented/libshogun/evaluation_cross_validation_mkl_weight_storage.cpp b/examples/undocumented/libshogun/evaluation_cross_validation_mkl_weight_storage.cpp index 7bc3e500f91..dd042371127 100644 --- a/examples/undocumented/libshogun/evaluation_cross_validation_mkl_weight_storage.cpp +++ b/examples/undocumented/libshogun/evaluation_cross_validation_mkl_weight_storage.cpp @@ -8,17 +8,16 @@ */ #include -#include -#include -#include -#include #include #include +#include #include -#include -#include #include -#include +#include +#include +#include +#include +#include #include using namespace shogun; @@ -50,6 +49,40 @@ void gen_rand_data(SGVector lab, SGMatrix feat, feat.display_matrix("feat"); } +SGMatrix calculate_weights( + CParameterObserverCV& obs, int32_t folds, int32_t run, int32_t len) +{ + int32_t column = 0; + SGMatrix weights(len, folds * run); + for (int o = 0; o < obs.get_num_observations(); o++) + { + auto obs_storage = obs.get_observation(o); + for (int i = 0; i < obs_storage->get_num_folds(); i++) + { + auto fold = obs_storage->get_fold(i); + CMKLClassification* machine = + (CMKLClassification*)fold->get_trained_machine(); + SG_REF(machine) + CCombinedKernel* k = (CCombinedKernel*)machine->get_kernel(); + auto w = k->get_subkernel_weights(); + + /* Copy the weights inside the matrix */ + /* Each of the columns will represent a set of weights */ + for (auto j = 0; j < w.size(); j++) + { + weights.set_element(w[j], j, column); + } + + SG_UNREF(k) + SG_UNREF(machine) + SG_UNREF(fold) + column++; + } + SG_UNREF(obs_storage) + } + return weights; +} + void test_mkl_cross_validation() { /* generate random data */ @@ -97,16 +130,14 @@ void test_mkl_cross_validation() CCrossValidation* cross=new CCrossValidation(svm, comb_features, labels, split, eval, false); /* add print output listener and mkl storage listener */ - cross->add_cross_validation_output(new CCrossValidationPrintOutput()); - CCrossValidationMKLStorage* mkl_storage=new CCrossValidationMKLStorage(); - cross->add_cross_validation_output(mkl_storage); + CParameterObserverCV mkl_obs{true}; + cross->subscribe_to_parameters(&mkl_obs); - /* perform cross-validation, this will print loads of information - * (caused by the CCrossValidationPrintOutput instance attached to it) */ + /* perform cross-validation, this will print loads of information */ CEvaluationResult* result=cross->evaluate(); /* print mkl weights */ - SGMatrix weights=mkl_storage->get_mkl_weights(); + auto weights = calculate_weights(mkl_obs, num_folds, 1, 3); weights.display_matrix("mkl weights"); /* print mean and variance of each kernel weight. These could for example @@ -115,6 +146,8 @@ void test_mkl_cross_validation() CStatistics::matrix_variance(weights, false).display_vector("variance per kernel"); CStatistics::matrix_std_deviation(weights, false).display_vector("std-dev per kernel"); + /* Clear */ + mkl_obs.clear(); SG_UNREF(result); /* again for two runs */ @@ -122,14 +155,17 @@ void test_mkl_cross_validation() result=cross->evaluate(); /* print mkl weights */ - weights=mkl_storage->get_mkl_weights(); - weights.display_matrix("mkl weights"); + SGMatrix weights_2 = calculate_weights(mkl_obs, num_folds, 2, 3); + weights_2.display_matrix("mkl weights"); /* print mean and variance of each kernel weight. These could for example * been used to compute confidence intervals */ - CStatistics::matrix_mean(weights, false).display_vector("mean per kernel"); - CStatistics::matrix_variance(weights, false).display_vector("variance per kernel"); - CStatistics::matrix_std_deviation(weights, false).display_vector("std-dev per kernel"); + CStatistics::matrix_mean(weights_2, false) + .display_vector("mean per kernel"); + CStatistics::matrix_variance(weights_2, false) + .display_vector("variance per kernel"); + CStatistics::matrix_std_deviation(weights_2, false) + .display_vector("std-dev per kernel"); /* clean up */ SG_UNREF(result); diff --git a/examples/undocumented/libshogun/evaluation_cross_validation_multiclass_mkl.cpp b/examples/undocumented/libshogun/evaluation_cross_validation_multiclass_mkl.cpp index eb83866eaf0..39e43638690 100644 --- a/examples/undocumented/libshogun/evaluation_cross_validation_multiclass_mkl.cpp +++ b/examples/undocumented/libshogun/evaluation_cross_validation_multiclass_mkl.cpp @@ -97,11 +97,12 @@ void test_multiclass_mkl_cv() /* perform x-val and print result */ CCrossValidationResult* result=(CCrossValidationResult*)cross->evaluate(); - SG_SPRINT("mean of %d %d-fold x-val runs: %f\n", n_runs, n_folds, - result->mean); + SG_SPRINT( + "mean of %d %d-fold x-val runs: %f\n", n_runs, n_folds, + result->get_mean()); /* assert high accuracy */ - ASSERT(result->mean>0.81); + ASSERT(result->get_mean() > 0.81); /* clean up */ SG_UNREF(features); diff --git a/examples/undocumented/libshogun/evaluation_cross_validation_regression.cpp b/examples/undocumented/libshogun/evaluation_cross_validation_regression.cpp index abd7490a82f..4b0faf98c33 100644 --- a/examples/undocumented/libshogun/evaluation_cross_validation_regression.cpp +++ b/examples/undocumented/libshogun/evaluation_cross_validation_regression.cpp @@ -105,7 +105,7 @@ void test_cross_validation() result->print_result(); /* same crude assertion as for above evaluation */ - ASSERT(result->mean<2); + ASSERT(result->get_mean() < 2); /* clean up */ SG_UNREF(result); diff --git a/examples/undocumented/libshogun/library_dyn_int.cpp b/examples/undocumented/libshogun/library_dyn_int.cpp deleted file mode 100644 index 82a0d33889f..00000000000 --- a/examples/undocumented/libshogun/library_dyn_int.cpp +++ /dev/null @@ -1,125 +0,0 @@ -/* - * This program is free software; you can redistribute it and/or modify - * it under the terms of the GNU General Public License as published by - * the Free Software Foundation; either version 3 of the License, or - * (at your option) any later version. - * - * Written (W) 2009 Soeren Sonnenburg - * Copyright (C) 2009 Fraunhofer Institute FIRST and Max-Planck-Society - */ - -#include -#include -#include -#include -#include -#include - -using namespace shogun; - -void print_message(FILE* target, const char* str) -{ - fprintf(target, "%s", str); -} - -void print_warning(FILE* target, const char* str) -{ - fprintf(target, "%s", str); -} - -void print_error(FILE* target, const char* str) -{ - fprintf(target, "%s", str); -} - -void gen_ints(uint256_t* &a, uint32_t* &b, uint32_t len) -{ - a=SG_MALLOC(uint256_t, len); - b=SG_MALLOC(uint32_t, len); - - CMath::init_random(17); - - for (uint32_t i=0; iset_loglevel(MSG_DEBUG); - - SG_SPRINT("gen data.."); - t.start(); - gen_ints(a,b, LEN); - t.cur_time_diff(true); - - SG_SPRINT("qsort.."); - t.start(); - CMath::qsort_index(a, b, LEN); - t.cur_time_diff(true); - - SG_SPRINT("\n\n"); - for (uint32_t i=0; i<10; i++) - { - SG_SPRINT("a[%d]=", i); - a[i].print_hex(); - SG_SPRINT("\n"); - } - - SG_SPRINT("\n\n"); - - uint64_t val1[4]={1,2,3,4}; - uint64_t val2[4]={5,6,7,8}; - a[0]=val1; - a[1]=val2; - a[2]=a[0]; - CMath::swap(a[0],a[1]); - - printf("a[0]==a[1] %d\n", (int) (a[0] == a[1])); - printf("a[0]a[1] %d\n", (int) (a[0] > a[1])); - printf("a[0]>=a[1] %d\n", (int) (a[0] >= a[1])); - - printf("a[0]==a[0] %d\n", (int) (a[0] == a[0])); - printf("a[0]a[0] %d\n", (int) (a[0] > a[0])); - printf("a[0]>=a[0] %d\n", (int) (a[0] >= a[0])); - - SG_SPRINT("\n\n"); - for (uint32_t i=0; i<10 ; i++) - { - SG_SPRINT("a[%d]=", i); - a[i].print_hex(); - printf("\n"); - } - - SG_FREE(a); - SG_FREE(b); - } - catch(ShogunException & sh) - { - SG_SPRINT("%s",sh.get_exception_string()); - } - - exit_shogun(); - - return 0; -} diff --git a/examples/undocumented/libshogun/library_indirect_object.cpp b/examples/undocumented/libshogun/library_indirect_object.cpp deleted file mode 100644 index edcd3344570..00000000000 --- a/examples/undocumented/libshogun/library_indirect_object.cpp +++ /dev/null @@ -1,45 +0,0 @@ -#include -#include -#include -#include -#include -#include - -#include - -using namespace shogun; - -const int l=10; - -int main(int argc, char** argv) -{ - init_shogun(); - - // create array a - int32_t* a=SG_MALLOC(int32_t, l); - for (int i=0; i INDIRECT; - // create array of indirect objects pointing to array a - INDIRECT::set_array(&a); - INDIRECT* x = SG_MALLOC(INDIRECT, l); - INDIRECT::init_slice(x, l); - - - printf("created array a and indirect object array x pointing to a.\n\n"); - for (int i=0; i -#include -#include -#include -#include -#include - -using namespace shogun; - -int32_t max=3; -const float64_t initial_value=1; -const float64_t another_value=2; - -void print_message(FILE* target, const char* str) -{ - fprintf(target, "%s", str); -} - -bool test_float_scalar() -{ - bool result=true; - - Parameter* original_parameter_list=new Parameter(); - float64_t original_parameter=initial_value; - original_parameter_list->add(&original_parameter, "param", ""); - - float64_t new_parameter=another_value; - Parameter* new_parameter_list=new Parameter(); - new_parameter_list->add(&new_parameter, "param", ""); - - original_parameter_list->set_from_parameters(new_parameter_list); - - result&=original_parameter==another_value; - - delete original_parameter_list; - delete new_parameter_list; - - return result; -} - -bool test_float_vector() -{ - bool result=true; - - Parameter* original_parameter_list=new Parameter(); - float64_t* original_parameter=SG_MALLOC(float64_t, max); - SGVector::fill_vector(original_parameter, max, initial_value); - - original_parameter_list->add_vector(&original_parameter, &max, "param", ""); - - float64_t* new_parameter=SG_MALLOC(float64_t, max); - SGVector::fill_vector(new_parameter, max, another_value); - - Parameter* new_parameter_list=new Parameter(); - new_parameter_list->add_vector(&new_parameter, &max, "param", ""); - - original_parameter_list->set_from_parameters(new_parameter_list); - - for (int32_t i=0; i::fill_vector(original_parameter, max*max, initial_value); - - original_parameter_list->add_matrix(&original_parameter, &max, &max, "param", ""); - - float64_t* new_parameter=SG_MALLOC(float64_t, max*max); - SGVector::fill_vector(new_parameter, max*max, another_value); - - Parameter* new_parameter_list=new Parameter(); - new_parameter_list->add_matrix(&new_parameter, &max, &max, "param", ""); - - original_parameter_list->set_from_parameters(new_parameter_list); - - for (int32_t i=0; iadd(&original_parameter, "kernel", ""); - - CSGObject* new_parameter=new CDistantSegmentsKernel(10, 10, 10); - Parameter* new_parameter_list=new Parameter(); - new_parameter_list->add(&new_parameter, "kernel", ""); - - /* note: old_parameter is SG_UNREF'ed, new one SG_REF'ed */ - original_parameter_list->set_from_parameters(new_parameter_list); - - result&=original_parameter==new_parameter; - - /* old original kernel was deleted by shogun's SG_UNREF */ - SG_UNREF(new_parameter); - delete original_parameter_list; - delete new_parameter_list; - - return result; -} - -bool test_sgobject_vector() -{ - bool result=true; - - Parameter* original_parameter_list=new Parameter(); - CSGObject** original_parameter=SG_MALLOC(CSGObject*, max); - for (int32_t i=0; iadd_vector(&original_parameter, &max, "param", ""); - - CSGObject** new_parameter=SG_MALLOC(CSGObject*, max); - for (int32_t i=0; iadd_vector(&new_parameter, &max, "param", ""); - - /* note: old_parameters are SG_UNREF'ed, new ones SG_REF'ed */ - original_parameter_list->set_from_parameters(new_parameter_list); - - for (int32_t i=0; iadd_matrix(&original_parameter, &max, &max, "param", ""); - - CSGObject** new_parameter=SG_MALLOC(CSGObject*, max*max); - for (int32_t i=0; iadd_matrix(&new_parameter, &max, &max, "param", ""); - - /* note: old_parameters are SG_UNREF'ed, new ones SG_REF'ed */ - original_parameter_list->set_from_parameters(new_parameter_list); - - for (int32_t i=0; i* features_test=new CDenseFeatures( feat_test); - CGaussianKernel* kernel=new CGaussianKernel(kernel_cache, rbf_width); + CGaussianKernel* kernel=new CGaussianKernel(rbf_width); kernel->init(features_train, features_train); // also epsilon svr possible here diff --git a/examples/undocumented/python/CMakeLists.txt b/examples/undocumented/python/CMakeLists.txt new file mode 100644 index 00000000000..3ded2f7410e --- /dev/null +++ b/examples/undocumented/python/CMakeLists.txt @@ -0,0 +1,25 @@ +INCLUDE(PythonEnvironment) +GET_PYTHON_ENV() + +FILE(GLOB PYTHON_EXAMPLES RELATIVE "${CMAKE_CURRENT_SOURCE_DIR}" "${CMAKE_CURRENT_SOURCE_DIR}/*.py") + +IF (NOT USE_GPL_SHOGUN) + LIST(REMOVE_ITEM PYTHON_EXAMPLES + "classifier_svmlin.py" + "kernel_distantsegments.py" + "modelselection_parameter_tree.py" + "variational_classifier.py" + "evaluation_cross_validation_multiclass_storage.py" + ) +ENDIF() + +FOREACH(EXAMPLE_PY ${PYTHON_EXAMPLES}) + STRING(REGEX REPLACE ".py\$" "" EXAMPLE "${EXAMPLE_PY}") + MESSAGE(${EXAMPLE_PY}) + + add_test(NAME python_legacy-${EXAMPLE} + WORKING_DIRECTORY ${CMAKE_CURRENT_SOURCE_DIR} + COMMAND ${PYTHON_EXECUTABLE} ${EXAMPLE_PY}) + set_property(TEST python_legacy-${EXAMPLE} PROPERTY + ENVIRONMENT "${PYTHON_ENV_VARS}") +ENDFOREACH() \ No newline at end of file diff --git a/examples/undocumented/python_modular/classifier_custom_kernel_modular.py b/examples/undocumented/python/classifier_custom_kernel.py similarity index 73% rename from examples/undocumented/python_modular/classifier_custom_kernel_modular.py rename to examples/undocumented/python/classifier_custom_kernel.py index 699926b2ba1..89e3ffef759 100644 --- a/examples/undocumented/python_modular/classifier_custom_kernel_modular.py +++ b/examples/undocumented/python/classifier_custom_kernel.py @@ -1,8 +1,8 @@ #!/usr/bin/env python parameter_list = [[1,7],[2,8]] -def classifier_custom_kernel_modular (C=1,dim=7): - from modshogun import RealFeatures, BinaryLabels, CustomKernel, LibSVM +def classifier_custom_kernel (C=1,dim=7): + from shogun import RealFeatures, BinaryLabels, CustomKernel, LibSVM from numpy import diag,ones,sign from numpy.random import rand,seed @@ -23,4 +23,4 @@ def classifier_custom_kernel_modular (C=1,dim=7): if __name__=='__main__': print('custom_kernel') - classifier_custom_kernel_modular(*parameter_list[0]) + classifier_custom_kernel(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/classifier_domainadaptationsvm_modular.py b/examples/undocumented/python/classifier_domainadaptationsvm.py similarity index 89% rename from examples/undocumented/python_modular/classifier_domainadaptationsvm_modular.py rename to examples/undocumented/python/classifier_domainadaptationsvm.py index 3b28ff00199..a92f7f98f98 100644 --- a/examples/undocumented/python_modular/classifier_domainadaptationsvm_modular.py +++ b/examples/undocumented/python/classifier_domainadaptationsvm.py @@ -1,17 +1,17 @@ #!/usr/bin/env python import numpy -from modshogun import StringCharFeatures, BinaryLabels, DNA -from modshogun import WeightedDegreeStringKernel -from modshogun import MSG_DEBUG +from shogun import StringCharFeatures, BinaryLabels, DNA +from shogun import WeightedDegreeStringKernel +from shogun import MSG_DEBUG try: - from modshogun import DomainAdaptationSVM + from shogun import DomainAdaptationSVM except ImportError: print("DomainAdaptationSVM not available") exit(0) try: - from modshogun import SVMLight + from shogun import SVMLight except ImportError: print("SVMLight not available") exit(0) @@ -67,7 +67,7 @@ testdna2,label_testdna2,1,3],[traindna,testdna,label_traindna,label_testdna,traindna2,label_traindna2, \ testdna2,label_testdna2,2,5]] -def classifier_domainadaptationsvm_modular (fm_train_dna=traindna,fm_test_dna=testdna, \ +def classifier_domainadaptationsvm (fm_train_dna=traindna,fm_test_dna=testdna, \ label_train_dna=label_traindna, \ label_test_dna=label_testdna,fm_train_dna2=traindna2,fm_test_dna2=testdna2, \ label_train_dna2=label_traindna2,label_test_dna2=label_testdna2,C=1,degree=3): @@ -102,4 +102,4 @@ def classifier_domainadaptationsvm_modular (fm_train_dna=traindna,fm_test_dna=te if __name__=='__main__': print('SVMLight') - classifier_domainadaptationsvm_modular(*parameter_list[0]) + classifier_domainadaptationsvm(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/classifier_featureblock_logistic_regression.py b/examples/undocumented/python/classifier_featureblock_logistic_regression.py similarity index 90% rename from examples/undocumented/python_modular/classifier_featureblock_logistic_regression.py rename to examples/undocumented/python/classifier_featureblock_logistic_regression.py index 2f52c02b1f0..1e4da7837fa 100644 --- a/examples/undocumented/python_modular/classifier_featureblock_logistic_regression.py +++ b/examples/undocumented/python/classifier_featureblock_logistic_regression.py @@ -12,9 +12,9 @@ def classifier_featureblock_logistic_regression (fm_train=traindat,fm_test=testdat,label_train=label_traindat): - from modshogun import BinaryLabels, RealFeatures, IndexBlock, IndexBlockGroup + from shogun import BinaryLabels, RealFeatures, IndexBlock, IndexBlockGroup try: - from modshogun import FeatureBlockLogisticRegression + from shogun import FeatureBlockLogisticRegression except ImportError: print("FeatureBlockLogisticRegression not available") exit(0) diff --git a/examples/undocumented/python_modular/classifier_gmnpsvm_modular.py b/examples/undocumented/python/classifier_gmnpsvm.py similarity index 69% rename from examples/undocumented/python_modular/classifier_gmnpsvm_modular.py rename to examples/undocumented/python/classifier_gmnpsvm.py index a0ca1149f7c..a3a1afb3f07 100644 --- a/examples/undocumented/python_modular/classifier_gmnpsvm_modular.py +++ b/examples/undocumented/python/classifier_gmnpsvm.py @@ -5,9 +5,9 @@ parameter_list = [[traindat,testdat,label_traindat,2.1,1,1e-5],[traindat,testdat,label_traindat,2.2,1,1e-5]] -def classifier_gmnpsvm_modular (train_fname=traindat,test_fname=testdat,label_fname=label_traindat,width=2.1,C=1,epsilon=1e-5): - from modshogun import RealFeatures, MulticlassLabels - from modshogun import GaussianKernel, GMNPSVM, CSVFile +def classifier_gmnpsvm (train_fname=traindat,test_fname=testdat,label_fname=label_traindat,width=2.1,C=1,epsilon=1e-5): + from shogun import RealFeatures, MulticlassLabels + from shogun import GaussianKernel, GMNPSVM, CSVFile feats_train=RealFeatures(CSVFile(train_fname)) feats_test=RealFeatures(CSVFile(test_fname)) @@ -23,4 +23,4 @@ def classifier_gmnpsvm_modular (train_fname=traindat,test_fname=testdat,label_fn return out,kernel if __name__=='__main__': print('GMNPSVM') - classifier_gmnpsvm_modular(*parameter_list[0]) + classifier_gmnpsvm(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/classifier_gpbtsvm_modular.py b/examples/undocumented/python/classifier_gpbtsvm.py similarity index 69% rename from examples/undocumented/python_modular/classifier_gpbtsvm_modular.py rename to examples/undocumented/python/classifier_gpbtsvm.py index f71b650ca29..f74d7ed26b0 100644 --- a/examples/undocumented/python_modular/classifier_gpbtsvm_modular.py +++ b/examples/undocumented/python/classifier_gpbtsvm.py @@ -5,12 +5,12 @@ parameter_list = [[traindat,testdat,label_traindat,2.1,1,1e-5],[traindat,testdat,label_traindat,2.2,1,1e-5]] -def classifier_gpbtsvm_modular (train_fname=traindat,test_fname=testdat,label_fname=label_traindat,width=2.1,C=1,epsilon=1e-5): - from modshogun import RealFeatures, BinaryLabels - from modshogun import GaussianKernel - from modshogun import CSVFile +def classifier_gpbtsvm (train_fname=traindat,test_fname=testdat,label_fname=label_traindat,width=2.1,C=1,epsilon=1e-5): + from shogun import RealFeatures, BinaryLabels + from shogun import GaussianKernel + from shogun import CSVFile try: - from modshogun import GPBTSVM + from shogun import GPBTSVM except ImportError: print("GPBTSVM not available") exit(0) @@ -30,4 +30,4 @@ def classifier_gpbtsvm_modular (train_fname=traindat,test_fname=testdat,label_fn if __name__=='__main__': print('GPBTSVM') - classifier_gpbtsvm_modular(*parameter_list[0]) + classifier_gpbtsvm(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/classifier_larank_modular.py b/examples/undocumented/python/classifier_larank.py similarity index 71% rename from examples/undocumented/python_modular/classifier_larank_modular.py rename to examples/undocumented/python/classifier_larank.py index 893c17884cd..a6a152d0f35 100644 --- a/examples/undocumented/python_modular/classifier_larank_modular.py +++ b/examples/undocumented/python/classifier_larank.py @@ -2,11 +2,11 @@ from numpy import * parameter_list = [[10,3,15,0.9,1,2000,1],[20,4,15,0.9,1,5000,2]] -def classifier_larank_modular (num_vec,num_class,distance,C=0.9,num_threads=1,num_iter=5,seed=1): - from modshogun import RealFeatures, MulticlassLabels - from modshogun import GaussianKernel - from modshogun import LaRank - from modshogun import Math_init_random +def classifier_larank (num_vec,num_class,distance,C=0.9,num_threads=1,num_iter=5,seed=1): + from shogun import RealFeatures, MulticlassLabels + from shogun import GaussianKernel + from shogun import LaRank + from shogun import Math_init_random # reproducible results Math_init_random(seed) @@ -18,8 +18,8 @@ def classifier_larank_modular (num_vec,num_class,distance,C=0.9,num_threads=1,nu fm_train=array(random.randn(num_class,num_vec)) fm_test=array(random.randn(num_class,num_vec)) for i in range(len(label_train)): - fm_train[label_train[i],i]+=distance - fm_test[label_test[i],i]+=distance + fm_train[int(label_train[i]),i]+=distance + fm_test[int(label_test[i]),i]+=distance feats_train=RealFeatures(fm_train) feats_test=RealFeatures(fm_test) @@ -43,4 +43,4 @@ def classifier_larank_modular (num_vec,num_class,distance,C=0.9,num_threads=1,nu if __name__=='__main__': print('LaRank') - [predictions, svm, labels] = classifier_larank_modular(*parameter_list[0]) + [predictions, svm, labels] = classifier_larank(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/classifier_lda_modular.py b/examples/undocumented/python/classifier_lda.py similarity index 72% rename from examples/undocumented/python_modular/classifier_lda_modular.py rename to examples/undocumented/python/classifier_lda.py index 2b1df46596d..917c1a8b2a7 100644 --- a/examples/undocumented/python_modular/classifier_lda_modular.py +++ b/examples/undocumented/python/classifier_lda.py @@ -5,8 +5,8 @@ parameter_list = [[traindat,testdat,label_traindat,3,1],[traindat,testdat,label_traindat,4,1]] -def classifier_lda_modular (train_fname=traindat,test_fname=testdat,label_fname=label_traindat,gamma=3,num_threads=1): - from modshogun import RealFeatures, BinaryLabels, LDA, CSVFile +def classifier_lda (train_fname=traindat,test_fname=testdat,label_fname=label_traindat,gamma=3,num_threads=1): + from shogun import RealFeatures, BinaryLabels, LDA, CSVFile feats_train=RealFeatures(CSVFile(train_fname)) feats_test=RealFeatures(CSVFile(test_fname)) @@ -22,4 +22,4 @@ def classifier_lda_modular (train_fname=traindat,test_fname=testdat,label_fname= if __name__=='__main__': print('LDA') - classifier_lda_modular(*parameter_list[0]) + classifier_lda(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/classifier_libsvmoneclass_modular.py b/examples/undocumented/python/classifier_libsvmoneclass.py similarity index 69% rename from examples/undocumented/python_modular/classifier_libsvmoneclass_modular.py rename to examples/undocumented/python/classifier_libsvmoneclass.py index d7aef5a8b8b..43126195910 100644 --- a/examples/undocumented/python_modular/classifier_libsvmoneclass_modular.py +++ b/examples/undocumented/python/classifier_libsvmoneclass.py @@ -4,8 +4,8 @@ parameter_list = [[traindat,testdat,2.2,1,1e-7],[traindat,testdat,2.1,1,1e-5]] -def classifier_libsvmoneclass_modular (train_fname=traindat,test_fname=testdat,width=2.1,C=1,epsilon=1e-5): - from modshogun import RealFeatures, GaussianKernel, LibSVMOneClass, CSVFile +def classifier_libsvmoneclass (train_fname=traindat,test_fname=testdat,width=2.1,C=1,epsilon=1e-5): + from shogun import RealFeatures, GaussianKernel, LibSVMOneClass, CSVFile feats_train=RealFeatures(CSVFile(train_fname)) feats_test=RealFeatures(CSVFile(test_fname)) @@ -21,4 +21,4 @@ def classifier_libsvmoneclass_modular (train_fname=traindat,test_fname=testdat,w if __name__=='__main__': print('LibSVMOneClass') - classifier_libsvmoneclass_modular(*parameter_list[0]) + classifier_libsvmoneclass(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/classifier_mpdsvm_modular.py b/examples/undocumented/python/classifier_mpdsvm.py similarity index 69% rename from examples/undocumented/python_modular/classifier_mpdsvm_modular.py rename to examples/undocumented/python/classifier_mpdsvm.py index 10a7ecc41ca..eb05a6ea52b 100644 --- a/examples/undocumented/python_modular/classifier_mpdsvm_modular.py +++ b/examples/undocumented/python/classifier_mpdsvm.py @@ -5,11 +5,11 @@ parameter_list = [[traindat,testdat,label_traindat,1,1e-5],[traindat,testdat,label_traindat,0.9,1e-5]] -def classifier_mpdsvm_modular (train_fname=traindat,test_fname=testdat,label_fname=label_traindat,C=1,epsilon=1e-5): +def classifier_mpdsvm (train_fname=traindat,test_fname=testdat,label_fname=label_traindat,C=1,epsilon=1e-5): - from modshogun import RealFeatures, BinaryLabels - from modshogun import GaussianKernel - from modshogun import MPDSVM, CSVFile + from shogun import RealFeatures, BinaryLabels + from shogun import GaussianKernel + from shogun import MPDSVM, CSVFile feats_train=RealFeatures(CSVFile(train_fname)) feats_test=RealFeatures(CSVFile(test_fname)) @@ -26,4 +26,4 @@ def classifier_mpdsvm_modular (train_fname=traindat,test_fname=testdat,label_fna if __name__=='__main__': print('MPDSVM') - classifier_mpdsvm_modular(*parameter_list[0]) + classifier_mpdsvm(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/classifier_multiclass_ecoc.py b/examples/undocumented/python/classifier_multiclass_ecoc.py similarity index 85% rename from examples/undocumented/python_modular/classifier_multiclass_ecoc.py rename to examples/undocumented/python/classifier_multiclass_ecoc.py index b80b4a9c557..4c6311c3b1e 100644 --- a/examples/undocumented/python_modular/classifier_multiclass_ecoc.py +++ b/examples/undocumented/python/classifier_multiclass_ecoc.py @@ -12,21 +12,21 @@ def classifier_multiclass_ecoc (fm_train_real=traindat,fm_test_real=testdat,label_train_multiclass=label_traindat,label_test_multiclass=label_testdat,lawidth=2.1,C=1,epsilon=1e-5): - import modshogun - from modshogun import ECOCStrategy, LibLinear, L2R_L2LOSS_SVC, LinearMulticlassMachine - from modshogun import MulticlassAccuracy - from modshogun import RealFeatures, MulticlassLabels + import shogun + from shogun import ECOCStrategy, LibLinear, L2R_L2LOSS_SVC, LinearMulticlassMachine + from shogun import MulticlassAccuracy + from shogun import RealFeatures, MulticlassLabels def nonabstract_class(name): try: - getattr(modshogun, name)() + getattr(shogun, name)() except TypeError: return False return True - encoders = [x for x in dir(modshogun) + encoders = [x for x in dir(shogun) if re.match(r'ECOC.+Encoder', x) and nonabstract_class(x)] - decoders = [x for x in dir(modshogun) + decoders = [x for x in dir(shogun) if re.match(r'ECOC.+Decoder', x) and nonabstract_class(x)] fea_train = RealFeatures(fm_train_real) @@ -46,8 +46,8 @@ def nonabstract_class(name): #print((format_str % ('s', 's', 's')) % ('encoder', 'decoder', 'codelen', 'time', 'accuracy')) def run_ecoc(ier, idr): - encoder = getattr(modshogun, encoders[ier])() - decoder = getattr(modshogun, decoders[idr])() + encoder = getattr(shogun, encoders[ier])() + decoder = getattr(shogun, decoders[idr])() # whether encoder is data dependent if hasattr(encoder, 'set_labels'): diff --git a/examples/undocumented/python_modular/classifier_multiclassliblinear_modular.py b/examples/undocumented/python/classifier_multiclassliblinear.py similarity index 68% rename from examples/undocumented/python_modular/classifier_multiclassliblinear_modular.py rename to examples/undocumented/python/classifier_multiclassliblinear.py index 1a0d09a0b26..cd6be65156e 100644 --- a/examples/undocumented/python_modular/classifier_multiclassliblinear_modular.py +++ b/examples/undocumented/python/classifier_multiclassliblinear.py @@ -5,9 +5,9 @@ parameter_list = [[traindat,testdat,label_traindat,label_testdat,2.1,1,1e-5],[traindat,testdat,label_traindat,label_testdat,2.2,1,1e-5]] -def classifier_multiclassliblinear_modular (fm_train_real=traindat,fm_test_real=testdat,label_train_multiclass=label_traindat,label_test_multiclass=label_testdat,width=2.1,C=1,epsilon=1e-5): - from modshogun import RealFeatures, MulticlassLabels - from modshogun import MulticlassLibLinear +def classifier_multiclassliblinear (fm_train_real=traindat,fm_test_real=testdat,label_train_multiclass=label_traindat,label_test_multiclass=label_testdat,width=2.1,C=1,epsilon=1e-5): + from shogun import RealFeatures, MulticlassLabels + from shogun import MulticlassLibLinear feats_train=RealFeatures(fm_train_real) feats_test=RealFeatures(fm_test_real) @@ -21,7 +21,7 @@ def classifier_multiclassliblinear_modular (fm_train_real=traindat,fm_test_real= out = label_pred.get_labels() if label_test_multiclass is not None: - from modshogun import MulticlassAccuracy + from shogun import MulticlassAccuracy labels_test = MulticlassLabels(label_test_multiclass) evaluator = MulticlassAccuracy() acc = evaluator.evaluate(label_pred, labels_test) @@ -31,4 +31,4 @@ def classifier_multiclassliblinear_modular (fm_train_real=traindat,fm_test_real= if __name__=='__main__': print('MulticlassLibLinear') - classifier_multiclassliblinear_modular(*parameter_list[0]) + classifier_multiclassliblinear(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/classifier_multiclassmachine_modular.py b/examples/undocumented/python/classifier_multiclassmachine.py similarity index 67% rename from examples/undocumented/python_modular/classifier_multiclassmachine_modular.py rename to examples/undocumented/python/classifier_multiclassmachine.py index 3c21f9168c2..1f670b084f6 100644 --- a/examples/undocumented/python_modular/classifier_multiclassmachine_modular.py +++ b/examples/undocumented/python/classifier_multiclassmachine.py @@ -5,10 +5,10 @@ parameter_list = [[traindat,testdat,label_traindat,2.1,1,1e-5],[traindat,testdat,label_traindat,2.2,1,1e-5]] -def classifier_multiclassmachine_modular (fm_train_real=traindat,fm_test_real=testdat,label_train_multiclass=label_traindat,width=2.1,C=1,epsilon=1e-5): - from modshogun import RealFeatures, MulticlassLabels - from modshogun import GaussianKernel - from modshogun import LibSVM, KernelMulticlassMachine, MulticlassOneVsRestStrategy +def classifier_multiclassmachine (fm_train_real=traindat,fm_test_real=testdat,label_train_multiclass=label_traindat,width=2.1,C=1,epsilon=1e-5): + from shogun import RealFeatures, MulticlassLabels + from shogun import GaussianKernel + from shogun import LibSVM, KernelMulticlassMachine, MulticlassOneVsRestStrategy feats_train=RealFeatures(fm_train_real) feats_test=RealFeatures(fm_test_real) @@ -28,4 +28,4 @@ def classifier_multiclassmachine_modular (fm_train_real=traindat,fm_test_real=te if __name__=='__main__': print('MulticlassMachine') - classifier_multiclassmachine_modular(*parameter_list[0]) + classifier_multiclassmachine(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/classifier_multiclassocas_modular.py b/examples/undocumented/python/classifier_multiclassocas.py similarity index 71% rename from examples/undocumented/python_modular/classifier_multiclassocas_modular.py rename to examples/undocumented/python/classifier_multiclassocas.py index 3f0cbb96b45..f4984854f0d 100644 --- a/examples/undocumented/python_modular/classifier_multiclassocas_modular.py +++ b/examples/undocumented/python/classifier_multiclassocas.py @@ -2,11 +2,11 @@ from numpy import * parameter_list = [[10,3,15,2.1,1,1e-5,1],[20,4,15,2.2,2,1e-5,2]] -def classifier_multiclassocas_modular (num_vec=10,num_class=3,distance=15,width=2.1,C=1,epsilon=1e-5,seed=1): - from modshogun import RealFeatures, MulticlassLabels - from modshogun import Math_init_random +def classifier_multiclassocas (num_vec=10,num_class=3,distance=15,width=2.1,C=1,epsilon=1e-5,seed=1): + from shogun import RealFeatures, MulticlassLabels + from shogun import Math_init_random try: - from modshogun import MulticlassOCAS + from shogun import MulticlassOCAS except ImportError: print("MulticlassOCAS not available") return @@ -21,8 +21,8 @@ def classifier_multiclassocas_modular (num_vec=10,num_class=3,distance=15,width= fm_train=array(random.randn(num_class,num_vec)) fm_test=array(random.randn(num_class,num_vec)) for i in range(len(label_train)): - fm_train[label_train[i],i]+=distance - fm_test[label_test[i],i]+=distance + fm_train[int(label_train[i]),i]+=distance + fm_test[int(label_test[i]),i]+=distance feats_train=RealFeatures(fm_train) feats_test=RealFeatures(fm_test) @@ -39,4 +39,4 @@ def classifier_multiclassocas_modular (num_vec=10,num_class=3,distance=15,width= if __name__=='__main__': print('MulticlassOCAS') - classifier_multiclassocas_modular(*parameter_list[0]) + classifier_multiclassocas(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/classifier_multilabeloutputliblinear_modular.py b/examples/undocumented/python/classifier_multilabeloutputliblinear.py similarity index 64% rename from examples/undocumented/python_modular/classifier_multilabeloutputliblinear_modular.py rename to examples/undocumented/python/classifier_multilabeloutputliblinear.py index 9d4cd603fd2..57099cf9ec2 100644 --- a/examples/undocumented/python_modular/classifier_multilabeloutputliblinear_modular.py +++ b/examples/undocumented/python/classifier_multilabeloutputliblinear.py @@ -5,9 +5,9 @@ parameter_list = [[traindat,testdat,label_traindat,label_testdat,2.1,1,1e-5],[traindat,testdat,label_traindat,label_testdat,2.2,1,1e-5]] -def classifier_multilabeloutputliblinear_modular (fm_train_real=traindat,fm_test_real=testdat,label_train_multiclass=label_traindat,label_test_multiclass=label_testdat,width=2.1,C=1,epsilon=1e-5): - from modshogun import RealFeatures, MulticlassLabels, MultilabelLabels - from modshogun import MulticlassLibLinear +def classifier_multilabeloutputliblinear (fm_train_real=traindat,fm_test_real=testdat,label_train_multiclass=label_traindat,label_test_multiclass=label_testdat,width=2.1,C=1,epsilon=1e-5): + from shogun import RealFeatures, MulticlassLabels, MultilabelLabels + from shogun import MulticlassLibLinear feats_train=RealFeatures(fm_train_real) feats_test=RealFeatures(fm_test_real) @@ -24,4 +24,4 @@ def classifier_multilabeloutputliblinear_modular (fm_train_real=traindat,fm_test if __name__=='__main__': print('MultilabelOutputLibLinear') - classifier_multilabeloutputliblinear_modular(*parameter_list[0]) + classifier_multilabeloutputliblinear(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/classifier_perceptron_modular.py b/examples/undocumented/python/classifier_perceptron.py similarity index 81% rename from examples/undocumented/python_modular/classifier_perceptron_modular.py rename to examples/undocumented/python/classifier_perceptron.py index 716903368fd..7d6743f40a5 100644 --- a/examples/undocumented/python_modular/classifier_perceptron_modular.py +++ b/examples/undocumented/python/classifier_perceptron.py @@ -4,9 +4,9 @@ parameter_list = [[100, 2, 5,1.,1000,1,1], [100, 2, 5,1.,1000,1,2]] -def classifier_perceptron_modular (n=100, dim=2, distance=5,learn_rate=1.,max_iter=1000,num_threads=1,seed=1): - from modshogun import RealFeatures, BinaryLabels - from modshogun import Perceptron +def classifier_perceptron (n=100, dim=2, distance=5,learn_rate=1.,max_iter=1000,num_threads=1,seed=1): + from shogun import RealFeatures, BinaryLabels + from shogun import Perceptron random.seed(seed) @@ -39,4 +39,4 @@ def classifier_perceptron_modular (n=100, dim=2, distance=5,learn_rate=1.,max_it if __name__=='__main__': print('Perceptron') - classifier_perceptron_modular(*parameter_list[0]) + classifier_perceptron(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/classifier_ssk_modular.py b/examples/undocumented/python/classifier_ssk.py similarity index 82% rename from examples/undocumented/python_modular/classifier_ssk_modular.py rename to examples/undocumented/python/classifier_ssk.py index 8c426a3e05a..ccb983a62b7 100644 --- a/examples/undocumented/python_modular/classifier_ssk_modular.py +++ b/examples/undocumented/python/classifier_ssk.py @@ -18,11 +18,11 @@ parameter_list = [[traindat,testdat,label_traindat,1,5,0.9]] -def classifier_ssk_modular (fm_train_dna=traindat,fm_test_dna=testdat, +def classifier_ssk (fm_train_dna=traindat,fm_test_dna=testdat, label_train_dna=label_traindat,C=1,maxlen=1,decay=1): - from modshogun import StringCharFeatures, BinaryLabels - from modshogun import LibSVM, SubsequenceStringKernel, DNA - from modshogun import ErrorRateMeasure + from shogun import StringCharFeatures, BinaryLabels + from shogun import LibSVM, SubsequenceStringKernel, DNA + from shogun import ErrorRateMeasure feats_train=StringCharFeatures(fm_train_dna, DNA) feats_test=StringCharFeatures(fm_test_dna, DNA) @@ -45,4 +45,4 @@ def classifier_ssk_modular (fm_train_dna=traindat,fm_test_dna=testdat, if __name__=='__main__': print('SringSubsequenceKernel classification DNA') - classifier_ssk_modular(*parameter_list[0]) + classifier_ssk(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/classifier_svmlight_modular.py b/examples/undocumented/python/classifier_svmlight.py similarity index 73% rename from examples/undocumented/python_modular/classifier_svmlight_modular.py rename to examples/undocumented/python/classifier_svmlight.py index 08ed0e49300..156c7be45e4 100644 --- a/examples/undocumented/python_modular/classifier_svmlight_modular.py +++ b/examples/undocumented/python/classifier_svmlight.py @@ -8,11 +8,11 @@ parameter_list = [[traindat,testdat,label_traindat,1.1,1e-5,1],[traindat,testdat,label_traindat,1.2,1e-5,1]] -def classifier_svmlight_modular (fm_train_dna=traindat,fm_test_dna=testdat,label_train_dna=label_traindat,C=1.2,epsilon=1e-5,num_threads=1): - from modshogun import StringCharFeatures, BinaryLabels, DNA - from modshogun import WeightedDegreeStringKernel +def classifier_svmlight (fm_train_dna=traindat,fm_test_dna=testdat,label_train_dna=label_traindat,C=1.2,epsilon=1e-5,num_threads=1): + from shogun import StringCharFeatures, BinaryLabels, DNA + from shogun import WeightedDegreeStringKernel try: - from modshogun import SVMLight + from shogun import SVMLight except ImportError: print('No support for SVMLight available.') return @@ -37,4 +37,4 @@ def classifier_svmlight_modular (fm_train_dna=traindat,fm_test_dna=testdat,label return kernel if __name__=='__main__': print('SVMLight') - classifier_svmlight_modular(*parameter_list[0]) + classifier_svmlight(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/classifier_svmlight_batch_linadd_modular.py b/examples/undocumented/python/classifier_svmlight_batch_linadd.py similarity index 81% rename from examples/undocumented/python_modular/classifier_svmlight_batch_linadd_modular.py rename to examples/undocumented/python/classifier_svmlight_batch_linadd.py index fd907279799..abe957423aa 100644 --- a/examples/undocumented/python_modular/classifier_svmlight_batch_linadd_modular.py +++ b/examples/undocumented/python/classifier_svmlight_batch_linadd.py @@ -9,13 +9,13 @@ parameter_list=[[train_dna, test_dna, label, 20, 0.9, 1e-7, 1], [train_dna, test_dna, label, 20, 2.3, 1e-7, 4]] -def classifier_svmlight_batch_linadd_modular (fm_train_dna, fm_test_dna, +def classifier_svmlight_batch_linadd (fm_train_dna, fm_test_dna, label_train_dna, degree, C, epsilon, num_threads): - from modshogun import StringCharFeatures, BinaryLabels, DNA - from modshogun import WeightedDegreeStringKernel, MSG_DEBUG + from shogun import StringCharFeatures, BinaryLabels, DNA + from shogun import WeightedDegreeStringKernel, MSG_DEBUG try: - from modshogun import SVMLight + from shogun import SVMLight except ImportError: print('No support for SVMLight available.') return @@ -51,4 +51,4 @@ def classifier_svmlight_batch_linadd_modular (fm_train_dna, fm_test_dna, if __name__=='__main__': print('SVMlight batch') - classifier_svmlight_batch_linadd_modular(*parameter_list[0]) + classifier_svmlight_batch_linadd(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/classifier_svmlight_linear_term_modular.py b/examples/undocumented/python/classifier_svmlight_linear_term.py similarity index 85% rename from examples/undocumented/python_modular/classifier_svmlight_linear_term_modular.py rename to examples/undocumented/python/classifier_svmlight_linear_term.py index 5d9fc08d213..fa319153f74 100644 --- a/examples/undocumented/python_modular/classifier_svmlight_linear_term_modular.py +++ b/examples/undocumented/python/classifier_svmlight_linear_term.py @@ -26,14 +26,14 @@ parameter_list = [[traindna,testdna,label_traindna,3,10,1e-5,1],[traindna,testdna,label_traindna,3,10,1e-5,1]] -def classifier_svmlight_linear_term_modular (fm_train_dna=traindna,fm_test_dna=testdna, \ +def classifier_svmlight_linear_term (fm_train_dna=traindna,fm_test_dna=testdna, \ label_train_dna=label_traindna,degree=3, \ C=10,epsilon=1e-5,num_threads=1): - from modshogun import StringCharFeatures, BinaryLabels, DNA - from modshogun import WeightedDegreeStringKernel + from shogun import StringCharFeatures, BinaryLabels, DNA + from shogun import WeightedDegreeStringKernel try: - from modshogun import SVMLight + from shogun import SVMLight except ImportError: print("SVMLight is not available") exit(0) @@ -60,4 +60,4 @@ def classifier_svmlight_linear_term_modular (fm_train_dna=traindna,fm_test_dna=t if __name__=='__main__': print('SVMLight') - classifier_svmlight_linear_term_modular(*parameter_list[0]) + classifier_svmlight_linear_term(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/classifier_svmlin_modular.py b/examples/undocumented/python/classifier_svmlin.py similarity index 71% rename from examples/undocumented/python_modular/classifier_svmlin_modular.py rename to examples/undocumented/python/classifier_svmlin.py index 038eb1c1bc4..c35990fe6bf 100644 --- a/examples/undocumented/python_modular/classifier_svmlin_modular.py +++ b/examples/undocumented/python/classifier_svmlin.py @@ -5,9 +5,9 @@ parameter_list = [[traindat,testdat,label_traindat,0.9,1e-5,1],[traindat,testdat,label_traindat,0.8,1e-5,1]] -def classifier_svmlin_modular (train_fname=traindat,test_fname=testdat,label_fname=label_traindat,C=0.9,epsilon=1e-5,num_threads=1): - from modshogun import RealFeatures, SparseRealFeatures, BinaryLabels - from modshogun import SVMLin, CSVFile +def classifier_svmlin (train_fname=traindat,test_fname=testdat,label_fname=label_traindat,C=0.9,epsilon=1e-5,num_threads=1): + from shogun import RealFeatures, SparseRealFeatures, BinaryLabels + from shogun import SVMLin, CSVFile feats_train=RealFeatures(CSVFile(train_fname)) feats_test=RealFeatures(CSVFile(test_fname)) @@ -26,4 +26,4 @@ def classifier_svmlin_modular (train_fname=traindat,test_fname=testdat,label_fna if __name__=='__main__': print('SVMLin') - classifier_svmlin_modular(*parameter_list[0]) + classifier_svmlin(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/classifier_svmocas_modular.py b/examples/undocumented/python/classifier_svmocas.py similarity index 72% rename from examples/undocumented/python_modular/classifier_svmocas_modular.py rename to examples/undocumented/python/classifier_svmocas.py index 057057b7ed8..69d9328d8fd 100644 --- a/examples/undocumented/python_modular/classifier_svmocas_modular.py +++ b/examples/undocumented/python/classifier_svmocas.py @@ -5,11 +5,11 @@ parameter_list = [[traindat,testdat,label_traindat,0.9,1e-5,1],[traindat,testdat,label_traindat,0.8,1e-5,1]] -def classifier_svmocas_modular (train_fname=traindat,test_fname=testdat,label_fname=label_traindat,C=0.9,epsilon=1e-5,num_threads=1): - from modshogun import RealFeatures, BinaryLabels - from modshogun import CSVFile +def classifier_svmocas (train_fname=traindat,test_fname=testdat,label_fname=label_traindat,C=0.9,epsilon=1e-5,num_threads=1): + from shogun import RealFeatures, BinaryLabels + from shogun import CSVFile try: - from modshogun import SVMOcas + from shogun import SVMOcas except ImportError: print("SVMOcas not available") return @@ -31,4 +31,4 @@ def classifier_svmocas_modular (train_fname=traindat,test_fname=testdat,label_fn if __name__=='__main__': print('SVMOcas') - classifier_svmocas_modular(*parameter_list[0]) + classifier_svmocas(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/classifier_svmsgd_modular.py b/examples/undocumented/python/classifier_svmsgd.py similarity index 69% rename from examples/undocumented/python_modular/classifier_svmsgd_modular.py rename to examples/undocumented/python/classifier_svmsgd.py index 183fcc405a8..de289376aff 100644 --- a/examples/undocumented/python_modular/classifier_svmsgd_modular.py +++ b/examples/undocumented/python/classifier_svmsgd.py @@ -5,9 +5,9 @@ parameter_list = [[traindat,testdat,label_traindat,0.9,1,6],[traindat,testdat,label_traindat,0.8,1,5]] -def classifier_svmsgd_modular (train_fname=traindat,test_fname=testdat,label_fname=label_traindat,C=0.9,num_threads=1,num_iter=5): - from modshogun import RealFeatures, SparseRealFeatures, BinaryLabels - from modshogun import SVMSGD, CSVFile +def classifier_svmsgd (train_fname=traindat,test_fname=testdat,label_fname=label_traindat,C=0.9,num_threads=1,num_iter=5): + from shogun import RealFeatures, SparseRealFeatures, BinaryLabels + from shogun import SVMSGD, CSVFile feats_train=RealFeatures(CSVFile(train_fname)) feats_test=RealFeatures(CSVFile(test_fname)) @@ -25,4 +25,4 @@ def classifier_svmsgd_modular (train_fname=traindat,test_fname=testdat,label_fna if __name__=='__main__': print('SVMSGD') - classifier_svmsgd_modular(*parameter_list[0]) + classifier_svmsgd(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/converter_diffusionmaps_modular.py b/examples/undocumented/python/converter_diffusionmaps.py similarity index 70% rename from examples/undocumented/python_modular/converter_diffusionmaps_modular.py rename to examples/undocumented/python/converter_diffusionmaps.py index 622433c8008..ce134aae07f 100644 --- a/examples/undocumented/python_modular/converter_diffusionmaps_modular.py +++ b/examples/undocumented/python/converter_diffusionmaps.py @@ -2,9 +2,9 @@ data = '../data/fm_train_real.dat' parameter_list = [[data,10],[data,20]] -def converter_diffusionmaps_modular (data_fname,t): +def converter_diffusionmaps (data_fname,t): try: - from modshogun import RealFeatures, DiffusionMaps, GaussianKernel, CSVFile + from shogun import RealFeatures, DiffusionMaps, GaussianKernel, CSVFile features = RealFeatures(CSVFile(data_fname)) @@ -20,5 +20,5 @@ def converter_diffusionmaps_modular (data_fname,t): if __name__=='__main__': print('DiffusionMaps') - converter_diffusionmaps_modular(*parameter_list[0]) + converter_diffusionmaps(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/converter_factoranalysis_modular.py b/examples/undocumented/python/converter_factoranalysis.py similarity index 71% rename from examples/undocumented/python_modular/converter_factoranalysis_modular.py rename to examples/undocumented/python/converter_factoranalysis.py index 17bfee2bae4..775a8aa09a1 100644 --- a/examples/undocumented/python_modular/converter_factoranalysis_modular.py +++ b/examples/undocumented/python/converter_factoranalysis.py @@ -2,10 +2,10 @@ data = '../data/fm_train_real.dat' parameter_list = [[data]] -def converter_factoranalysis_modular(data_fname): +def converter_factoranalysis(data_fname): try: import numpy - from modshogun import RealFeatures, FactorAnalysis, EuclideanDistance, CSVFile + from shogun import RealFeatures, FactorAnalysis, EuclideanDistance, CSVFile features = RealFeatures(CSVFile(data_fname)) @@ -22,4 +22,4 @@ def converter_factoranalysis_modular(data_fname): if __name__=='__main__': print('Factor Analysis') - converter_factoranalysis_modular(*parameter_list[0]) + converter_factoranalysis(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/converter_hasheddoc_modular.py b/examples/undocumented/python/converter_hasheddoc.py similarity index 80% rename from examples/undocumented/python_modular/converter_hasheddoc_modular.py rename to examples/undocumented/python/converter_hasheddoc.py index 880763a6c18..99299a3a964 100644 --- a/examples/undocumented/python_modular/converter_hasheddoc_modular.py +++ b/examples/undocumented/python/converter_hasheddoc.py @@ -4,10 +4,10 @@ parameter_list=[[strings]] -def converter_hasheddoc_modular(strings): - from modshogun import SparseRealFeatures, RAWBYTE, StringCharFeatures, Features, HashedDocDotFeatures - from modshogun import NGramTokenizer - from modshogun import HashedDocConverter +def converter_hasheddoc(strings): + from shogun import SparseRealFeatures, RAWBYTE, StringCharFeatures, Features, HashedDocDotFeatures + from shogun import NGramTokenizer + from shogun import HashedDocConverter from numpy import array #create string features @@ -43,6 +43,6 @@ def converter_hasheddoc_modular(strings): if __name__=='__main__': print('HashedDocConverter') - converter_hasheddoc_modular(*parameter_list[0]) + converter_hasheddoc(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/converter_hessianlocallylinearembedding_modular.py b/examples/undocumented/python/converter_hessianlocallylinearembedding.py similarity index 67% rename from examples/undocumented/python_modular/converter_hessianlocallylinearembedding_modular.py rename to examples/undocumented/python/converter_hessianlocallylinearembedding.py index f97323fb452..61c07f43926 100644 --- a/examples/undocumented/python_modular/converter_hessianlocallylinearembedding_modular.py +++ b/examples/undocumented/python/converter_hessianlocallylinearembedding.py @@ -2,11 +2,11 @@ data = '../data/fm_train_real.dat' parameter_list = [[data,20],[data,30]] -def converter_hessianlocallylinearembedding_modular (data_fname,k): +def converter_hessianlocallylinearembedding (data_fname,k): try: - from modshogun import RealFeatures, CSVFile + from shogun import RealFeatures, CSVFile try: - from modshogun import HessianLocallyLinearEmbedding + from shogun import HessianLocallyLinearEmbedding except ImportError: print("HessianLocallyLinearEmbedding not available") exit(0) @@ -24,5 +24,5 @@ def converter_hessianlocallylinearembedding_modular (data_fname,k): if __name__=='__main__': print('HessianLocallyLinearEmbedding') - converter_hessianlocallylinearembedding_modular(*parameter_list[0]) + converter_hessianlocallylinearembedding(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/converter_isomap_modular.py b/examples/undocumented/python/converter_isomap.py similarity index 63% rename from examples/undocumented/python_modular/converter_isomap_modular.py rename to examples/undocumented/python/converter_isomap.py index 0c633e8082f..fb41999abd9 100644 --- a/examples/undocumented/python_modular/converter_isomap_modular.py +++ b/examples/undocumented/python/converter_isomap.py @@ -2,9 +2,9 @@ data = '../data/fm_train_real.dat' parameter_list = [[data]] -def converter_isomap_modular (data_fname): - from modshogun import RealFeatures, CSVFile - from modshogun import Isomap +def converter_isomap (data_fname): + from shogun import RealFeatures, CSVFile + from shogun import Isomap features = RealFeatures(CSVFile(data)) @@ -17,5 +17,5 @@ def converter_isomap_modular (data_fname): if __name__=='__main__': print('Isomap') - #converter_isomap_modular(*parameter_list[0]) + #converter_isomap(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/converter_kernellocallylinearembedding_modular.py b/examples/undocumented/python/converter_kernellocallylinearembedding.py similarity index 68% rename from examples/undocumented/python_modular/converter_kernellocallylinearembedding_modular.py rename to examples/undocumented/python/converter_kernellocallylinearembedding.py index b157d00275e..b8cf370b7a8 100644 --- a/examples/undocumented/python_modular/converter_kernellocallylinearembedding_modular.py +++ b/examples/undocumented/python/converter_kernellocallylinearembedding.py @@ -2,11 +2,11 @@ data = '../data/fm_train_real.dat' parameter_list = [[data,20],[data,30]] -def converter_kernellocallylinearembedding_modular (data_fname,k): +def converter_kernellocallylinearembedding (data_fname,k): try: - from modshogun import RealFeatures, LinearKernel, CSVFile + from shogun import RealFeatures, LinearKernel, CSVFile try: - from modshogun import KernelLocallyLinearEmbedding + from shogun import KernelLocallyLinearEmbedding except ImportError: print("KernelLocallyLinearEmbedding not available") exit(0) @@ -26,5 +26,5 @@ def converter_kernellocallylinearembedding_modular (data_fname,k): if __name__=='__main__': print('KernelLocallyLinearEmbedding') - converter_kernellocallylinearembedding_modular(*parameter_list[0]) + converter_kernellocallylinearembedding(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/converter_laplacianeigenmaps_modular.py b/examples/undocumented/python/converter_laplacianeigenmaps.py similarity index 71% rename from examples/undocumented/python_modular/converter_laplacianeigenmaps_modular.py rename to examples/undocumented/python/converter_laplacianeigenmaps.py index b2126757fe0..810ded4cb55 100644 --- a/examples/undocumented/python_modular/converter_laplacianeigenmaps_modular.py +++ b/examples/undocumented/python/converter_laplacianeigenmaps.py @@ -2,11 +2,11 @@ data = '../data/fm_train_real.dat' parameter_list = [[data,20],[data,30]] -def converter_laplacianeigenmaps_modular (data_fname,k): +def converter_laplacianeigenmaps (data_fname,k): try: - from modshogun import RealFeatures, CSVFile + from shogun import RealFeatures, CSVFile try: - from modshogun import LaplacianEigenmaps + from shogun import LaplacianEigenmaps except ImportError: print("LaplacianEigenmaps not available") exit(0) @@ -25,5 +25,5 @@ def converter_laplacianeigenmaps_modular (data_fname,k): if __name__=='__main__': print('LaplacianEigenmaps') - converter_laplacianeigenmaps_modular(*parameter_list[0]) + converter_laplacianeigenmaps(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/converter_linearlocaltangentspacealignment_modular.py b/examples/undocumented/python/converter_linearlocaltangentspacealignment.py similarity index 67% rename from examples/undocumented/python_modular/converter_linearlocaltangentspacealignment_modular.py rename to examples/undocumented/python/converter_linearlocaltangentspacealignment.py index 669dcf5b4b1..65e7218c963 100644 --- a/examples/undocumented/python_modular/converter_linearlocaltangentspacealignment_modular.py +++ b/examples/undocumented/python/converter_linearlocaltangentspacealignment.py @@ -2,11 +2,11 @@ data = '../data/fm_train_real.dat' parameter_list = [[data,20],[data,30]] -def converter_linearlocaltangentspacealignment_modular (data_fname,k): +def converter_linearlocaltangentspacealignment (data_fname,k): try: - from modshogun import RealFeatures, CSVFile + from shogun import RealFeatures, CSVFile try: - from modshogun import LinearLocalTangentSpaceAlignment + from shogun import LinearLocalTangentSpaceAlignment except ImportError: print("LinearLocalTangentSpaceAlignment not available") exit(0) @@ -24,5 +24,5 @@ def converter_linearlocaltangentspacealignment_modular (data_fname,k): if __name__=='__main__': print('LinearLocalTangentSpaceAlignment') - converter_linearlocaltangentspacealignment_modular(*parameter_list[0]) + converter_linearlocaltangentspacealignment(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/converter_localitypreservingprojections_modular.py b/examples/undocumented/python/converter_localitypreservingprojections.py similarity index 61% rename from examples/undocumented/python_modular/converter_localitypreservingprojections_modular.py rename to examples/undocumented/python/converter_localitypreservingprojections.py index b240abb2283..2ff78bae35f 100644 --- a/examples/undocumented/python_modular/converter_localitypreservingprojections_modular.py +++ b/examples/undocumented/python/converter_localitypreservingprojections.py @@ -2,9 +2,9 @@ data = '../data/fm_train_real.dat' parameter_list = [[data,20],[data,30]] -def converter_localitypreservingprojections_modular (data_fname,k): - from modshogun import RealFeatures, CSVFile - from modshogun import LocalityPreservingProjections +def converter_localitypreservingprojections (data_fname,k): + from shogun import RealFeatures, CSVFile + from shogun import LocalityPreservingProjections features = RealFeatures(CSVFile(data_fname)) converter = LocalityPreservingProjections() @@ -17,5 +17,5 @@ def converter_localitypreservingprojections_modular (data_fname,k): if __name__=='__main__': print('LocalityPreservingProjections') - #converter_localitypreservingprojections_modular(*parameter_list[0]) + #converter_localitypreservingprojections(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/converter_locallylinearembedding_modular.py b/examples/undocumented/python/converter_locallylinearembedding.py similarity index 69% rename from examples/undocumented/python_modular/converter_locallylinearembedding_modular.py rename to examples/undocumented/python/converter_locallylinearembedding.py index cf6510f563d..1c85d3ffd11 100644 --- a/examples/undocumented/python_modular/converter_locallylinearembedding_modular.py +++ b/examples/undocumented/python/converter_locallylinearembedding.py @@ -2,11 +2,11 @@ data = '../data/fm_train_real.dat' parameter_list = [[data,20],[data,30]] -def converter_locallylinearembedding_modular (data_fname,k): +def converter_locallylinearembedding (data_fname,k): try: - from modshogun import RealFeatures, CSVFile + from shogun import RealFeatures, CSVFile try: - from modshogun import LocallyLinearEmbedding + from shogun import LocallyLinearEmbedding except ImportError: print("LocallyLinearEmbedding not available") exit(0) @@ -24,5 +24,5 @@ def converter_locallylinearembedding_modular (data_fname,k): if __name__=='__main__': print('LocallyLinearEmbedding') - converter_locallylinearembedding_modular(*parameter_list[0]) + converter_locallylinearembedding(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/converter_localtangentspacealignment_modular.py b/examples/undocumented/python/converter_localtangentspacealignment.py similarity index 68% rename from examples/undocumented/python_modular/converter_localtangentspacealignment_modular.py rename to examples/undocumented/python/converter_localtangentspacealignment.py index 42cde892c78..d47e706b5d7 100644 --- a/examples/undocumented/python_modular/converter_localtangentspacealignment_modular.py +++ b/examples/undocumented/python/converter_localtangentspacealignment.py @@ -2,11 +2,11 @@ data = '../data/fm_train_real.dat' parameter_list = [[data,20],[data,30]] -def converter_localtangentspacealignment_modular (data_fname,k): +def converter_localtangentspacealignment (data_fname,k): try: - from modshogun import RealFeatures, CSVFile + from shogun import RealFeatures, CSVFile try: - from modshogun import LocalTangentSpaceAlignment + from shogun import LocalTangentSpaceAlignment except ImportError: print("LocalTangentSpaceAlignment not available") exit(0) @@ -25,5 +25,5 @@ def converter_localtangentspacealignment_modular (data_fname,k): if __name__=='__main__': print('LocalTangentSpaceAlignment') - converter_localtangentspacealignment_modular(*parameter_list[0]) + converter_localtangentspacealignment(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/converter_multidimensionalscaling_modular.py b/examples/undocumented/python/converter_multidimensionalscaling.py similarity index 79% rename from examples/undocumented/python_modular/converter_multidimensionalscaling_modular.py rename to examples/undocumented/python/converter_multidimensionalscaling.py index 54b8d9d944e..8acaac5de4a 100644 --- a/examples/undocumented/python_modular/converter_multidimensionalscaling_modular.py +++ b/examples/undocumented/python/converter_multidimensionalscaling.py @@ -2,10 +2,10 @@ data = '../data/fm_train_real.dat' parameter_list = [[data]] -def converter_multidimensionalscaling_modular (data_fname): +def converter_multidimensionalscaling (data_fname): try: import numpy - from modshogun import RealFeatures, MultidimensionalScaling, EuclideanDistance, CSVFile + from shogun import RealFeatures, MultidimensionalScaling, EuclideanDistance, CSVFile features = RealFeatures(CSVFile(data_fname)) @@ -29,4 +29,4 @@ def converter_multidimensionalscaling_modular (data_fname): if __name__=='__main__': print('MultidimensionalScaling') - converter_multidimensionalscaling_modular(*parameter_list[0]) + converter_multidimensionalscaling(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/converter_stochasticproximityembedding_modular.py b/examples/undocumented/python/converter_stochasticproximityembedding.py similarity index 71% rename from examples/undocumented/python_modular/converter_stochasticproximityembedding_modular.py rename to examples/undocumented/python/converter_stochasticproximityembedding.py index c1bcced509b..16684987d13 100644 --- a/examples/undocumented/python_modular/converter_stochasticproximityembedding_modular.py +++ b/examples/undocumented/python/converter_stochasticproximityembedding.py @@ -2,9 +2,9 @@ data = '../data/fm_train_real.dat' parameter_list = [[data, 20]] -def converter_stochasticproximityembedding_modular (data_fname, k): +def converter_stochasticproximityembedding (data_fname, k): try: - from modshogun import RealFeatures,StochasticProximityEmbedding, SPE_GLOBAL, SPE_LOCAL, CSVFile + from shogun import RealFeatures,StochasticProximityEmbedding, SPE_GLOBAL, SPE_LOCAL, CSVFile features = RealFeatures(CSVFile(data_fname)) @@ -25,4 +25,4 @@ def converter_stochasticproximityembedding_modular (data_fname, k): if __name__=='__main__': print('StochasticProximityEmbedding') - converter_stochasticproximityembedding_modular(*parameter_list[0]) + converter_stochasticproximityembedding(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/converter_tdistributedstochasticneighborembedding_modular.py b/examples/undocumented/python/converter_tdistributedstochasticneighborembedding.py similarity index 61% rename from examples/undocumented/python_modular/converter_tdistributedstochasticneighborembedding_modular.py rename to examples/undocumented/python/converter_tdistributedstochasticneighborembedding.py index 16fdf12f726..9d93083ec32 100644 --- a/examples/undocumented/python_modular/converter_tdistributedstochasticneighborembedding_modular.py +++ b/examples/undocumented/python/converter_tdistributedstochasticneighborembedding.py @@ -2,10 +2,10 @@ data = '../data/fm_train_real.dat' parameter_list = [[data]] -def converter_tdistributedstochasticneighborembedding_modular(data_fname, seed=1): +def converter_tdistributedstochasticneighborembedding(data_fname, seed=1): try: - from modshogun import RealFeatures, TDistributedStochasticNeighborEmbedding - from modshogun import Math_init_random, CSVFile + from shogun import RealFeatures, TDistributedStochasticNeighborEmbedding + from shogun import Math_init_random, CSVFile # reproducible results Math_init_random(seed) @@ -22,4 +22,4 @@ def converter_tdistributedstochasticneighborembedding_modular(data_fname, seed=1 if __name__=='__main__': print('TDistributedStochasticNeighborEmbedding') - converter_tdistributedstochasticneighborembedding_modular(*parameter_list[0]) + converter_tdistributedstochasticneighborembedding(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/distance_canberra_modular.py b/examples/undocumented/python/distance_canberra.py similarity index 73% rename from examples/undocumented/python_modular/distance_canberra_modular.py rename to examples/undocumented/python/distance_canberra.py index 7c65bf1dfa5..6c006827a54 100644 --- a/examples/undocumented/python_modular/distance_canberra_modular.py +++ b/examples/undocumented/python/distance_canberra.py @@ -4,8 +4,8 @@ parameter_list = [[traindat,testdat],[traindat,testdat]] -def distance_canberra_modular (train_fname=traindat,test_fname=testdat): - from modshogun import RealFeatures, CanberraMetric, CSVFile +def distance_canberra (train_fname=traindat,test_fname=testdat): + from shogun import RealFeatures, CanberraMetric, CSVFile feats_train=RealFeatures(CSVFile(train_fname)) feats_test=RealFeatures(CSVFile(test_fname)) @@ -19,4 +19,4 @@ def distance_canberra_modular (train_fname=traindat,test_fname=testdat): if __name__=='__main__': print('CanberaMetric') - distance_canberra_modular(*parameter_list[0]) + distance_canberra(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/distance_canberraword_modular.py b/examples/undocumented/python/distance_canberraword.py similarity index 78% rename from examples/undocumented/python_modular/distance_canberraword_modular.py rename to examples/undocumented/python/distance_canberraword.py index ba7bcd876a8..b01198783b6 100644 --- a/examples/undocumented/python_modular/distance_canberraword_modular.py +++ b/examples/undocumented/python/distance_canberraword.py @@ -8,10 +8,10 @@ parameter_list = [[traindna,testdna,3,0,False],[traindna,testdna,3,0,False]] -def distance_canberraword_modular (fm_train_dna=traindna,fm_test_dna=testdna,order=3,gap=0,reverse=False): - from modshogun import StringCharFeatures, StringWordFeatures, DNA - from modshogun import SortWordString - from modshogun import CanberraWordDistance +def distance_canberraword (fm_train_dna=traindna,fm_test_dna=testdna,order=3,gap=0,reverse=False): + from shogun import StringCharFeatures, StringWordFeatures, DNA + from shogun import SortWordString + from shogun import CanberraWordDistance charfeat=StringCharFeatures(DNA) charfeat.set_features(fm_train_dna) @@ -38,4 +38,4 @@ def distance_canberraword_modular (fm_train_dna=traindna,fm_test_dna=testdna,ord if __name__=='__main__': print('CanberraWordDistance') - distance_canberraword_modular(*parameter_list[0]) + distance_canberraword(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/distance_chebyshew_modular.py b/examples/undocumented/python/distance_chebyshew.py similarity index 73% rename from examples/undocumented/python_modular/distance_chebyshew_modular.py rename to examples/undocumented/python/distance_chebyshew.py index fe33d770cb0..1a0a0bb6164 100644 --- a/examples/undocumented/python_modular/distance_chebyshew_modular.py +++ b/examples/undocumented/python/distance_chebyshew.py @@ -4,8 +4,8 @@ parameter_list = [[traindat,testdat],[traindat,testdat]] -def distance_chebyshew_modular (train_fname=traindat,test_fname=testdat): - from modshogun import RealFeatures, ChebyshewMetric, CSVFile +def distance_chebyshew (train_fname=traindat,test_fname=testdat): + from shogun import RealFeatures, ChebyshewMetric, CSVFile feats_train=RealFeatures(CSVFile(train_fname)) feats_test=RealFeatures(CSVFile(test_fname)) @@ -19,4 +19,4 @@ def distance_chebyshew_modular (train_fname=traindat,test_fname=testdat): if __name__=='__main__': print('ChebyshewMetric') - distance_chebyshew_modular(*parameter_list[0]) + distance_chebyshew(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/distance_chisquare_modular.py b/examples/undocumented/python/distance_chisquare.py similarity index 73% rename from examples/undocumented/python_modular/distance_chisquare_modular.py rename to examples/undocumented/python/distance_chisquare.py index 0e30bc6d9b2..ce8588b4a34 100644 --- a/examples/undocumented/python_modular/distance_chisquare_modular.py +++ b/examples/undocumented/python/distance_chisquare.py @@ -4,8 +4,8 @@ parameter_list = [[traindat,testdat,],[traindat,testdat]] -def distance_chisquare_modular (train_fname=traindat,test_fname=testdat): - from modshogun import RealFeatures, ChiSquareDistance, CSVFile +def distance_chisquare (train_fname=traindat,test_fname=testdat): + from shogun import RealFeatures, ChiSquareDistance, CSVFile feats_train=RealFeatures(CSVFile(train_fname)) feats_test=RealFeatures(CSVFile(test_fname)) @@ -18,4 +18,4 @@ def distance_chisquare_modular (train_fname=traindat,test_fname=testdat): if __name__=='__main__': print('ChiSquareDistance') - distance_chisquare_modular(*parameter_list[0]) + distance_chisquare(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/distance_director_euclidean_modular.py b/examples/undocumented/python/distance_director_euclidean.py similarity index 80% rename from examples/undocumented/python_modular/distance_director_euclidean_modular.py rename to examples/undocumented/python/distance_director_euclidean.py index 9945f8dcb37..deab1aa6aee 100644 --- a/examples/undocumented/python_modular/distance_director_euclidean_modular.py +++ b/examples/undocumented/python/distance_director_euclidean.py @@ -1,15 +1,15 @@ #!/usr/bin/env python import numpy -from modshogun import RealFeatures, MSG_DEBUG +from shogun import RealFeatures, MSG_DEBUG numpy.random.seed(17) traindat = numpy.random.random_sample((10,10)) testdat = numpy.random.random_sample((10,10)) parameter_list=[[traindat,testdat,1.2],[traindat,testdat,1.4]] -def distance_director_euclidean_modular (fm_train_real=traindat,fm_test_real=testdat,scale=1.2): +def distance_director_euclidean (fm_train_real=traindat,fm_test_real=testdat,scale=1.2): try: - from modshogun import DirectorDistance + from shogun import DirectorDistance except ImportError: print("recompile shogun with --enable-swig-directors") return @@ -22,8 +22,8 @@ def distance_function(self, idx_a, idx_b): seq2 = self.get_rhs().get_feature_vector(idx_b) return numpy.linalg.norm(seq1-seq2) - from modshogun import EuclideanDistance - from modshogun import Time + from shogun import EuclideanDistance + from shogun import Time feats_train=RealFeatures(fm_train_real) #feats_train.io.set_loglevel(MSG_DEBUG) @@ -53,4 +53,4 @@ def distance_function(self, idx_a, idx_b): if __name__=='__main__': print('DirectorEuclideanDistance') - distance_director_euclidean_modular(*parameter_list[0]) + distance_director_euclidean(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/distance_geodesic_modular.py b/examples/undocumented/python/distance_geodesic.py similarity index 73% rename from examples/undocumented/python_modular/distance_geodesic_modular.py rename to examples/undocumented/python/distance_geodesic.py index 1d33867bc4f..eab7ba89811 100644 --- a/examples/undocumented/python_modular/distance_geodesic_modular.py +++ b/examples/undocumented/python/distance_geodesic.py @@ -4,9 +4,9 @@ parameter_list = [[traindat,testdat],[traindat,testdat]] -def distance_geodesic_modular (train_fname=traindat,test_fname=testdat): +def distance_geodesic (train_fname=traindat,test_fname=testdat): - from modshogun import RealFeatures, GeodesicMetric, CSVFile + from shogun import RealFeatures, GeodesicMetric, CSVFile feats_train=RealFeatures(CSVFile(train_fname)) feats_test=RealFeatures(CSVFile(test_fname)) @@ -22,4 +22,4 @@ def distance_geodesic_modular (train_fname=traindat,test_fname=testdat): if __name__=='__main__': print('GeodesicMetric') - distance_geodesic_modular(*parameter_list[0]) + distance_geodesic(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/distance_hammingword_modular.py b/examples/undocumented/python/distance_hammingword.py similarity index 82% rename from examples/undocumented/python_modular/distance_hammingword_modular.py rename to examples/undocumented/python/distance_hammingword.py index f853da7087f..e0534c0cbf9 100644 --- a/examples/undocumented/python_modular/distance_hammingword_modular.py +++ b/examples/undocumented/python/distance_hammingword.py @@ -9,12 +9,12 @@ parameter_list = [[traindna,testdna,testdat,4,0,False,False], [traindna,testdna,testdat,3,0,False,False]] -def distance_hammingword_modular (fm_train_dna=traindna,fm_test_dna=testdna, +def distance_hammingword (fm_train_dna=traindna,fm_test_dna=testdna, fm_test_real=testdat,order=3,gap=0,reverse=False,use_sign=False): - from modshogun import StringCharFeatures, StringWordFeatures, DNA - from modshogun import SortWordString - from modshogun import HammingWordDistance + from shogun import StringCharFeatures, StringWordFeatures, DNA + from shogun import SortWordString + from shogun import HammingWordDistance charfeat=StringCharFeatures(DNA) charfeat.set_features(fm_train_dna) @@ -41,4 +41,4 @@ def distance_hammingword_modular (fm_train_dna=traindna,fm_test_dna=testdna, if __name__=='__main__': print('HammingWordDistance') - distance_hammingword_modular(*parameter_list[0]) + distance_hammingword(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/distance_jensen_modular.py b/examples/undocumented/python/distance_jensen.py similarity index 74% rename from examples/undocumented/python_modular/distance_jensen_modular.py rename to examples/undocumented/python/distance_jensen.py index 6b4205f0eb6..43cd0233a95 100644 --- a/examples/undocumented/python_modular/distance_jensen_modular.py +++ b/examples/undocumented/python/distance_jensen.py @@ -4,9 +4,9 @@ parameter_list = [[traindat,testdat],[traindat,testdat]] -def distance_jensen_modular (train_fname=traindat,test_fname=testdat): +def distance_jensen (train_fname=traindat,test_fname=testdat): - from modshogun import RealFeatures, JensenMetric, CSVFile + from shogun import RealFeatures, JensenMetric, CSVFile feats_train=RealFeatures(CSVFile(train_fname)) feats_test=RealFeatures(CSVFile(test_fname)) @@ -20,4 +20,4 @@ def distance_jensen_modular (train_fname=traindat,test_fname=testdat): if __name__=='__main__': print('JensenMetric') - distance_jensen_modular(*parameter_list[0]) + distance_jensen(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/distance_manhattenword_modular.py b/examples/undocumented/python/distance_manhattenword.py similarity index 77% rename from examples/undocumented/python_modular/distance_manhattenword_modular.py rename to examples/undocumented/python/distance_manhattenword.py index c639dd72186..e377f757285 100644 --- a/examples/undocumented/python_modular/distance_manhattenword_modular.py +++ b/examples/undocumented/python/distance_manhattenword.py @@ -4,9 +4,9 @@ parameter_list = [[traindna,testdna,3,0,False],[traindna,testdna,4,0,False]] -def distance_manhattenword_modular (train_fname=traindna,test_fname=testdna,order=3,gap=0,reverse=False): - from modshogun import StringCharFeatures, StringWordFeatures, DNA - from modshogun import SortWordString, ManhattanWordDistance, CSVFile +def distance_manhattenword (train_fname=traindna,test_fname=testdna,order=3,gap=0,reverse=False): + from shogun import StringCharFeatures, StringWordFeatures, DNA + from shogun import SortWordString, ManhattanWordDistance, CSVFile charfeat=StringCharFeatures(CSVFile(train_fname), DNA) feats_train=StringWordFeatures(charfeat.get_alphabet()) @@ -31,4 +31,4 @@ def distance_manhattenword_modular (train_fname=traindna,test_fname=testdna,orde if __name__=='__main__': print('ManhattanWordDistance') - distance_manhattenword_modular(*parameter_list[0]) + distance_manhattenword(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/distance_minkowski_modular.py b/examples/undocumented/python/distance_minkowski.py similarity index 73% rename from examples/undocumented/python_modular/distance_minkowski_modular.py rename to examples/undocumented/python/distance_minkowski.py index 76250567fff..765e9818e9b 100644 --- a/examples/undocumented/python_modular/distance_minkowski_modular.py +++ b/examples/undocumented/python/distance_minkowski.py @@ -4,8 +4,8 @@ parameter_list = [[traindat,testdat,3],[traindat,testdat,4]] -def distance_minkowski_modular (train_fname=traindat,test_fname=testdat,k=3): - from modshogun import RealFeatures, MinkowskiMetric, CSVFile +def distance_minkowski (train_fname=traindat,test_fname=testdat,k=3): + from shogun import RealFeatures, MinkowskiMetric, CSVFile feats_train=RealFeatures(CSVFile(train_fname)) feats_test=RealFeatures(CSVFile(test_fname)) @@ -20,5 +20,5 @@ def distance_minkowski_modular (train_fname=traindat,test_fname=testdat,k=3): if __name__=='__main__': print('MinkowskiMetric') - distance_minkowski_modular(*parameter_list[0]) + distance_minkowski(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/distance_normsquared_modular.py b/examples/undocumented/python/distance_normsquared.py similarity index 74% rename from examples/undocumented/python_modular/distance_normsquared_modular.py rename to examples/undocumented/python/distance_normsquared.py index ede13312457..0e1d435514b 100644 --- a/examples/undocumented/python_modular/distance_normsquared_modular.py +++ b/examples/undocumented/python/distance_normsquared.py @@ -4,8 +4,8 @@ parameter_list = [[traindat,testdat],[traindat,testdat]] -def distance_normsquared_modular (train_fname=traindat,test_fname=testdat): - from modshogun import RealFeatures, EuclideanDistance, CSVFile +def distance_normsquared (train_fname=traindat,test_fname=testdat): + from shogun import RealFeatures, EuclideanDistance, CSVFile feats_train=RealFeatures(CSVFile(train_fname)) feats_test=RealFeatures(CSVFile(test_fname)) @@ -22,4 +22,4 @@ def distance_normsquared_modular (train_fname=traindat,test_fname=testdat): if __name__=='__main__': print('EuclideanDistance - NormSquared') - distance_normsquared_modular(*parameter_list[0]) + distance_normsquared(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/distance_sparseeuclidean_modular.py b/examples/undocumented/python/distance_sparseeuclidean.py similarity index 75% rename from examples/undocumented/python_modular/distance_sparseeuclidean_modular.py rename to examples/undocumented/python/distance_sparseeuclidean.py index d1221cd4f5e..9f2c48eac01 100644 --- a/examples/undocumented/python_modular/distance_sparseeuclidean_modular.py +++ b/examples/undocumented/python/distance_sparseeuclidean.py @@ -4,8 +4,8 @@ parameter_list = [[traindat,testdat],[traindat,testdat]] -def distance_sparseeuclidean_modular (train_fname=traindat,test_fname=testdat): - from modshogun import RealFeatures, SparseRealFeatures, SparseEuclideanDistance, CSVFile +def distance_sparseeuclidean (train_fname=traindat,test_fname=testdat): + from shogun import RealFeatures, SparseRealFeatures, SparseEuclideanDistance, CSVFile realfeat=RealFeatures(CSVFile(train_fname)) feats_train=SparseRealFeatures() @@ -24,4 +24,4 @@ def distance_sparseeuclidean_modular (train_fname=traindat,test_fname=testdat): if __name__=='__main__': print('SparseEuclideanDistance') - distance_sparseeuclidean_modular(*parameter_list[0]) + distance_sparseeuclidean(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/distance_tanimoto_modular.py b/examples/undocumented/python/distance_tanimoto.py similarity index 73% rename from examples/undocumented/python_modular/distance_tanimoto_modular.py rename to examples/undocumented/python/distance_tanimoto.py index c71fb49b2e0..471ec4bec2a 100644 --- a/examples/undocumented/python_modular/distance_tanimoto_modular.py +++ b/examples/undocumented/python/distance_tanimoto.py @@ -4,8 +4,8 @@ parameter_list = [[traindat,testdat],[traindat,testdat]] -def distance_tanimoto_modular (train_fname=traindat,test_fname=testdat): - from modshogun import RealFeatures, TanimotoDistance, CSVFile +def distance_tanimoto (train_fname=traindat,test_fname=testdat): + from shogun import RealFeatures, TanimotoDistance, CSVFile feats_train=RealFeatures(CSVFile(train_fname)) feats_test=RealFeatures(CSVFile(test_fname)) @@ -20,4 +20,4 @@ def distance_tanimoto_modular (train_fname=traindat,test_fname=testdat): if __name__=='__main__': print('TanimotoDistance') - distance_tanimoto_modular(*parameter_list[0]) + distance_tanimoto(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/distribution_histogram_modular.py b/examples/undocumented/python/distribution_histogram.py similarity index 80% rename from examples/undocumented/python_modular/distribution_histogram_modular.py rename to examples/undocumented/python/distribution_histogram.py index bacfd045b78..52f1752c1f4 100644 --- a/examples/undocumented/python_modular/distribution_histogram_modular.py +++ b/examples/undocumented/python/distribution_histogram.py @@ -6,9 +6,9 @@ parameter_list = [[traindna,3,0,False],[traindna,4,0,False]] -def distribution_histogram_modular (fm_dna=traindna,order=3,gap=0,reverse=False): - from modshogun import StringWordFeatures, StringCharFeatures, DNA - from modshogun import Histogram +def distribution_histogram (fm_dna=traindna,order=3,gap=0,reverse=False): + from shogun import StringWordFeatures, StringCharFeatures, DNA + from shogun import Histogram charfeat=StringCharFeatures(DNA) charfeat.set_features(fm_dna) @@ -35,5 +35,5 @@ def distribution_histogram_modular (fm_dna=traindna,order=3,gap=0,reverse=False) if __name__=='__main__': print('Histogram') - distribution_histogram_modular(*parameter_list[0]) + distribution_histogram(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/distribution_hmm_modular.py b/examples/undocumented/python/distribution_hmm.py similarity index 82% rename from examples/undocumented/python_modular/distribution_hmm_modular.py rename to examples/undocumented/python/distribution_hmm.py index c95e937b61e..a3019d33c8e 100644 --- a/examples/undocumented/python_modular/distribution_hmm_modular.py +++ b/examples/undocumented/python/distribution_hmm.py @@ -5,9 +5,9 @@ parameter_list=[[data, 1, 64, 1e-5, 2, 0, False, 5], [data, 3, 6, 1e-1, 1, 0, False, 2]] -def distribution_hmm_modular(fm_cube, N, M, pseudo, order, gap, reverse, num_examples): - from modshogun import StringWordFeatures, StringCharFeatures, CUBE - from modshogun import HMM, BW_NORMAL +def distribution_hmm(fm_cube, N, M, pseudo, order, gap, reverse, num_examples): + from shogun import StringWordFeatures, StringCharFeatures, CUBE + from shogun import HMM, BW_NORMAL charfeat=StringCharFeatures(CUBE) charfeat.set_features(fm_cube) @@ -42,4 +42,4 @@ def distribution_hmm_modular(fm_cube, N, M, pseudo, order, gap, reverse, num_exa if __name__=='__main__': print('HMM') - distribution_hmm_modular(*parameter_list[0]) + distribution_hmm(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/distribution_linearhmm_modular.py b/examples/undocumented/python/distribution_linearhmm.py similarity index 80% rename from examples/undocumented/python_modular/distribution_linearhmm_modular.py rename to examples/undocumented/python/distribution_linearhmm.py index 721f6e847af..defb8e0f366 100644 --- a/examples/undocumented/python_modular/distribution_linearhmm_modular.py +++ b/examples/undocumented/python/distribution_linearhmm.py @@ -6,10 +6,10 @@ parameter_list = [[traindna,3,0,False],[traindna,4,0,False]] -def distribution_linearhmm_modular (fm_dna=traindna,order=3,gap=0,reverse=False): +def distribution_linearhmm (fm_dna=traindna,order=3,gap=0,reverse=False): - from modshogun import StringWordFeatures, StringCharFeatures, DNA - from modshogun import LinearHMM + from shogun import StringWordFeatures, StringCharFeatures, DNA + from shogun import LinearHMM charfeat=StringCharFeatures(DNA) charfeat.set_features(fm_dna) @@ -36,5 +36,5 @@ def distribution_linearhmm_modular (fm_dna=traindna,order=3,gap=0,reverse=False) ########################################################################### if __name__=='__main__': - distribution_linearhmm_modular(*parameter_list[0]) + distribution_linearhmm(*parameter_list[0]) print('LinearHMM') diff --git a/examples/undocumented/python_modular/distribution_ppwm_modular.py b/examples/undocumented/python/distribution_ppwm.py similarity index 86% rename from examples/undocumented/python_modular/distribution_ppwm_modular.py rename to examples/undocumented/python/distribution_ppwm.py index 01f6fcb6513..11097a4f631 100644 --- a/examples/undocumented/python_modular/distribution_ppwm_modular.py +++ b/examples/undocumented/python/distribution_ppwm.py @@ -6,9 +6,9 @@ parameter_list = [[traindna,3],[traindna,4]] -def distribution_ppwm_modular (fm_dna=traindna, order=3): - from modshogun import StringByteFeatures, StringCharFeatures, DNA - from modshogun import PositionalPWM +def distribution_ppwm (fm_dna=traindna, order=3): + from shogun import StringByteFeatures, StringCharFeatures, DNA + from shogun import PositionalPWM from numpy import array,e,log,exp @@ -62,4 +62,4 @@ def distribution_ppwm_modular (fm_dna=traindna, order=3): if __name__=='__main__': print('PositionalPWM') - distribution_ppwm_modular(*parameter_list[0]) + distribution_ppwm(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/evaluation_clustering.py b/examples/undocumented/python/evaluation_clustering.py similarity index 84% rename from examples/undocumented/python_modular/evaluation_clustering.py rename to examples/undocumented/python/evaluation_clustering.py index 22c5388eb3f..b0376772240 100644 --- a/examples/undocumented/python_modular/evaluation_clustering.py +++ b/examples/undocumented/python/evaluation_clustering.py @@ -26,9 +26,9 @@ def prepare_data(): def run_clustering(data, k): - from modshogun import KMeans - from modshogun import EuclideanDistance - from modshogun import RealFeatures + from shogun import KMeans + from shogun import EuclideanDistance + from shogun import RealFeatures fea = RealFeatures(data) distance = EuclideanDistance(fea, fea) @@ -40,9 +40,9 @@ def run_clustering(data, k): return kmeans.get_cluster_centers() def assign_labels(data, centroids, ncenters): - from modshogun import EuclideanDistance - from modshogun import RealFeatures, MulticlassLabels - from modshogun import KNN + from shogun import EuclideanDistance + from shogun import RealFeatures, MulticlassLabels + from shogun import KNN from numpy import arange labels = MulticlassLabels(arange(0.,ncenters)) @@ -54,9 +54,9 @@ def assign_labels(data, centroids, ncenters): return knn.apply(fea) def evaluation_clustering (features=fea, ground_truth=gnd_raw, ncenters=10): - from modshogun import ClusteringAccuracy, ClusteringMutualInformation - from modshogun import MulticlassLabels - from modshogun import Math + from shogun import ClusteringAccuracy, ClusteringMutualInformation + from shogun import MulticlassLabels + from shogun import Math # reproducable results Math.init_random(1) diff --git a/examples/undocumented/python_modular/evaluation_clustering_simple.py b/examples/undocumented/python/evaluation_clustering_simple.py similarity index 83% rename from examples/undocumented/python_modular/evaluation_clustering_simple.py rename to examples/undocumented/python/evaluation_clustering_simple.py index a98c9d2c138..fbce11084ad 100644 --- a/examples/undocumented/python_modular/evaluation_clustering_simple.py +++ b/examples/undocumented/python/evaluation_clustering_simple.py @@ -5,10 +5,10 @@ #from pylab import * def run_clustering(data, k): - from modshogun import KMeans - from modshogun import Math_init_random - from modshogun import EuclideanDistance - from modshogun import RealFeatures + from shogun import KMeans + from shogun import Math_init_random + from shogun import EuclideanDistance + from shogun import RealFeatures fea = RealFeatures(data) distance = EuclideanDistance(fea, fea) @@ -20,9 +20,9 @@ def run_clustering(data, k): return kmeans.get_cluster_centers() def assign_labels(data, centroids, ncenters): - from modshogun import EuclideanDistance - from modshogun import RealFeatures, MulticlassLabels - from modshogun import KNN + from shogun import EuclideanDistance + from shogun import RealFeatures, MulticlassLabels + from shogun import KNN from numpy import arange labels = MulticlassLabels(arange(0.,ncenters)) @@ -34,9 +34,9 @@ def assign_labels(data, centroids, ncenters): return knn.apply(fea) def evaluation_clustering_simple (n_data=100, sqrt_num_blobs=4, distance=5): - from modshogun import ClusteringAccuracy, ClusteringMutualInformation - from modshogun import MulticlassLabels, GaussianBlobsDataGenerator - from modshogun import Math + from shogun import ClusteringAccuracy, ClusteringMutualInformation + from shogun import MulticlassLabels, GaussianBlobsDataGenerator + from shogun import Math # reproducable results Math.init_random(1) diff --git a/examples/undocumented/python_modular/evaluation_contingencytableevaluation_modular.py b/examples/undocumented/python/evaluation_contingencytableevaluation.py similarity index 78% rename from examples/undocumented/python_modular/evaluation_contingencytableevaluation_modular.py rename to examples/undocumented/python/evaluation_contingencytableevaluation.py index 44c0ae06f89..8868e2ca310 100644 --- a/examples/undocumented/python_modular/evaluation_contingencytableevaluation_modular.py +++ b/examples/undocumented/python/evaluation_contingencytableevaluation.py @@ -9,12 +9,12 @@ parameter_list = [[ground_truth,predicted]] -def evaluation_contingencytableevaluation_modular (ground_truth, predicted): - from modshogun import BinaryLabels - from modshogun import ContingencyTableEvaluation - from modshogun import AccuracyMeasure,ErrorRateMeasure,BALMeasure - from modshogun import WRACCMeasure,F1Measure,CrossCorrelationMeasure - from modshogun import RecallMeasure,PrecisionMeasure,SpecificityMeasure +def evaluation_contingencytableevaluation (ground_truth, predicted): + from shogun import BinaryLabels + from shogun import ContingencyTableEvaluation + from shogun import AccuracyMeasure,ErrorRateMeasure,BALMeasure + from shogun import WRACCMeasure,F1Measure,CrossCorrelationMeasure + from shogun import RecallMeasure,PrecisionMeasure,SpecificityMeasure ground_truth_labels = BinaryLabels(ground_truth) predicted_labels = BinaryLabels(predicted) @@ -54,5 +54,5 @@ def evaluation_contingencytableevaluation_modular (ground_truth, predicted): if __name__=='__main__': print('EvaluationContingencyTableEvaluation') - evaluation_contingencytableevaluation_modular(*parameter_list[0]) + evaluation_contingencytableevaluation(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/evaluation_cross_validation_classification.py b/examples/undocumented/python/evaluation_cross_validation_classification.py similarity index 85% rename from examples/undocumented/python_modular/evaluation_cross_validation_classification.py rename to examples/undocumented/python/evaluation_cross_validation_classification.py index deafd88a9e6..5a7def4bace 100644 --- a/examples/undocumented/python_modular/evaluation_cross_validation_classification.py +++ b/examples/undocumented/python/evaluation_cross_validation_classification.py @@ -22,12 +22,12 @@ parameter_list = [[traindat,label_traindat]] def evaluation_cross_validation_classification (traindat=traindat, label_traindat=label_traindat): - from modshogun import CrossValidation, CrossValidationResult - from modshogun import ContingencyTableEvaluation, ACCURACY - from modshogun import StratifiedCrossValidationSplitting - from modshogun import BinaryLabels - from modshogun import RealFeatures - from modshogun import LibLinear, L2R_L2LOSS_SVC + from shogun import CrossValidation, CrossValidationResult + from shogun import ContingencyTableEvaluation, ACCURACY + from shogun import StratifiedCrossValidationSplitting + from shogun import BinaryLabels + from shogun import RealFeatures + from shogun import LibLinear, L2R_L2LOSS_SVC # training data features=RealFeatures(traindat) diff --git a/examples/undocumented/python_modular/evaluation_cross_validation_mkl_weight_storage.py b/examples/undocumented/python/evaluation_cross_validation_mkl_weight_storage.py similarity index 71% rename from examples/undocumented/python_modular/evaluation_cross_validation_mkl_weight_storage.py rename to examples/undocumented/python/evaluation_cross_validation_mkl_weight_storage.py index 9c8bd61d398..27a0378dbfb 100644 --- a/examples/undocumented/python_modular/evaluation_cross_validation_mkl_weight_storage.py +++ b/examples/undocumented/python/evaluation_cross_validation_mkl_weight_storage.py @@ -22,15 +22,14 @@ parameter_list = [[traindat,label_traindat]] def evaluation_cross_validation_mkl_weight_storage(traindat=traindat, label_traindat=label_traindat): - from modshogun import CrossValidation, CrossValidationResult - from modshogun import CrossValidationPrintOutput - from modshogun import CrossValidationMKLStorage - from modshogun import ContingencyTableEvaluation, ACCURACY - from modshogun import StratifiedCrossValidationSplitting - from modshogun import BinaryLabels - from modshogun import RealFeatures, CombinedFeatures - from modshogun import GaussianKernel, CombinedKernel - from modshogun import LibSVM, MKLClassification + from shogun import CrossValidation, CrossValidationResult + from shogun import ParameterObserverCV + from shogun import ContingencyTableEvaluation, ACCURACY + from shogun import StratifiedCrossValidationSplitting + from shogun import BinaryLabels + from shogun import RealFeatures, CombinedFeatures + from shogun import GaussianKernel, CombinedKernel + from shogun import LibSVM, MKLClassification # training data, combined features all on same data features=RealFeatures(traindat) @@ -65,18 +64,25 @@ def evaluation_cross_validation_mkl_weight_storage(traindat=traindat, label_trai cross_validation.set_autolock(False) # append cross vlaidation output classes - #cross_validation.add_cross_validation_output(CrossValidationPrintOutput()) - mkl_storage=CrossValidationMKLStorage() - cross_validation.add_cross_validation_output(mkl_storage) + mkl_storage=ParameterObserverCV() + cross_validation.subscribe_to_parameters(mkl_storage) cross_validation.set_num_runs(3) # perform cross-validation result=cross_validation.evaluate() # print mkl weights - weights=mkl_storage.get_mkl_weights() - #print "mkl weights during cross--validation" - #print weights + weights = [] + for obs_index in range(mkl_storage.get_num_observations()): + obs = mkl_storage.get_observation(obs_index) + for fold_index in range(obs.get_num_folds()): + fold = obs.get_fold(fold_index) + machine = MKLClassification.obtain_from_generic(fold.get_trained_machine()) + w = machine.get_kernel().get_subkernel_weights() + weights.append(w) + + print("mkl weights during cross--validation") + print(weights) if __name__=='__main__': print('Evaluation CrossValidationClassification') diff --git a/examples/undocumented/python_modular/evaluation_cross_validation_multiclass_storage.py b/examples/undocumented/python/evaluation_cross_validation_multiclass_storage.py similarity index 64% rename from examples/undocumented/python_modular/evaluation_cross_validation_multiclass_storage.py rename to examples/undocumented/python/evaluation_cross_validation_multiclass_storage.py index 03dad281383..8a622c731f5 100644 --- a/examples/undocumented/python_modular/evaluation_cross_validation_multiclass_storage.py +++ b/examples/undocumented/python/evaluation_cross_validation_multiclass_storage.py @@ -23,16 +23,16 @@ parameter_list = [[traindat,label_traindat]] def evaluation_cross_validation_multiclass_storage (traindat=traindat, label_traindat=label_traindat): - from modshogun import CrossValidation, CrossValidationResult - from modshogun import CrossValidationPrintOutput - from modshogun import CrossValidationMKLStorage, CrossValidationMulticlassStorage - from modshogun import MulticlassAccuracy, F1Measure - from modshogun import StratifiedCrossValidationSplitting - from modshogun import MulticlassLabels - from modshogun import RealFeatures, CombinedFeatures - from modshogun import GaussianKernel, CombinedKernel - from modshogun import MKLMulticlass - from modshogun import Statistics, MSG_DEBUG, Math + from shogun import CrossValidation, CrossValidationResult + from shogun import ParameterObserverCV + from shogun import MulticlassAccuracy, F1Measure + from shogun import StratifiedCrossValidationSplitting + from shogun import MulticlassLabels + from shogun import RealFeatures, CombinedFeatures + from shogun import GaussianKernel, CombinedKernel + from shogun import MKLMulticlass + from shogun import Statistics, MSG_DEBUG, Math + from shogun import ROCEvaluation Math.init_random(1) @@ -67,23 +67,28 @@ def evaluation_cross_validation_multiclass_storage (traindat=traindat, label_tra splitting_strategy, evaluation_criterium) cross_validation.set_autolock(False) - # append cross vlaidation output classes - #cross_validation.add_cross_validation_output(CrossValidationPrintOutput()) - #mkl_storage=CrossValidationMKLStorage() - #cross_validation.add_cross_validation_output(mkl_storage) - multiclass_storage=CrossValidationMulticlassStorage() - multiclass_storage.append_binary_evaluation(F1Measure()) - cross_validation.add_cross_validation_output(multiclass_storage) + # append cross validation parameter observer + multiclass_storage=ParameterObserverCV() + cross_validation.subscribe_to_parameters(multiclass_storage) cross_validation.set_num_runs(3) # perform cross-validation result=cross_validation.evaluate() - roc_0_0_0 = multiclass_storage.get_fold_ROC(0,0,0) - #print roc_0_0_0 - auc_0_0_0 = multiclass_storage.get_fold_evaluation_result(0,0,0,0) - #print auc_0_0_0 - return roc_0_0_0, auc_0_0_0 + # get first observation and first fold + obs = multiclass_storage.get_observations()[0] + fold = obs.get_folds_results()[0] + + # get fold ROC for first class + eval_ROC = ROCEvaluation() + pred_lab_binary = MulticlassLabels.obtain_from_generic(fold.get_test_result()).get_binary_for_class(0) + true_lab_binary = MulticlassLabels.obtain_from_generic(fold.get_test_true_result()).get_binary_for_class(0) + eval_ROC.evaluate(pred_lab_binary, true_lab_binary) + print eval_ROC.get_ROC() + + # get fold evaluation result + acc_measure = F1Measure() + print acc_measure.evaluate(pred_lab_binary, true_lab_binary) if __name__=='__main__': diff --git a/examples/undocumented/python_modular/evaluation_cross_validation_regression.py b/examples/undocumented/python/evaluation_cross_validation_regression.py similarity index 87% rename from examples/undocumented/python_modular/evaluation_cross_validation_regression.py rename to examples/undocumented/python/evaluation_cross_validation_regression.py index 45ce5bd5c4b..f124809b6ca 100644 --- a/examples/undocumented/python_modular/evaluation_cross_validation_regression.py +++ b/examples/undocumented/python/evaluation_cross_validation_regression.py @@ -14,10 +14,10 @@ parameter_list = [[traindat,label_traindat,0.8,1e-6],[traindat,label_traindat,0.9,1e-7]] def evaluation_cross_validation_regression (train_fname=traindat,label_fname=label_traindat,width=0.8,tau=1e-6): - from modshogun import CrossValidation, CrossValidationResult - from modshogun import MeanSquaredError, CrossValidationSplitting - from modshogun import RegressionLabels, RealFeatures - from modshogun import GaussianKernel, KernelRidgeRegression, CSVFile + from shogun import CrossValidation, CrossValidationResult + from shogun import MeanSquaredError, CrossValidationSplitting + from shogun import RegressionLabels, RealFeatures + from shogun import GaussianKernel, KernelRidgeRegression, CSVFile # training data features=RealFeatures(CSVFile(train_fname)) diff --git a/examples/undocumented/python_modular/evaluation_director_contingencytableevaluation_modular.py b/examples/undocumented/python/evaluation_director_contingencytableevaluation.py similarity index 77% rename from examples/undocumented/python_modular/evaluation_director_contingencytableevaluation_modular.py rename to examples/undocumented/python/evaluation_director_contingencytableevaluation.py index 09447426d6d..2f839bd3b54 100644 --- a/examples/undocumented/python_modular/evaluation_director_contingencytableevaluation_modular.py +++ b/examples/undocumented/python/evaluation_director_contingencytableevaluation.py @@ -9,9 +9,9 @@ parameter_list = [[ground_truth,predicted]] -def evaluation_director_contingencytableevaluation_modular (ground_truth, predicted): +def evaluation_director_contingencytableevaluation (ground_truth, predicted): try: - from modshogun import DirectorContingencyTableEvaluation, ED_MAXIMIZE + from shogun import DirectorContingencyTableEvaluation, ED_MAXIMIZE except ImportError: print("recompile shogun with --enable-swig-directors") return @@ -24,7 +24,7 @@ def get_custom_direction(self): def get_custom_score(self): return self.get_WRACC()+self.get_BAL() - from modshogun import BinaryLabels + from shogun import BinaryLabels evaluator = SimpleWeightedBinaryEvaluator() r = evaluator.evaluate(BinaryLabels(ground_truth), BinaryLabels(predicted)) @@ -35,5 +35,5 @@ def get_custom_score(self): if __name__=='__main__': print('EvaluationDirectorContingencyTableEvaluation') - evaluation_director_contingencytableevaluation_modular(*parameter_list[0]) + evaluation_director_contingencytableevaluation(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/evaluation_meansquarederror_modular.py b/examples/undocumented/python/evaluation_meansquarederror.py similarity index 70% rename from examples/undocumented/python_modular/evaluation_meansquarederror_modular.py rename to examples/undocumented/python/evaluation_meansquarederror.py index be46a716477..141b281f845 100644 --- a/examples/undocumented/python_modular/evaluation_meansquarederror_modular.py +++ b/examples/undocumented/python/evaluation_meansquarederror.py @@ -11,9 +11,9 @@ parameter_list = [[ground_truth,predicted]] -def evaluation_meansquarederror_modular (ground_truth, predicted): - from modshogun import RegressionLabels - from modshogun import MeanSquaredError +def evaluation_meansquarederror (ground_truth, predicted): + from shogun import RegressionLabels + from shogun import MeanSquaredError ground_truth_labels = RegressionLabels(ground_truth) predicted_labels = RegressionLabels(predicted) @@ -26,5 +26,5 @@ def evaluation_meansquarederror_modular (ground_truth, predicted): if __name__=='__main__': print('MeanSquaredError') - evaluation_meansquarederror_modular(*parameter_list[0]) + evaluation_meansquarederror(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/evaluation_meansquaredlogerror_modular.py b/examples/undocumented/python/evaluation_meansquaredlogerror.py similarity index 71% rename from examples/undocumented/python_modular/evaluation_meansquaredlogerror_modular.py rename to examples/undocumented/python/evaluation_meansquaredlogerror.py index fde8efa5c77..b15c4a00b2d 100644 --- a/examples/undocumented/python_modular/evaluation_meansquaredlogerror_modular.py +++ b/examples/undocumented/python/evaluation_meansquaredlogerror.py @@ -11,9 +11,9 @@ parameter_list = [[ground_truth,predicted]] -def evaluation_meansquaredlogerror_modular (ground_truth, predicted): - from modshogun import RegressionLabels - from modshogun import MeanSquaredLogError +def evaluation_meansquaredlogerror (ground_truth, predicted): + from shogun import RegressionLabels + from shogun import MeanSquaredLogError ground_truth_labels = RegressionLabels(ground_truth) predicted_labels = RegressionLabels(predicted) @@ -26,5 +26,5 @@ def evaluation_meansquaredlogerror_modular (ground_truth, predicted): if __name__=='__main__': print('EvaluationMeanSquaredLogError') - evaluation_meansquaredlogerror_modular(*parameter_list[0]) + evaluation_meansquaredlogerror(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/evaluation_multiclassaccuracy_modular.py b/examples/undocumented/python/evaluation_multiclassaccuracy.py similarity index 73% rename from examples/undocumented/python_modular/evaluation_multiclassaccuracy_modular.py rename to examples/undocumented/python/evaluation_multiclassaccuracy.py index aab08d4540a..3c38486fed9 100644 --- a/examples/undocumented/python_modular/evaluation_multiclassaccuracy_modular.py +++ b/examples/undocumented/python/evaluation_multiclassaccuracy.py @@ -9,9 +9,9 @@ parameter_list = [[ground_truth,predicted]] -def evaluation_multiclassaccuracy_modular (ground_truth, predicted): - from modshogun import MulticlassLabels - from modshogun import MulticlassAccuracy +def evaluation_multiclassaccuracy (ground_truth, predicted): + from shogun import MulticlassLabels + from shogun import MulticlassAccuracy ground_truth_labels = MulticlassLabels(ground_truth) predicted_labels = MulticlassLabels(predicted) @@ -24,5 +24,5 @@ def evaluation_multiclassaccuracy_modular (ground_truth, predicted): if __name__=='__main__': print('MulticlassAccuracy') - evaluation_multiclassaccuracy_modular(*parameter_list[0]) + evaluation_multiclassaccuracy(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/evaluation_multiclassovrevaluation_modular.py b/examples/undocumented/python/evaluation_multiclassovrevaluation.py similarity index 71% rename from examples/undocumented/python_modular/evaluation_multiclassovrevaluation_modular.py rename to examples/undocumented/python/evaluation_multiclassovrevaluation.py index 4452611e04d..0e2f8d6a959 100644 --- a/examples/undocumented/python_modular/evaluation_multiclassovrevaluation_modular.py +++ b/examples/undocumented/python/evaluation_multiclassovrevaluation.py @@ -4,10 +4,10 @@ parameter_list = [[traindat, label_traindat]] -def evaluation_multiclassovrevaluation_modular(train_fname=traindat, label_fname=label_traindat): - from modshogun import MulticlassOVREvaluation,ROCEvaluation - from modshogun import MulticlassLibLinear,RealFeatures,ContingencyTableEvaluation,ACCURACY - from modshogun import MulticlassLabels, Math, CSVFile +def evaluation_multiclassovrevaluation(train_fname=traindat, label_fname=label_traindat): + from shogun import MulticlassOVREvaluation,ROCEvaluation + from shogun import MulticlassLibLinear,RealFeatures,ContingencyTableEvaluation,ACCURACY + from shogun import MulticlassLabels, Math, CSVFile Math.init_random(1) ground_truth_labels = MulticlassLabels(CSVFile(label_fname)) @@ -29,5 +29,5 @@ def evaluation_multiclassovrevaluation_modular(train_fname=traindat, label_fname if __name__=='__main__': print('MulticlassOVREvaluation') - evaluation_multiclassovrevaluation_modular(*parameter_list[0]) + evaluation_multiclassovrevaluation(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/evaluation_prcevaluation_modular.py b/examples/undocumented/python/evaluation_prcevaluation.py similarity index 74% rename from examples/undocumented/python_modular/evaluation_prcevaluation_modular.py rename to examples/undocumented/python/evaluation_prcevaluation.py index 9472157f1d7..6c5325ac66d 100644 --- a/examples/undocumented/python_modular/evaluation_prcevaluation_modular.py +++ b/examples/undocumented/python/evaluation_prcevaluation.py @@ -9,9 +9,9 @@ parameter_list = [[ground_truth,predicted]] -def evaluation_prcevaluation_modular (ground_truth, predicted): - from modshogun import BinaryLabels - from modshogun import PRCEvaluation +def evaluation_prcevaluation (ground_truth, predicted): + from shogun import BinaryLabels + from shogun import PRCEvaluation ground_truth_labels = BinaryLabels(ground_truth) predicted_labels = BinaryLabels(predicted) @@ -24,5 +24,5 @@ def evaluation_prcevaluation_modular (ground_truth, predicted): if __name__=='__main__': print('PRCEvaluation') - evaluation_prcevaluation_modular(*parameter_list[0]) + evaluation_prcevaluation(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/evaluation_rocevaluation_modular.py b/examples/undocumented/python/evaluation_rocevaluation.py similarity index 74% rename from examples/undocumented/python_modular/evaluation_rocevaluation_modular.py rename to examples/undocumented/python/evaluation_rocevaluation.py index b56c4987b01..932c2e009f5 100644 --- a/examples/undocumented/python_modular/evaluation_rocevaluation_modular.py +++ b/examples/undocumented/python/evaluation_rocevaluation.py @@ -9,9 +9,9 @@ parameter_list = [[ground_truth,predicted]] -def evaluation_rocevaluation_modular (ground_truth, predicted): - from modshogun import BinaryLabels - from modshogun import ROCEvaluation +def evaluation_rocevaluation (ground_truth, predicted): + from shogun import BinaryLabels + from shogun import ROCEvaluation ground_truth_labels = BinaryLabels(ground_truth) predicted_labels = BinaryLabels(predicted) @@ -24,5 +24,5 @@ def evaluation_rocevaluation_modular (ground_truth, predicted): if __name__=='__main__': print('ROCEvaluation') - evaluation_rocevaluation_modular(*parameter_list[0]) + evaluation_rocevaluation(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/evaluation_thresholds_modular.py b/examples/undocumented/python/evaluation_thresholds.py similarity index 83% rename from examples/undocumented/python_modular/evaluation_thresholds_modular.py rename to examples/undocumented/python/evaluation_thresholds.py index 2089a7ed10b..cd6f1e516b6 100644 --- a/examples/undocumented/python_modular/evaluation_thresholds_modular.py +++ b/examples/undocumented/python/evaluation_thresholds.py @@ -1,8 +1,8 @@ #!/usr/bin/env python parameter_list = [[1000]] -def evaluation_thresholds_modular (index): - from modshogun import BinaryLabels, ROCEvaluation +def evaluation_thresholds (index): + from shogun import BinaryLabels, ROCEvaluation import numpy numpy.random.seed(17) output=numpy.arange(-1,1,0.001) @@ -29,4 +29,4 @@ def evaluation_thresholds_modular (index): if __name__=='__main__': print('Evaluation with Thresholds') - evaluation_thresholds_modular(*parameter_list[0]) + evaluation_thresholds(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/features_binned_dot_modular.py b/examples/undocumented/python/features_binned_dot.py similarity index 81% rename from examples/undocumented/python_modular/features_binned_dot_modular.py rename to examples/undocumented/python/features_binned_dot.py index 4605bf6e91b..a4ef9b18141 100644 --- a/examples/undocumented/python_modular/features_binned_dot_modular.py +++ b/examples/undocumented/python/features_binned_dot.py @@ -7,8 +7,8 @@ parameter_list = [(matrix,bins)] -def features_binned_dot_modular (matrix, bins): - from modshogun import RealFeatures, BinnedDotFeatures +def features_binned_dot (matrix, bins): + from shogun import RealFeatures, BinnedDotFeatures rf=RealFeatures(matrix) #print(rf.get_feature_matrix()) @@ -29,4 +29,4 @@ def features_binned_dot_modular (matrix, bins): if __name__=='__main__': print('BinnedDotFeatures') - features_binned_dot_modular(*parameter_list[0]) + features_binned_dot(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/features_dense_modular.py b/examples/undocumented/python/features_dense.py similarity index 85% rename from examples/undocumented/python_modular/features_dense_modular.py rename to examples/undocumented/python/features_dense.py index ae671ae57b4..136bc216d4a 100644 --- a/examples/undocumented/python_modular/features_dense_modular.py +++ b/examples/undocumented/python/features_dense.py @@ -1,5 +1,5 @@ #!/usr/bin/env python -from modshogun import RealFeatures, LongIntFeatures, ByteFeatures +from shogun import RealFeatures, LongIntFeatures, ByteFeatures from numpy import array, float64, int64, uint8, all # create dense matrices A,B,C @@ -11,7 +11,7 @@ # ... of type Real, LongInt and Byte parameter_list = [[matrixA,matrixB,matrixC]] -def features_dense_modular (A=matrixA,B=matrixB,C=matrixC): +def features_dense (A=matrixA,B=matrixB,C=matrixC): a=RealFeatures(A) b=LongIntFeatures(B) @@ -42,4 +42,4 @@ def features_dense_modular (A=matrixA,B=matrixB,C=matrixC): if __name__=='__main__': print('dense') - features_dense_modular(*parameter_list[0]) + features_dense(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/features_dense_byte_modular.py b/examples/undocumented/python/features_dense_byte.py similarity index 84% rename from examples/undocumented/python_modular/features_dense_byte_modular.py rename to examples/undocumented/python/features_dense_byte.py index abe348f9b2f..f85645529c9 100644 --- a/examples/undocumented/python_modular/features_dense_byte_modular.py +++ b/examples/undocumented/python/features_dense_byte.py @@ -6,8 +6,8 @@ parameter_list=[[A]] -def features_dense_byte_modular (A): - from modshogun import ByteFeatures +def features_dense_byte (A): + from shogun import ByteFeatures # create dense features a # ... of type Byte @@ -31,4 +31,4 @@ def features_dense_byte_modular (A): if __name__=='__main__': print('ByteFeatures') - features_dense_byte_modular(*parameter_list[0]) + features_dense_byte(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/features_dense_io_modular.py b/examples/undocumented/python/features_dense_io.py similarity index 63% rename from examples/undocumented/python_modular/features_dense_io_modular.py rename to examples/undocumented/python/features_dense_io.py index 0acf352ddbf..630468c0d3c 100644 --- a/examples/undocumented/python_modular/features_dense_io_modular.py +++ b/examples/undocumented/python/features_dense_io.py @@ -1,8 +1,8 @@ #!/usr/bin/env python parameter_list=[[]] -def features_dense_io_modular(): - from modshogun import RealFeatures, CSVFile +def features_dense_io(): + from shogun import RealFeatures, CSVFile feats=RealFeatures() f=CSVFile("../data/fm_train_real.dat","r") f.set_delimiter(" ") @@ -11,4 +11,4 @@ def features_dense_io_modular(): if __name__=='__main__': print('Dense Real Features IO') - features_dense_io_modular(*parameter_list[0]) + features_dense_io(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/features_dense_longint_modular.py b/examples/undocumented/python/features_dense_longint.py similarity index 77% rename from examples/undocumented/python_modular/features_dense_longint_modular.py rename to examples/undocumented/python/features_dense_longint.py index 330dfcd5114..5a21f5b1f42 100644 --- a/examples/undocumented/python_modular/features_dense_longint_modular.py +++ b/examples/undocumented/python/features_dense_longint.py @@ -1,5 +1,5 @@ #!/usr/bin/env python -from modshogun import LongIntFeatures +from shogun import LongIntFeatures from numpy import array, int64, all # create dense matrix A @@ -8,7 +8,7 @@ parameter_list = [[matrix]] # ... of type LongInt -def features_dense_longint_modular (A=matrix): +def features_dense_longint (A=matrix): a=LongIntFeatures(A) # get first feature vector and set it @@ -22,4 +22,4 @@ def features_dense_longint_modular (A=matrix): if __name__=='__main__': print('dense_longint') - features_dense_longint_modular(*parameter_list[0]) + features_dense_longint(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/features_dense_protocols_modular.py b/examples/undocumented/python/features_dense_protocols.py similarity index 88% rename from examples/undocumented/python_modular/features_dense_protocols_modular.py rename to examples/undocumented/python/features_dense_protocols.py index 9b200a23458..a11a8639fc3 100644 --- a/examples/undocumented/python_modular/features_dense_protocols_modular.py +++ b/examples/undocumented/python/features_dense_protocols.py @@ -1,7 +1,7 @@ #!/usr/bin/env python import numpy -from modshogun import RealFeatures -from modshogun import LongIntFeatures +from shogun import RealFeatures +from shogun import LongIntFeatures from numpy import array, float64, int64 @@ -10,7 +10,7 @@ parameter_list = [[data]] -def features_dense_protocols_modular (in_data=data): +def features_dense_protocols (in_data=data): m_real=array(in_data, dtype=float64, order='F') f_real=RealFeatures(m_real) @@ -81,4 +81,4 @@ def features_dense_protocols_modular (in_data=data): if __name__=='__main__': print('dense_protocols') - features_dense_protocols_modular(*parameter_list[0]) + features_dense_protocols(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/features_dense_real_modular.py b/examples/undocumented/python/features_dense_real.py similarity index 83% rename from examples/undocumented/python_modular/features_dense_real_modular.py rename to examples/undocumented/python/features_dense_real.py index 2d25c25eac8..169f8011abd 100644 --- a/examples/undocumented/python_modular/features_dense_real_modular.py +++ b/examples/undocumented/python/features_dense_real.py @@ -1,5 +1,5 @@ #!/usr/bin/env python -from modshogun import RealFeatures +from shogun import RealFeatures from numpy import array, float64, all # create dense matrices A,B,C @@ -8,7 +8,7 @@ parameter_list = [[matrix]] # ... of type LongInt -def features_dense_real_modular (A=matrix): +def features_dense_real (A=matrix): # ... of type Real, LongInt and Byte a=RealFeatures(A) @@ -30,4 +30,4 @@ def features_dense_real_modular (A=matrix): if __name__=='__main__': print('dense_real') - features_dense_real_modular(*parameter_list[0]) + features_dense_real(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/features_dense_zero_copy_modular.py b/examples/undocumented/python/features_dense_zero_copy.py similarity index 82% rename from examples/undocumented/python_modular/features_dense_zero_copy_modular.py rename to examples/undocumented/python/features_dense_zero_copy.py index 0e28d2f48c6..9e7e0c5a40c 100644 --- a/examples/undocumented/python_modular/features_dense_zero_copy_modular.py +++ b/examples/undocumented/python/features_dense_zero_copy.py @@ -1,6 +1,6 @@ #!/usr/bin/env python import numpy -from modshogun import RealFeatures +from shogun import RealFeatures from numpy import array, float64, int64 # create dense matrice @@ -8,7 +8,7 @@ parameter_list = [[data]] -def features_dense_zero_copy_modular (in_data=data): +def features_dense_zero_copy (in_data=data): feats = None if numpy.__version__ >= '1.5': feats=numpy.array(in_data, dtype=float64, order='F') @@ -38,4 +38,4 @@ def features_dense_zero_copy_modular (in_data=data): if __name__=='__main__': print('dense_zero_copy') - features_dense_zero_copy_modular(*parameter_list[0]) + features_dense_zero_copy(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/features_director_dot_modular.py b/examples/undocumented/python/features_director_dot.py similarity index 88% rename from examples/undocumented/python_modular/features_director_dot_modular.py rename to examples/undocumented/python/features_director_dot.py index d09952ab2fc..047e2c2fda3 100644 --- a/examples/undocumented/python_modular/features_director_dot_modular.py +++ b/examples/undocumented/python/features_director_dot.py @@ -9,11 +9,11 @@ parameter_list = [[traindat,testdat,label_traindat,0.9,1e-3],[traindat,testdat,label_traindat,0.8,1e-2]] -def features_director_dot_modular (fm_train_real, fm_test_real, +def features_director_dot (fm_train_real, fm_test_real, label_train_twoclass, C, epsilon): try: - from modshogun import DirectorDotFeatures - from modshogun import RealVector + from shogun import DirectorDotFeatures + from shogun import RealVector except ImportError: print("recompile shogun with --enable-swig-directors") return @@ -61,9 +61,9 @@ def get_dim_feature_space(self): # return NumpyFeatures(self.data-other.data) - #from modshogun import RealFeatures, SparseRealFeatures, BinaryLabels - #from modshogun import LibLinear, L2R_L2LOSS_SVC_DUAL - #from modshogun import Math_init_random + #from shogun import RealFeatures, SparseRealFeatures, BinaryLabels + #from shogun import LibLinear, L2R_L2LOSS_SVC_DUAL + #from shogun import Math_init_random #Math_init_random(17) #feats_train=RealFeatures(fm_train_real) @@ -105,4 +105,4 @@ def get_dim_feature_space(self): if __name__=='__main__': print('DirectorLinear') - features_director_dot_modular(*parameter_list[0]) + features_director_dot(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/features_hasheddocdot_modular.py b/examples/undocumented/python/features_hasheddocdot.py similarity index 76% rename from examples/undocumented/python_modular/features_hasheddocdot_modular.py rename to examples/undocumented/python/features_hasheddocdot.py index bc20aebb69e..06b4bc78695 100644 --- a/examples/undocumented/python_modular/features_hasheddocdot_modular.py +++ b/examples/undocumented/python/features_hasheddocdot.py @@ -3,10 +3,10 @@ parameter_list=[[strings]] -def features_hasheddocdot_modular(strings): - from modshogun import StringCharFeatures, RAWBYTE - from modshogun import HashedDocDotFeatures - from modshogun import NGramTokenizer +def features_hasheddocdot(strings): + from shogun import StringCharFeatures, RAWBYTE + from shogun import HashedDocDotFeatures + from shogun import NGramTokenizer from numpy import array #create string features @@ -34,4 +34,4 @@ def features_hasheddocdot_modular(strings): if __name__=='__main__': print('HashedDocDotFeatures') - features_hasheddocdot_modular(*parameter_list[0]) + features_hasheddocdot(*parameter_list[0]) diff --git a/examples/undocumented/python/features_io.py b/examples/undocumented/python/features_io.py new file mode 100644 index 00000000000..ba41c42d3d0 --- /dev/null +++ b/examples/undocumented/python/features_io.py @@ -0,0 +1,87 @@ +#!/usr/bin/env python +from tools.load import LoadMatrix +lm=LoadMatrix() +data=lm.load_numbers('../data/fm_train_real.dat') +label=lm.load_numbers('../data/label_train_twoclass.dat') + +parameter_list=[[data,label]] + +def features_io (fm_train_real, label_train_twoclass): + import numpy + from shogun import SparseRealFeatures, RealFeatures, MulticlassLabels + from shogun import GaussianKernel + from shogun import LibSVMFile, CSVFile, BinaryFile, HDF5File + from tempfile import NamedTemporaryFile + + feats=SparseRealFeatures(fm_train_real) + feats2=SparseRealFeatures() + + tmp_fm_train_sparsereal_bin = NamedTemporaryFile(suffix='sparsereal.bin') + f=BinaryFile(tmp_fm_train_sparsereal_bin.name, "w") + feats.save(f) + + tmp_fm_train_sparsereal_ascii = NamedTemporaryFile(suffix='sparsereal.ascii') + f=LibSVMFile(tmp_fm_train_sparsereal_ascii.name, "w") + feats.save(f) + + f=BinaryFile(tmp_fm_train_sparsereal_bin.name) + feats2.load(f) + + f=LibSVMFile(tmp_fm_train_sparsereal_ascii.name) + feats2.load(f) + + feats=RealFeatures(fm_train_real) + feats2=RealFeatures() + + tmp_fm_train_real_bin = NamedTemporaryFile(suffix='real.bin') + f=BinaryFile(tmp_fm_train_real_bin.name, "w") + feats.save(f) + + tmp_fm_train_real_h5 = NamedTemporaryFile(suffix='real.h5') + f=HDF5File(tmp_fm_train_real_h5.name, "w", "/data/doubles") + feats.save(f) + + tmp_fm_train_real_ascii = NamedTemporaryFile(suffix='real.ascii') + f=CSVFile(tmp_fm_train_real_ascii.name, "w") + feats.save(f) + + f=BinaryFile(tmp_fm_train_real_bin.name) + feats2.load(f) + #print("diff binary", numpy.max(numpy.abs(feats2.get_feature_matrix().flatten()-fm_train_real.flatten()))) + + f=CSVFile(tmp_fm_train_real_ascii.name) + feats2.load(f) + #print("diff ascii", numpy.max(numpy.abs(feats2.get_feature_matrix().flatten()-fm_train_real.flatten()))) + + lab=MulticlassLabels(numpy.array([0.0,1.0,2.0,3.0])) + lab2=MulticlassLabels() + tmp_label_train_twoclass_ascii = NamedTemporaryFile(suffix='twoclass.ascii') + f=CSVFile(tmp_label_train_twoclass_ascii.name, "w") + lab.save(f) + + tmp_label_train_twoclass_bin = NamedTemporaryFile(suffix='twoclass.bin') + f=BinaryFile(tmp_label_train_twoclass_bin.name, "w") + lab.save(f) + + tmp_label_train_real_h5 = NamedTemporaryFile(suffix='real.h5') + f=HDF5File(tmp_label_train_real_h5.name, "w", "/data/labels") + lab.save(f) + + f=CSVFile(tmp_label_train_twoclass_ascii.name) + lab2.load(f) + + f=BinaryFile(tmp_label_train_twoclass_bin.name) + lab2.load(f) + + f=HDF5File(tmp_fm_train_real_h5.name, "r", "/data/doubles") + feats2.load(f) + #print(feats2.get_feature_matrix()) + f=HDF5File(tmp_label_train_real_h5.name, "r", "/data/labels") + lab2.load(f) + #print(lab2.get_labels()) + + return feats, feats2, lab, lab2 + +if __name__=='__main__': + print('Features IO') + features_io(*parameter_list[0]) diff --git a/examples/undocumented/python/features_read_svmlight_format.py b/examples/undocumented/python/features_read_svmlight_format.py new file mode 100644 index 00000000000..741fd59571b --- /dev/null +++ b/examples/undocumented/python/features_read_svmlight_format.py @@ -0,0 +1,16 @@ +#!/usr/bin/env python +parameter_list=[['../data/train_sparsereal.light']] + +def features_read_svmlight_format (fname): + from tempfile import NamedTemporaryFile + from shogun import SparseRealFeatures + from shogun import LibSVMFile + + f=SparseRealFeatures() + lab=f.load_with_labels(LibSVMFile(fname)) + tmp_file = NamedTemporaryFile(suffix='svmlight') + f.save_with_labels(LibSVMFile(tmp_file.name, 'w'), lab) + +if __name__=='__main__': + print('Reading SVMLIGHT format') + features_read_svmlight_format(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/features_snp_modular.py b/examples/undocumented/python/features_snp.py similarity index 71% rename from examples/undocumented/python_modular/features_snp_modular.py rename to examples/undocumented/python/features_snp.py index aae5d7fcec5..afaa31aecf3 100644 --- a/examples/undocumented/python_modular/features_snp_modular.py +++ b/examples/undocumented/python/features_snp.py @@ -1,8 +1,8 @@ #!/usr/bin/env python parameter_list=[['../data/snps.dat']] -def features_snp_modular (fname): - from modshogun import StringByteFeatures, SNPFeatures, SNP +def features_snp (fname): + from shogun import StringByteFeatures, SNPFeatures, SNP sf=StringByteFeatures(SNP) sf.load_ascii_file(fname, False, SNP, SNP) @@ -14,4 +14,4 @@ def features_snp_modular (fname): if __name__=='__main__': print('SNP Features') - features_snp_modular(*parameter_list[0]) + features_snp(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/features_sparse_modular.py b/examples/undocumented/python/features_sparse.py similarity index 90% rename from examples/undocumented/python_modular/features_sparse_modular.py rename to examples/undocumented/python/features_sparse.py index 46cd441c2fb..8c80f69f461 100644 --- a/examples/undocumented/python_modular/features_sparse_modular.py +++ b/examples/undocumented/python/features_sparse.py @@ -4,9 +4,9 @@ A=numpy.array([[1,2,3],[4,0,0],[0,0,0],[0,5,0],[0,0,6],[9,9,9]], dtype=numpy.float64) parameter_list=[[A]] -def features_sparse_modular (A): +def features_sparse (A): from scipy.sparse import csc_matrix - from modshogun import SparseRealFeatures + from shogun import SparseRealFeatures from numpy import array, float64, all # sparse representation X of dense matrix A @@ -42,4 +42,4 @@ def features_sparse_modular (A): if __name__=='__main__': print('Sparse Features') - features_sparse_modular(*parameter_list[0]) + features_sparse(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/features_string_char_modular.py b/examples/undocumented/python/features_string_char.py similarity index 82% rename from examples/undocumented/python_modular/features_string_char_modular.py rename to examples/undocumented/python/features_string_char.py index 58d87b53464..cab77c02af4 100644 --- a/examples/undocumented/python_modular/features_string_char_modular.py +++ b/examples/undocumented/python/features_string_char.py @@ -3,8 +3,8 @@ parameter_list=[[strings]] -def features_string_char_modular (strings): - from modshogun import StringCharFeatures, RAWBYTE +def features_string_char (strings): + from shogun import StringCharFeatures, RAWBYTE from numpy import array #create string features @@ -25,4 +25,4 @@ def features_string_char_modular (strings): if __name__=='__main__': print('StringCharFeatures') - features_string_char_modular(*parameter_list[0]) + features_string_char(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/features_string_char_compressed_modular.py b/examples/undocumented/python/features_string_char_compressed.py similarity index 89% rename from examples/undocumented/python_modular/features_string_char_compressed_modular.py rename to examples/undocumented/python/features_string_char_compressed.py index c7ef1e5d0a6..462809c66bc 100644 --- a/examples/undocumented/python_modular/features_string_char_compressed_modular.py +++ b/examples/undocumented/python/features_string_char_compressed.py @@ -1,10 +1,10 @@ #!/usr/bin/env python -parameter_list = [['features_string_char_compressed_modular.py']] +parameter_list = [['features_string_char_compressed.py']] -def features_string_char_compressed_modular (fname): - from modshogun import StringCharFeatures, StringFileCharFeatures, RAWBYTE - from modshogun import UNCOMPRESSED,SNAPPY,LZO,GZIP,BZIP2,LZMA, MSG_DEBUG - from modshogun import DecompressCharString +def features_string_char_compressed (fname): + from shogun import StringCharFeatures, StringFileCharFeatures, RAWBYTE + from shogun import UNCOMPRESSED,SNAPPY,LZO,GZIP,BZIP2,LZMA, MSG_DEBUG + from shogun import DecompressCharString f=StringFileCharFeatures(fname, RAWBYTE) @@ -93,4 +93,4 @@ def features_string_char_compressed_modular (fname): if __name__=='__main__': print('Compressing StringCharFileFeatures') - features_string_char_compressed_modular(*parameter_list[0]) + features_string_char_compressed(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/features_string_file_modular.py b/examples/undocumented/python/features_string_file.py similarity index 75% rename from examples/undocumented/python_modular/features_string_file_modular.py rename to examples/undocumented/python/features_string_file.py index 5917e55c12e..d0e633d1226 100644 --- a/examples/undocumented/python_modular/features_string_file_modular.py +++ b/examples/undocumented/python/features_string_file.py @@ -1,9 +1,9 @@ #!/usr/bin/env python -parameter_list=[[".", "features_string_char_modular.py"]] +parameter_list=[[".", "features_string_char.py"]] -def features_string_file_modular (directory, fname): - from modshogun import StringCharFeatures, RAWBYTE - from modshogun import CSVFile +def features_string_file (directory, fname): + from shogun import StringCharFeatures, RAWBYTE + from shogun import CSVFile # load features from directory f=StringCharFeatures(RAWBYTE) @@ -29,4 +29,4 @@ def features_string_file_modular (directory, fname): if __name__=='__main__': print('StringWordFeatures') - features_string_file_modular(*parameter_list[0]) + features_string_file(*parameter_list[0]) diff --git a/examples/undocumented/python/features_string_file_char.py b/examples/undocumented/python/features_string_file_char.py new file mode 100644 index 00000000000..3335b668535 --- /dev/null +++ b/examples/undocumented/python/features_string_file_char.py @@ -0,0 +1,12 @@ +#!/usr/bin/env python +parameter_list = [['features_string_file_char.py']] + +def features_string_file_char (fname): + from shogun import StringFileCharFeatures, RAWBYTE + f = StringFileCharFeatures(fname, RAWBYTE) + #print("strings", f.get_features()) + return f + +if __name__=='__main__': + print('Compressing StringCharFileFeatures') + features_string_file_char(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/features_string_hashed_wd_modular.py b/examples/undocumented/python/features_string_hashed_wd.py similarity index 69% rename from examples/undocumented/python_modular/features_string_hashed_wd_modular.py rename to examples/undocumented/python/features_string_hashed_wd.py index d430c9d8049..7ee0679fcee 100644 --- a/examples/undocumented/python_modular/features_string_hashed_wd_modular.py +++ b/examples/undocumented/python/features_string_hashed_wd.py @@ -1,5 +1,5 @@ #!/usr/bin/env python -from modshogun import LongIntFeatures +from shogun import LongIntFeatures from numpy import array, int64, all # create dense matrix A @@ -8,12 +8,12 @@ parameter_list = [[matrix,3,1,2],[matrix,3,1,2]] # ... of type LongInt -def features_string_hashed_wd_modular (A=matrix,order=3,start_order=1,hash_bits=2): +def features_string_hashed_wd (A=matrix,order=3,start_order=1,hash_bits=2): a=LongIntFeatures(A) from numpy import array, uint8 - from modshogun import HashedWDFeatures, StringByteFeatures, RAWDNA - from modshogun import MSG_DEBUG + from shogun import HashedWDFeatures, StringByteFeatures, RAWDNA + from shogun import MSG_DEBUG x=[array([0,1,2,3,0,1,2,3,3,2,2,1,1],dtype=uint8)] from_order=order @@ -28,4 +28,4 @@ def features_string_hashed_wd_modular (A=matrix,order=3,start_order=1,hash_bits= if __name__=='__main__': print('string_hashed_wd') - features_string_hashed_wd_modular(*parameter_list[0]) + features_string_hashed_wd(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/features_string_sliding_window_modular.py b/examples/undocumented/python/features_string_sliding_window.py similarity index 85% rename from examples/undocumented/python_modular/features_string_sliding_window_modular.py rename to examples/undocumented/python/features_string_sliding_window.py index 8d6ff6831a0..05ac8de69c7 100644 --- a/examples/undocumented/python_modular/features_string_sliding_window_modular.py +++ b/examples/undocumented/python/features_string_sliding_window.py @@ -4,9 +4,9 @@ parameter_list=[[s]] -def features_string_sliding_window_modular (strings): - from modshogun import StringCharFeatures, DNA - from modshogun import DynamicIntArray +def features_string_sliding_window (strings): + from shogun import StringCharFeatures, DNA + from shogun import DynamicIntArray f=StringCharFeatures([strings], DNA) @@ -45,4 +45,4 @@ def features_string_sliding_window_modular (strings): if __name__=='__main__': print('Sliding Window') - features_string_sliding_window_modular(*parameter_list[0]) + features_string_sliding_window(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/features_string_ulong_modular.py b/examples/undocumented/python/features_string_ulong.py similarity index 71% rename from examples/undocumented/python_modular/features_string_ulong_modular.py rename to examples/undocumented/python/features_string_ulong.py index 7ea478e34b9..ca0808299fa 100644 --- a/examples/undocumented/python_modular/features_string_ulong_modular.py +++ b/examples/undocumented/python/features_string_ulong.py @@ -2,9 +2,9 @@ parameter_list = [[0,2,0,False],[0,3,0,False]] -def features_string_ulong_modular (start=0,order=2,gap=0,rev=False): +def features_string_ulong (start=0,order=2,gap=0,rev=False): - from modshogun import StringCharFeatures, StringUlongFeatures, RAWBYTE + from shogun import StringCharFeatures, StringUlongFeatures, RAWBYTE from numpy import array, uint64 #create string features @@ -21,4 +21,4 @@ def features_string_ulong_modular (start=0,order=2,gap=0,rev=False): if __name__=='__main__': print('simple_longint') - features_string_ulong_modular(*parameter_list[0]) + features_string_ulong(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/features_string_word_modular.py b/examples/undocumented/python/features_string_word.py similarity index 80% rename from examples/undocumented/python_modular/features_string_word_modular.py rename to examples/undocumented/python/features_string_word.py index dfd58099d39..0bba571d6a8 100644 --- a/examples/undocumented/python_modular/features_string_word_modular.py +++ b/examples/undocumented/python/features_string_word.py @@ -3,8 +3,8 @@ parameter_list=[[strings,0,2,0,False]] -def features_string_word_modular (strings, start, order, gap, rev): - from modshogun import StringCharFeatures, StringWordFeatures, RAWBYTE +def features_string_word (strings, start, order, gap, rev): + from shogun import StringCharFeatures, StringWordFeatures, RAWBYTE from numpy import array, uint16 #create string features @@ -28,4 +28,4 @@ def features_string_word_modular (strings, start, order, gap, rev): if __name__=='__main__': print('StringWordFeatures') - features_string_word_modular(*parameter_list[0]) + features_string_word(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/graphical/classifier_gaussian_process_binary_classification.py b/examples/undocumented/python/graphical/classifier_gaussian_process_binary_classification.py similarity index 97% rename from examples/undocumented/python_modular/graphical/classifier_gaussian_process_binary_classification.py rename to examples/undocumented/python/graphical/classifier_gaussian_process_binary_classification.py index ce8e06feb56..2d88b9693c6 100644 --- a/examples/undocumented/python_modular/graphical/classifier_gaussian_process_binary_classification.py +++ b/examples/undocumented/python/graphical/classifier_gaussian_process_binary_classification.py @@ -35,7 +35,7 @@ def gaussian_process_binary_classification_laplace(X_train, y_train, n_test=50): # import all necessary modules from Shogun (some of them require Eigen3) try: - from modshogun import RealFeatures, BinaryLabels, GaussianKernel, \ + from shogun import RealFeatures, BinaryLabels, GaussianKernel, \ LogitLikelihood, ProbitLikelihood, ZeroMean, SingleLaplacianInferenceMethod, \ EPInferenceMethod, GaussianProcessClassification except ImportError: diff --git a/examples/undocumented/python_modular/graphical/classifier_perceptron_graphical.py b/examples/undocumented/python/graphical/classifier_perceptron_graphical.py similarity index 94% rename from examples/undocumented/python_modular/graphical/classifier_perceptron_graphical.py rename to examples/undocumented/python/graphical/classifier_perceptron_graphical.py index d7efc844432..2064921b835 100644 --- a/examples/undocumented/python_modular/graphical/classifier_perceptron_graphical.py +++ b/examples/undocumented/python/graphical/classifier_perceptron_graphical.py @@ -7,9 +7,9 @@ parameter_list = [[20, 5, 1., 1000, 1, None, 5], [100, 5, 1., 1000, 1, None, 10]] def classifier_perceptron_graphical(n=100, distance=5, learn_rate=1., max_iter=1000, num_threads=1, seed=None, nperceptrons=5): - from modshogun import RealFeatures, BinaryLabels - from modshogun import Perceptron - from modshogun import MSG_INFO + from shogun import RealFeatures, BinaryLabels + from shogun import Perceptron + from shogun import MSG_INFO # 2D data _DIM = 2 diff --git a/examples/undocumented/python_modular/graphical/cluster_kmeans.py b/examples/undocumented/python/graphical/cluster_kmeans.py similarity index 97% rename from examples/undocumented/python_modular/graphical/cluster_kmeans.py rename to examples/undocumented/python/graphical/cluster_kmeans.py index e549ab3bbbf..63a82911954 100644 --- a/examples/undocumented/python_modular/graphical/cluster_kmeans.py +++ b/examples/undocumented/python/graphical/cluster_kmeans.py @@ -2,7 +2,7 @@ from numpy import ones,zeros,cos,sin,concatenate from numpy.random import randn -from modshogun import * +from shogun import * k=4 num=1000 diff --git a/examples/undocumented/python_modular/graphical/cluster_kpp.py b/examples/undocumented/python/graphical/cluster_kpp.py similarity index 98% rename from examples/undocumented/python_modular/graphical/cluster_kpp.py rename to examples/undocumented/python/graphical/cluster_kpp.py index c0c0f6f9b9a..70b76cdeeeb 100644 --- a/examples/undocumented/python_modular/graphical/cluster_kpp.py +++ b/examples/undocumented/python/graphical/cluster_kpp.py @@ -11,7 +11,7 @@ from numpy import array,ones,zeros,cos,sin,concatenate from numpy.random import randn -from modshogun import * +from shogun import * k=2 num=500 diff --git a/examples/undocumented/python_modular/graphical/converter_fastica_bss.py b/examples/undocumented/python/graphical/converter_fastica_bss.py similarity index 100% rename from examples/undocumented/python_modular/graphical/converter_fastica_bss.py rename to examples/undocumented/python/graphical/converter_fastica_bss.py diff --git a/examples/undocumented/python_modular/graphical/converter_ffsep_bss.py b/examples/undocumented/python/graphical/converter_ffsep_bss.py similarity index 94% rename from examples/undocumented/python_modular/graphical/converter_ffsep_bss.py rename to examples/undocumented/python/graphical/converter_ffsep_bss.py index 64e01a94a3a..cfb4527e568 100644 --- a/examples/undocumented/python_modular/graphical/converter_ffsep_bss.py +++ b/examples/undocumented/python/graphical/converter_ffsep_bss.py @@ -11,8 +11,8 @@ import numpy as np import pylab as pl -from modshogun import RealFeatures -from modshogun import FFSep +from shogun import RealFeatures +from shogun import FFSep # Generate sample data np.random.seed(0) diff --git a/examples/undocumented/python_modular/graphical/converter_jade_bss.py b/examples/undocumented/python/graphical/converter_jade_bss.py similarity index 94% rename from examples/undocumented/python_modular/graphical/converter_jade_bss.py rename to examples/undocumented/python/graphical/converter_jade_bss.py index 973508a27c0..f06b2efe033 100644 --- a/examples/undocumented/python_modular/graphical/converter_jade_bss.py +++ b/examples/undocumented/python/graphical/converter_jade_bss.py @@ -11,8 +11,8 @@ import numpy as np import pylab as pl -from modshogun import RealFeatures -from modshogun import Jade +from shogun import RealFeatures +from shogun import Jade # Generate sample data np.random.seed(0) diff --git a/examples/undocumented/python_modular/graphical/converter_jedi_bss.py b/examples/undocumented/python/graphical/converter_jedi_bss.py similarity index 94% rename from examples/undocumented/python_modular/graphical/converter_jedi_bss.py rename to examples/undocumented/python/graphical/converter_jedi_bss.py index 184f86f5a33..b110ccc493b 100644 --- a/examples/undocumented/python_modular/graphical/converter_jedi_bss.py +++ b/examples/undocumented/python/graphical/converter_jedi_bss.py @@ -11,8 +11,8 @@ import numpy as np import pylab as pl -from modshogun import RealFeatures -from modshogun import JediSep +from shogun import RealFeatures +from shogun import JediSep # Generate sample data np.random.seed(0) diff --git a/examples/undocumented/python_modular/graphical/converter_sobi_bss.py b/examples/undocumented/python/graphical/converter_sobi_bss.py similarity index 94% rename from examples/undocumented/python_modular/graphical/converter_sobi_bss.py rename to examples/undocumented/python/graphical/converter_sobi_bss.py index 808f0880ad8..67b11eaab4e 100644 --- a/examples/undocumented/python_modular/graphical/converter_sobi_bss.py +++ b/examples/undocumented/python/graphical/converter_sobi_bss.py @@ -11,8 +11,8 @@ import numpy as np import pylab as pl -from modshogun import RealFeatures -from modshogun import SOBI +from shogun import RealFeatures +from shogun import SOBI # Generate sample data np.random.seed(0) diff --git a/examples/undocumented/python_modular/graphical/converter_spe_helix.py b/examples/undocumented/python/graphical/converter_spe_helix.py similarity index 94% rename from examples/undocumented/python_modular/graphical/converter_spe_helix.py rename to examples/undocumented/python/graphical/converter_spe_helix.py index c3c79aaf6d3..08164b1d2d0 100644 --- a/examples/undocumented/python_modular/graphical/converter_spe_helix.py +++ b/examples/undocumented/python/graphical/converter_spe_helix.py @@ -19,9 +19,9 @@ import pylab import util -from modshogun import RealFeatures -from modshogun import StochasticProximityEmbedding, SPE_GLOBAL -from modshogun import SPE_LOCAL, Isomap +from shogun import RealFeatures +from shogun import StochasticProximityEmbedding, SPE_GLOBAL +from shogun import SPE_LOCAL, Isomap # Number of data points N = 500 diff --git a/examples/undocumented/python_modular/graphical/converter_uwedge_bss.py b/examples/undocumented/python/graphical/converter_uwedge_bss.py similarity index 93% rename from examples/undocumented/python_modular/graphical/converter_uwedge_bss.py rename to examples/undocumented/python/graphical/converter_uwedge_bss.py index 0450d7fee1f..bbab4c6ccaa 100644 --- a/examples/undocumented/python_modular/graphical/converter_uwedge_bss.py +++ b/examples/undocumented/python/graphical/converter_uwedge_bss.py @@ -8,8 +8,8 @@ import numpy as np import pylab as pl -from modshogun import RealFeatures -from modshogun import UWedgeSep +from shogun import RealFeatures +from shogun import UWedgeSep # Generate sample data np.random.seed(0) diff --git a/examples/undocumented/python_modular/graphical/eigenfaces.py b/examples/undocumented/python/graphical/eigenfaces.py similarity index 98% rename from examples/undocumented/python_modular/graphical/eigenfaces.py rename to examples/undocumented/python/graphical/eigenfaces.py index d4a888f4627..60253957680 100644 --- a/examples/undocumented/python_modular/graphical/eigenfaces.py +++ b/examples/undocumented/python/graphical/eigenfaces.py @@ -41,9 +41,9 @@ import numpy as np from numpy import random -from modshogun import RealFeatures -from modshogun import PCA -from modshogun import EuclideanDistance +from shogun import RealFeatures +from shogun import PCA +from shogun import EuclideanDistance import math import os import pylab as pl diff --git a/examples/undocumented/python_modular/graphical/em_1d_gmm.py b/examples/undocumented/python/graphical/em_1d_gmm.py similarity index 96% rename from examples/undocumented/python_modular/graphical/em_1d_gmm.py rename to examples/undocumented/python/graphical/em_1d_gmm.py index cb00495acb2..1e826e3a514 100644 --- a/examples/undocumented/python_modular/graphical/em_1d_gmm.py +++ b/examples/undocumented/python/graphical/em_1d_gmm.py @@ -1,7 +1,7 @@ from pylab import figure,show,connect,hist,plot,legend from numpy import array, append, arange, empty, exp -from modshogun import Gaussian, GMM -from modshogun import RealFeatures +from shogun import Gaussian, GMM +from shogun import RealFeatures import util util.set_title('EM for 1d GMM example') diff --git a/examples/undocumented/python_modular/graphical/em_2d_gmm.py b/examples/undocumented/python/graphical/em_2d_gmm.py similarity index 97% rename from examples/undocumented/python_modular/graphical/em_2d_gmm.py rename to examples/undocumented/python/graphical/em_2d_gmm.py index 92cb298de37..90551c7b440 100644 --- a/examples/undocumented/python_modular/graphical/em_2d_gmm.py +++ b/examples/undocumented/python/graphical/em_2d_gmm.py @@ -1,7 +1,7 @@ from pylab import figure,scatter,contour,show,legend,connect from numpy import array, append, arange, reshape, empty, exp -from modshogun import Gaussian, GMM -from modshogun import RealFeatures +from shogun import Gaussian, GMM +from shogun import RealFeatures import util util.set_title('EM for 2d GMM example') diff --git a/examples/undocumented/python_modular/graphical/group_lasso.py b/examples/undocumented/python/graphical/group_lasso.py similarity index 98% rename from examples/undocumented/python_modular/graphical/group_lasso.py rename to examples/undocumented/python/graphical/group_lasso.py index fb27e2361c9..1de54d418b6 100644 --- a/examples/undocumented/python_modular/graphical/group_lasso.py +++ b/examples/undocumented/python/graphical/group_lasso.py @@ -4,7 +4,7 @@ import matplotlib.pyplot as plt from numpy.random import rand, randn, permutation, multivariate_normal -from modshogun import BinaryLabels, RealFeatures, IndexBlock, IndexBlockGroup, FeatureBlockLogisticRegression +from shogun import BinaryLabels, RealFeatures, IndexBlock, IndexBlockGroup, FeatureBlockLogisticRegression def generate_synthetic_logistic_data(n, p, L, blk_nnz, gcov, nstd): diff --git a/examples/undocumented/python_modular/graphical/interactive_clustering_demo.py b/examples/undocumented/python/graphical/interactive_clustering_demo.py similarity index 99% rename from examples/undocumented/python_modular/graphical/interactive_clustering_demo.py rename to examples/undocumented/python/graphical/interactive_clustering_demo.py index ad434521619..38c14597cae 100644 --- a/examples/undocumented/python_modular/graphical/interactive_clustering_demo.py +++ b/examples/undocumented/python/graphical/interactive_clustering_demo.py @@ -16,9 +16,9 @@ from matplotlib.backends.backend_qt4agg import NavigationToolbar2QT as NavigationToolbar from matplotlib.figure import Figure -from modshogun import * -from modshogun import * -from modshogun import * +from shogun import * +from shogun import * +from shogun import * import util class Form(QMainWindow): diff --git a/examples/undocumented/python_modular/graphical/interactive_gp_demo.py b/examples/undocumented/python/graphical/interactive_gp_demo.py similarity index 99% rename from examples/undocumented/python_modular/graphical/interactive_gp_demo.py rename to examples/undocumented/python/graphical/interactive_gp_demo.py index ebbff94d0e2..9e1846eb52c 100644 --- a/examples/undocumented/python_modular/graphical/interactive_gp_demo.py +++ b/examples/undocumented/python/graphical/interactive_gp_demo.py @@ -29,9 +29,9 @@ from matplotlib.backends.backend_qt4agg import NavigationToolbar2QT as NavigationToolbar from matplotlib.figure import Figure -from modshogun import * -from modshogun import * -from modshogun import * +from shogun import * +from shogun import * +from shogun import * import util class Form(QMainWindow): diff --git a/examples/undocumented/python_modular/graphical/interactive_kmm_demo.py b/examples/undocumented/python/graphical/interactive_kmm_demo.py similarity index 99% rename from examples/undocumented/python_modular/graphical/interactive_kmm_demo.py rename to examples/undocumented/python/graphical/interactive_kmm_demo.py index 80f957201b7..9eae0833371 100644 --- a/examples/undocumented/python_modular/graphical/interactive_kmm_demo.py +++ b/examples/undocumented/python/graphical/interactive_kmm_demo.py @@ -25,9 +25,9 @@ from matplotlib.backends.backend_qt4agg import NavigationToolbar2QT as NavigationToolbar from matplotlib.figure import Figure -from modshogun import * -from modshogun import KernelMeanMatching -from modshogun import Math +from shogun import * +from shogun import KernelMeanMatching +from shogun import Math import util class Form(QMainWindow): diff --git a/examples/undocumented/python_modular/graphical/interactive_svm_demo.py b/examples/undocumented/python/graphical/interactive_svm_demo.py similarity index 99% rename from examples/undocumented/python_modular/graphical/interactive_svm_demo.py rename to examples/undocumented/python/graphical/interactive_svm_demo.py index 14b00292d29..1732f89394d 100644 --- a/examples/undocumented/python_modular/graphical/interactive_svm_demo.py +++ b/examples/undocumented/python/graphical/interactive_svm_demo.py @@ -16,7 +16,7 @@ from matplotlib.backends.backend_qt4agg import NavigationToolbar2QT as NavigationToolbar from matplotlib.figure import Figure -from modshogun import * +from shogun import * import util class Form(QMainWindow): diff --git a/examples/undocumented/python_modular/graphical/interactive_svr_demo.py b/examples/undocumented/python/graphical/interactive_svr_demo.py similarity index 99% rename from examples/undocumented/python_modular/graphical/interactive_svr_demo.py rename to examples/undocumented/python/graphical/interactive_svr_demo.py index b0276ae3765..302e9e86c82 100644 --- a/examples/undocumented/python_modular/graphical/interactive_svr_demo.py +++ b/examples/undocumented/python/graphical/interactive_svr_demo.py @@ -16,7 +16,7 @@ from matplotlib.backends.backend_qt4agg import NavigationToolbar2QT as NavigationToolbar from matplotlib.figure import Figure -from modshogun import * +from shogun import * class Form(QMainWindow): def __init__(self, parent=None): diff --git a/examples/undocumented/python_modular/graphical/inverse_covariance_estimation_demo.py b/examples/undocumented/python/graphical/inverse_covariance_estimation_demo.py similarity index 96% rename from examples/undocumented/python_modular/graphical/inverse_covariance_estimation_demo.py rename to examples/undocumented/python/graphical/inverse_covariance_estimation_demo.py index 09faf32ac6f..91848758b03 100755 --- a/examples/undocumented/python_modular/graphical/inverse_covariance_estimation_demo.py +++ b/examples/undocumented/python/graphical/inverse_covariance_estimation_demo.py @@ -5,7 +5,7 @@ from pylab import show, imshow def simulate_data (n,p): - from modshogun import SparseInverseCovariance + from shogun import SparseInverseCovariance import numpy as np #create a random pxp covariance matrix @@ -18,7 +18,7 @@ def simulate_data (n,p): return data def inverse_covariance (data,lc): - from modshogun import SparseInverseCovariance + from shogun import SparseInverseCovariance from numpy import dot sic = SparseInverseCovariance() diff --git a/examples/undocumented/python_modular/graphical/kernel_ridge_regression.py b/examples/undocumented/python/graphical/kernel_ridge_regression.py similarity index 97% rename from examples/undocumented/python_modular/graphical/kernel_ridge_regression.py rename to examples/undocumented/python/graphical/kernel_ridge_regression.py index 7d9a9227b32..be72f275076 100644 --- a/examples/undocumented/python_modular/graphical/kernel_ridge_regression.py +++ b/examples/undocumented/python/graphical/kernel_ridge_regression.py @@ -1,7 +1,7 @@ from pylab import figure,pcolor,scatter,contour,colorbar,show,subplot,plot,connect from numpy import array,meshgrid,reshape,linspace,min,max from numpy import concatenate,transpose,ravel -from modshogun import * +from shogun import * import util util.set_title('KernelRidgeRegression') diff --git a/examples/undocumented/python_modular/graphical/kernel_ridge_regression_sinc.py b/examples/undocumented/python/graphical/kernel_ridge_regression_sinc.py similarity index 96% rename from examples/undocumented/python_modular/graphical/kernel_ridge_regression_sinc.py rename to examples/undocumented/python/graphical/kernel_ridge_regression_sinc.py index 2ea9a0aa311..7f9d68d8284 100644 --- a/examples/undocumented/python_modular/graphical/kernel_ridge_regression_sinc.py +++ b/examples/undocumented/python/graphical/kernel_ridge_regression_sinc.py @@ -1,5 +1,5 @@ from pylab import figure,pcolor,scatter,contour,colorbar,show,subplot,plot,legend,connect -from modshogun import * +from shogun import * import util util.set_title('KernelRidgeRegression on Sine') diff --git a/examples/undocumented/python_modular/graphical/latex_plot_inits.py b/examples/undocumented/python/graphical/latex_plot_inits.py similarity index 100% rename from examples/undocumented/python_modular/graphical/latex_plot_inits.py rename to examples/undocumented/python/graphical/latex_plot_inits.py diff --git a/examples/undocumented/python_modular/graphical/lda.py b/examples/undocumented/python/graphical/lda.py similarity index 96% rename from examples/undocumented/python_modular/graphical/lda.py rename to examples/undocumented/python/graphical/lda.py index 14b4ccf9c47..9eef3796f2f 100644 --- a/examples/undocumented/python_modular/graphical/lda.py +++ b/examples/undocumented/python/graphical/lda.py @@ -1,5 +1,5 @@ from pylab import figure,pcolor,scatter,contour,colorbar,show,subplot,plot,connect -from modshogun import * +from shogun import * import util util.set_title('LDA') diff --git a/examples/undocumented/python_modular/graphical/mclda.py b/examples/undocumented/python/graphical/mclda.py similarity index 93% rename from examples/undocumented/python_modular/graphical/mclda.py rename to examples/undocumented/python/graphical/mclda.py index 961b93dfe7d..eae10a4ec65 100644 --- a/examples/undocumented/python_modular/graphical/mclda.py +++ b/examples/undocumented/python/graphical/mclda.py @@ -1,6 +1,6 @@ -from modshogun import RealFeatures -from modshogun import MulticlassLabels -from modshogun import MCLDA +from shogun import RealFeatures +from shogun import MulticlassLabels +from shogun import MCLDA from pylab import pcolor, contour, colorbar, connect, show, plot, axis import numpy as np diff --git a/examples/undocumented/python_modular/graphical/metric_lmnn_objective.py b/examples/undocumented/python/graphical/metric_lmnn_objective.py similarity index 92% rename from examples/undocumented/python_modular/graphical/metric_lmnn_objective.py rename to examples/undocumented/python/graphical/metric_lmnn_objective.py index 1fc83932eb0..bae796fd351 100644 --- a/examples/undocumented/python_modular/graphical/metric_lmnn_objective.py +++ b/examples/undocumented/python/graphical/metric_lmnn_objective.py @@ -36,10 +36,10 @@ def load_compressed_features(fname_features): def metric_lmnn_statistics(k=3, fname_features='../../data/fm_train_multiclass_digits.dat.gz', fname_labels='../../data/label_train_multiclass_digits.dat'): try: - from modshogun import LMNN, CSVFile, RealFeatures, MulticlassLabels, MSG_DEBUG + from shogun import LMNN, CSVFile, RealFeatures, MulticlassLabels, MSG_DEBUG import matplotlib.pyplot as pyplot except ImportError: - print 'Error importing modshogun or other required modules. Please, verify their installation.' + print 'Error importing shogun or other required modules. Please, verify their installation.' return features = RealFeatures(load_compressed_features(fname_features).T) diff --git a/examples/undocumented/python_modular/graphical/multiclass_qda.py b/examples/undocumented/python/graphical/multiclass_qda.py similarity index 97% rename from examples/undocumented/python_modular/graphical/multiclass_qda.py rename to examples/undocumented/python/graphical/multiclass_qda.py index 91afb419078..53419f45e57 100644 --- a/examples/undocumented/python_modular/graphical/multiclass_qda.py +++ b/examples/undocumented/python/graphical/multiclass_qda.py @@ -10,8 +10,8 @@ import util from scipy import linalg -from modshogun import QDA -from modshogun import RealFeatures, MulticlassLabels +from shogun import QDA +from shogun import RealFeatures, MulticlassLabels # colormap cmap = mpl.colors.LinearSegmentedColormap('color_classes', diff --git a/examples/undocumented/python_modular/graphical/multiple_smvs.py b/examples/undocumented/python/graphical/multiple_smvs.py similarity index 98% rename from examples/undocumented/python_modular/graphical/multiple_smvs.py rename to examples/undocumented/python/graphical/multiple_smvs.py index 21ad567584e..a728ba5b03a 100644 --- a/examples/undocumented/python_modular/graphical/multiple_smvs.py +++ b/examples/undocumented/python/graphical/multiple_smvs.py @@ -4,7 +4,7 @@ from pylab import figure,pcolor,scatter,contour,colorbar,show,subplot,connect,axis from numpy import concatenate from numpy.random import randn -from modshogun import * +from shogun import * import util util.set_title('Multiple SVMS') diff --git a/examples/undocumented/python_modular/graphical/prc.py b/examples/undocumented/python/graphical/prc.py similarity index 92% rename from examples/undocumented/python_modular/graphical/prc.py rename to examples/undocumented/python/graphical/prc.py index d13eda7ada1..79ef6c0c942 100644 --- a/examples/undocumented/python_modular/graphical/prc.py +++ b/examples/undocumented/python/graphical/prc.py @@ -1,7 +1,7 @@ from pylab import plot,grid,title,subplot,xlabel,ylabel,text,subplots_adjust,fill_between,mean,connect,show -from modshogun import GaussianKernel -from modshogun import LibSVM, LDA -from modshogun import PRCEvaluation +from shogun import GaussianKernel +from shogun import LibSVM, LDA +from shogun import PRCEvaluation import util util.set_title('PRC example') diff --git a/examples/undocumented/python_modular/graphical/preprocessor_kpca_graphical.py b/examples/undocumented/python/graphical/preprocessor_kpca_graphical.py similarity index 94% rename from examples/undocumented/python_modular/graphical/preprocessor_kpca_graphical.py rename to examples/undocumented/python/graphical/preprocessor_kpca_graphical.py index a2344ef8629..c6ef1437235 100644 --- a/examples/undocumented/python_modular/graphical/preprocessor_kpca_graphical.py +++ b/examples/undocumented/python/graphical/preprocessor_kpca_graphical.py @@ -25,9 +25,9 @@ parameter_list = [[data,0.01,1.0], [data,0.05,2.0]] def preprocessor_kernelpca_modular (data, threshold, width): - from modshogun import RealFeatures - from modshogun import KernelPCA - from modshogun import GaussianKernel + from shogun import RealFeatures + from shogun import KernelPCA + from shogun import GaussianKernel features = RealFeatures(data) kernel=GaussianKernel(features,features,width) preprocessor=KernelPCA(kernel) diff --git a/examples/undocumented/python_modular/graphical/qda.py b/examples/undocumented/python/graphical/qda.py similarity index 93% rename from examples/undocumented/python_modular/graphical/qda.py rename to examples/undocumented/python/graphical/qda.py index 3361cf12378..e8486410e5d 100644 --- a/examples/undocumented/python_modular/graphical/qda.py +++ b/examples/undocumented/python/graphical/qda.py @@ -1,6 +1,6 @@ -from modshogun import RealFeatures -from modshogun import MulticlassLabels -from modshogun import QDA +from shogun import RealFeatures +from shogun import MulticlassLabels +from shogun import QDA from pylab import pcolor, contour, colorbar, connect, show, plot, axis import numpy as np diff --git a/examples/undocumented/python_modular/graphical/regression_gaussian_process_demo.py b/examples/undocumented/python/graphical/regression_gaussian_process_demo.py similarity index 98% rename from examples/undocumented/python_modular/graphical/regression_gaussian_process_demo.py rename to examples/undocumented/python/graphical/regression_gaussian_process_demo.py index eb24bf4f1b9..132d6a71ebc 100644 --- a/examples/undocumented/python_modular/graphical/regression_gaussian_process_demo.py +++ b/examples/undocumented/python/graphical/regression_gaussian_process_demo.py @@ -6,14 +6,14 @@ ########################################################################### from numpy import * from numpy.random import randn -from modshogun import * +from shogun import * import pylab as PL import matplotlib import logging as LG import scipy as SP -from modshogun import GradientModelSelection -from modshogun import ModelSelectionParameters, R_EXP, R_LINEAR -from modshogun import ParameterCombination +from shogun import GradientModelSelection +from shogun import ModelSelectionParameters, R_EXP, R_LINEAR +from shogun import ParameterCombination def plot_training_data(x, y, shift=None, diff --git a/examples/undocumented/python_modular/graphical/regression_gaussian_process_modelselection.py b/examples/undocumented/python/graphical/regression_gaussian_process_modelselection.py similarity index 91% rename from examples/undocumented/python_modular/graphical/regression_gaussian_process_modelselection.py rename to examples/undocumented/python/graphical/regression_gaussian_process_modelselection.py index d9621f259cc..f916e7a4ea8 100644 --- a/examples/undocumented/python_modular/graphical/regression_gaussian_process_modelselection.py +++ b/examples/undocumented/python/graphical/regression_gaussian_process_modelselection.py @@ -6,10 +6,10 @@ def regression_gaussian_process_modelselection (n=100, n_test=100, \ x_range=5, x_range_test=10, noise_var=0.4): - from modshogun import RealFeatures, RegressionLabels - from modshogun import GaussianKernel - from modshogun import GradientModelSelection, ModelSelectionParameters - from modshogun import GaussianLikelihood, ZeroMean, \ + from shogun import RealFeatures, RegressionLabels + from shogun import GaussianKernel + from shogun import GradientModelSelection, ModelSelectionParameters + from shogun import GaussianLikelihood, ZeroMean, \ ExactInferenceMethod, GaussianProcessRegression, GradientCriterion, \ GradientEvaluation diff --git a/examples/undocumented/python_modular/graphical/regression_lars.py b/examples/undocumented/python/graphical/regression_lars.py similarity index 94% rename from examples/undocumented/python_modular/graphical/regression_lars.py rename to examples/undocumented/python/graphical/regression_lars.py index d78aef010c0..bab8189130a 100644 --- a/examples/undocumented/python_modular/graphical/regression_lars.py +++ b/examples/undocumented/python/graphical/regression_lars.py @@ -3,9 +3,9 @@ import numpy as np import matplotlib.pyplot as plt -from modshogun import RegressionLabels, RealFeatures -from modshogun import LeastAngleRegression, LinearRidgeRegression, LeastSquaresRegression -from modshogun import MeanSquaredError +from shogun import RegressionLabels, RealFeatures +from shogun import LeastAngleRegression, LinearRidgeRegression, LeastSquaresRegression +from shogun import MeanSquaredError # we compare LASSO with ordinary least-squares (OLE) # in the ideal case, the MSE of OLE should coincide diff --git a/examples/undocumented/python_modular/graphical/roc.py b/examples/undocumented/python/graphical/roc.py similarity index 92% rename from examples/undocumented/python_modular/graphical/roc.py rename to examples/undocumented/python/graphical/roc.py index aeb16c4275e..0c0b6c60438 100644 --- a/examples/undocumented/python_modular/graphical/roc.py +++ b/examples/undocumented/python/graphical/roc.py @@ -1,7 +1,7 @@ from pylab import plot,grid,title,subplot,xlabel,ylabel,text,subplots_adjust,fill_between,mean,connect,show -from modshogun import GaussianKernel -from modshogun import LibSVM, LDA -from modshogun import ROCEvaluation +from shogun import GaussianKernel +from shogun import LibSVM, LDA +from shogun import ROCEvaluation import util util.set_title('ROC example') diff --git a/examples/undocumented/python_modular/graphical/smem_1d_gmm.py b/examples/undocumented/python/graphical/smem_1d_gmm.py similarity index 96% rename from examples/undocumented/python_modular/graphical/smem_1d_gmm.py rename to examples/undocumented/python/graphical/smem_1d_gmm.py index 46daf184344..256404c93f7 100644 --- a/examples/undocumented/python_modular/graphical/smem_1d_gmm.py +++ b/examples/undocumented/python/graphical/smem_1d_gmm.py @@ -1,7 +1,7 @@ from pylab import figure,show,connect,hist,plot,legend from numpy import array, append, arange, empty, exp -from modshogun import Gaussian, GMM -from modshogun import RealFeatures +from shogun import Gaussian, GMM +from shogun import RealFeatures import util util.set_title('SMEM for 1d GMM example') diff --git a/examples/undocumented/python_modular/graphical/smem_2d_gmm.py b/examples/undocumented/python/graphical/smem_2d_gmm.py similarity index 98% rename from examples/undocumented/python_modular/graphical/smem_2d_gmm.py rename to examples/undocumented/python/graphical/smem_2d_gmm.py index 1b113a96989..6655aceee30 100644 --- a/examples/undocumented/python_modular/graphical/smem_2d_gmm.py +++ b/examples/undocumented/python/graphical/smem_2d_gmm.py @@ -1,7 +1,7 @@ from pylab import figure,scatter,contour,show,legend,connect from numpy import array, append, arange, reshape, empty, exp -from modshogun import Gaussian, GMM -from modshogun import RealFeatures +from shogun import Gaussian, GMM +from shogun import RealFeatures import util util.set_title('SMEM for 2d GMM example') diff --git a/examples/undocumented/python_modular/graphical/so_multiclass_BMRM.py b/examples/undocumented/python/graphical/so_multiclass_BMRM.py similarity index 92% rename from examples/undocumented/python_modular/graphical/so_multiclass_BMRM.py rename to examples/undocumented/python/graphical/so_multiclass_BMRM.py index 64295040669..8c59826e060 100644 --- a/examples/undocumented/python_modular/graphical/so_multiclass_BMRM.py +++ b/examples/undocumented/python/graphical/so_multiclass_BMRM.py @@ -3,10 +3,10 @@ import numpy as np import matplotlib.pyplot as plt -from modshogun import RealFeatures -from modshogun import MulticlassModel, MulticlassSOLabels, RealNumber, DualLibQPBMSOSVM -from modshogun import BMRM, PPBMRM, P3BMRM -from modshogun import StructuredAccuracy +from shogun import RealFeatures +from shogun import MulticlassModel, MulticlassSOLabels, RealNumber, DualLibQPBMSOSVM +from shogun import BMRM, PPBMRM, P3BMRM +from shogun import StructuredAccuracy def fill_data(cnt, minv, maxv): x1 = np.linspace(minv, maxv, cnt) diff --git a/examples/undocumented/python_modular/graphical/so_multiclass_director_BMRM.py b/examples/undocumented/python/graphical/so_multiclass_director_BMRM.py similarity index 94% rename from examples/undocumented/python_modular/graphical/so_multiclass_director_BMRM.py rename to examples/undocumented/python/graphical/so_multiclass_director_BMRM.py index e49e28987eb..0edbe80b833 100644 --- a/examples/undocumented/python_modular/graphical/so_multiclass_director_BMRM.py +++ b/examples/undocumented/python/graphical/so_multiclass_director_BMRM.py @@ -3,10 +3,10 @@ import numpy as np import matplotlib.pyplot as plt -from modshogun import RealFeatures -from modshogun import MulticlassModel, MulticlassSOLabels, RealNumber, DualLibQPBMSOSVM, DirectorStructuredModel -from modshogun import BMRM, PPBMRM, P3BMRM, ResultSet, RealVector -from modshogun import StructuredAccuracy +from shogun import RealFeatures +from shogun import MulticlassModel, MulticlassSOLabels, RealNumber, DualLibQPBMSOSVM, DirectorStructuredModel +from shogun import BMRM, PPBMRM, P3BMRM, ResultSet, RealVector +from shogun import StructuredAccuracy class MulticlassStructuredModel(DirectorStructuredModel): def __init__(self,features,labels): diff --git a/examples/undocumented/python_modular/graphical/statistics_hsic.py b/examples/undocumented/python/graphical/statistics_hsic.py similarity index 95% rename from examples/undocumented/python_modular/graphical/statistics_hsic.py rename to examples/undocumented/python/graphical/statistics_hsic.py index c5b76658179..a0e44f24d3b 100644 --- a/examples/undocumented/python_modular/graphical/statistics_hsic.py +++ b/examples/undocumented/python/graphical/statistics_hsic.py @@ -10,13 +10,13 @@ from pylab import * from scipy import * -from modshogun import RealFeatures -from modshogun import DataGenerator -from modshogun import GaussianKernel -from modshogun import HSIC -from modshogun import PERMUTATION, HSIC_GAMMA -from modshogun import EuclideanDistance -from modshogun import Statistics, Math +from shogun import RealFeatures +from shogun import DataGenerator +from shogun import GaussianKernel +from shogun import HSIC +from shogun import PERMUTATION, HSIC_GAMMA +from shogun import EuclideanDistance +from shogun import Statistics, Math # for nice plotting that fits into our shogun tutorial import latex_plot_inits diff --git a/examples/undocumented/python_modular/graphical/statistics_linear_time_mmd.py b/examples/undocumented/python/graphical/statistics_linear_time_mmd.py similarity index 94% rename from examples/undocumented/python_modular/graphical/statistics_linear_time_mmd.py rename to examples/undocumented/python/graphical/statistics_linear_time_mmd.py index f57ef905866..c2e001dc399 100644 --- a/examples/undocumented/python_modular/graphical/statistics_linear_time_mmd.py +++ b/examples/undocumented/python/graphical/statistics_linear_time_mmd.py @@ -10,13 +10,13 @@ from pylab import * from scipy import * -from modshogun import RealFeatures -from modshogun import MeanShiftDataGenerator -from modshogun import GaussianKernel, CombinedKernel -from modshogun import LinearTimeMMD, MMDKernelSelectionOpt -from modshogun import PERMUTATION, MMD1_GAUSSIAN -from modshogun import EuclideanDistance -from modshogun import Statistics, Math +from shogun import RealFeatures +from shogun import MeanShiftDataGenerator +from shogun import GaussianKernel, CombinedKernel +from shogun import LinearTimeMMD, MMDKernelSelectionOpt +from shogun import PERMUTATION, MMD1_GAUSSIAN +from shogun import EuclideanDistance +from shogun import Statistics, Math # for nice plotting that fits into our shogun tutorial import latex_plot_inits diff --git a/examples/undocumented/python_modular/graphical/statistics_quadratic_time_mmd.py b/examples/undocumented/python/graphical/statistics_quadratic_time_mmd.py similarity index 95% rename from examples/undocumented/python_modular/graphical/statistics_quadratic_time_mmd.py rename to examples/undocumented/python/graphical/statistics_quadratic_time_mmd.py index f70459583fe..1fc33ff41f6 100644 --- a/examples/undocumented/python_modular/graphical/statistics_quadratic_time_mmd.py +++ b/examples/undocumented/python/graphical/statistics_quadratic_time_mmd.py @@ -10,13 +10,13 @@ from pylab import * from scipy import * -from modshogun import RealFeatures -from modshogun import MeanShiftDataGenerator -from modshogun import GaussianKernel, CombinedKernel -from modshogun import QuadraticTimeMMD, MMDKernelSelectionMax -from modshogun import PERMUTATION, MMD2_SPECTRUM, MMD2_GAMMA, BIASED, UNBIASED -from modshogun import EuclideanDistance -from modshogun import Statistics, Math +from shogun import RealFeatures +from shogun import MeanShiftDataGenerator +from shogun import GaussianKernel, CombinedKernel +from shogun import QuadraticTimeMMD, MMDKernelSelectionMax +from shogun import PERMUTATION, MMD2_SPECTRUM, MMD2_GAMMA, BIASED, UNBIASED +from shogun import EuclideanDistance +from shogun import Statistics, Math # for nice plotting that fits into our shogun tutorial import latex_plot_inits diff --git a/examples/undocumented/python_modular/graphical/svm.py b/examples/undocumented/python/graphical/svm.py similarity index 96% rename from examples/undocumented/python_modular/graphical/svm.py rename to examples/undocumented/python/graphical/svm.py index 688aeea0637..9f6146fa259 100644 --- a/examples/undocumented/python_modular/graphical/svm.py +++ b/examples/undocumented/python/graphical/svm.py @@ -1,6 +1,6 @@ from pylab import figure,pcolor,scatter,contour,colorbar,show,subplot,plot,connect,axis from numpy.random import randn -from modshogun import * +from shogun import * import util util.set_title('SVM') diff --git a/examples/undocumented/python_modular/graphical/svmlin.py b/examples/undocumented/python/graphical/svmlin.py similarity index 98% rename from examples/undocumented/python_modular/graphical/svmlin.py rename to examples/undocumented/python/graphical/svmlin.py index 402edcdf85d..b380b7cf4ae 100644 --- a/examples/undocumented/python_modular/graphical/svmlin.py +++ b/examples/undocumented/python/graphical/svmlin.py @@ -1,5 +1,5 @@ from pylab import figure,pcolor,scatter,contour,colorbar,show,subplot,plot,axis, connect -from modshogun import * +from shogun import * import util util.set_title('SVM Linear 1') diff --git a/examples/undocumented/python_modular/graphical/svr_sinc.py b/examples/undocumented/python/graphical/svr_sinc.py similarity index 96% rename from examples/undocumented/python_modular/graphical/svr_sinc.py rename to examples/undocumented/python/graphical/svr_sinc.py index d1245f71aee..54ff5b96b3b 100644 --- a/examples/undocumented/python_modular/graphical/svr_sinc.py +++ b/examples/undocumented/python/graphical/svr_sinc.py @@ -1,5 +1,5 @@ from pylab import figure,pcolor,scatter,contour,colorbar,show,subplot,plot,legend, connect -from modshogun import * +from shogun import * import util util.set_title('SVR on Sinus') diff --git a/examples/undocumented/python_modular/graphical/util.py b/examples/undocumented/python/graphical/util.py similarity index 96% rename from examples/undocumented/python_modular/graphical/util.py rename to examples/undocumented/python/graphical/util.py index e4b749e014c..00bf51f917d 100644 --- a/examples/undocumented/python_modular/graphical/util.py +++ b/examples/undocumented/python/graphical/util.py @@ -4,7 +4,7 @@ from numpy import ones, array, double, meshgrid, reshape, linspace, \ concatenate, ravel, pi, sinc from numpy.random import randn, rand -from modshogun import BinaryLabels, RegressionLabels, RealFeatures, SparseRealFeatures +from shogun import BinaryLabels, RegressionLabels, RealFeatures, SparseRealFeatures QUITKEY='q' NUM_EXAMPLES=100 diff --git a/examples/undocumented/python_modular/kernel_anova_modular.py b/examples/undocumented/python/kernel_anova.py similarity index 72% rename from examples/undocumented/python_modular/kernel_anova_modular.py rename to examples/undocumented/python/kernel_anova.py index 2da2547cbb9..e1ab024add1 100644 --- a/examples/undocumented/python_modular/kernel_anova_modular.py +++ b/examples/undocumented/python/kernel_anova.py @@ -3,8 +3,8 @@ testdat = '../data/fm_test_real.dat' parameter_list = [[traindat,testdat,2,10], [traindat,testdat,5,10]] -def kernel_anova_modular (train_fname=traindat,test_fname=testdat,cardinality=2, size_cache=10): - from modshogun import ANOVAKernel,RealFeatures,CSVFile +def kernel_anova (train_fname=traindat,test_fname=testdat,cardinality=2, size_cache=10): + from shogun import ANOVAKernel,RealFeatures,CSVFile feats_train=RealFeatures(CSVFile(train_fname)) feats_test=RealFeatures(CSVFile(test_fname)) @@ -18,4 +18,4 @@ def kernel_anova_modular (train_fname=traindat,test_fname=testdat,cardinality=2, if __name__=='__main__': print('ANOVA') - kernel_anova_modular(*parameter_list[0]) + kernel_anova(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/kernel_auc_modular.py b/examples/undocumented/python/kernel_auc.py similarity index 68% rename from examples/undocumented/python_modular/kernel_auc_modular.py rename to examples/undocumented/python/kernel_auc.py index c5acf276a6e..eb62666c13b 100644 --- a/examples/undocumented/python_modular/kernel_auc_modular.py +++ b/examples/undocumented/python/kernel_auc.py @@ -3,9 +3,9 @@ label_traindat = '../data/label_train_twoclass.dat' parameter_list = [[traindat,label_traindat,1.7], [traindat,label_traindat,1.6]] -def kernel_auc_modular (train_fname=traindat,label_fname=label_traindat,width=1.7): - from modshogun import GaussianKernel, AUCKernel, RealFeatures - from modshogun import BinaryLabels, CSVFile +def kernel_auc (train_fname=traindat,label_fname=label_traindat,width=1.7): + from shogun import GaussianKernel, AUCKernel, RealFeatures + from shogun import BinaryLabels, CSVFile feats_train=RealFeatures(CSVFile(train_fname)) subkernel=GaussianKernel(feats_train, feats_train, width) @@ -17,4 +17,4 @@ def kernel_auc_modular (train_fname=traindat,label_fname=label_traindat,width=1. if __name__=='__main__': print('AUC') - kernel_auc_modular(*parameter_list[0]) + kernel_auc(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/kernel_cauchy_modular.py b/examples/undocumented/python/kernel_cauchy.py similarity index 73% rename from examples/undocumented/python_modular/kernel_cauchy_modular.py rename to examples/undocumented/python/kernel_cauchy.py index b0da43cc6ec..10814ebf1eb 100644 --- a/examples/undocumented/python_modular/kernel_cauchy_modular.py +++ b/examples/undocumented/python/kernel_cauchy.py @@ -4,8 +4,8 @@ parameter_list=[[traindat,testdat, 1.0],[traindat,testdat, 10.0]] -def kernel_cauchy_modular (train_fname=traindat,test_fname=testdat, sigma=1.0): - from modshogun import RealFeatures, CauchyKernel, CSVFile, EuclideanDistance +def kernel_cauchy (train_fname=traindat,test_fname=testdat, sigma=1.0): + from shogun import RealFeatures, CauchyKernel, CSVFile, EuclideanDistance feats_train=RealFeatures(CSVFile(train_fname)) feats_test=RealFeatures(CSVFile(test_fname)) @@ -20,4 +20,4 @@ def kernel_cauchy_modular (train_fname=traindat,test_fname=testdat, sigma=1.0): if __name__=='__main__': print('Cauchy') - kernel_cauchy_modular(*parameter_list[0]) + kernel_cauchy(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/kernel_chi2_modular.py b/examples/undocumented/python/kernel_chi2.py similarity index 72% rename from examples/undocumented/python_modular/kernel_chi2_modular.py rename to examples/undocumented/python/kernel_chi2.py index c157f23be49..95c4b4a62b8 100644 --- a/examples/undocumented/python_modular/kernel_chi2_modular.py +++ b/examples/undocumented/python/kernel_chi2.py @@ -4,8 +4,8 @@ parameter_list = [[traindat,testdat,1.4,10], [traindat,testdat,1.5,10]] -def kernel_chi2_modular (train_fname=traindat,test_fname=testdat,width=1.4, size_cache=10): - from modshogun import RealFeatures, Chi2Kernel, CSVFile, NormOne +def kernel_chi2 (train_fname=traindat,test_fname=testdat,width=1.4, size_cache=10): + from shogun import RealFeatures, Chi2Kernel, CSVFile, NormOne feats_train=RealFeatures(CSVFile(train_fname)) feats_test=RealFeatures(CSVFile(test_fname)) @@ -19,4 +19,4 @@ def kernel_chi2_modular (train_fname=traindat,test_fname=testdat,width=1.4, size if __name__=='__main__': print('Chi2') - kernel_chi2_modular(*parameter_list[0]) + kernel_chi2(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/kernel_circular_modular.py b/examples/undocumented/python/kernel_circular.py similarity index 73% rename from examples/undocumented/python_modular/kernel_circular_modular.py rename to examples/undocumented/python/kernel_circular.py index 17d1c41d36f..872a523c365 100644 --- a/examples/undocumented/python_modular/kernel_circular_modular.py +++ b/examples/undocumented/python/kernel_circular.py @@ -4,8 +4,8 @@ parameter_list=[[traindat,testdat, 1.0],[traindat,testdat, 5.0]] -def kernel_circular_modular(train_fname=traindat,test_fname=testdat, sigma=1.0): - from modshogun import RealFeatures, CircularKernel, EuclideanDistance, CSVFile +def kernel_circular(train_fname=traindat,test_fname=testdat, sigma=1.0): + from shogun import RealFeatures, CircularKernel, EuclideanDistance, CSVFile feats_train=RealFeatures(CSVFile(train_fname)) feats_test=RealFeatures(CSVFile(test_fname)) @@ -21,4 +21,4 @@ def kernel_circular_modular(train_fname=traindat,test_fname=testdat, sigma=1.0): if __name__=='__main__': print('Circular') - kernel_circular_modular(*parameter_list[0]) + kernel_circular(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/kernel_combined_modular.py b/examples/undocumented/python/kernel_combined.py similarity index 81% rename from examples/undocumented/python_modular/kernel_combined_modular.py rename to examples/undocumented/python/kernel_combined.py index bf35563caa7..50227811a12 100644 --- a/examples/undocumented/python_modular/kernel_combined_modular.py +++ b/examples/undocumented/python/kernel_combined.py @@ -9,9 +9,9 @@ testdna = lm.load_dna('../data/fm_test_dna.dat') parameter_list = [[traindat,testdat,traindna,testdna],[traindat,testdat,traindna,testdna]] -def kernel_combined_modular (fm_train_real=traindat,fm_test_real=testdat,fm_train_dna=traindna,fm_test_dna=testdna ): - from modshogun import CombinedKernel, GaussianKernel, FixedDegreeStringKernel, LocalAlignmentStringKernel - from modshogun import RealFeatures, StringCharFeatures, CombinedFeatures, DNA +def kernel_combined (fm_train_real=traindat,fm_test_real=testdat,fm_train_dna=traindna,fm_test_dna=testdna ): + from shogun import CombinedKernel, GaussianKernel, FixedDegreeStringKernel, LocalAlignmentStringKernel + from shogun import RealFeatures, StringCharFeatures, CombinedFeatures, DNA kernel=CombinedKernel() feats_train=CombinedFeatures() @@ -47,6 +47,6 @@ def kernel_combined_modular (fm_train_real=traindat,fm_test_real=testdat,fm_trai if __name__=='__main__': print('Combined') - kernel_combined_modular(*parameter_list[0]) + kernel_combined(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/kernel_combined_custom_poly_modular.py b/examples/undocumented/python/kernel_combined_custom_poly.py similarity index 80% rename from examples/undocumented/python_modular/kernel_combined_custom_poly_modular.py rename to examples/undocumented/python/kernel_combined_custom_poly.py index ebb57d1884b..490f7fecac9 100644 --- a/examples/undocumented/python_modular/kernel_combined_custom_poly_modular.py +++ b/examples/undocumented/python/kernel_combined_custom_poly.py @@ -7,10 +7,10 @@ parameter_list= [[traindat,testdat,label_traindat],[traindat,testdat,label_traindat]] -def kernel_combined_custom_poly_modular (train_fname = traindat,test_fname = testdat,train_label_fname=label_traindat): - from modshogun import CombinedFeatures, RealFeatures, BinaryLabels - from modshogun import CombinedKernel, PolyKernel, CustomKernel - from modshogun import LibSVM, CSVFile +def kernel_combined_custom_poly (train_fname = traindat,test_fname = testdat,train_label_fname=label_traindat): + from shogun import CombinedFeatures, RealFeatures, BinaryLabels + from shogun import CombinedKernel, PolyKernel, CustomKernel + from shogun import LibSVM, CSVFile kernel = CombinedKernel() feats_train = CombinedFeatures() @@ -53,4 +53,4 @@ def kernel_combined_custom_poly_modular (train_fname = traindat,test_fname = tes return km_train,kernel if __name__=='__main__': - kernel_combined_custom_poly_modular(*parameter_list[0]) + kernel_combined_custom_poly(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/kernel_comm_ulong_string_modular.py b/examples/undocumented/python/kernel_comm_ulong_string.py similarity index 77% rename from examples/undocumented/python_modular/kernel_comm_ulong_string_modular.py rename to examples/undocumented/python/kernel_comm_ulong_string.py index 5c7c33c8c69..f6f04e45521 100644 --- a/examples/undocumented/python_modular/kernel_comm_ulong_string_modular.py +++ b/examples/undocumented/python/kernel_comm_ulong_string.py @@ -6,11 +6,11 @@ testdat = lm.load_dna('../data/fm_test_dna.dat') parameter_list = [[traindat,testdat,3,0,False ],[traindat,testdat,4,0,False]] -def kernel_comm_ulong_string_modular (fm_train_dna=traindat,fm_test_dna=testdat, order=3, gap=0, reverse = False): +def kernel_comm_ulong_string (fm_train_dna=traindat,fm_test_dna=testdat, order=3, gap=0, reverse = False): - from modshogun import CommUlongStringKernel - from modshogun import StringUlongFeatures, StringCharFeatures, DNA - from modshogun import SortUlongString + from shogun import CommUlongStringKernel + from shogun import StringUlongFeatures, StringCharFeatures, DNA + from shogun import SortUlongString charfeat=StringCharFeatures(DNA) charfeat.set_features(fm_train_dna) @@ -40,4 +40,4 @@ def kernel_comm_ulong_string_modular (fm_train_dna=traindat,fm_test_dna=testdat, if __name__=='__main__': print('CommUlongString') - kernel_comm_ulong_string_modular(*parameter_list[0]) + kernel_comm_ulong_string(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/kernel_comm_word_string_modular.py b/examples/undocumented/python/kernel_comm_word_string.py similarity index 76% rename from examples/undocumented/python_modular/kernel_comm_word_string_modular.py rename to examples/undocumented/python/kernel_comm_word_string.py index e9179dd23be..8054abb9081 100644 --- a/examples/undocumented/python_modular/kernel_comm_word_string_modular.py +++ b/examples/undocumented/python/kernel_comm_word_string.py @@ -6,11 +6,11 @@ testdat = lm.load_dna('../data/fm_test_dna.dat') parameter_list = [[traindat,testdat,4,0,False, False],[traindat,testdat,4,0,False,False]] -def kernel_comm_word_string_modular (fm_train_dna=traindat, fm_test_dna=testdat, order=3, gap=0, reverse = False, use_sign = False): +def kernel_comm_word_string (fm_train_dna=traindat, fm_test_dna=testdat, order=3, gap=0, reverse = False, use_sign = False): - from modshogun import CommWordStringKernel - from modshogun import StringWordFeatures, StringCharFeatures, DNA - from modshogun import SortWordString + from shogun import CommWordStringKernel + from shogun import StringWordFeatures, StringCharFeatures, DNA + from shogun import SortWordString charfeat=StringCharFeatures(DNA) charfeat.set_features(fm_train_dna) @@ -37,4 +37,4 @@ def kernel_comm_word_string_modular (fm_train_dna=traindat, fm_test_dna=testdat, if __name__=='__main__': print('CommWordString') - kernel_comm_word_string_modular(*parameter_list[0]) + kernel_comm_word_string(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/kernel_const_modular.py b/examples/undocumented/python/kernel_const.py similarity index 70% rename from examples/undocumented/python_modular/kernel_const_modular.py rename to examples/undocumented/python/kernel_const.py index a9fc552dbd9..5c3f0658df2 100644 --- a/examples/undocumented/python_modular/kernel_const_modular.py +++ b/examples/undocumented/python/kernel_const.py @@ -1,9 +1,9 @@ #!/usr/bin/env python parameter_list =[[23],[24]] -def kernel_const_modular (c=23): - from modshogun import DummyFeatures - from modshogun import ConstKernel +def kernel_const (c=23): + from shogun import DummyFeatures + from shogun import ConstKernel feats_train=DummyFeatures(10) feats_test=DummyFeatures(17) @@ -17,4 +17,4 @@ def kernel_const_modular (c=23): if __name__=='__main__': print('Const') - kernel_const_modular(*parameter_list[0]) + kernel_const(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/kernel_custom_modular.py b/examples/undocumented/python/kernel_custom.py similarity index 89% rename from examples/undocumented/python_modular/kernel_custom_modular.py rename to examples/undocumented/python/kernel_custom.py index 0264719871e..c77d1017166 100644 --- a/examples/undocumented/python_modular/kernel_custom_modular.py +++ b/examples/undocumented/python/kernel_custom.py @@ -4,12 +4,12 @@ parameter_list=[[7],[8]] -def kernel_custom_modular (dim=7): +def kernel_custom (dim=7): from numpy.random import rand, seed from numpy import array, float32, int32 - from modshogun import RealFeatures - from modshogun import CustomKernel - from modshogun import IndexFeatures + from shogun import RealFeatures + from shogun import CustomKernel + from shogun import IndexFeatures seed(17) data=rand(dim, dim) @@ -56,5 +56,5 @@ def kernel_custom_modular (dim=7): if __name__=='__main__': print('Custom') - kernel_custom_modular(*parameter_list[0]) + kernel_custom(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/kernel_diag_modular.py b/examples/undocumented/python/kernel_diag.py similarity index 70% rename from examples/undocumented/python_modular/kernel_diag_modular.py rename to examples/undocumented/python/kernel_diag.py index 88894facc1a..6c24a4e9614 100644 --- a/examples/undocumented/python_modular/kernel_diag_modular.py +++ b/examples/undocumented/python/kernel_diag.py @@ -1,8 +1,8 @@ #!/usr/bin/env python parameter_list =[[23],[24]] -def kernel_diag_modular (diag=23): - from modshogun import DummyFeatures - from modshogun import DiagKernel +def kernel_diag (diag=23): + from shogun import DummyFeatures + from shogun import DiagKernel feats_train=DummyFeatures(10) feats_test=DummyFeatures(17) @@ -16,5 +16,5 @@ def kernel_diag_modular (diag=23): if __name__=='__main__': print('Diag') - kernel_diag_modular(*parameter_list[0]) + kernel_diag(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/kernel_director_linear_modular.py b/examples/undocumented/python/kernel_director_linear.py similarity index 80% rename from examples/undocumented/python_modular/kernel_director_linear_modular.py rename to examples/undocumented/python/kernel_director_linear.py index 42d560ed7d8..6892d7e9d24 100644 --- a/examples/undocumented/python_modular/kernel_director_linear_modular.py +++ b/examples/undocumented/python/kernel_director_linear.py @@ -1,13 +1,13 @@ #!/usr/bin/env python import numpy -from modshogun import RealFeatures, MSG_DEBUG +from shogun import RealFeatures, MSG_DEBUG traindat = numpy.random.random_sample((10,10)) testdat = numpy.random.random_sample((10,10)) parameter_list=[[traindat,testdat,1.2],[traindat,testdat,1.4]] -def kernel_director_linear_modular (fm_train_real=traindat,fm_test_real=testdat,scale=1.2): +def kernel_director_linear (fm_train_real=traindat,fm_test_real=testdat,scale=1.2): try: - from modshogun import DirectorKernel + from shogun import DirectorKernel except ImportError: print("recompile shogun with --enable-swig-directors") return @@ -21,8 +21,8 @@ def kernel_function(self, idx_a, idx_b): return numpy.dot(seq1, seq2) - from modshogun import LinearKernel, AvgDiagKernelNormalizer - from modshogun import Time + from shogun import LinearKernel, AvgDiagKernelNormalizer + from shogun import Time feats_train=RealFeatures(fm_train_real) #feats_train.io.set_loglevel(MSG_DEBUG) @@ -54,4 +54,4 @@ def kernel_function(self, idx_a, idx_b): if __name__=='__main__': print('DirectorLinear') - kernel_director_linear_modular(*parameter_list[0]) + kernel_director_linear(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/kernel_distance_modular.py b/examples/undocumented/python/kernel_distance.py similarity index 72% rename from examples/undocumented/python_modular/kernel_distance_modular.py rename to examples/undocumented/python/kernel_distance.py index b1081077557..29ce767eefa 100644 --- a/examples/undocumented/python_modular/kernel_distance_modular.py +++ b/examples/undocumented/python/kernel_distance.py @@ -4,8 +4,8 @@ parameter_list=[[traindat,testdat,1.7],[traindat,testdat,1.8]] -def kernel_distance_modular (train_fname=traindat,test_fname=testdat,width=1.7): - from modshogun import RealFeatures, DistanceKernel, EuclideanDistance, CSVFile +def kernel_distance (train_fname=traindat,test_fname=testdat,width=1.7): + from shogun import RealFeatures, DistanceKernel, EuclideanDistance, CSVFile feats_train=RealFeatures(CSVFile(train_fname)) feats_test=RealFeatures(CSVFile(test_fname)) @@ -20,4 +20,4 @@ def kernel_distance_modular (train_fname=traindat,test_fname=testdat,width=1.7): if __name__=='__main__': print('Distance') - kernel_distance_modular(*parameter_list[0]) + kernel_distance(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/kernel_distantsegments_modular.py b/examples/undocumented/python/kernel_distantsegments.py similarity index 71% rename from examples/undocumented/python_modular/kernel_distantsegments_modular.py rename to examples/undocumented/python/kernel_distantsegments.py index cbbf14bf411..082ec2f7fde 100644 --- a/examples/undocumented/python_modular/kernel_distantsegments_modular.py +++ b/examples/undocumented/python/kernel_distantsegments.py @@ -6,9 +6,9 @@ parameter_list = [[traindat,testdat,5,5],[traindat,testdat,6,6]] -def kernel_distantsegments_modular (fm_train_dna=traindat,fm_test_dna=testdat,delta=5, theta=5): - from modshogun import StringCharFeatures, DNA - from modshogun import DistantSegmentsKernel +def kernel_distantsegments (fm_train_dna=traindat,fm_test_dna=testdat,delta=5, theta=5): + from shogun import StringCharFeatures, DNA + from shogun import DistantSegmentsKernel feats_train=StringCharFeatures(fm_train_dna, DNA) feats_test=StringCharFeatures(fm_test_dna, DNA) @@ -24,4 +24,4 @@ def kernel_distantsegments_modular (fm_train_dna=traindat,fm_test_dna=testdat,de if __name__=='__main__': print('DistantSegments') - kernel_distantsegments_modular(*parameter_list[0]) + kernel_distantsegments(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/kernel_exponential_modular.py b/examples/undocumented/python/kernel_exponential.py similarity index 72% rename from examples/undocumented/python_modular/kernel_exponential_modular.py rename to examples/undocumented/python/kernel_exponential.py index 41aa3e8f741..4bf809a8642 100644 --- a/examples/undocumented/python_modular/kernel_exponential_modular.py +++ b/examples/undocumented/python/kernel_exponential.py @@ -4,8 +4,8 @@ parameter_list=[[traindat,testdat, 1.0],[traindat,testdat, 5.0]] -def kernel_exponential_modular (train_fname=traindat,test_fname=testdat, tau_coef=1.0): - from modshogun import RealFeatures, ExponentialKernel, EuclideanDistance, CSVFile +def kernel_exponential (train_fname=traindat,test_fname=testdat, tau_coef=1.0): + from shogun import RealFeatures, ExponentialKernel, EuclideanDistance, CSVFile feats_train=RealFeatures(CSVFile(train_fname)) feats_test=RealFeatures(CSVFile(test_fname)) @@ -21,4 +21,4 @@ def kernel_exponential_modular (train_fname=traindat,test_fname=testdat, tau_coe if __name__=='__main__': print('Exponential') - kernel_exponential_modular(*parameter_list[0]) + kernel_exponential(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/kernel_fisher_modular.py b/examples/undocumented/python/kernel_fisher.py similarity index 89% rename from examples/undocumented/python_modular/kernel_fisher_modular.py rename to examples/undocumented/python/kernel_fisher.py index 0397542ce01..99a452e07ea 100644 --- a/examples/undocumented/python_modular/kernel_fisher_modular.py +++ b/examples/undocumented/python/kernel_fisher.py @@ -11,14 +11,14 @@ fm_hmm_pos=[ traindat[i] for i in where([label_traindat==1])[1] ] fm_hmm_neg=[ traindat[i] for i in where([label_traindat==-1])[1] ] -def kernel_fisher_modular (fm_train_dna=traindat, fm_test_dna=testdat, +def kernel_fisher (fm_train_dna=traindat, fm_test_dna=testdat, label_train_dna=label_traindat, N=1,M=4,pseudo=1e-1,order=1,gap=0,reverse=False, kargs=[1,False,True]): - from modshogun import StringCharFeatures, StringWordFeatures, FKFeatures, DNA - from modshogun import PolyKernel - from modshogun import HMM, BW_NORMAL#, MSG_DEBUG + from shogun import StringCharFeatures, StringWordFeatures, FKFeatures, DNA + from shogun import PolyKernel + from shogun import HMM, BW_NORMAL#, MSG_DEBUG # train HMM for positive class charfeat=StringCharFeatures(fm_hmm_pos, DNA) @@ -66,4 +66,4 @@ def kernel_fisher_modular (fm_train_dna=traindat, fm_test_dna=testdat, if __name__=='__main__': print("Fisher Kernel") - kernel_fisher_modular(*parameter_list[0]) + kernel_fisher(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/kernel_fixed_degree_string_modular.py b/examples/undocumented/python/kernel_fixed_degree_string.py similarity index 71% rename from examples/undocumented/python_modular/kernel_fixed_degree_string_modular.py rename to examples/undocumented/python/kernel_fixed_degree_string.py index ff46c982c2a..51c8a5d42f0 100644 --- a/examples/undocumented/python_modular/kernel_fixed_degree_string_modular.py +++ b/examples/undocumented/python/kernel_fixed_degree_string.py @@ -6,9 +6,9 @@ testdat = lm.load_dna('../data/fm_test_dna.dat') parameter_list=[[traindat, testdat,3],[traindat,testdat,4]] -def kernel_fixed_degree_string_modular (fm_train_dna=traindat, fm_test_dna=testdat,degree=3): - from modshogun import StringCharFeatures, DNA - from modshogun import FixedDegreeStringKernel +def kernel_fixed_degree_string (fm_train_dna=traindat, fm_test_dna=testdat,degree=3): + from shogun import StringCharFeatures, DNA + from shogun import FixedDegreeStringKernel feats_train=StringCharFeatures(fm_train_dna, DNA) feats_test=StringCharFeatures(fm_test_dna, DNA) @@ -22,4 +22,4 @@ def kernel_fixed_degree_string_modular (fm_train_dna=traindat, fm_test_dna=testd return km_train,km_test,kernel if __name__=='__main__': print('FixedDegreeString') - kernel_fixed_degree_string_modular(*parameter_list[0]) + kernel_fixed_degree_string(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/kernel_gaussian_modular.py b/examples/undocumented/python/kernel_gaussian.py similarity index 72% rename from examples/undocumented/python_modular/kernel_gaussian_modular.py rename to examples/undocumented/python/kernel_gaussian.py index 12aa031204a..181465ffe2d 100644 --- a/examples/undocumented/python_modular/kernel_gaussian_modular.py +++ b/examples/undocumented/python/kernel_gaussian.py @@ -4,8 +4,8 @@ parameter_list=[[traindat,testdat, 1.3],[traindat,testdat, 1.4]] -def kernel_gaussian_modular (train_fname=traindat,test_fname=testdat, width=1.3): - from modshogun import RealFeatures, GaussianKernel, CSVFile +def kernel_gaussian (train_fname=traindat,test_fname=testdat, width=1.3): + from shogun import RealFeatures, GaussianKernel, CSVFile feats_train=RealFeatures(CSVFile(train_fname)) feats_test=RealFeatures(CSVFile(test_fname)) @@ -20,4 +20,4 @@ def kernel_gaussian_modular (train_fname=traindat,test_fname=testdat, width=1.3) if __name__=='__main__': print('Gaussian') - kernel_gaussian_modular(*parameter_list[0]) + kernel_gaussian(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/kernel_gaussian_shift_modular.py b/examples/undocumented/python/kernel_gaussian_shift.py similarity index 72% rename from examples/undocumented/python_modular/kernel_gaussian_shift_modular.py rename to examples/undocumented/python/kernel_gaussian_shift.py index a025b317b61..460dd43bedb 100644 --- a/examples/undocumented/python_modular/kernel_gaussian_shift_modular.py +++ b/examples/undocumented/python/kernel_gaussian_shift.py @@ -7,8 +7,8 @@ parameter_list=[[traindat,testdat,1.8,2,1],[traindat,testdat,1.9,2,1]] -def kernel_gaussian_shift_modular (train_fname=traindat,test_fname=testdat,width=1.8,max_shift=2,shift_step=1): - from modshogun import RealFeatures, GaussianShiftKernel, CSVFile +def kernel_gaussian_shift (train_fname=traindat,test_fname=testdat,width=1.8,max_shift=2,shift_step=1): + from shogun import RealFeatures, GaussianShiftKernel, CSVFile feats_train=RealFeatures(CSVFile(train_fname)) feats_test=RealFeatures(CSVFile(test_fname)) @@ -23,4 +23,4 @@ def kernel_gaussian_shift_modular (train_fname=traindat,test_fname=testdat,width if __name__=='__main__': print('GaussianShift') - kernel_gaussian_shift_modular(*parameter_list[0]) + kernel_gaussian_shift(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/kernel_histogram_word_string_modular.py b/examples/undocumented/python/kernel_histogram_word_string.py similarity index 74% rename from examples/undocumented/python_modular/kernel_histogram_word_string_modular.py rename to examples/undocumented/python/kernel_histogram_word_string.py index 140bbf3c52d..92b955b63e2 100644 --- a/examples/undocumented/python_modular/kernel_histogram_word_string_modular.py +++ b/examples/undocumented/python/kernel_histogram_word_string.py @@ -7,11 +7,11 @@ label_traindat = lm.load_labels('../data/label_train_dna.dat') parameter_list=[[traindat,testdat,label_traindat,1,1e1, 1e0],[traindat,testdat,label_traindat,1,1e4,1e4]] -def kernel_histogram_word_string_modular (fm_train_dna=traindat,fm_test_dna=testdat,label_train_dna=label_traindat,order=3,ppseudo_count=1,npseudo_count=1): +def kernel_histogram_word_string (fm_train_dna=traindat,fm_test_dna=testdat,label_train_dna=label_traindat,order=3,ppseudo_count=1,npseudo_count=1): - from modshogun import StringCharFeatures, StringWordFeatures, DNA, BinaryLabels - from modshogun import HistogramWordStringKernel, AvgDiagKernelNormalizer - from modshogun import PluginEstimate#, MSG_DEBUG + from shogun import StringCharFeatures, StringWordFeatures, DNA, BinaryLabels + from shogun import HistogramWordStringKernel, AvgDiagKernelNormalizer + from shogun import PluginEstimate#, MSG_DEBUG charfeat=StringCharFeatures(DNA) #charfeat.io.set_loglevel(MSG_DEBUG) @@ -40,4 +40,4 @@ def kernel_histogram_word_string_modular (fm_train_dna=traindat,fm_test_dna=test if __name__=='__main__': print('PluginEstimate w/ HistogramWord') - kernel_histogram_word_string_modular(*parameter_list[0]) + kernel_histogram_word_string(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/kernel_inversemultiquadric_modular.py b/examples/undocumented/python/kernel_inversemultiquadric.py similarity index 70% rename from examples/undocumented/python_modular/kernel_inversemultiquadric_modular.py rename to examples/undocumented/python/kernel_inversemultiquadric.py index 4b9490232e7..4efefd7f4e1 100644 --- a/examples/undocumented/python_modular/kernel_inversemultiquadric_modular.py +++ b/examples/undocumented/python/kernel_inversemultiquadric.py @@ -4,8 +4,8 @@ parameter_list=[[traindat,testdat, 1.0],[traindat,testdat, 5.0]] -def kernel_inversemultiquadric_modular (train_fname=traindat,test_fname=testdat, shift_coef=1.0): - from modshogun import RealFeatures, InverseMultiQuadricKernel, EuclideanDistance, CSVFile +def kernel_inversemultiquadric (train_fname=traindat,test_fname=testdat, shift_coef=1.0): + from shogun import RealFeatures, InverseMultiQuadricKernel, EuclideanDistance, CSVFile feats_train=RealFeatures(CSVFile(train_fname)) feats_test=RealFeatures(CSVFile(test_fname)) @@ -21,4 +21,4 @@ def kernel_inversemultiquadric_modular (train_fname=traindat,test_fname=testdat, if __name__=='__main__': print('InverseMultiquadric') - kernel_inversemultiquadric_modular(*parameter_list[0]) + kernel_inversemultiquadric(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/kernel_io_modular.py b/examples/undocumented/python/kernel_io.py similarity index 59% rename from examples/undocumented/python_modular/kernel_io_modular.py rename to examples/undocumented/python/kernel_io.py index 20597fa839f..720cb738763 100644 --- a/examples/undocumented/python_modular/kernel_io_modular.py +++ b/examples/undocumented/python/kernel_io.py @@ -4,31 +4,29 @@ parameter_list=[[traindat,testdat,1.9],[traindat,testdat,1.7]] -def kernel_io_modular (train_fname=traindat,test_fname=testdat,width=1.9): - from modshogun import RealFeatures, GaussianKernel, CSVFile +def kernel_io (train_fname=traindat,test_fname=testdat,width=1.9): + from shogun import RealFeatures, GaussianKernel, CSVFile + from tempfile import NamedTemporaryFile feats_train=RealFeatures(CSVFile(train_fname)) feats_test=RealFeatures(CSVFile(test_fname)) kernel=GaussianKernel(feats_train, feats_train, width) km_train=kernel.get_kernel_matrix() - f=CSVFile("tmp/gaussian_train.csv","w") + tmp_train_csv = NamedTemporaryFile(suffix='train.csv') + f=CSVFile(tmp_train_csv.name, "w") kernel.save(f) del f kernel.init(feats_train, feats_test) km_test=kernel.get_kernel_matrix() - f=CSVFile("tmp/gaussian_test.csv","w") + tmp_test_csv = NamedTemporaryFile(suffix='test.csv') + f=CSVFile(tmp_test_csv.name,"w") kernel.save(f) del f - #clean up - import os - os.unlink("tmp/gaussian_test.csv") - os.unlink("tmp/gaussian_train.csv") - return km_train, km_test, kernel if __name__=='__main__': print('Gaussian') - kernel_io_modular(*parameter_list[0]) + kernel_io(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/kernel_linear_modular.py b/examples/undocumented/python/kernel_linear.py similarity index 73% rename from examples/undocumented/python_modular/kernel_linear_modular.py rename to examples/undocumented/python/kernel_linear.py index e4c88b16078..2f668b453b4 100644 --- a/examples/undocumented/python_modular/kernel_linear_modular.py +++ b/examples/undocumented/python/kernel_linear.py @@ -4,9 +4,9 @@ parameter_list=[[traindat,testdat,1.2],[traindat,testdat,1.4]] -def kernel_linear_modular (train_fname=traindat,test_fname=testdat,scale=1.2): +def kernel_linear (train_fname=traindat,test_fname=testdat,scale=1.2): - from modshogun import RealFeatures, LinearKernel, AvgDiagKernelNormalizer, CSVFile + from shogun import RealFeatures, LinearKernel, AvgDiagKernelNormalizer, CSVFile feats_train=RealFeatures(CSVFile(train_fname)) feats_test=RealFeatures(CSVFile(test_fname)) @@ -22,4 +22,4 @@ def kernel_linear_modular (train_fname=traindat,test_fname=testdat,scale=1.2): if __name__=='__main__': print('Linear') - kernel_linear_modular(*parameter_list[0]) + kernel_linear(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/kernel_linear_byte_modular.py b/examples/undocumented/python/kernel_linear_byte.py similarity index 72% rename from examples/undocumented/python_modular/kernel_linear_byte_modular.py rename to examples/undocumented/python/kernel_linear_byte.py index e0f8068cad9..41ebbee28a9 100644 --- a/examples/undocumented/python_modular/kernel_linear_byte_modular.py +++ b/examples/undocumented/python/kernel_linear_byte.py @@ -4,8 +4,8 @@ parameter_list=[[traindat,testdat],[traindat,testdat]] -def kernel_linear_byte_modular (train_fname=traindat,test_fname=testdat): - from modshogun import LinearKernel, ByteFeatures, CSVFile +def kernel_linear_byte (train_fname=traindat,test_fname=testdat): + from shogun import LinearKernel, ByteFeatures, CSVFile feats_train=ByteFeatures(CSVFile(train_fname)) feats_test=ByteFeatures(CSVFile(test_fname)) @@ -19,4 +19,4 @@ def kernel_linear_byte_modular (train_fname=traindat,test_fname=testdat): if __name__=='__main__': print('LinearByte') - kernel_linear_byte_modular(*parameter_list[0]) + kernel_linear_byte(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/kernel_linear_string_modular.py b/examples/undocumented/python/kernel_linear_string.py similarity index 73% rename from examples/undocumented/python_modular/kernel_linear_string_modular.py rename to examples/undocumented/python/kernel_linear_string.py index 3c4b4f1d168..22eeb1d63ef 100644 --- a/examples/undocumented/python_modular/kernel_linear_string_modular.py +++ b/examples/undocumented/python/kernel_linear_string.py @@ -6,9 +6,9 @@ parameter_list=[[traindat,testdat],[traindat,testdat]] -def kernel_linear_string_modular (fm_train_dna=traindat,fm_test_dna=testdat): - from modshogun import StringCharFeatures, DNA - from modshogun import LinearStringKernel +def kernel_linear_string (fm_train_dna=traindat,fm_test_dna=testdat): + from shogun import StringCharFeatures, DNA + from shogun import LinearStringKernel feats_train=StringCharFeatures(fm_train_dna, DNA) feats_test=StringCharFeatures(fm_test_dna, DNA) @@ -23,4 +23,4 @@ def kernel_linear_string_modular (fm_train_dna=traindat,fm_test_dna=testdat): if __name__=='__main__': from tools.load import LoadMatrix print('LinearString') - kernel_linear_string_modular(*parameter_list[0]) + kernel_linear_string(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/kernel_linear_word_modular.py b/examples/undocumented/python/kernel_linear_word.py similarity index 74% rename from examples/undocumented/python_modular/kernel_linear_word_modular.py rename to examples/undocumented/python/kernel_linear_word.py index e484a84d275..1ac78017dd0 100644 --- a/examples/undocumented/python_modular/kernel_linear_word_modular.py +++ b/examples/undocumented/python/kernel_linear_word.py @@ -8,10 +8,10 @@ parameter_list=[[traindat,testdat,1.2],[traindat,testdat,1.2]] -def kernel_linear_word_modular (fm_train_word=traindat,fm_test_word=testdat,scale=1.2): +def kernel_linear_word (fm_train_word=traindat,fm_test_word=testdat,scale=1.2): - from modshogun import LinearKernel, AvgDiagKernelNormalizer - from modshogun import WordFeatures + from shogun import LinearKernel, AvgDiagKernelNormalizer + from shogun import WordFeatures feats_train=WordFeatures(fm_train_word) feats_test=WordFeatures(fm_test_word) @@ -27,4 +27,4 @@ def kernel_linear_word_modular (fm_train_word=traindat,fm_test_word=testdat,scal if __name__=='__main__': print('LinearWord') - kernel_linear_word_modular(*parameter_list[0]) + kernel_linear_word(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/kernel_local_alignment_string_modular.py b/examples/undocumented/python/kernel_local_alignment_string.py similarity index 70% rename from examples/undocumented/python_modular/kernel_local_alignment_string_modular.py rename to examples/undocumented/python/kernel_local_alignment_string.py index 0a4990932b3..e2ab9aeb245 100644 --- a/examples/undocumented/python_modular/kernel_local_alignment_string_modular.py +++ b/examples/undocumented/python/kernel_local_alignment_string.py @@ -6,10 +6,10 @@ testdat = lm.load_dna('../data/fm_test_dna.dat') parameter_list=[[traindat,testdat],[traindat,testdat]] -def kernel_local_alignment_string_modular (fm_train_dna=traindat,fm_test_dna=testdat): +def kernel_local_alignment_string (fm_train_dna=traindat,fm_test_dna=testdat): - from modshogun import StringCharFeatures, DNA - from modshogun import LocalAlignmentStringKernel + from shogun import StringCharFeatures, DNA + from shogun import LocalAlignmentStringKernel feats_train=StringCharFeatures(fm_train_dna, DNA) feats_test=StringCharFeatures(fm_test_dna, DNA) @@ -23,4 +23,4 @@ def kernel_local_alignment_string_modular (fm_train_dna=traindat,fm_test_dna=tes if __name__=='__main__': print('LocalAlignmentString') - kernel_local_alignment_string_modular(*parameter_list[0]) + kernel_local_alignment_string(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/kernel_locality_improved_string_modular.py b/examples/undocumented/python/kernel_locality_improved_string.py similarity index 69% rename from examples/undocumented/python_modular/kernel_locality_improved_string_modular.py rename to examples/undocumented/python/kernel_locality_improved_string.py index 581c3b64179..95d01f499b0 100644 --- a/examples/undocumented/python_modular/kernel_locality_improved_string_modular.py +++ b/examples/undocumented/python/kernel_locality_improved_string.py @@ -6,10 +6,10 @@ testdat = lm.load_dna('../data/fm_test_dna.dat') parameter_list=[[traindat,testdat,5,5,7],[traindat,testdat,5,5,7]] -def kernel_locality_improved_string_modular (fm_train_dna=traindat,fm_test_dna=testdat,length=5,inner_degree=5,outer_degree=7): +def kernel_locality_improved_string (fm_train_dna=traindat,fm_test_dna=testdat,length=5,inner_degree=5,outer_degree=7): - from modshogun import StringCharFeatures, DNA - from modshogun import LocalityImprovedStringKernel + from shogun import StringCharFeatures, DNA + from shogun import LocalityImprovedStringKernel feats_train=StringCharFeatures(fm_train_dna, DNA) feats_test=StringCharFeatures(fm_test_dna, DNA) @@ -24,4 +24,4 @@ def kernel_locality_improved_string_modular (fm_train_dna=traindat,fm_test_dna=t if __name__=='__main__': print('LocalityImprovedString') - kernel_locality_improved_string_modular(*parameter_list[0]) + kernel_locality_improved_string(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/kernel_log_modular.py b/examples/undocumented/python/kernel_log.py similarity index 74% rename from examples/undocumented/python_modular/kernel_log_modular.py rename to examples/undocumented/python/kernel_log.py index 95d93e06d8a..3c6cc15f84b 100644 --- a/examples/undocumented/python_modular/kernel_log_modular.py +++ b/examples/undocumented/python/kernel_log.py @@ -4,8 +4,8 @@ parameter_list=[[traindat,testdat, 2.0],[traindat,testdat, 3.0]] -def kernel_log_modular (train_fname=traindat,test_fname=testdat, degree=2.0): - from modshogun import RealFeatures, LogKernel, EuclideanDistance, CSVFile +def kernel_log (train_fname=traindat,test_fname=testdat, degree=2.0): + from shogun import RealFeatures, LogKernel, EuclideanDistance, CSVFile feats_train=RealFeatures(CSVFile(train_fname)) feats_test=RealFeatures(CSVFile(test_fname)) @@ -22,4 +22,4 @@ def kernel_log_modular (train_fname=traindat,test_fname=testdat, degree=2.0): if __name__=='__main__': print('Log') - kernel_log_modular(*parameter_list[0]) + kernel_log(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/kernel_match_word_string_modular.py b/examples/undocumented/python/kernel_match_word_string.py similarity index 78% rename from examples/undocumented/python_modular/kernel_match_word_string_modular.py rename to examples/undocumented/python/kernel_match_word_string.py index 71e86adc933..f7a81e02162 100644 --- a/examples/undocumented/python_modular/kernel_match_word_string_modular.py +++ b/examples/undocumented/python/kernel_match_word_string.py @@ -7,10 +7,10 @@ parameter_list = [[traindat,testdat, 3,1.4,10,3,0,False],[ traindat,testdat, 3,1.4,10,3,0,False]] -def kernel_match_word_string_modular (fm_train_dna=traindat,fm_test_dna=testdat, +def kernel_match_word_string (fm_train_dna=traindat,fm_test_dna=testdat, degree=3,scale=1.4,size_cache=10,order=3,gap=0,reverse=False): - from modshogun import MatchWordStringKernel, AvgDiagKernelNormalizer - from modshogun import StringWordFeatures, StringCharFeatures, DNA + from shogun import MatchWordStringKernel, AvgDiagKernelNormalizer + from shogun import StringWordFeatures, StringCharFeatures, DNA charfeat=StringCharFeatures(fm_train_dna, DNA) feats_train=StringWordFeatures(DNA) @@ -31,5 +31,5 @@ def kernel_match_word_string_modular (fm_train_dna=traindat,fm_test_dna=testdat, if __name__=='__main__': print('MatchWordString') - kernel_match_word_string_modular(*parameter_list[0]) + kernel_match_word_string(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/kernel_multiquadric_modular.py b/examples/undocumented/python/kernel_multiquadric.py similarity index 72% rename from examples/undocumented/python_modular/kernel_multiquadric_modular.py rename to examples/undocumented/python/kernel_multiquadric.py index c09c8b18038..f377e35ebc7 100644 --- a/examples/undocumented/python_modular/kernel_multiquadric_modular.py +++ b/examples/undocumented/python/kernel_multiquadric.py @@ -5,8 +5,8 @@ parameter_list=[[traindat,testdat, 1.0],[traindat,testdat, 5.0]] -def kernel_multiquadric_modular (train_fname=traindat,test_fname=testdat, shift_coef=1.0): - from modshogun import RealFeatures, MultiquadricKernel, EuclideanDistance, CSVFile +def kernel_multiquadric (train_fname=traindat,test_fname=testdat, shift_coef=1.0): + from shogun import RealFeatures, MultiquadricKernel, EuclideanDistance, CSVFile feats_train=RealFeatures(CSVFile(train_fname)) feats_test=RealFeatures(CSVFile(test_fname)) @@ -23,4 +23,4 @@ def kernel_multiquadric_modular (train_fname=traindat,test_fname=testdat, shift_ if __name__=='__main__': print('Multiquadric') - kernel_multiquadric_modular(*parameter_list[0]) + kernel_multiquadric(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/kernel_oligo_string_modular.py b/examples/undocumented/python/kernel_oligo_string.py similarity index 72% rename from examples/undocumented/python_modular/kernel_oligo_string_modular.py rename to examples/undocumented/python/kernel_oligo_string.py index bb7d6b0b301..466a739ef0d 100644 --- a/examples/undocumented/python_modular/kernel_oligo_string_modular.py +++ b/examples/undocumented/python/kernel_oligo_string.py @@ -6,9 +6,9 @@ parameter_list = [[traindat,testdat,3,1.2,10],[traindat,testdat,4,1.3,10]] -def kernel_oligo_string_modular (fm_train_dna=traindat,fm_test_dna=testdat,k=3,width=1.2,size_cache=10): - from modshogun import StringCharFeatures, DNA - from modshogun import OligoStringKernel +def kernel_oligo_string (fm_train_dna=traindat,fm_test_dna=testdat,k=3,width=1.2,size_cache=10): + from shogun import StringCharFeatures, DNA + from shogun import OligoStringKernel feats_train=StringCharFeatures(fm_train_dna, DNA) feats_test=StringCharFeatures(fm_test_dna, DNA) @@ -23,4 +23,4 @@ def kernel_oligo_string_modular (fm_train_dna=traindat,fm_test_dna=testdat,k=3,w if __name__=='__main__': print('OligoString') - kernel_oligo_string_modular(*parameter_list[0]) + kernel_oligo_string(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/kernel_poly_modular.py b/examples/undocumented/python/kernel_poly.py similarity index 75% rename from examples/undocumented/python_modular/kernel_poly_modular.py rename to examples/undocumented/python/kernel_poly.py index b8adae1d85d..56b481a59be 100644 --- a/examples/undocumented/python_modular/kernel_poly_modular.py +++ b/examples/undocumented/python/kernel_poly.py @@ -4,9 +4,9 @@ parameter_list = [[traindat,testdat,4,False,True],[traindat,testdat,5,False,True]] -def kernel_poly_modular (train_fname=traindat,test_fname=testdat,degree=4,inhomogene=False, +def kernel_poly (train_fname=traindat,test_fname=testdat,degree=4,inhomogene=False, use_normalization=True): - from modshogun import RealFeatures, PolyKernel, CSVFile + from shogun import RealFeatures, PolyKernel, CSVFile feats_train=RealFeatures(CSVFile(train_fname)) feats_test=RealFeatures(CSVFile(test_fname)) @@ -20,4 +20,4 @@ def kernel_poly_modular (train_fname=traindat,test_fname=testdat,degree=4,inhomo return km_train,km_test,kernel if __name__=='__main__': print('Poly') - kernel_poly_modular (*parameter_list[0]) + kernel_poly (*parameter_list[0]) diff --git a/examples/undocumented/python_modular/kernel_poly_match_string_modular.py b/examples/undocumented/python/kernel_poly_match_string.py similarity index 70% rename from examples/undocumented/python_modular/kernel_poly_match_string_modular.py rename to examples/undocumented/python/kernel_poly_match_string.py index 2fb76de93d4..2fc6bb0ce5a 100644 --- a/examples/undocumented/python_modular/kernel_poly_match_string_modular.py +++ b/examples/undocumented/python/kernel_poly_match_string.py @@ -5,9 +5,9 @@ testdat = lm.load_dna('../data/fm_test_dna.dat') parameter_list = [[traindat,testdat,3,False],[traindat,testdat,4,False]] -def kernel_poly_match_string_modular (fm_train_dna=traindat,fm_test_dna=testdat,degree=3,inhomogene=False): - from modshogun import PolyMatchStringKernel - from modshogun import StringCharFeatures, DNA +def kernel_poly_match_string (fm_train_dna=traindat,fm_test_dna=testdat,degree=3,inhomogene=False): + from shogun import PolyMatchStringKernel + from shogun import StringCharFeatures, DNA feats_train=StringCharFeatures(fm_train_dna, DNA) feats_test=StringCharFeatures(fm_train_dna, DNA) @@ -21,4 +21,4 @@ def kernel_poly_match_string_modular (fm_train_dna=traindat,fm_test_dna=testdat, if __name__=='__main__': print('PolyMatchString') - kernel_poly_match_string_modular(*parameter_list[0]) + kernel_poly_match_string(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/kernel_poly_match_word_string_modular.py b/examples/undocumented/python/kernel_poly_match_word_string.py similarity index 77% rename from examples/undocumented/python_modular/kernel_poly_match_word_string_modular.py rename to examples/undocumented/python/kernel_poly_match_word_string.py index 51ed2ce57f2..c2ac3017603 100644 --- a/examples/undocumented/python_modular/kernel_poly_match_word_string_modular.py +++ b/examples/undocumented/python/kernel_poly_match_word_string.py @@ -6,10 +6,10 @@ parameter_list = [[traindat,testdat,2,True,3,0,False],[traindat,testdat,2,True,3,0,False]] -def kernel_poly_match_word_string_modular (fm_train_dna=traindat,fm_test_dna=testdat, +def kernel_poly_match_word_string (fm_train_dna=traindat,fm_test_dna=testdat, degree=2,inhomogene=True,order=3,gap=0,reverse=False): - from modshogun import PolyMatchWordStringKernel - from modshogun import StringWordFeatures, StringCharFeatures, DNA + from shogun import PolyMatchWordStringKernel + from shogun import StringWordFeatures, StringCharFeatures, DNA @@ -30,4 +30,4 @@ def kernel_poly_match_word_string_modular (fm_train_dna=traindat,fm_test_dna=tes if __name__=='__main__': print('PolyMatchWordString') - kernel_poly_match_word_string_modular(*parameter_list[0]) + kernel_poly_match_word_string(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/kernel_power_modular.py b/examples/undocumented/python/kernel_power.py similarity index 74% rename from examples/undocumented/python_modular/kernel_power_modular.py rename to examples/undocumented/python/kernel_power.py index 072b788eb8b..52c7c471808 100644 --- a/examples/undocumented/python_modular/kernel_power_modular.py +++ b/examples/undocumented/python/kernel_power.py @@ -4,8 +4,8 @@ testdat = '../data/fm_test_real.dat' parameter_list=[[traindat,testdat, 2.0],[traindat,testdat, 3.0]] -def kernel_power_modular (train_fname=traindat,test_fname=testdat, degree=2.0): - from modshogun import RealFeatures, PowerKernel, EuclideanDistance, CSVFile +def kernel_power (train_fname=traindat,test_fname=testdat, degree=2.0): + from shogun import RealFeatures, PowerKernel, EuclideanDistance, CSVFile feats_train=RealFeatures(CSVFile(train_fname)) feats_test=RealFeatures(CSVFile(test_fname)) @@ -22,4 +22,4 @@ def kernel_power_modular (train_fname=traindat,test_fname=testdat, degree=2.0): if __name__=='__main__': print('Power') - kernel_power_modular(*parameter_list[0]) + kernel_power(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/kernel_rationalquadratic_modular.py b/examples/undocumented/python/kernel_rationalquadratic.py similarity index 71% rename from examples/undocumented/python_modular/kernel_rationalquadratic_modular.py rename to examples/undocumented/python/kernel_rationalquadratic.py index 87119a13696..f2e4a991c7a 100644 --- a/examples/undocumented/python_modular/kernel_rationalquadratic_modular.py +++ b/examples/undocumented/python/kernel_rationalquadratic.py @@ -4,8 +4,8 @@ testdat = '../data/fm_test_real.dat' parameter_list=[[traindat,testdat, 1.0],[traindat,testdat, 5.0]] -def kernel_rationalquadratic_modular (train_fname=traindat,test_fname=testdat, shift_coef=1.0): - from modshogun import RealFeatures, RationalQuadraticKernel, EuclideanDistance, CSVFile +def kernel_rationalquadratic (train_fname=traindat,test_fname=testdat, shift_coef=1.0): + from shogun import RealFeatures, RationalQuadraticKernel, EuclideanDistance, CSVFile feats_train=RealFeatures(CSVFile(train_fname)) feats_test=RealFeatures(CSVFile(test_fname)) @@ -22,4 +22,4 @@ def kernel_rationalquadratic_modular (train_fname=traindat,test_fname=testdat, s if __name__=='__main__': print('RationalQuadratic') - kernel_rationalquadratic_modular(*parameter_list[0]) + kernel_rationalquadratic(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/kernel_salzberg_word_string_modular.py b/examples/undocumented/python/kernel_salzberg_word_string.py similarity index 77% rename from examples/undocumented/python_modular/kernel_salzberg_word_string_modular.py rename to examples/undocumented/python/kernel_salzberg_word_string.py index 4063f1dc5f2..88a4773896e 100644 --- a/examples/undocumented/python_modular/kernel_salzberg_word_string_modular.py +++ b/examples/undocumented/python/kernel_salzberg_word_string.py @@ -6,11 +6,11 @@ label_traindat = lm.load_labels('../data/label_train_dna.dat') parameter_list = [[traindat,testdat,label_traindat,3,0,False],[traindat,testdat,label_traindat,3,0,False]] -def kernel_salzberg_word_string_modular (fm_train_dna=traindat,fm_test_dna=testdat,label_train_dna=label_traindat, +def kernel_salzberg_word_string (fm_train_dna=traindat,fm_test_dna=testdat,label_train_dna=label_traindat, order=3,gap=0,reverse=False): - from modshogun import StringCharFeatures, StringWordFeatures, DNA, BinaryLabels - from modshogun import SalzbergWordStringKernel - from modshogun import PluginEstimate + from shogun import StringCharFeatures, StringWordFeatures, DNA, BinaryLabels + from shogun import SalzbergWordStringKernel + from shogun import PluginEstimate charfeat=StringCharFeatures(fm_train_dna, DNA) feats_train=StringWordFeatures(charfeat.get_alphabet()) @@ -37,5 +37,5 @@ def kernel_salzberg_word_string_modular (fm_train_dna=traindat,fm_test_dna=testd if __name__=='__main__': print('PluginEstimate w/ SalzbergWord') - kernel_salzberg_word_string_modular(*parameter_list[0]) + kernel_salzberg_word_string(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/kernel_sigmoid_modular.py b/examples/undocumented/python/kernel_sigmoid.py similarity index 72% rename from examples/undocumented/python_modular/kernel_sigmoid_modular.py rename to examples/undocumented/python/kernel_sigmoid.py index f85d810c2a4..64ad32595ba 100644 --- a/examples/undocumented/python_modular/kernel_sigmoid_modular.py +++ b/examples/undocumented/python/kernel_sigmoid.py @@ -4,8 +4,8 @@ parameter_list = [[traindat,testdat,10,1.2,1.3],[traindat,testdat,10,1.2,1.3]] -def kernel_sigmoid_modular (train_fname=traindat,test_fname=testdat,size_cache=10,gamma=1.2,coef0=1.3): - from modshogun import RealFeatures, SigmoidKernel, CSVFile +def kernel_sigmoid (train_fname=traindat,test_fname=testdat,size_cache=10,gamma=1.2,coef0=1.3): + from shogun import RealFeatures, SigmoidKernel, CSVFile feats_train=RealFeatures(CSVFile(train_fname)) feats_test=RealFeatures(CSVFile(test_fname)) @@ -19,4 +19,4 @@ def kernel_sigmoid_modular (train_fname=traindat,test_fname=testdat,size_cache=1 if __name__=='__main__': print('Sigmoid') - kernel_sigmoid_modular(*parameter_list[0]) + kernel_sigmoid(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/kernel_simple_locality_improved_string_modular.py b/examples/undocumented/python/kernel_simple_locality_improved_string.py similarity index 72% rename from examples/undocumented/python_modular/kernel_simple_locality_improved_string_modular.py rename to examples/undocumented/python/kernel_simple_locality_improved_string.py index 672405e181c..cd9e33053d8 100644 --- a/examples/undocumented/python_modular/kernel_simple_locality_improved_string_modular.py +++ b/examples/undocumented/python/kernel_simple_locality_improved_string.py @@ -6,11 +6,11 @@ parameter_list = [[traindat,testdat,5,5,1],[traindat,testdat,5,3,2]] -def kernel_simple_locality_improved_string_modular (fm_train_dna=traindat,fm_test_dna=testdat, +def kernel_simple_locality_improved_string (fm_train_dna=traindat,fm_test_dna=testdat, length=5,inner_degree=5,outer_degree=1 ): - from modshogun import StringCharFeatures, DNA - from modshogun import SimpleLocalityImprovedStringKernel, MSG_DEBUG + from shogun import StringCharFeatures, DNA + from shogun import SimpleLocalityImprovedStringKernel, MSG_DEBUG feats_train=StringCharFeatures(fm_train_dna, DNA) #feats_train.io.set_loglevel(MSG_DEBUG) @@ -27,4 +27,4 @@ def kernel_simple_locality_improved_string_modular (fm_train_dna=traindat,fm_tes if __name__=='__main__': print('SimpleLocalityImprovedString') - kernel_simple_locality_improved_string_modular(*parameter_list[0]) + kernel_simple_locality_improved_string(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/kernel_sparse_gaussian_modular.py b/examples/undocumented/python/kernel_sparse_gaussian.py similarity index 72% rename from examples/undocumented/python_modular/kernel_sparse_gaussian_modular.py rename to examples/undocumented/python/kernel_sparse_gaussian.py index 62628b24893..1ad4623d420 100644 --- a/examples/undocumented/python_modular/kernel_sparse_gaussian_modular.py +++ b/examples/undocumented/python/kernel_sparse_gaussian.py @@ -6,9 +6,9 @@ parameter_list = [[traindat,testdat,1.1],[traindat,testdat,1.2]] -def kernel_sparse_gaussian_modular (fm_train_real=traindat,fm_test_real=testdat,width=1.1 ): - from modshogun import SparseRealFeatures - from modshogun import GaussianKernel +def kernel_sparse_gaussian (fm_train_real=traindat,fm_test_real=testdat,width=1.1 ): + from shogun import SparseRealFeatures + from shogun import GaussianKernel feats_train=SparseRealFeatures(fm_train_real) feats_test=SparseRealFeatures(fm_test_real) @@ -23,4 +23,4 @@ def kernel_sparse_gaussian_modular (fm_train_real=traindat,fm_test_real=testdat, if __name__=='__main__': print('SparseGaussian') - kernel_sparse_gaussian_modular (*parameter_list[0]) + kernel_sparse_gaussian (*parameter_list[0]) diff --git a/examples/undocumented/python_modular/kernel_sparse_linear_modular.py b/examples/undocumented/python/kernel_sparse_linear.py similarity index 72% rename from examples/undocumented/python_modular/kernel_sparse_linear_modular.py rename to examples/undocumented/python/kernel_sparse_linear.py index 794092c7c61..8ca21aa1831 100644 --- a/examples/undocumented/python_modular/kernel_sparse_linear_modular.py +++ b/examples/undocumented/python/kernel_sparse_linear.py @@ -7,9 +7,9 @@ parameter_list = [[traindat,testdat,1.1],[traindat,testdat,1.2]] -def kernel_sparse_linear_modular (fm_train_real=traindat,fm_test_real=testdat,scale=1.1): - from modshogun import SparseRealFeatures - from modshogun import LinearKernel, AvgDiagKernelNormalizer +def kernel_sparse_linear (fm_train_real=traindat,fm_test_real=testdat,scale=1.1): + from shogun import SparseRealFeatures + from shogun import LinearKernel, AvgDiagKernelNormalizer feats_train=SparseRealFeatures(fm_train_real) feats_test=SparseRealFeatures(fm_test_real) @@ -25,4 +25,4 @@ def kernel_sparse_linear_modular (fm_train_real=traindat,fm_test_real=testdat,sc if __name__=='__main__': print('SparseLinear') - kernel_sparse_linear_modular(*parameter_list[0]) + kernel_sparse_linear(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/kernel_sparse_poly_modular.py b/examples/undocumented/python/kernel_sparse_poly.py similarity index 77% rename from examples/undocumented/python_modular/kernel_sparse_poly_modular.py rename to examples/undocumented/python/kernel_sparse_poly.py index 13c11f00b80..b881012da50 100644 --- a/examples/undocumented/python_modular/kernel_sparse_poly_modular.py +++ b/examples/undocumented/python/kernel_sparse_poly.py @@ -6,11 +6,11 @@ parameter_list = [[traindat,testdat,10,3,True],[traindat,testdat,10,4,True]] -def kernel_sparse_poly_modular (fm_train_real=traindat,fm_test_real=testdat, +def kernel_sparse_poly (fm_train_real=traindat,fm_test_real=testdat, size_cache=10,degree=3,inhomogene=True ): - from modshogun import SparseRealFeatures - from modshogun import PolyKernel + from shogun import SparseRealFeatures + from shogun import PolyKernel feats_train=SparseRealFeatures(fm_train_real) feats_test=SparseRealFeatures(fm_test_real) @@ -27,4 +27,4 @@ def kernel_sparse_poly_modular (fm_train_real=traindat,fm_test_real=testdat, if __name__=='__main__': print('SparsePoly') - kernel_sparse_poly_modular(*parameter_list[0]) + kernel_sparse_poly(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/kernel_spherical_modular.py b/examples/undocumented/python/kernel_spherical.py similarity index 72% rename from examples/undocumented/python_modular/kernel_spherical_modular.py rename to examples/undocumented/python/kernel_spherical.py index 76e9f4432eb..52a429e8a99 100644 --- a/examples/undocumented/python_modular/kernel_spherical_modular.py +++ b/examples/undocumented/python/kernel_spherical.py @@ -8,10 +8,10 @@ parameter_list=[[traindat,testdat, 1.0],[traindat,testdat, 5.0]] -def kernel_spherical_modular (fm_train_real=traindat,fm_test_real=testdat, sigma=1.0): - from modshogun import RealFeatures - from modshogun import MultiquadricKernel - from modshogun import EuclideanDistance +def kernel_spherical (fm_train_real=traindat,fm_test_real=testdat, sigma=1.0): + from shogun import RealFeatures + from shogun import MultiquadricKernel + from shogun import EuclideanDistance feats_train=RealFeatures(fm_train_real) feats_test=RealFeatures(fm_test_real) @@ -28,4 +28,4 @@ def kernel_spherical_modular (fm_train_real=traindat,fm_test_real=testdat, sigma if __name__=='__main__': print('Spherical') - kernel_spherical_modular(*parameter_list[0]) + kernel_spherical(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/kernel_spline_modular.py b/examples/undocumented/python/kernel_spline.py similarity index 75% rename from examples/undocumented/python_modular/kernel_spline_modular.py rename to examples/undocumented/python/kernel_spline.py index 2b4dcbc9a1b..26aae43c802 100644 --- a/examples/undocumented/python_modular/kernel_spline_modular.py +++ b/examples/undocumented/python/kernel_spline.py @@ -8,9 +8,9 @@ parameter_list=[[traindat,testdat],[traindat,testdat]] -def kernel_spline_modular (fm_train_real=traindat,fm_test_real=testdat): - from modshogun import RealFeatures - from modshogun import SplineKernel +def kernel_spline (fm_train_real=traindat,fm_test_real=testdat): + from shogun import RealFeatures + from shogun import SplineKernel feats_train=RealFeatures(fm_train_real) feats_test=RealFeatures(fm_test_real) @@ -25,4 +25,4 @@ def kernel_spline_modular (fm_train_real=traindat,fm_test_real=testdat): if __name__=='__main__': print('Spline') - kernel_spline_modular(*parameter_list[0]) + kernel_spline(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/kernel_ssk_string_modular.py b/examples/undocumented/python/kernel_ssk_string.py similarity index 77% rename from examples/undocumented/python_modular/kernel_ssk_string_modular.py rename to examples/undocumented/python/kernel_ssk_string.py index e5694c81c5a..bfa5e28ed1a 100644 --- a/examples/undocumented/python_modular/kernel_ssk_string_modular.py +++ b/examples/undocumented/python/kernel_ssk_string.py @@ -17,9 +17,9 @@ parameter_list = [[traindat,testdat,2,0.75],[traindat,testdat,3,0.75]] -def kernel_ssk_string_modular (fm_train_dna=traindat, fm_test_dna=testdat, maxlen=1, decay=1): - from modshogun import SubsequenceStringKernel - from modshogun import StringCharFeatures, DNA +def kernel_ssk_string (fm_train_dna=traindat, fm_test_dna=testdat, maxlen=1, decay=1): + from shogun import SubsequenceStringKernel + from shogun import StringCharFeatures, DNA feats_train=StringCharFeatures(fm_train_dna, DNA) feats_test=StringCharFeatures(fm_test_dna, DNA) @@ -35,5 +35,5 @@ def kernel_ssk_string_modular (fm_train_dna=traindat, fm_test_dna=testdat, maxle if __name__=='__main__': print('SubsequenceStringKernel DNA') - kernel_ssk_string_modular(*parameter_list[0]) - kernel_ssk_string_modular(*parameter_list[1]) + kernel_ssk_string(*parameter_list[0]) + kernel_ssk_string(*parameter_list[1]) diff --git a/examples/undocumented/python_modular/kernel_top_modular.py b/examples/undocumented/python/kernel_top.py similarity index 88% rename from examples/undocumented/python_modular/kernel_top_modular.py rename to examples/undocumented/python/kernel_top.py index 337b63a1771..5f76324121e 100644 --- a/examples/undocumented/python_modular/kernel_top_modular.py +++ b/examples/undocumented/python/kernel_top.py @@ -13,11 +13,11 @@ parameter_list = [[traindat,testdat,label_traindat,1e-1,1,0,False,[1, False, True]], \ [traindat,testdat,label_traindat,1e-1,1,0,False,[1, False, True] ]] -def kernel_top_modular (fm_train_dna=traindat,fm_test_dna=testdat,label_train_dna=label_traindat,pseudo=1e-1, +def kernel_top (fm_train_dna=traindat,fm_test_dna=testdat,label_train_dna=label_traindat,pseudo=1e-1, order=1,gap=0,reverse=False,kargs=[1, False, True]): - from modshogun import StringCharFeatures, StringWordFeatures, TOPFeatures, DNA - from modshogun import PolyKernel - from modshogun import HMM, BW_NORMAL + from shogun import StringCharFeatures, StringWordFeatures, TOPFeatures, DNA + from shogun import PolyKernel + from shogun import HMM, BW_NORMAL N=1 # toy HMM with 1 state M=4 # 4 observations -> DNA @@ -66,4 +66,4 @@ def kernel_top_modular (fm_train_dna=traindat,fm_test_dna=testdat,label_train_dn if __name__=='__main__': print("TOP Kernel") - kernel_top_modular(*parameter_list[0]) + kernel_top(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/kernel_tstudent_modular.py b/examples/undocumented/python/kernel_tstudent.py similarity index 73% rename from examples/undocumented/python_modular/kernel_tstudent_modular.py rename to examples/undocumented/python/kernel_tstudent.py index f88b34b0bf8..3cca9640459 100644 --- a/examples/undocumented/python_modular/kernel_tstudent_modular.py +++ b/examples/undocumented/python/kernel_tstudent.py @@ -8,10 +8,10 @@ parameter_list=[[traindat,testdat, 2.0],[traindat,testdat, 3.0]] -def kernel_tstudent_modular (fm_train_real=traindat,fm_test_real=testdat, degree=2.0): - from modshogun import RealFeatures - from modshogun import TStudentKernel - from modshogun import EuclideanDistance +def kernel_tstudent (fm_train_real=traindat,fm_test_real=testdat, degree=2.0): + from shogun import RealFeatures + from shogun import TStudentKernel + from shogun import EuclideanDistance feats_train=RealFeatures(fm_train_real) feats_test=RealFeatures(fm_test_real) @@ -28,4 +28,4 @@ def kernel_tstudent_modular (fm_train_real=traindat,fm_test_real=testdat, degree if __name__=='__main__': print('TStudent') - kernel_tstudent_modular(*parameter_list[0]) + kernel_tstudent(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/kernel_wave_modular.py b/examples/undocumented/python/kernel_wave.py similarity index 73% rename from examples/undocumented/python_modular/kernel_wave_modular.py rename to examples/undocumented/python/kernel_wave.py index e225bf9dcff..accf91cbe96 100644 --- a/examples/undocumented/python_modular/kernel_wave_modular.py +++ b/examples/undocumented/python/kernel_wave.py @@ -8,10 +8,10 @@ parameter_list=[[traindat,testdat, 1.0],[traindat,testdat, 10.0]] -def kernel_wave_modular (fm_train_real=traindat,fm_test_real=testdat, theta=1.0): - from modshogun import RealFeatures - from modshogun import WaveKernel - from modshogun import EuclideanDistance +def kernel_wave (fm_train_real=traindat,fm_test_real=testdat, theta=1.0): + from shogun import RealFeatures + from shogun import WaveKernel + from shogun import EuclideanDistance feats_train=RealFeatures(fm_train_real) feats_test=RealFeatures(fm_test_real) @@ -28,4 +28,4 @@ def kernel_wave_modular (fm_train_real=traindat,fm_test_real=testdat, theta=1.0) if __name__=='__main__': print('Wave') - kernel_wave_modular(*parameter_list[0]) + kernel_wave(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/kernel_wavelet_modular.py b/examples/undocumented/python/kernel_wavelet.py similarity index 73% rename from examples/undocumented/python_modular/kernel_wavelet_modular.py rename to examples/undocumented/python/kernel_wavelet.py index 01d9e4ea5fc..f18df45b22b 100644 --- a/examples/undocumented/python_modular/kernel_wavelet_modular.py +++ b/examples/undocumented/python/kernel_wavelet.py @@ -8,9 +8,9 @@ parameter_list=[[traindat,testdat, 1.5, 1.0],[traindat,testdat, 1.0, 1.5]] -def kernel_wavelet_modular (fm_train_real=traindat,fm_test_real=testdat, dilation=1.5, translation=1.0): - from modshogun import RealFeatures - from modshogun import WaveletKernel +def kernel_wavelet (fm_train_real=traindat,fm_test_real=testdat, dilation=1.5, translation=1.0): + from shogun import RealFeatures + from shogun import WaveletKernel feats_train=RealFeatures(fm_train_real) feats_test=RealFeatures(fm_test_real) @@ -25,4 +25,4 @@ def kernel_wavelet_modular (fm_train_real=traindat,fm_test_real=testdat, dilatio if __name__=='__main__': print('Wavelet') - kernel_wavelet_modular(*parameter_list[0]) + kernel_wavelet(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/kernel_weighted_comm_word_string_modular.py b/examples/undocumented/python/kernel_weighted_comm_word_string.py similarity index 76% rename from examples/undocumented/python_modular/kernel_weighted_comm_word_string_modular.py rename to examples/undocumented/python/kernel_weighted_comm_word_string.py index 512d45a3ee8..763d4705b27 100644 --- a/examples/undocumented/python_modular/kernel_weighted_comm_word_string_modular.py +++ b/examples/undocumented/python/kernel_weighted_comm_word_string.py @@ -6,10 +6,10 @@ parameter_list = [[traindat,testdat],[traindat,testdat]] -def kernel_weighted_comm_word_string_modular (fm_train_dna=traindat,fm_test_dna=testdat,order=3,gap=0,reverse=True ): - from modshogun import WeightedCommWordStringKernel - from modshogun import StringWordFeatures, StringCharFeatures, DNA - from modshogun import SortWordString +def kernel_weighted_comm_word_string (fm_train_dna=traindat,fm_test_dna=testdat,order=3,gap=0,reverse=True ): + from shogun import WeightedCommWordStringKernel + from shogun import StringWordFeatures, StringCharFeatures, DNA + from shogun import SortWordString charfeat=StringCharFeatures(fm_train_dna, DNA) feats_train=StringWordFeatures(charfeat.get_alphabet()) @@ -35,4 +35,4 @@ def kernel_weighted_comm_word_string_modular (fm_train_dna=traindat,fm_test_dna= if __name__=='__main__': print('WeightedCommWordString') - kernel_weighted_comm_word_string_modular(*parameter_list[0]) + kernel_weighted_comm_word_string(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/kernel_weighted_degree_position_string_modular.py b/examples/undocumented/python/kernel_weighted_degree_position_string.py similarity index 74% rename from examples/undocumented/python_modular/kernel_weighted_degree_position_string_modular.py rename to examples/undocumented/python/kernel_weighted_degree_position_string.py index 47ade02b538..9c515b12ecf 100644 --- a/examples/undocumented/python_modular/kernel_weighted_degree_position_string_modular.py +++ b/examples/undocumented/python/kernel_weighted_degree_position_string.py @@ -6,9 +6,9 @@ testdat = lm.load_dna('../data/fm_test_dna.dat') parameter_list = [[traindat,testdat,20],[traindat,testdat,22]] -def kernel_weighted_degree_position_string_modular (fm_train_dna=traindat,fm_test_dna=testdat,degree=20): - from modshogun import StringCharFeatures, DNA - from modshogun import WeightedDegreePositionStringKernel, MSG_DEBUG +def kernel_weighted_degree_position_string (fm_train_dna=traindat,fm_test_dna=testdat,degree=20): + from shogun import StringCharFeatures, DNA + from shogun import WeightedDegreePositionStringKernel, MSG_DEBUG feats_train=StringCharFeatures(fm_train_dna, DNA) #feats_train.io.set_loglevel(MSG_DEBUG) @@ -27,4 +27,4 @@ def kernel_weighted_degree_position_string_modular (fm_train_dna=traindat,fm_tes if __name__=='__main__': print('WeightedDegreePositionString') - kernel_weighted_degree_position_string_modular(*parameter_list[0]) + kernel_weighted_degree_position_string(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/kernel_weighted_degree_string_modular.py b/examples/undocumented/python/kernel_weighted_degree_string.py similarity index 80% rename from examples/undocumented/python_modular/kernel_weighted_degree_string_modular.py rename to examples/undocumented/python/kernel_weighted_degree_string.py index ed063c2b6e5..96c131ceba0 100644 --- a/examples/undocumented/python_modular/kernel_weighted_degree_string_modular.py +++ b/examples/undocumented/python/kernel_weighted_degree_string.py @@ -6,9 +6,9 @@ parameter_list = [[traindat,testdat,3],[traindat,testdat,20]] -def kernel_weighted_degree_string_modular (fm_train_dna=traindat,fm_test_dna=testdat,degree=20): - from modshogun import StringCharFeatures, DNA - from modshogun import WeightedDegreeStringKernel, MSG_DEBUG +def kernel_weighted_degree_string (fm_train_dna=traindat,fm_test_dna=testdat,degree=20): + from shogun import StringCharFeatures, DNA + from shogun import WeightedDegreeStringKernel, MSG_DEBUG feats_train=StringCharFeatures(fm_train_dna, DNA) #feats_train.io.set_loglevel(MSG_DEBUG) @@ -38,4 +38,4 @@ def kernel_weighted_degree_string_modular (fm_train_dna=traindat,fm_test_dna=tes if __name__=='__main__': print('WeightedDegreeString') - kernel_weighted_degree_string_modular(*parameter_list[0]) + kernel_weighted_degree_string(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/labels_io_modular.py b/examples/undocumented/python/labels_io.py similarity index 67% rename from examples/undocumented/python_modular/labels_io_modular.py rename to examples/undocumented/python/labels_io.py index 52673cc1a67..b98179d0d6f 100644 --- a/examples/undocumented/python_modular/labels_io_modular.py +++ b/examples/undocumented/python/labels_io.py @@ -1,8 +1,8 @@ #!/usr/bin/env python parameter_list=[[]] -def labels_io_modular(): - from modshogun import RegressionLabels, CSVFile +def labels_io(): + from shogun import RegressionLabels, CSVFile lab=RegressionLabels() f=CSVFile("../data/label_train_regression.dat","r") f.set_delimiter(" ") @@ -12,4 +12,4 @@ def labels_io_modular(): if __name__=='__main__': print('Labels IO') - labels_io_modular(*parameter_list[0]) + labels_io(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/library_fisher2x3_modular.py b/examples/undocumented/python/library_fisher2x3.py similarity index 76% rename from examples/undocumented/python_modular/library_fisher2x3_modular.py rename to examples/undocumented/python/library_fisher2x3.py index ce023ecd793..fc6de163164 100644 --- a/examples/undocumented/python_modular/library_fisher2x3_modular.py +++ b/examples/undocumented/python/library_fisher2x3.py @@ -1,6 +1,6 @@ #!/usr/bin/env python from numpy import * -from modshogun import * +from shogun import * x=array([[20.0,15,15],[10,20,20]]) y=array([[21.0,21,18],[19,19,22]]) @@ -9,11 +9,11 @@ parameter_list = [[x,concatenate((x,y,z),1)]] -def library_fisher2x3_modular (table, tables): +def library_fisher2x3 (table, tables): pval=Statistics_fishers_exact_test_for_2x3_table(table) pvals=Statistics_fishers_exact_test_for_multiple_2x3_tables(tables) return (pval,pvals) if __name__=='__main__': print('Fisher 2x3') - library_fisher2x3_modular(*parameter_list[0]) + library_fisher2x3(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/library_time.py b/examples/undocumented/python/library_time.py similarity index 94% rename from examples/undocumented/python_modular/library_time.py rename to examples/undocumented/python/library_time.py index 58ef6ea6779..cf7fde8309e 100644 --- a/examples/undocumented/python_modular/library_time.py +++ b/examples/undocumented/python/library_time.py @@ -1,6 +1,6 @@ #!/usr/bin/env python import time -from modshogun import Time +from shogun import Time parameter_list = [[5],[1.0]] def library_time (sleep_secs): diff --git a/examples/undocumented/python_modular/mathematics_linsolver_cg.py b/examples/undocumented/python/mathematics_linsolver_cg.py similarity index 100% rename from examples/undocumented/python_modular/mathematics_linsolver_cg.py rename to examples/undocumented/python/mathematics_linsolver_cg.py diff --git a/examples/undocumented/python_modular/mathematics_logdet.py b/examples/undocumented/python/mathematics_logdet.py similarity index 100% rename from examples/undocumented/python_modular/mathematics_logdet.py rename to examples/undocumented/python/mathematics_logdet.py diff --git a/examples/undocumented/python_modular/mathematics_sparseinversecovariance_modular.py b/examples/undocumented/python/mathematics_sparseinversecovariance.py similarity index 71% rename from examples/undocumented/python_modular/mathematics_sparseinversecovariance_modular.py rename to examples/undocumented/python/mathematics_sparseinversecovariance.py index aa51ba94f58..3f82d661005 100644 --- a/examples/undocumented/python_modular/mathematics_sparseinversecovariance_modular.py +++ b/examples/undocumented/python/mathematics_sparseinversecovariance.py @@ -6,9 +6,9 @@ parameter_list = [[data,0.0],[data,1.0]] -def mathematics_sparseinversecovariance_modular (data,lc): +def mathematics_sparseinversecovariance (data,lc): try: - from modshogun import SparseInverseCovariance + from shogun import SparseInverseCovariance except ImportError: print("SparseInverseCovariance not available") exit(0) @@ -24,6 +24,6 @@ def mathematics_sparseinversecovariance_modular (data,lc): if __name__=='__main__': print('SparseInverseCovariance') - mathematics_sparseinversecovariance_modular(*parameter_list[0]) + mathematics_sparseinversecovariance(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/metric_lmnn_modular.py b/examples/undocumented/python/metric_lmnn.py similarity index 77% rename from examples/undocumented/python_modular/metric_lmnn_modular.py rename to examples/undocumented/python/metric_lmnn.py index 1768e627de3..f9aa05d1184 100644 --- a/examples/undocumented/python_modular/metric_lmnn_modular.py +++ b/examples/undocumented/python/metric_lmnn.py @@ -6,9 +6,9 @@ parameter_list = [[traindat,testdat,label_traindat,3]] -def metric_lmnn_modular(train_fname=traindat,test_fname=testdat,label_train_fname=label_traindat,k=3): +def metric_lmnn(train_fname=traindat,test_fname=testdat,label_train_fname=label_traindat,k=3): try: - from modshogun import RealFeatures,MulticlassLabels,LMNN,KNN,CSVFile + from shogun import RealFeatures,MulticlassLabels,LMNN,KNN,CSVFile except ImportError: return @@ -31,4 +31,4 @@ def metric_lmnn_modular(train_fname=traindat,test_fname=testdat,label_train_fnam if __name__=='__main__': print('LMNN') - metric_lmnn_modular(*parameter_list[0]) + metric_lmnn(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/mkl_binclass_modular.py b/examples/undocumented/python/mkl_binclass.py similarity index 85% rename from examples/undocumented/python_modular/mkl_binclass_modular.py rename to examples/undocumented/python/mkl_binclass.py index 90ab575b541..c5211faa5e4 100644 --- a/examples/undocumented/python_modular/mkl_binclass_modular.py +++ b/examples/undocumented/python/mkl_binclass.py @@ -1,13 +1,13 @@ #!/usr/bin/env python -from modshogun import CombinedFeatures, RealFeatures, BinaryLabels -from modshogun import CombinedKernel, PolyKernel, CustomKernel -from modshogun import MKLClassification +from shogun import CombinedFeatures, RealFeatures, BinaryLabels +from shogun import CombinedKernel, PolyKernel, CustomKernel +from shogun import MKLClassification from tools.load import LoadMatrix lm=LoadMatrix() #only run example if SVMLight is included as LibSVM solver crashes in MKLClassification try: - from modshogun import SVMLight + from shogun import SVMLight except ImportError: print("SVMLight not available") exit(0) @@ -21,7 +21,7 @@ # fm_test_real.shape # combined_custom() -def mkl_binclass_modular (fm_train_real=traindat,fm_test_real=testdat,fm_label_twoclass = label_traindat): +def mkl_binclass (fm_train_real=traindat,fm_test_real=testdat,fm_label_twoclass = label_traindat): ################################## # set up and train @@ -85,4 +85,4 @@ def mkl_binclass_modular (fm_train_real=traindat,fm_test_real=testdat,fm_label_t return mkl.apply(),kernel if __name__=='__main__': - mkl_binclass_modular (*parameter_list[0]) + mkl_binclass (*parameter_list[0]) diff --git a/examples/undocumented/python_modular/mkl_multiclass_modular.py b/examples/undocumented/python/mkl_multiclass.py similarity index 85% rename from examples/undocumented/python_modular/mkl_multiclass_modular.py rename to examples/undocumented/python/mkl_multiclass.py index 19312976be2..d88cc7bdf58 100644 --- a/examples/undocumented/python_modular/mkl_multiclass_modular.py +++ b/examples/undocumented/python/mkl_multiclass.py @@ -9,12 +9,12 @@ [ fm_train_real, fm_test_real, label_train_multiclass, 1.2, 1.2, 1e-5, 1, 0.001, 1.5], [ fm_train_real, fm_test_real, label_train_multiclass, 5, 1.2, 1e-2, 1, 0.001, 2]] -def mkl_multiclass_modular (fm_train_real, fm_test_real, label_train_multiclass, +def mkl_multiclass (fm_train_real, fm_test_real, label_train_multiclass, width, C, epsilon, num_threads, mkl_epsilon, mkl_norm): - from modshogun import CombinedFeatures, RealFeatures, MulticlassLabels - from modshogun import CombinedKernel, GaussianKernel, LinearKernel,PolyKernel - from modshogun import MKLMulticlass + from shogun import CombinedFeatures, RealFeatures, MulticlassLabels + from shogun import CombinedKernel, GaussianKernel, LinearKernel,PolyKernel + from shogun import MKLMulticlass kernel = CombinedKernel() feats_train = CombinedFeatures() @@ -61,4 +61,4 @@ def mkl_multiclass_modular (fm_train_real, fm_test_real, label_train_multiclass, if __name__ == '__main__': print('mkl_multiclass') - mkl_multiclass_modular(*parameter_list[0]) + mkl_multiclass(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/modelselection_grid_search_kernel.py b/examples/undocumented/python/modelselection_grid_search_kernel.py similarity index 86% rename from examples/undocumented/python_modular/modelselection_grid_search_kernel.py rename to examples/undocumented/python/modelselection_grid_search_kernel.py index 6f41f34b28a..3163ed5acce 100644 --- a/examples/undocumented/python_modular/modelselection_grid_search_kernel.py +++ b/examples/undocumented/python/modelselection_grid_search_kernel.py @@ -12,18 +12,18 @@ from numpy import random import math -from modshogun import CrossValidation, CrossValidationResult -from modshogun import ContingencyTableEvaluation, ACCURACY -from modshogun import StratifiedCrossValidationSplitting -from modshogun import BinaryLabels -from modshogun import RealFeatures -from modshogun import GaussianKernel, PowerKernel -from modshogun import LibSVM -from modshogun import MinkowskiMetric -from modshogun import GridSearchModelSelection -from modshogun import ModelSelectionParameters, R_EXP, R_LINEAR -from modshogun import ParameterCombination -from modshogun import Math +from shogun import CrossValidation, CrossValidationResult +from shogun import ContingencyTableEvaluation, ACCURACY +from shogun import StratifiedCrossValidationSplitting +from shogun import BinaryLabels +from shogun import RealFeatures +from shogun import GaussianKernel, PowerKernel +from shogun import LibSVM +from shogun import MinkowskiMetric +from shogun import GridSearchModelSelection +from shogun import ModelSelectionParameters, R_EXP, R_LINEAR +from shogun import ParameterCombination +from shogun import Math def create_param_tree(): root=ModelSelectionParameters() @@ -132,7 +132,7 @@ def modelselection_grid_search_kernel (num_subsets, num_vectors, dim_vectors): casted=CrossValidationResult.obtain_from_generic(result); #print "result mean:", casted.mean - return classifier,result,casted.mean + return classifier,result,casted.get_mean() if __name__=='__main__': print('ModelselectionGridSearchKernel') diff --git a/examples/undocumented/python_modular/modelselection_grid_search_krr_modular.py b/examples/undocumented/python/modelselection_grid_search_krr.py similarity index 85% rename from examples/undocumented/python_modular/modelselection_grid_search_krr_modular.py rename to examples/undocumented/python/modelselection_grid_search_krr.py index 1a88381c2a2..dd673c53ede 100644 --- a/examples/undocumented/python_modular/modelselection_grid_search_krr_modular.py +++ b/examples/undocumented/python/modelselection_grid_search_krr.py @@ -22,16 +22,16 @@ parameter_list = [[traindat,testdat,label_traindat,2.1,1,1e-5,1e-2], \ [traindat,testdat,label_traindat,2.1,1,1e-5,1e-2]] -def modelselection_grid_search_krr_modular (fm_train=traindat,fm_test=testdat,label_train=label_traindat,\ +def modelselection_grid_search_krr (fm_train=traindat,fm_test=testdat,label_train=label_traindat,\ width=2.1,C=1,epsilon=1e-5,tube_epsilon=1e-2): - from modshogun import CrossValidation, CrossValidationResult - from modshogun import MeanSquaredError - from modshogun import CrossValidationSplitting - from modshogun import RegressionLabels - from modshogun import RealFeatures - from modshogun import KernelRidgeRegression - from modshogun import GridSearchModelSelection - from modshogun import ModelSelectionParameters + from shogun import CrossValidation, CrossValidationResult + from shogun import MeanSquaredError + from shogun import CrossValidationSplitting + from shogun import RegressionLabels + from shogun import RealFeatures + from shogun import KernelRidgeRegression + from shogun import GridSearchModelSelection + from shogun import ModelSelectionParameters # training data features_train=RealFeatures(traindat) @@ -91,9 +91,9 @@ def modelselection_grid_search_krr_modular (fm_train=traindat,fm_test=testdat,la # creates all the parameters to optimize def create_param_tree(): - from modshogun import ModelSelectionParameters, R_EXP, R_LINEAR - from modshogun import ParameterCombination - from modshogun import GaussianKernel, PolyKernel + from shogun import ModelSelectionParameters, R_EXP, R_LINEAR + from shogun import ParameterCombination + from shogun import GaussianKernel, PolyKernel import math root=ModelSelectionParameters() @@ -142,4 +142,4 @@ def create_param_tree(): if __name__=='__main__': print('ModelselectionGridSearchKRR') - modelselection_grid_search_krr_modular(*parameter_list[0]) + modelselection_grid_search_krr(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/modelselection_grid_search_liblinear_modular.py b/examples/undocumented/python/modelselection_grid_search_liblinear.py similarity index 78% rename from examples/undocumented/python_modular/modelselection_grid_search_liblinear_modular.py rename to examples/undocumented/python/modelselection_grid_search_liblinear.py index ac733591473..dd267c97f32 100644 --- a/examples/undocumented/python_modular/modelselection_grid_search_liblinear_modular.py +++ b/examples/undocumented/python/modelselection_grid_search_liblinear.py @@ -21,16 +21,16 @@ parameter_list = [[traindat,label_traindat]] -def modelselection_grid_search_liblinear_modular (traindat=traindat, label_traindat=label_traindat): - from modshogun import CrossValidation, CrossValidationResult - from modshogun import ContingencyTableEvaluation, ACCURACY - from modshogun import StratifiedCrossValidationSplitting - from modshogun import GridSearchModelSelection - from modshogun import ModelSelectionParameters, R_EXP - from modshogun import ParameterCombination - from modshogun import BinaryLabels - from modshogun import RealFeatures - from modshogun import LibLinear, L2R_L2LOSS_SVC +def modelselection_grid_search_liblinear (traindat=traindat, label_traindat=label_traindat): + from shogun import CrossValidation, CrossValidationResult + from shogun import ContingencyTableEvaluation, ACCURACY + from shogun import StratifiedCrossValidationSplitting + from shogun import GridSearchModelSelection + from shogun import ModelSelectionParameters, R_EXP + from shogun import ParameterCombination + from shogun import BinaryLabels + from shogun import RealFeatures + from shogun import LibLinear, L2R_L2LOSS_SVC # build parameter tree to select C1 and C2 param_tree_root=ModelSelectionParameters() @@ -83,4 +83,4 @@ def modelselection_grid_search_liblinear_modular (traindat=traindat, label_train if __name__=='__main__': print('ModelSelectionGridSearchLibLinear') - modelselection_grid_search_liblinear_modular(*parameter_list[0]) + modelselection_grid_search_liblinear(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/modelselection_grid_search_libsvr_modular.py b/examples/undocumented/python/modelselection_grid_search_libsvr.py similarity index 84% rename from examples/undocumented/python_modular/modelselection_grid_search_libsvr_modular.py rename to examples/undocumented/python/modelselection_grid_search_libsvr.py index cabe6ae75ba..deaf18a7495 100644 --- a/examples/undocumented/python_modular/modelselection_grid_search_libsvr_modular.py +++ b/examples/undocumented/python/modelselection_grid_search_libsvr.py @@ -22,18 +22,18 @@ parameter_list = [[traindat,testdat,label_traindat,2.1,1,1e-5,1e-2], \ [traindat,testdat,label_traindat,2.1,1,1e-5,1e-2]] -def modelselection_grid_search_libsvr_modular (fm_train=traindat,fm_test=testdat,label_train=label_traindat,\ +def modelselection_grid_search_libsvr (fm_train=traindat,fm_test=testdat,label_train=label_traindat,\ width=2.1,C=1,epsilon=1e-5,tube_epsilon=1e-2): - from modshogun import CrossValidation, CrossValidationResult - from modshogun import MeanSquaredError - from modshogun import CrossValidationSplitting - from modshogun import RegressionLabels - from modshogun import RealFeatures - from modshogun import GaussianKernel - from modshogun import LibSVR - from modshogun import GridSearchModelSelection - from modshogun import ModelSelectionParameters, R_EXP - from modshogun import ParameterCombination + from shogun import CrossValidation, CrossValidationResult + from shogun import MeanSquaredError + from shogun import CrossValidationSplitting + from shogun import RegressionLabels + from shogun import RealFeatures + from shogun import GaussianKernel + from shogun import LibSVR + from shogun import GridSearchModelSelection + from shogun import ModelSelectionParameters, R_EXP + from shogun import ParameterCombination # training data features_train=RealFeatures(traindat) @@ -109,4 +109,4 @@ def modelselection_grid_search_libsvr_modular (fm_train=traindat,fm_test=testdat if __name__=='__main__': print('ModelselectionGridSearchLibSVR') - modelselection_grid_search_libsvr_modular(*parameter_list[0]) + modelselection_grid_search_libsvr(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/modelselection_parameter_tree_modular.py b/examples/undocumented/python/modelselection_parameter_tree.py similarity index 88% rename from examples/undocumented/python_modular/modelselection_parameter_tree_modular.py rename to examples/undocumented/python/modelselection_parameter_tree.py index b55a73beea6..d1b221bbdb8 100644 --- a/examples/undocumented/python_modular/modelselection_parameter_tree_modular.py +++ b/examples/undocumented/python/modelselection_parameter_tree.py @@ -11,13 +11,13 @@ parameter_list=[[None]] -def modelselection_parameter_tree_modular (dummy): - from modshogun import ParameterCombination - from modshogun import ModelSelectionParameters, R_EXP, R_LINEAR - from modshogun import PowerKernel - from modshogun import GaussianKernel - from modshogun import DistantSegmentsKernel - from modshogun import MinkowskiMetric +def modelselection_parameter_tree (dummy): + from shogun import ParameterCombination + from shogun import ModelSelectionParameters, R_EXP, R_LINEAR + from shogun import PowerKernel + from shogun import GaussianKernel + from shogun import DistantSegmentsKernel + from shogun import MinkowskiMetric import math root=ModelSelectionParameters() @@ -99,7 +99,7 @@ def modelselection_parameter_tree_modular (dummy): if __name__=='__main__': print('ModelSelection ParameterTree') - modelselection_parameter_tree_modular(*parameter_list[0]) + modelselection_parameter_tree(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/modelselection_random_search_liblinear_modular.py b/examples/undocumented/python/modelselection_random_search_liblinear.py similarity index 78% rename from examples/undocumented/python_modular/modelselection_random_search_liblinear_modular.py rename to examples/undocumented/python/modelselection_random_search_liblinear.py index a2b42f3ad72..3eef55e44ba 100644 --- a/examples/undocumented/python_modular/modelselection_random_search_liblinear_modular.py +++ b/examples/undocumented/python/modelselection_random_search_liblinear.py @@ -19,16 +19,16 @@ parameter_list = [[traindat,label_traindat]] -def modelselection_random_search_liblinear_modular (traindat=traindat, label_traindat=label_traindat): - from modshogun import CrossValidation, CrossValidationResult - from modshogun import ContingencyTableEvaluation, ACCURACY - from modshogun import StratifiedCrossValidationSplitting - from modshogun import RandomSearchModelSelection - from modshogun import ModelSelectionParameters, R_EXP - from modshogun import ParameterCombination - from modshogun import BinaryLabels - from modshogun import RealFeatures - from modshogun import LibLinear, L2R_L2LOSS_SVC +def modelselection_random_search_liblinear (traindat=traindat, label_traindat=label_traindat): + from shogun import CrossValidation, CrossValidationResult + from shogun import ContingencyTableEvaluation, ACCURACY + from shogun import StratifiedCrossValidationSplitting + from shogun import RandomSearchModelSelection + from shogun import ModelSelectionParameters, R_EXP + from shogun import ParameterCombination + from shogun import BinaryLabels + from shogun import RealFeatures + from shogun import LibLinear, L2R_L2LOSS_SVC # build parameter tree to select C1 and C2 param_tree_root=ModelSelectionParameters() @@ -81,4 +81,4 @@ def modelselection_random_search_liblinear_modular (traindat=traindat, label_tra if __name__=='__main__': print('ModelSelectionRandomSearchLibLinear') - modelselection_random_search_liblinear_modular(*parameter_list[0]) + modelselection_random_search_liblinear(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/multiclass_c45classifiertree_modular.py b/examples/undocumented/python/multiclass_c45classifiertree.py similarity index 81% rename from examples/undocumented/python_modular/multiclass_c45classifiertree_modular.py rename to examples/undocumented/python/multiclass_c45classifiertree.py index 2b48b4415c1..aca26ef09b3 100644 --- a/examples/undocumented/python_modular/multiclass_c45classifiertree_modular.py +++ b/examples/undocumented/python/multiclass_c45classifiertree.py @@ -10,9 +10,9 @@ parameter_list = [[traindat,testdat,label_traindat,feattypes]] -def multiclass_c45classifiertree_modular(train=traindat,test=testdat,labels=label_traindat,ft=feattypes): +def multiclass_c45classifiertree(train=traindat,test=testdat,labels=label_traindat,ft=feattypes): try: - from modshogun import RealFeatures, MulticlassLabels, CSVFile, C45ClassifierTree + from shogun import RealFeatures, MulticlassLabels, CSVFile, C45ClassifierTree from numpy import random, int32 except ImportError: print("Could not import Shogun and/or numpy modules") @@ -25,8 +25,8 @@ def multiclass_c45classifiertree_modular(train=traindat,test=testdat,labels=labe # divide train dataset into training and validation subsets in the ratio 2/3 to 1/3 subset=int32(random.permutation(feats_train.get_num_vectors())) - vsubset=subset[1:subset.size/3] - trsubset=subset[1+subset.size/3:subset.size] + vsubset=subset[1:int(subset.size/3)] + trsubset=subset[1+int(subset.size/3):subset.size] # C4.5 Tree formation using training subset train_labels.add_subset(trsubset) @@ -57,4 +57,4 @@ def multiclass_c45classifiertree_modular(train=traindat,test=testdat,labels=labe if __name__=='__main__': print('C45ClassifierTree') - multiclass_c45classifiertree_modular(*parameter_list[0]) + multiclass_c45classifiertree(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/multiclass_id3classifiertree_modular.py b/examples/undocumented/python/multiclass_id3classifiertree.py similarity index 83% rename from examples/undocumented/python_modular/multiclass_id3classifiertree_modular.py rename to examples/undocumented/python/multiclass_id3classifiertree.py index bc7902fd4fb..58f7cb0ed74 100644 --- a/examples/undocumented/python_modular/multiclass_id3classifiertree_modular.py +++ b/examples/undocumented/python/multiclass_id3classifiertree.py @@ -16,9 +16,9 @@ parameter_list = [[train_data, train_labels, test_data]] -def multiclass_id3classifiertree_modular(train=train_data,labels=train_labels,test=test_data): +def multiclass_id3classifiertree(train=train_data,labels=train_labels,test=test_data): try: - from modshogun import RealFeatures, MulticlassLabels, ID3ClassifierTree + from shogun import RealFeatures, MulticlassLabels, ID3ClassifierTree except ImportError: return @@ -39,4 +39,4 @@ def multiclass_id3classifiertree_modular(train=train_data,labels=train_labels,te if __name__=='__main__': print('ID3ClassifierTree') - multiclass_id3classifiertree_modular(*parameter_list[0]) + multiclass_id3classifiertree(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/multiclass_randomforest_modular.py b/examples/undocumented/python/multiclass_randomforest.py similarity index 79% rename from examples/undocumented/python_modular/multiclass_randomforest_modular.py rename to examples/undocumented/python/multiclass_randomforest.py index 55f9358492f..a110b58e739 100644 --- a/examples/undocumented/python_modular/multiclass_randomforest_modular.py +++ b/examples/undocumented/python/multiclass_randomforest.py @@ -10,9 +10,9 @@ parameter_list = [[traindat,testdat,label_traindat,feattypes]] -def multiclass_randomforest_modular(train=traindat,test=testdat,labels=label_traindat,ft=feattypes): +def multiclass_randomforest(train=traindat,test=testdat,labels=label_traindat,ft=feattypes): try: - from modshogun import RealFeatures, MulticlassLabels, CSVFile, RandomForest, MajorityVote + from shogun import RealFeatures, MulticlassLabels, CSVFile, RandomForest, MajorityVote except ImportError: print("Could not import Shogun modules") return @@ -35,4 +35,4 @@ def multiclass_randomforest_modular(train=traindat,test=testdat,labels=label_tra if __name__=='__main__': print('RandomForest') - multiclass_randomforest_modular(*parameter_list[0]) + multiclass_randomforest(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/preprocessor_dimensionreductionpreprocessor_modular.py b/examples/undocumented/python/preprocessor_dimensionreductionpreprocessor.py similarity index 67% rename from examples/undocumented/python_modular/preprocessor_dimensionreductionpreprocessor_modular.py rename to examples/undocumented/python/preprocessor_dimensionreductionpreprocessor.py index ccfe6a735f0..577def4dc9a 100644 --- a/examples/undocumented/python_modular/preprocessor_dimensionreductionpreprocessor_modular.py +++ b/examples/undocumented/python/preprocessor_dimensionreductionpreprocessor.py @@ -6,11 +6,11 @@ parameter_list = [[data, 20], [data, 30]] -def preprocessor_dimensionreductionpreprocessor_modular (data, k): - from modshogun import RealFeatures - from modshogun import DimensionReductionPreprocessor +def preprocessor_dimensionreductionpreprocessor (data, k): + from shogun import RealFeatures + from shogun import DimensionReductionPreprocessor try: - from modshogun import LocallyLinearEmbedding + from shogun import LocallyLinearEmbedding except ImportError: print("LocallyLinearEmbedding not available") exit(0) @@ -29,5 +29,5 @@ def preprocessor_dimensionreductionpreprocessor_modular (data, k): if __name__=='__main__': print('DimensionReductionPreprocessor') - preprocessor_dimensionreductionpreprocessor_modular(*parameter_list[0]) + preprocessor_dimensionreductionpreprocessor(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/preprocessor_fisherlda_modular.py b/examples/undocumented/python/preprocessor_fisherlda.py similarity index 65% rename from examples/undocumented/python_modular/preprocessor_fisherlda_modular.py rename to examples/undocumented/python/preprocessor_fisherlda.py index 0e263c194e5..5ff12da25dd 100644 --- a/examples/undocumented/python_modular/preprocessor_fisherlda_modular.py +++ b/examples/undocumented/python/preprocessor_fisherlda.py @@ -1,6 +1,6 @@ #!/usr/bin/env python from tools.load import LoadMatrix -from modshogun import * +from shogun import * lm=LoadMatrix() @@ -8,11 +8,11 @@ labels = lm.load_numbers('../data/label_train_multiclass.dat') parameter_list = [[data, labels, CANVAR_FLDA], [data, labels, CLASSIC_FLDA]] -def preprocessor_fisherlda_modular (data, labels, method): +def preprocessor_fisherlda (data, labels, method): - from modshogun import RealFeatures, MulticlassLabels, CANVAR_FLDA - from modshogun import FisherLda - from modshogun import MulticlassLabels + from shogun import RealFeatures, MulticlassLabels, CANVAR_FLDA + from shogun import FisherLda + from shogun import MulticlassLabels sg_features = RealFeatures(data) sg_labels = MulticlassLabels(labels) @@ -26,5 +26,5 @@ def preprocessor_fisherlda_modular (data, labels, method): if __name__=='__main__': print('FisherLda') - preprocessor_fisherlda_modular(*parameter_list[0]) + preprocessor_fisherlda(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/preprocessor_kernelpca_modular.py b/examples/undocumented/python/preprocessor_kernelpca.py similarity index 68% rename from examples/undocumented/python_modular/preprocessor_kernelpca_modular.py rename to examples/undocumented/python/preprocessor_kernelpca.py index a4efcd3c815..40905167fb2 100644 --- a/examples/undocumented/python_modular/preprocessor_kernelpca_modular.py +++ b/examples/undocumented/python/preprocessor_kernelpca.py @@ -6,10 +6,10 @@ parameter_list = [[data, 0.01, 1.0], [data, 0.05, 2.0]] -def preprocessor_kernelpca_modular (data, threshold, width): - from modshogun import RealFeatures - from modshogun import KernelPCA - from modshogun import GaussianKernel +def preprocessor_kernelpca (data, threshold, width): + from shogun import RealFeatures + from shogun import KernelPCA + from shogun import GaussianKernel features = RealFeatures(data) @@ -25,5 +25,5 @@ def preprocessor_kernelpca_modular (data, threshold, width): if __name__=='__main__': print('KernelPCA') - preprocessor_kernelpca_modular(*parameter_list[0]) + preprocessor_kernelpca(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/preprocessor_logplusone_modular.py b/examples/undocumented/python/preprocessor_logplusone.py similarity index 75% rename from examples/undocumented/python_modular/preprocessor_logplusone_modular.py rename to examples/undocumented/python/preprocessor_logplusone.py index 457db1471af..11115edac6b 100644 --- a/examples/undocumented/python_modular/preprocessor_logplusone_modular.py +++ b/examples/undocumented/python/preprocessor_logplusone.py @@ -7,11 +7,11 @@ parameter_list = [[traindat+10,testdat+10,1.4,10],[traindat+10,testdat+10,1.5,10]] -def preprocessor_logplusone_modular (fm_train_real=traindat,fm_test_real=testdat,width=1.4,size_cache=10): +def preprocessor_logplusone (fm_train_real=traindat,fm_test_real=testdat,width=1.4,size_cache=10): - from modshogun import Chi2Kernel - from modshogun import RealFeatures - from modshogun import LogPlusOne + from shogun import Chi2Kernel + from shogun import RealFeatures + from shogun import LogPlusOne feats_train=RealFeatures(fm_train_real) feats_test=RealFeatures(fm_test_real) @@ -34,4 +34,4 @@ def preprocessor_logplusone_modular (fm_train_real=traindat,fm_test_real=testdat if __name__=='__main__': print('LogPlusOne') - preprocessor_logplusone_modular(*parameter_list[0]) + preprocessor_logplusone(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/preprocessor_normone_modular.py b/examples/undocumented/python/preprocessor_normone.py similarity index 75% rename from examples/undocumented/python_modular/preprocessor_normone_modular.py rename to examples/undocumented/python/preprocessor_normone.py index 3a6a3453b8d..24afe917cdd 100644 --- a/examples/undocumented/python_modular/preprocessor_normone_modular.py +++ b/examples/undocumented/python/preprocessor_normone.py @@ -7,11 +7,11 @@ parameter_list = [[traindat,testdat,1.4,10],[traindat,testdat,1.5,10]] -def preprocessor_normone_modular (fm_train_real=traindat,fm_test_real=testdat,width=1.4,size_cache=10): +def preprocessor_normone (fm_train_real=traindat,fm_test_real=testdat,width=1.4,size_cache=10): - from modshogun import Chi2Kernel - from modshogun import RealFeatures - from modshogun import NormOne + from shogun import Chi2Kernel + from shogun import RealFeatures + from shogun import NormOne feats_train=RealFeatures(fm_train_real) feats_test=RealFeatures(fm_test_real) @@ -33,4 +33,4 @@ def preprocessor_normone_modular (fm_train_real=traindat,fm_test_real=testdat,wi if __name__=='__main__': print('NormOne') - preprocessor_normone_modular(*parameter_list[0]) + preprocessor_normone(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/preprocessor_pca_modular.py b/examples/undocumented/python/preprocessor_pca.py similarity index 70% rename from examples/undocumented/python_modular/preprocessor_pca_modular.py rename to examples/undocumented/python/preprocessor_pca.py index f5363d6019a..23149502390 100644 --- a/examples/undocumented/python_modular/preprocessor_pca_modular.py +++ b/examples/undocumented/python/preprocessor_pca.py @@ -6,9 +6,9 @@ parameter_list = [[data]] -def preprocessor_pca_modular (data): - from modshogun import RealFeatures - from modshogun import PCA +def preprocessor_pca (data): + from shogun import RealFeatures + from shogun import PCA features = RealFeatures(data) @@ -21,5 +21,5 @@ def preprocessor_pca_modular (data): if __name__=='__main__': print('PCA') - preprocessor_pca_modular(*parameter_list[0]) + preprocessor_pca(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/preprocessor_prunevarsubmean_modular.py b/examples/undocumented/python/preprocessor_prunevarsubmean.py similarity index 74% rename from examples/undocumented/python_modular/preprocessor_prunevarsubmean_modular.py rename to examples/undocumented/python/preprocessor_prunevarsubmean.py index 42696ca32b7..91bcc0376a4 100644 --- a/examples/undocumented/python_modular/preprocessor_prunevarsubmean_modular.py +++ b/examples/undocumented/python/preprocessor_prunevarsubmean.py @@ -7,10 +7,10 @@ parameter_list = [[traindat,testdat,1.5,10],[traindat,testdat,1.5,10]] -def preprocessor_prunevarsubmean_modular (fm_train_real=traindat,fm_test_real=testdat,width=1.4,size_cache=10): - from modshogun import Chi2Kernel - from modshogun import RealFeatures - from modshogun import PruneVarSubMean +def preprocessor_prunevarsubmean (fm_train_real=traindat,fm_test_real=testdat,width=1.4,size_cache=10): + from shogun import Chi2Kernel + from shogun import RealFeatures + from shogun import PruneVarSubMean feats_train=RealFeatures(fm_train_real) feats_test=RealFeatures(fm_test_real) @@ -32,4 +32,4 @@ def preprocessor_prunevarsubmean_modular (fm_train_real=traindat,fm_test_real=te if __name__=='__main__': print('PruneVarSubMean') - preprocessor_prunevarsubmean_modular(*parameter_list[0]) + preprocessor_prunevarsubmean(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/preprocessor_randomfouriergausspreproc_modular.py b/examples/undocumented/python/preprocessor_randomfouriergausspreproc.py similarity index 70% rename from examples/undocumented/python_modular/preprocessor_randomfouriergausspreproc_modular.py rename to examples/undocumented/python/preprocessor_randomfouriergausspreproc.py index c72ed33e4bd..f1e9a0dc656 100644 --- a/examples/undocumented/python_modular/preprocessor_randomfouriergausspreproc_modular.py +++ b/examples/undocumented/python/preprocessor_randomfouriergausspreproc.py @@ -7,13 +7,13 @@ parameter_list = [[traindat,testdat,1.5,10],[traindat,testdat,1.5,10]] -from modshogun import Math_init_random; +from shogun import Math_init_random; Math_init_random(12345); -def preprocessor_randomfouriergausspreproc_modular (fm_train_real=traindat,fm_test_real=testdat,width=1.4,size_cache=10): - from modshogun import Chi2Kernel - from modshogun import RealFeatures - from modshogun import RandomFourierGaussPreproc +def preprocessor_randomfouriergausspreproc (fm_train_real=traindat,fm_test_real=testdat,width=1.4,size_cache=10): + from shogun import Chi2Kernel + from shogun import RealFeatures + from shogun import RandomFourierGaussPreproc feats_train=RealFeatures(fm_train_real) feats_test=RealFeatures(fm_test_real) @@ -35,4 +35,4 @@ def preprocessor_randomfouriergausspreproc_modular (fm_train_real=traindat,fm_te if __name__=='__main__': print('RandomFourierGaussPreproc') - preprocessor_randomfouriergausspreproc_modular(*parameter_list[0]) + preprocessor_randomfouriergausspreproc(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/preprocessor_sortulongstring_modular.py b/examples/undocumented/python/preprocessor_sortulongstring.py similarity index 76% rename from examples/undocumented/python_modular/preprocessor_sortulongstring_modular.py rename to examples/undocumented/python/preprocessor_sortulongstring.py index 5f87eb93930..4b817b2996f 100644 --- a/examples/undocumented/python_modular/preprocessor_sortulongstring_modular.py +++ b/examples/undocumented/python/preprocessor_sortulongstring.py @@ -7,11 +7,11 @@ parameter_list = [[traindna,testdna,4,0,False,False],[traindna,testdna,3,0,False,False]] -def preprocessor_sortulongstring_modular (fm_train_dna=traindna,fm_test_dna=testdna,order=3,gap=0,reverse=False,use_sign=False): +def preprocessor_sortulongstring (fm_train_dna=traindna,fm_test_dna=testdna,order=3,gap=0,reverse=False,use_sign=False): - from modshogun import CommUlongStringKernel - from modshogun import StringCharFeatures, StringUlongFeatures, DNA - from modshogun import SortUlongString + from shogun import CommUlongStringKernel + from shogun import StringCharFeatures, StringUlongFeatures, DNA + from shogun import SortUlongString charfeat=StringCharFeatures(DNA) @@ -40,4 +40,4 @@ def preprocessor_sortulongstring_modular (fm_train_dna=traindna,fm_test_dna=test if __name__=='__main__': print('CommUlongString') - preprocessor_sortulongstring_modular(*parameter_list[0]) + preprocessor_sortulongstring(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/preprocessor_sortwordstring_modular.py b/examples/undocumented/python/preprocessor_sortwordstring.py similarity index 76% rename from examples/undocumented/python_modular/preprocessor_sortwordstring_modular.py rename to examples/undocumented/python/preprocessor_sortwordstring.py index fdc7f36a62e..231d334298a 100644 --- a/examples/undocumented/python_modular/preprocessor_sortwordstring_modular.py +++ b/examples/undocumented/python/preprocessor_sortwordstring.py @@ -7,11 +7,11 @@ parameter_list = [[traindna,testdna,3,0,False,False],[traindna,testdna,3,0,False,False]] -def preprocessor_sortwordstring_modular (fm_train_dna=traindna,fm_test_dna=testdna,order=3,gap=0,reverse=False,use_sign=False): +def preprocessor_sortwordstring (fm_train_dna=traindna,fm_test_dna=testdna,order=3,gap=0,reverse=False,use_sign=False): - from modshogun import CommWordStringKernel - from modshogun import StringCharFeatures, StringWordFeatures, DNA - from modshogun import SortWordString + from shogun import CommWordStringKernel + from shogun import StringCharFeatures, StringWordFeatures, DNA + from shogun import SortWordString charfeat=StringCharFeatures(fm_train_dna, DNA) feats_train=StringWordFeatures(charfeat.get_alphabet()) @@ -37,4 +37,4 @@ def preprocessor_sortwordstring_modular (fm_train_dna=traindna,fm_test_dna=testd if __name__=='__main__': print('CommWordString') - preprocessor_sortwordstring_modular(*parameter_list[0]) + preprocessor_sortwordstring(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/regression_cartree_modular.py b/examples/undocumented/python/regression_cartree.py similarity index 80% rename from examples/undocumented/python_modular/regression_cartree_modular.py rename to examples/undocumented/python/regression_cartree.py index 94221969dcb..ad89191af11 100644 --- a/examples/undocumented/python_modular/regression_cartree_modular.py +++ b/examples/undocumented/python/regression_cartree.py @@ -6,9 +6,9 @@ parameter_list = [[50,5,15,0.2,feattypes]] -def regression_cartree_modular(num_train=500,num_test=50,x_range=15,noise_var=0.2,ft=feattypes): +def regression_cartree(num_train=500,num_test=50,x_range=15,noise_var=0.2,ft=feattypes): try: - from modshogun import RealFeatures, RegressionLabels, CSVFile, CARTree, PT_REGRESSION + from shogun import RealFeatures, RegressionLabels, CSVFile, CARTree, PT_REGRESSION from numpy import random except ImportError: print("Could not import Shogun and/or numpy modules") @@ -40,4 +40,4 @@ def regression_cartree_modular(num_train=500,num_test=50,x_range=15,noise_var=0. if __name__=='__main__': print('CARTree') - regression_cartree_modular(*parameter_list[0]) + regression_cartree(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/regression_chaidtree_modular.py b/examples/undocumented/python/regression_chaidtree.py similarity index 80% rename from examples/undocumented/python_modular/regression_chaidtree_modular.py rename to examples/undocumented/python/regression_chaidtree.py index 1bdef04488d..60b7f883ca6 100644 --- a/examples/undocumented/python_modular/regression_chaidtree_modular.py +++ b/examples/undocumented/python/regression_chaidtree.py @@ -6,9 +6,9 @@ parameter_list = [[500,50,15,0.2,feattypes]] -def regression_chaidtree_modular(num_train=500,num_test=50,x_range=15,noise_var=0.2,ft=feattypes): +def regression_chaidtree(num_train=500,num_test=50,x_range=15,noise_var=0.2,ft=feattypes): try: - from modshogun import RealFeatures, RegressionLabels, CSVFile, CHAIDTree, PT_REGRESSION + from shogun import RealFeatures, RegressionLabels, CSVFile, CHAIDTree, PT_REGRESSION from numpy import random except ImportError: print("Could not import Shogun and/or numpy modules") @@ -40,4 +40,4 @@ def regression_chaidtree_modular(num_train=500,num_test=50,x_range=15,noise_var= if __name__=='__main__': print('CHAIDTree') - regression_chaidtree_modular(*parameter_list[0]) + regression_chaidtree(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/regression_randomforest_modular.py b/examples/undocumented/python/regression_randomforest.py similarity index 82% rename from examples/undocumented/python_modular/regression_randomforest_modular.py rename to examples/undocumented/python/regression_randomforest.py index b907b8f939f..acfc0d2073e 100644 --- a/examples/undocumented/python_modular/regression_randomforest_modular.py +++ b/examples/undocumented/python/regression_randomforest.py @@ -10,9 +10,9 @@ parameter_list = [[500,50,15,0.2,feattypes]] -def regression_randomforest_modular(num_train=500,num_test=50,x_range=15,noise_var=0.2,ft=feattypes): +def regression_randomforest(num_train=500,num_test=50,x_range=15,noise_var=0.2,ft=feattypes): try: - from modshogun import RealFeatures, RegressionLabels, CSVFile, RandomForest, MeanRule, PT_REGRESSION + from shogun import RealFeatures, RegressionLabels, CSVFile, RandomForest, MeanRule, PT_REGRESSION except ImportError: print("Could not import Shogun modules") return @@ -45,4 +45,4 @@ def regression_randomforest_modular(num_train=500,num_test=50,x_range=15,noise_v if __name__=='__main__': print('RandomForest') - regression_randomforest_modular(*parameter_list[0]) + regression_randomforest(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/regression_svrlight_modular.py b/examples/undocumented/python/regression_svrlight.py similarity index 81% rename from examples/undocumented/python_modular/regression_svrlight_modular.py rename to examples/undocumented/python/regression_svrlight.py index 752e17c980a..e9126fc8c7d 100644 --- a/examples/undocumented/python_modular/regression_svrlight_modular.py +++ b/examples/undocumented/python/regression_svrlight.py @@ -13,14 +13,14 @@ parameter_list = [[traindat,testdat,label_traindat,1.2,1,1e-5,1e-2,1],[traindat,testdat,label_traindat,2.3,0.5,1e-5,1e-6,1]] -def regression_svrlight_modular (fm_train=traindat,fm_test=testdat,label_train=label_traindat, \ +def regression_svrlight (fm_train=traindat,fm_test=testdat,label_train=label_traindat, \ width=1.2,C=1,epsilon=1e-5,tube_epsilon=1e-2,num_threads=3): - from modshogun import RegressionLabels, RealFeatures - from modshogun import GaussianKernel + from shogun import RegressionLabels, RealFeatures + from shogun import GaussianKernel try: - from modshogun import SVRLight + from shogun import SVRLight except ImportError: print('No support for SVRLight available.') return @@ -44,4 +44,4 @@ def regression_svrlight_modular (fm_train=traindat,fm_test=testdat,label_train=l if __name__=='__main__': print('SVRLight') - regression_svrlight_modular(*parameter_list[0]) + regression_svrlight(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/serialization_complex_example.py b/examples/undocumented/python/serialization_complex_example.py similarity index 70% rename from examples/undocumented/python_modular/serialization_complex_example.py rename to examples/undocumented/python/serialization_complex_example.py index 2e1a17b226f..33966a40452 100644 --- a/examples/undocumented/python_modular/serialization_complex_example.py +++ b/examples/undocumented/python/serialization_complex_example.py @@ -9,12 +9,13 @@ def serialization_complex_example (num=5, dist=1, dim=10, C=2.0, width=10): import os from numpy import concatenate, zeros, ones from numpy.random import randn, seed - from modshogun import RealFeatures, MulticlassLabels - from modshogun import GMNPSVM - from modshogun import GaussianKernel - from modshogun import SerializableHdf5File,SerializableAsciiFile, \ + from shogun import RealFeatures, MulticlassLabels + from shogun import GMNPSVM + from shogun import GaussianKernel + from shogun import SerializableHdf5File,SerializableAsciiFile, \ SerializableJsonFile,SerializableXmlFile,MSG_DEBUG - from modshogun import NormOne, LogPlusOne + from shogun import NormOne, LogPlusOne + from tempfile import NamedTemporaryFile seed(17) @@ -40,57 +41,57 @@ def serialization_complex_example (num=5, dist=1, dim=10, C=2.0, width=10): #svm.print_serializable() - fstream = SerializableHdf5File("tmp/blaah.h5", "w") + tmp_h5 = NamedTemporaryFile(suffix='h5') + fstream = SerializableHdf5File(tmp_h5.name, "w") status = svm.save_serializable(fstream) check_status(status,'h5') - fstream = SerializableAsciiFile("tmp/blaah.asc", "w") + tmp_asc = NamedTemporaryFile(suffix='asc') + fstream = SerializableAsciiFile(tmp_asc.name, "w") status = svm.save_serializable(fstream) check_status(status,'asc') - fstream = SerializableJsonFile("tmp/blaah.json", "w") + tmp_json = NamedTemporaryFile(suffix='json') + fstream = SerializableJsonFile(tmp_json.name, "w") status = svm.save_serializable(fstream) check_status(status,'json') - fstream = SerializableXmlFile("tmp/blaah.xml", "w") + tmp_xml = NamedTemporaryFile(suffix='xml') + fstream = SerializableXmlFile(tmp_xml.name, "w") status = svm.save_serializable(fstream) check_status(status,'xml') - fstream = SerializableHdf5File("tmp/blaah.h5", "r") + fstream = SerializableHdf5File(tmp_h5.name, "r") new_svm=GMNPSVM() status = new_svm.load_serializable(fstream) check_status(status,'h5') new_svm.train() bias_h5 = new_svm.get_svm(0).get_bias() - fstream = SerializableAsciiFile("tmp/blaah.asc", "r") + fstream = SerializableAsciiFile(tmp_asc.name, "r") new_svm=GMNPSVM() status = new_svm.load_serializable(fstream) check_status(status,'asc') new_svm.train() bias_asc = new_svm.get_svm(0).get_bias() - fstream = SerializableJsonFile("tmp/blaah.json", "r") + fstream = SerializableJsonFile(tmp_json.name, "r") new_svm=GMNPSVM() status = new_svm.load_serializable(fstream) check_status(status,'json') new_svm.train() bias_json = new_svm.get_svm(0).get_bias() - fstream = SerializableXmlFile("tmp/blaah.xml", "r") + fstream = SerializableXmlFile(tmp_xml.name, "r") new_svm=GMNPSVM() status = new_svm.load_serializable(fstream) check_status(status,'xml') new_svm.train() bias_xml = new_svm.get_svm(0).get_bias() - os.unlink("tmp/blaah.h5") - os.unlink("tmp/blaah.asc") - os.unlink("tmp/blaah.json") - os.unlink("tmp/blaah.xml") return svm,new_svm, bias_ref, bias_h5, bias_asc, bias_json, bias_xml if __name__=='__main__': - print('Serialization SVMLight') + print('Serialization') serialization_complex_example(*parameter_list[0]) diff --git a/examples/undocumented/python/serialization_matrix.py b/examples/undocumented/python/serialization_matrix.py new file mode 100644 index 00000000000..6d6be7ea5b3 --- /dev/null +++ b/examples/undocumented/python/serialization_matrix.py @@ -0,0 +1,22 @@ +#!/usr/bin/env python +from shogun import * +from numpy import array + +parameter_list=[[[[1.0,2,3],[4,5,6]]]] + +def serialization_matrix (m): + from tempfile import NamedTemporaryFile + feats=RealFeatures(array(m)) + #feats.io.set_loglevel(0) + tmp_asc_1 = NamedTemporaryFile(suffix='1.asc') + fstream = SerializableAsciiFile(tmp_asc_1.name, "w") + feats.save_serializable(fstream) + + tmp_asc_2 = NamedTemporaryFile(suffix='2.asc') + l=MulticlassLabels(array([1.0,2,3])) + fstream = SerializableAsciiFile(tmp_asc_2.name, "w") + l.save_serializable(fstream) + +if __name__=='__main__': + print('Serialization Matrix Modular') + serialization_matrix(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/serialization_string_kernels_modular.py b/examples/undocumented/python/serialization_string_kernels.py similarity index 85% rename from examples/undocumented/python_modular/serialization_string_kernels_modular.py rename to examples/undocumented/python/serialization_string_kernels.py index 9ca556d6379..9971b8532b8 100644 --- a/examples/undocumented/python_modular/serialization_string_kernels_modular.py +++ b/examples/undocumented/python/serialization_string_kernels.py @@ -1,17 +1,17 @@ #!/usr/bin/env python -from modshogun import WeightedDegreeStringKernel, LinearKernel, PolyKernel, GaussianKernel, CTaxonomy -from modshogun import CombinedKernel, WeightedDegreeRBFKernel -from modshogun import StringCharFeatures, RealFeatures, CombinedFeatures, StringWordFeatures, SortWordString -from modshogun import DNA, PROTEIN, Labels -from modshogun import WeightedDegreeStringKernel, CombinedKernel, WeightedCommWordStringKernel, WeightedDegreePositionStringKernel -from modshogun import StringCharFeatures, DNA, StringWordFeatures, CombinedFeatures - -from modshogun import MSG_DEBUG -from modshogun import RealFeatures, BinaryLabels, DNA, Alphabet -from modshogun import WeightedDegreeStringKernel, GaussianKernel +from shogun import WeightedDegreeStringKernel, LinearKernel, PolyKernel, GaussianKernel, CTaxonomy +from shogun import CombinedKernel, WeightedDegreeRBFKernel +from shogun import StringCharFeatures, RealFeatures, CombinedFeatures, StringWordFeatures, SortWordString +from shogun import DNA, PROTEIN, Labels +from shogun import WeightedDegreeStringKernel, CombinedKernel, WeightedCommWordStringKernel, WeightedDegreePositionStringKernel +from shogun import StringCharFeatures, DNA, StringWordFeatures, CombinedFeatures + +from shogun import MSG_DEBUG +from shogun import RealFeatures, BinaryLabels, DNA, Alphabet +from shogun import WeightedDegreeStringKernel, GaussianKernel try: - from modshogun import SVMLight + from shogun import SVMLight except ImportError: print("SVMLight is not available") exit(0) @@ -149,7 +149,7 @@ def construct_features(features): parameter_list = [[200, 1, 100]] -def serialization_string_kernels_modular(n_data, num_shifts, size): +def serialization_string_kernels(n_data, num_shifts, size): """ serialize svm with string kernels """ @@ -215,5 +215,5 @@ def serialization_string_kernels_modular(n_data, num_shifts, size): if __name__=='__main__': - serialization_string_kernels_modular(*parameter_list[0]) + serialization_string_kernels(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/serialization_svmlight_modular.py b/examples/undocumented/python/serialization_svmlight.py similarity index 89% rename from examples/undocumented/python_modular/serialization_svmlight_modular.py rename to examples/undocumented/python/serialization_svmlight.py index e9e231896c6..f130c61a071 100644 --- a/examples/undocumented/python_modular/serialization_svmlight_modular.py +++ b/examples/undocumented/python/serialization_svmlight.py @@ -1,12 +1,12 @@ #!/usr/bin/env python parameter_list=[[10, 1, 2.1, 2.0]] -def serialization_svmlight_modular (num, dist, width, C): - from modshogun import MSG_DEBUG - from modshogun import RealFeatures, BinaryLabels, DNA, Alphabet - from modshogun import WeightedDegreeStringKernel, GaussianKernel +def serialization_svmlight (num, dist, width, C): + from shogun import MSG_DEBUG + from shogun import RealFeatures, BinaryLabels, DNA, Alphabet + from shogun import WeightedDegreeStringKernel, GaussianKernel try: - from modshogun import SVMLight + from shogun import SVMLight except ImportError: print("SVMLight not available") exit(0) @@ -119,4 +119,4 @@ def load(filename): if __name__=='__main__': print('Serialization SVMLight') - serialization_svmlight_modular(*parameter_list[0]) + serialization_svmlight(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/so_multiclass.py b/examples/undocumented/python/so_multiclass.py similarity index 92% rename from examples/undocumented/python_modular/so_multiclass.py rename to examples/undocumented/python/so_multiclass.py index 4923ea553c8..b6ba20bd0a9 100644 --- a/examples/undocumented/python_modular/so_multiclass.py +++ b/examples/undocumented/python/so_multiclass.py @@ -26,8 +26,8 @@ def gen_data(num_classes,num_samples,dim): def so_multiclass (fm_train_real=traindat,label_train_multiclass=label_traindat): try: - from modshogun import RealFeatures - from modshogun import MulticlassModel, MulticlassSOLabels, PrimalMosekSOSVM, RealNumber + from shogun import RealFeatures + from shogun import MulticlassModel, MulticlassSOLabels, PrimalMosekSOSVM, RealNumber except ImportError: print("Mosek not available") return diff --git a/examples/undocumented/python_modular/stochasticgbmachine_modular.py b/examples/undocumented/python/stochasticgbmachine.py similarity index 73% rename from examples/undocumented/python_modular/stochasticgbmachine_modular.py rename to examples/undocumented/python/stochasticgbmachine.py index e1acf8f1246..dae5d67ffce 100644 --- a/examples/undocumented/python_modular/stochasticgbmachine_modular.py +++ b/examples/undocumented/python/stochasticgbmachine.py @@ -9,9 +9,9 @@ parameter_list = [[traindat,label_traindat,feat_types]] -def stochasticgbmachine_modular(train=traindat,train_labels=label_traindat,ft=feat_types): +def stochasticgbmachine(train=traindat,train_labels=label_traindat,ft=feat_types): try: - from modshogun import RealFeatures, RegressionLabels, CSVFile, CARTree, StochasticGBMachine, SquaredLoss + from shogun import RealFeatures, RegressionLabels, CSVFile, CARTree, StochasticGBMachine, SquaredLoss except ImportError: print("Could not import Shogun modules") return @@ -31,16 +31,16 @@ def stochasticgbmachine_modular(train=traindat,train_labels=label_traindat,ft=fe s=StochasticGBMachine(cart,loss,500,0.01,0.6) # train - feats.add_subset(np.int32(p[0:num])) - labels.add_subset(np.int32(p[0:num])) + feats.add_subset(np.int32(p[0:int(num)])) + labels.add_subset(np.int32(p[0:int(num)])) s.set_labels(labels) s.train(feats) feats.remove_subset() labels.remove_subset() # apply - feats.add_subset(np.int32(p[num:len(p)])) - labels.add_subset(np.int32(p[num:len(p)])) + feats.add_subset(np.int32(p[int(num):len(p)])) + labels.add_subset(np.int32(p[int(num):len(p)])) output=s.apply_regression(feats) feats.remove_subset() @@ -50,4 +50,4 @@ def stochasticgbmachine_modular(train=traindat,train_labels=label_traindat,ft=fe if __name__=='__main__': print('StochasticGBMachine') - stochasticgbmachine_modular(*parameter_list[0]) + stochasticgbmachine(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/streaming_vw_modular.py b/examples/undocumented/python/streaming_vw.py similarity index 74% rename from examples/undocumented/python_modular/streaming_vw_modular.py rename to examples/undocumented/python/streaming_vw.py index aea41f328a4..8e21dbdbccb 100644 --- a/examples/undocumented/python_modular/streaming_vw_modular.py +++ b/examples/undocumented/python/streaming_vw.py @@ -1,12 +1,12 @@ #!/usr/bin/env python -from modshogun import StreamingVwFile -from modshogun import T_SVMLIGHT -from modshogun import StreamingVwFeatures -from modshogun import VowpalWabbit +from shogun import StreamingVwFile +from shogun import T_SVMLIGHT +from shogun import StreamingVwFeatures +from shogun import VowpalWabbit parameter_list=[[None]] -def streaming_vw_modular (dummy): +def streaming_vw (dummy): """Runs the VW algorithm on a toy dataset in SVMLight format.""" # Open the input file as a StreamingVwFile @@ -28,4 +28,4 @@ def streaming_vw_modular (dummy): ##return vw if __name__ == "__main__": - streaming_vw_modular(*parameter_list[0]) + streaming_vw(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/streaming_vw_createcache_modular.py b/examples/undocumented/python/streaming_vw_createcache.py similarity index 80% rename from examples/undocumented/python_modular/streaming_vw_createcache_modular.py rename to examples/undocumented/python/streaming_vw_createcache.py index bbdff3fbc44..2404f0350b7 100644 --- a/examples/undocumented/python_modular/streaming_vw_createcache_modular.py +++ b/examples/undocumented/python/streaming_vw_createcache.py @@ -1,13 +1,13 @@ #!/usr/bin/env python -from modshogun import StreamingVwFile -from modshogun import StreamingVwCacheFile -from modshogun import T_SVMLIGHT -from modshogun import StreamingVwFeatures -from modshogun import VowpalWabbit +from shogun import StreamingVwFile +from shogun import StreamingVwCacheFile +from shogun import T_SVMLIGHT +from shogun import StreamingVwFeatures +from shogun import VowpalWabbit parameter_list=[['../data/fm_train_sparsereal.dat']] -def streaming_vw_createcache_modular (fname): +def streaming_vw_createcache (fname): # First creates a binary cache from an ascii data file. # and then trains using the StreamingVwCacheFile as input @@ -42,4 +42,4 @@ def streaming_vw_createcache_modular (fname): ##return vw if __name__ == "__main__": - streaming_vw_createcache_modular(*parameter_list[0]) + streaming_vw_createcache(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/structure_discrete_hmsvm_bmrm.py b/examples/undocumented/python/structure_discrete_hmsvm_bmrm.py similarity index 84% rename from examples/undocumented/python_modular/structure_discrete_hmsvm_bmrm.py rename to examples/undocumented/python/structure_discrete_hmsvm_bmrm.py index 9fee3bd2826..59cf961f4b8 100644 --- a/examples/undocumented/python_modular/structure_discrete_hmsvm_bmrm.py +++ b/examples/undocumented/python/structure_discrete_hmsvm_bmrm.py @@ -9,10 +9,10 @@ parameter_list=[[data_dict]] def structure_discrete_hmsvm_bmrm (m_data_dict=data_dict): - from modshogun import RealMatrixFeatures, SequenceLabels, HMSVMModel, Sequence, TwoStateModel - from modshogun import StructuredAccuracy, SMT_TWO_STATE + from shogun import RealMatrixFeatures, SequenceLabels, HMSVMModel, Sequence, TwoStateModel + from shogun import StructuredAccuracy, SMT_TWO_STATE try: - from modshogun import DualLibQPBMSOSVM + from shogun import DualLibQPBMSOSVM except ImportError: print("DualLibQPBMSOSVM not available") exit(0) diff --git a/examples/undocumented/python_modular/structure_discrete_hmsvm_mosek.py b/examples/undocumented/python/structure_discrete_hmsvm_mosek.py similarity index 83% rename from examples/undocumented/python_modular/structure_discrete_hmsvm_mosek.py rename to examples/undocumented/python/structure_discrete_hmsvm_mosek.py index 14cecf439bf..47ac6e4b161 100644 --- a/examples/undocumented/python_modular/structure_discrete_hmsvm_mosek.py +++ b/examples/undocumented/python/structure_discrete_hmsvm_mosek.py @@ -9,11 +9,11 @@ parameter_list=[[data_dict]] def structure_discrete_hmsvm_mosek (m_data_dict=data_dict): - from modshogun import RealMatrixFeatures, SequenceLabels, HMSVMModel, Sequence, TwoStateModel - from modshogun import StructuredAccuracy, SMT_TWO_STATE + from shogun import RealMatrixFeatures, SequenceLabels, HMSVMModel, Sequence, TwoStateModel + from shogun import StructuredAccuracy, SMT_TWO_STATE try: - from modshogun import PrimalMosekSOSVM + from shogun import PrimalMosekSOSVM except ImportError: print("Mosek not available") return diff --git a/examples/undocumented/python_modular/structure_dynprog_modular.py b/examples/undocumented/python/structure_dynprog.py similarity index 98% rename from examples/undocumented/python_modular/structure_dynprog_modular.py rename to examples/undocumented/python/structure_dynprog.py index b9298489128..e97ddbf5954 100644 --- a/examples/undocumented/python_modular/structure_dynprog_modular.py +++ b/examples/undocumented/python/structure_dynprog.py @@ -4,7 +4,7 @@ parameter_list=[['../data/DynProg_example_py.pickle.gz']] -from modshogun import * +from shogun import * import numpy from numpy import array,Inf,float64,matrix,frompyfunc,zeros @@ -64,7 +64,7 @@ def loads(str): unpickler.dispatch[pickle.GLOBAL] = mapped_load_global return unpickler.load() -def structure_dynprog_modular (fname): +def structure_dynprog (fname): import sys #pickle is not compatible between python2 -> 3 @@ -190,4 +190,4 @@ def structure_dynprog_modular (fname): if __name__ == '__main__': print("Structure") - structure_dynprog_modular(*parameter_list[0]) + structure_dynprog(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/structure_factor_graph_model.py b/examples/undocumented/python/structure_factor_graph_model.py similarity index 90% rename from examples/undocumented/python_modular/structure_factor_graph_model.py rename to examples/undocumented/python/structure_factor_graph_model.py index 2c37f7582cc..face601195c 100644 --- a/examples/undocumented/python_modular/structure_factor_graph_model.py +++ b/examples/undocumented/python/structure_factor_graph_model.py @@ -1,7 +1,7 @@ #!/usr/bin/env python import numpy as np -from modshogun import TableFactorType +from shogun import TableFactorType # create the factor type with GT parameters tid = 0 @@ -20,10 +20,10 @@ fac_type_b = TableFactorType(tid_b, cards_b, w_gt_b) def gen_data(ftype, num_samples, show_data = False): - from modshogun import Math - from modshogun import FactorType, Factor, TableFactorType, FactorGraph - from modshogun import FactorGraphObservation, FactorGraphLabels, FactorGraphFeatures - from modshogun import MAPInference, TREE_MAX_PROD + from shogun import Math + from shogun import FactorType, Factor, TableFactorType, FactorGraph + from shogun import FactorGraphObservation, FactorGraphLabels, FactorGraphFeatures + from shogun import MAPInference, TREE_MAX_PROD Math.init_random(17) @@ -95,11 +95,11 @@ def gen_data(ftype, num_samples, show_data = False): parameter_list = [[samples,labels,w_all,ftype_all]] def structure_factor_graph_model(tr_samples = samples, tr_labels = labels, w = w_all, ftype = ftype_all): - from modshogun import SOSVMHelper, LabelsFactory - from modshogun import FactorGraphModel, MAPInference, TREE_MAX_PROD - from modshogun import StochasticSOSVM, FWSOSVM + from shogun import SOSVMHelper, LabelsFactory + from shogun import FactorGraphModel, MAPInference, TREE_MAX_PROD + from shogun import StochasticSOSVM, FWSOSVM try: - from modshogun import DualLibQPBMSOSVM + from shogun import DualLibQPBMSOSVM except ImportError: print("DualLibQPBMSOSVM not available") exit(0) diff --git a/examples/undocumented/python_modular/structure_graphcuts.py b/examples/undocumented/python/structure_graphcuts.py similarity index 96% rename from examples/undocumented/python_modular/structure_graphcuts.py rename to examples/undocumented/python/structure_graphcuts.py index 3615f90c6e8..59174ea05a1 100644 --- a/examples/undocumented/python_modular/structure_graphcuts.py +++ b/examples/undocumented/python/structure_graphcuts.py @@ -3,11 +3,11 @@ import numpy as np import itertools -from modshogun import Factor, TableFactorType, FactorGraph -from modshogun import FactorGraphObservation, FactorGraphLabels, FactorGraphFeatures -from modshogun import FactorGraphModel, GRAPH_CUT -from modshogun import GraphCut -from modshogun import StochasticSOSVM +from shogun import Factor, TableFactorType, FactorGraph +from shogun import FactorGraphObservation, FactorGraphLabels, FactorGraphFeatures +from shogun import FactorGraphModel, GRAPH_CUT +from shogun import GraphCut +from shogun import StochasticSOSVM def generate_data(num_train_samples, len_label, len_feat): """ Generate synthetic dataset diff --git a/examples/undocumented/python_modular/structure_hierarchical_multilabel_classification.py b/examples/undocumented/python/structure_hierarchical_multilabel_classification.py similarity index 95% rename from examples/undocumented/python_modular/structure_hierarchical_multilabel_classification.py rename to examples/undocumented/python/structure_hierarchical_multilabel_classification.py index 05392a2fab1..05c3e70e3ee 100644 --- a/examples/undocumented/python_modular/structure_hierarchical_multilabel_classification.py +++ b/examples/undocumented/python/structure_hierarchical_multilabel_classification.py @@ -7,10 +7,10 @@ http://kt.ijs.si/DragiKocev/PhD/resources/doku.php?id=hmc_classification#imageclef07d """ -from modshogun import MultilabelSOLabels, HierarchicalMultilabelModel -from modshogun import RealFeatures -from modshogun import StochasticSOSVM -from modshogun import StructuredAccuracy, LabelsFactory +from shogun import MultilabelSOLabels, HierarchicalMultilabelModel +from shogun import RealFeatures +from shogun import StochasticSOSVM +from shogun import StructuredAccuracy, LabelsFactory import numpy as np import time diff --git a/examples/undocumented/python_modular/structure_multiclass_bmrm.py b/examples/undocumented/python/structure_multiclass_bmrm.py similarity index 92% rename from examples/undocumented/python_modular/structure_multiclass_bmrm.py rename to examples/undocumented/python/structure_multiclass_bmrm.py index 1a293763b79..d7dbd034a47 100644 --- a/examples/undocumented/python_modular/structure_multiclass_bmrm.py +++ b/examples/undocumented/python/structure_multiclass_bmrm.py @@ -25,15 +25,15 @@ def gen_data(num_classes,num_samples,dim): parameter_list = [[traindat,label_traindat]] def structure_multiclass_bmrm(fm_train_real=traindat,label_train_multiclass=label_traindat): - from modshogun import MulticlassSOLabels, LabelsFactory - from modshogun import RealFeatures - from modshogun import SOSVMHelper + from shogun import MulticlassSOLabels, LabelsFactory + from shogun import RealFeatures + from shogun import SOSVMHelper try: - from modshogun import BMRM, PPBMRM, P3BMRM, DualLibQPBMSOSVM + from shogun import BMRM, PPBMRM, P3BMRM, DualLibQPBMSOSVM except ImportError: print("At least one of BMRM, PPBMRM, P3BMRM, DualLibQPBMSOSVM not available") exit(0) - from modshogun import MulticlassModel, RealNumber + from shogun import MulticlassModel, RealNumber labels = MulticlassSOLabels(label_train_multiclass) features = RealFeatures(fm_train_real.T) diff --git a/examples/undocumented/python_modular/structure_plif_hmsvm_bmrm.py b/examples/undocumented/python/structure_plif_hmsvm_bmrm.py similarity index 86% rename from examples/undocumented/python_modular/structure_plif_hmsvm_bmrm.py rename to examples/undocumented/python/structure_plif_hmsvm_bmrm.py index 91822d366a6..bd7542607a0 100644 --- a/examples/undocumented/python_modular/structure_plif_hmsvm_bmrm.py +++ b/examples/undocumented/python/structure_plif_hmsvm_bmrm.py @@ -3,9 +3,9 @@ parameter_list=[[50, 125, 10, 2]] def structure_plif_hmsvm_bmrm (num_examples, example_length, num_features, num_noise_features): - from modshogun import RealMatrixFeatures, TwoStateModel, StructuredAccuracy + from shogun import RealMatrixFeatures, TwoStateModel, StructuredAccuracy try: - from modshogun import DualLibQPBMSOSVM + from shogun import DualLibQPBMSOSVM except ImportError: print("DualLibQPBMSOSVM not available") exit(0) diff --git a/examples/undocumented/python_modular/structure_plif_hmsvm_mosek.py b/examples/undocumented/python/structure_plif_hmsvm_mosek.py similarity index 85% rename from examples/undocumented/python_modular/structure_plif_hmsvm_mosek.py rename to examples/undocumented/python/structure_plif_hmsvm_mosek.py index 2ef2e0dc40c..8e918f08512 100644 --- a/examples/undocumented/python_modular/structure_plif_hmsvm_mosek.py +++ b/examples/undocumented/python/structure_plif_hmsvm_mosek.py @@ -3,10 +3,10 @@ parameter_list=[[100, 250, 10, 2]] def structure_plif_hmsvm_mosek (num_examples, example_length, num_features, num_noise_features): - from modshogun import RealMatrixFeatures, TwoStateModel, StructuredAccuracy + from shogun import RealMatrixFeatures, TwoStateModel, StructuredAccuracy try: - from modshogun import PrimalMosekSOSVM + from shogun import PrimalMosekSOSVM except ImportError: print("Mosek not available") return diff --git a/examples/undocumented/python_modular/tests_check_commwordkernel_memleak_modular.py b/examples/undocumented/python/tests_check_commwordkernel_memleak.py similarity index 92% rename from examples/undocumented/python_modular/tests_check_commwordkernel_memleak_modular.py rename to examples/undocumented/python/tests_check_commwordkernel_memleak.py index 790e75bbc8e..5da5b8b07df 100644 --- a/examples/undocumented/python_modular/tests_check_commwordkernel_memleak_modular.py +++ b/examples/undocumented/python/tests_check_commwordkernel_memleak.py @@ -1,11 +1,11 @@ #!/usr/bin/env python parameter_list=[[10,7,0,False]] -def tests_check_commwordkernel_memleak_modular (num, order, gap, reverse): +def tests_check_commwordkernel_memleak (num, order, gap, reverse): import gc - from modshogun import Alphabet,StringCharFeatures,StringWordFeatures,DNA - from modshogun import SortWordString, MSG_DEBUG - from modshogun import CommWordStringKernel, IdentityKernelNormalizer + from shogun import Alphabet,StringCharFeatures,StringWordFeatures,DNA + from shogun import SortWordString, MSG_DEBUG + from shogun import CommWordStringKernel, IdentityKernelNormalizer from numpy import mat POS=[num*'ACGT', num*'ACGT', num*'ACGT',num*'ACGT', num*'ACGT', @@ -83,4 +83,4 @@ def tests_check_commwordkernel_memleak_modular (num, order, gap, reverse): if __name__=='__main__': print('Leak Check Comm Word Kernel') - tests_check_commwordkernel_memleak_modular(*parameter_list[0]) + tests_check_commwordkernel_memleak(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/tools/__init__.py b/examples/undocumented/python/tools/__init__.py similarity index 100% rename from examples/undocumented/python_modular/tools/__init__.py rename to examples/undocumented/python/tools/__init__.py diff --git a/examples/undocumented/python_modular/tools/generate_circle_data.py b/examples/undocumented/python/tools/generate_circle_data.py similarity index 100% rename from examples/undocumented/python_modular/tools/generate_circle_data.py rename to examples/undocumented/python/tools/generate_circle_data.py diff --git a/examples/undocumented/python_modular/tools/load.py b/examples/undocumented/python/tools/load.py similarity index 100% rename from examples/undocumented/python_modular/tools/load.py rename to examples/undocumented/python/tools/load.py diff --git a/examples/undocumented/python_modular/tools/multiclass_shared.py b/examples/undocumented/python/tools/multiclass_shared.py similarity index 100% rename from examples/undocumented/python_modular/tools/multiclass_shared.py rename to examples/undocumented/python/tools/multiclass_shared.py diff --git a/examples/undocumented/python_modular/transfer_multitask_clustered_logistic_regression.py b/examples/undocumented/python/transfer_multitask_clustered_logistic_regression.py similarity index 91% rename from examples/undocumented/python_modular/transfer_multitask_clustered_logistic_regression.py rename to examples/undocumented/python/transfer_multitask_clustered_logistic_regression.py index 87ac204ab44..e3dc8b2a64b 100644 --- a/examples/undocumented/python_modular/transfer_multitask_clustered_logistic_regression.py +++ b/examples/undocumented/python/transfer_multitask_clustered_logistic_regression.py @@ -11,9 +11,9 @@ parameter_list = [[traindat,testdat,label_traindat]] def transfer_multitask_clustered_logistic_regression (fm_train=traindat,fm_test=testdat,label_train=label_traindat): - from modshogun import BinaryLabels, RealFeatures, Task, TaskGroup, MSG_DEBUG + from shogun import BinaryLabels, RealFeatures, Task, TaskGroup, MSG_DEBUG try: - from modshogun import MultitaskClusteredLogisticRegression + from shogun import MultitaskClusteredLogisticRegression except ImportError: print("MultitaskClusteredLogisticRegression not available") exit() diff --git a/examples/undocumented/python_modular/transfer_multitask_l12_logistic_regression.py b/examples/undocumented/python/transfer_multitask_l12_logistic_regression.py similarity index 91% rename from examples/undocumented/python_modular/transfer_multitask_l12_logistic_regression.py rename to examples/undocumented/python/transfer_multitask_l12_logistic_regression.py index 56c7c84e57b..22c47c0c2f8 100644 --- a/examples/undocumented/python_modular/transfer_multitask_l12_logistic_regression.py +++ b/examples/undocumented/python/transfer_multitask_l12_logistic_regression.py @@ -11,9 +11,9 @@ parameter_list = [[traindat,testdat,label_traindat]] def transfer_multitask_l12_logistic_regression (fm_train=traindat,fm_test=testdat,label_train=label_traindat): - from modshogun import BinaryLabels, RealFeatures, Task, TaskGroup + from shogun import BinaryLabels, RealFeatures, Task, TaskGroup try: - from modshogun import MultitaskL12LogisticRegression + from shogun import MultitaskL12LogisticRegression except ImportError: print("MultitaskL12LogisticRegression not available") exit(0) diff --git a/examples/undocumented/python_modular/transfer_multitask_leastsquares_regression.py b/examples/undocumented/python/transfer_multitask_leastsquares_regression.py similarity index 90% rename from examples/undocumented/python_modular/transfer_multitask_leastsquares_regression.py rename to examples/undocumented/python/transfer_multitask_leastsquares_regression.py index 8ec07e61d5c..1b1357d96a5 100644 --- a/examples/undocumented/python_modular/transfer_multitask_leastsquares_regression.py +++ b/examples/undocumented/python/transfer_multitask_leastsquares_regression.py @@ -11,9 +11,9 @@ parameter_list = [[traindat,testdat,label_traindat]] def transfer_multitask_leastsquares_regression (fm_train=traindat,fm_test=testdat,label_train=label_traindat): - from modshogun import RegressionLabels, RealFeatures, Task, TaskGroup + from shogun import RegressionLabels, RealFeatures, Task, TaskGroup try: - from modshogun import MultitaskLeastSquaresRegression + from shogun import MultitaskLeastSquaresRegression except ImportError: print("MultitaskLeastSquaresRegression not available") exit(0) diff --git a/examples/undocumented/python_modular/transfer_multitask_logistic_regression.py b/examples/undocumented/python/transfer_multitask_logistic_regression.py similarity index 91% rename from examples/undocumented/python_modular/transfer_multitask_logistic_regression.py rename to examples/undocumented/python/transfer_multitask_logistic_regression.py index d475cb7757f..5c97b6f6a2d 100644 --- a/examples/undocumented/python_modular/transfer_multitask_logistic_regression.py +++ b/examples/undocumented/python/transfer_multitask_logistic_regression.py @@ -11,9 +11,9 @@ parameter_list = [[traindat,testdat,label_traindat]] def transfer_multitask_logistic_regression (fm_train=traindat,fm_test=testdat,label_train=label_traindat): - from modshogun import BinaryLabels, RealFeatures, Task, TaskGroup + from shogun import BinaryLabels, RealFeatures, Task, TaskGroup try: - from modshogun import MultitaskLogisticRegression + from shogun import MultitaskLogisticRegression except ImportError: print("MultitaskLogisticRegression not available") exit() diff --git a/examples/undocumented/python_modular/transfer_multitask_trace_logistic_regression.py b/examples/undocumented/python/transfer_multitask_trace_logistic_regression.py similarity index 91% rename from examples/undocumented/python_modular/transfer_multitask_trace_logistic_regression.py rename to examples/undocumented/python/transfer_multitask_trace_logistic_regression.py index a128a969cbf..c72912dd63a 100644 --- a/examples/undocumented/python_modular/transfer_multitask_trace_logistic_regression.py +++ b/examples/undocumented/python/transfer_multitask_trace_logistic_regression.py @@ -11,9 +11,9 @@ parameter_list = [[traindat,testdat,label_traindat]] def transfer_multitask_trace_logistic_regression (fm_train=traindat,fm_test=testdat,label_train=label_traindat): - from modshogun import BinaryLabels, RealFeatures, Task, TaskGroup + from shogun import BinaryLabels, RealFeatures, Task, TaskGroup try: - from modshogun import MultitaskTraceLogisticRegression + from shogun import MultitaskTraceLogisticRegression except ImportError: print("MultitaskTraceLogisticRegression not available") exit(0) diff --git a/examples/undocumented/python_modular/variational_classifier_modular.py b/examples/undocumented/python/variational_classifier.py similarity index 94% rename from examples/undocumented/python_modular/variational_classifier_modular.py rename to examples/undocumented/python/variational_classifier.py index c922674c35c..7461d660b24 100644 --- a/examples/undocumented/python_modular/variational_classifier_modular.py +++ b/examples/undocumented/python/variational_classifier.py @@ -37,12 +37,12 @@ label_binary_traindat = '%s/label_train_twoclass.dat'%path try: - from modshogun import GaussianProcessClassification + from shogun import GaussianProcessClassification except ImportError: print("GaussianProcessClassification is not available") exit(0) -from modshogun import * +from shogun import * parameter_list=[ [KLCholeskyInferenceMethod,traindat,testdat,label_binary_traindat,0,0,1e-5,1e-2,0], [KLCovarianceInferenceMethod,traindat,testdat,label_binary_traindat,0,0,1e-5,1e-2,0], @@ -50,7 +50,7 @@ [KLDualInferenceMethod,traindat,testdat,label_binary_traindat,0,0,1e-5,1e-2,0], [SingleLaplaceInferenceMethod,traindat,testdat,label_binary_traindat,0,0], ] -def variational_classifier_modular(kl_inference,train_fname=traindat,test_fname=testdat, +def variational_classifier(kl_inference,train_fname=traindat,test_fname=testdat, label_fname=label_binary_traindat,kernel_log_sigma=0,kernel_log_scale=0,noise_factor=1e-5, min_coeff_kernel=1e-2,max_attempt=0): from math import exp @@ -84,4 +84,4 @@ def variational_classifier_modular(kl_inference,train_fname=traindat,test_fname= if __name__=="__main__": print("variational_classifier") for parameter in parameter_list: - variational_classifier_modular(*parameter) + variational_classifier(*parameter) diff --git a/examples/undocumented/python_modular/CMakeLists.txt b/examples/undocumented/python_modular/CMakeLists.txt deleted file mode 100644 index a2250418693..00000000000 --- a/examples/undocumented/python_modular/CMakeLists.txt +++ /dev/null @@ -1,13 +0,0 @@ -INCLUDE(PythonEnvironment) -GET_PYTHON_ENV() - -FILE(GLOB PYTHON_EXAMPLES *.py) -FOREACH(EXAMPLE ${PYTHON_EXAMPLES}) - STRING(REGEX REPLACE ".*/(.*).py" "\\1" EXAMPLE_NAME ${EXAMPLE}) - add_test(NAME python_modular-${EXAMPLE_NAME} - WORKING_DIRECTORY ${CMAKE_CURRENT_SOURCE_DIR} - COMMAND ${PYTHON_EXECUTABLE} ${EXAMPLE}) - set_property(TEST python_modular-${EXAMPLE_NAME} PROPERTY - ENVIRONMENT "${PYTHON_ENV_VARS}") - -ENDFOREACH() \ No newline at end of file diff --git a/examples/undocumented/python_modular/features_io_modular.py b/examples/undocumented/python_modular/features_io_modular.py deleted file mode 100644 index 0b354c0f4c1..00000000000 --- a/examples/undocumented/python_modular/features_io_modular.py +++ /dev/null @@ -1,84 +0,0 @@ -#!/usr/bin/env python -from tools.load import LoadMatrix -lm=LoadMatrix() -data=lm.load_numbers('../data/fm_train_real.dat') -label=lm.load_numbers('../data/label_train_twoclass.dat') - -parameter_list=[[data,label]] - -def features_io_modular (fm_train_real, label_train_twoclass): - import numpy - from modshogun import SparseRealFeatures, RealFeatures, MulticlassLabels - from modshogun import GaussianKernel - from modshogun import LibSVMFile, CSVFile, BinaryFile, HDF5File - - feats=SparseRealFeatures(fm_train_real) - feats2=SparseRealFeatures() - - f=BinaryFile("tmp/fm_train_sparsereal.bin","w") - feats.save(f) - - f=LibSVMFile("tmp/fm_train_sparsereal.ascii","w") - feats.save(f) - - f=BinaryFile("tmp/fm_train_sparsereal.bin") - feats2.load(f) - - f=LibSVMFile("tmp/fm_train_sparsereal.ascii") - feats2.load(f) - - feats=RealFeatures(fm_train_real) - feats2=RealFeatures() - - f=BinaryFile("tmp/fm_train_real.bin","w") - feats.save(f) - - f=HDF5File("tmp/fm_train_real.h5","w", "/data/doubles") - feats.save(f) - - f=CSVFile("tmp/fm_train_real.ascii","w") - feats.save(f) - - f=BinaryFile("tmp/fm_train_real.bin") - feats2.load(f) - #print("diff binary", numpy.max(numpy.abs(feats2.get_feature_matrix().flatten()-fm_train_real.flatten()))) - - f=CSVFile("tmp/fm_train_real.ascii") - feats2.load(f) - #print("diff ascii", numpy.max(numpy.abs(feats2.get_feature_matrix().flatten()-fm_train_real.flatten()))) - - lab=MulticlassLabels(numpy.array([0.0,1.0,2.0,3.0])) - lab2=MulticlassLabels() - f=CSVFile("tmp/label_train_twoclass.ascii","w") - lab.save(f) - - f=BinaryFile("tmp/label_train_twoclass.bin","w") - lab.save(f) - - f=HDF5File("tmp/label_train_real.h5","w", "/data/labels") - lab.save(f) - - f=CSVFile("tmp/label_train_twoclass.ascii") - lab2.load(f) - - f=BinaryFile("tmp/label_train_twoclass.bin") - lab2.load(f) - - f=HDF5File("tmp/fm_train_real.h5","r", "/data/doubles") - feats2.load(f) - #print(feats2.get_feature_matrix()) - f=HDF5File("tmp/label_train_real.h5","r", "/data/labels") - lab2.load(f) - #print(lab2.get_labels()) - - #clean up - import os - for f in ['tmp/fm_train_sparsereal.bin','tmp/fm_train_sparsereal.ascii', - 'tmp/fm_train_real.bin','tmp/fm_train_real.h5','tmp/fm_train_real.ascii', - 'tmp/label_train_real.h5', 'tmp/label_train_twoclass.ascii','tmp/label_train_twoclass.bin']: - os.unlink(f) - return feats, feats2, lab, lab2 - -if __name__=='__main__': - print('Features IO') - features_io_modular(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/features_read_svmlight_format_modular.py b/examples/undocumented/python_modular/features_read_svmlight_format_modular.py deleted file mode 100644 index 5a3d2b8283b..00000000000 --- a/examples/undocumented/python_modular/features_read_svmlight_format_modular.py +++ /dev/null @@ -1,16 +0,0 @@ -#!/usr/bin/env python -parameter_list=[['../data/train_sparsereal.light']] - -def features_read_svmlight_format_modular (fname): - import os - from modshogun import SparseRealFeatures - from modshogun import LibSVMFile - - f=SparseRealFeatures() - lab=f.load_with_labels(LibSVMFile(fname)) - f.save_with_labels(LibSVMFile('tmp/testwrite.light', 'w'), lab) - os.unlink('tmp/testwrite.light') - -if __name__=='__main__': - print('Reading SVMLIGHT format') - features_read_svmlight_format_modular(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/features_string_file_char_modular.py b/examples/undocumented/python_modular/features_string_file_char_modular.py deleted file mode 100644 index 4bedccf6e50..00000000000 --- a/examples/undocumented/python_modular/features_string_file_char_modular.py +++ /dev/null @@ -1,12 +0,0 @@ -#!/usr/bin/env python -parameter_list = [['features_string_file_char_modular.py']] - -def features_string_file_char_modular (fname): - from modshogun import StringFileCharFeatures, RAWBYTE - f = StringFileCharFeatures(fname, RAWBYTE) - #print("strings", f.get_features()) - return f - -if __name__=='__main__': - print('Compressing StringCharFileFeatures') - features_string_file_char_modular(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/serialization_matrix_modular.py b/examples/undocumented/python_modular/serialization_matrix_modular.py deleted file mode 100644 index 1d0f6948ff8..00000000000 --- a/examples/undocumented/python_modular/serialization_matrix_modular.py +++ /dev/null @@ -1,23 +0,0 @@ -#!/usr/bin/env python -from modshogun import * -from numpy import array -import os - -parameter_list=[[[[1.0,2,3],[4,5,6]]]] - -def serialization_matrix_modular (m): - feats=RealFeatures(array(m)) - #feats.io.set_loglevel(0) - fstream = SerializableAsciiFile("tmp/foo.asc", "w") - feats.save_serializable(fstream) - - l=MulticlassLabels(array([1.0,2,3])) - fstream = SerializableAsciiFile("tmp/foo2.asc", "w") - l.save_serializable(fstream) - - os.unlink("tmp/foo.asc") - os.unlink("tmp/foo2.asc") - -if __name__=='__main__': - print('Serialization Matrix Modular') - serialization_matrix_modular(*parameter_list[0]) diff --git a/examples/undocumented/python_modular/tmp/.gitkeep b/examples/undocumented/python_modular/tmp/.gitkeep deleted file mode 100644 index e69de29bb2d..00000000000 diff --git a/scripts/check_format.sh b/scripts/check_format.sh new file mode 100755 index 00000000000..f1d4c0cec9b --- /dev/null +++ b/scripts/check_format.sh @@ -0,0 +1,76 @@ +#!/usr/bin/env bash +# This script checks code style by using clang-format +# on a git diff made between a base branch and a test branch, +# which is the one we want to check. +# +# This script was originally inspired by: +# https://github.com/root-project/root/blob/master/.travis.yml + +# Unofficial strict bash mode +# See: http://redsymbol.net/articles/unofficial-bash-strict-mode/ +set -euo pipefail +IFS=$'\n\t' + +function check_shogun_style { + + # Prevent cases in which we are testing branches + # different from develop + if [ ! `git branch --list ${2:-}` ] + then + echo "Branch ${2:-} does not exists locally. Fetching it." + git fetch origin "${2:-}:${2:-}" + + fi + + BASE_COMMIT=$(git rev-parse ${2:-}) + + + echo "-----" + echo "Shogun Style Checker" + echo "-----" + echo "Running clang-format-3.8 against branch ${2:-}, with hash $BASE_COMMIT" + + COMMIT_FILES=$(git diff --name-only $BASE_COMMIT) + + # Use clang-format only on existent files + LIST=("") + for file in $COMMIT_FILES + do + if [ -f $file ]; then + LIST=("${LIST[@]}" "$file") + fi + done + + RESULT_OUTPUT="$(git clang-format-3.8 --commit $BASE_COMMIT --diff --binary `which clang-format-3.8` $LIST)" + + if [ "$RESULT_OUTPUT" == "no modified files to format" ] \ + || [ "$RESULT_OUTPUT" == "clang-format-3.8 did not modify any files" ] \ + || [ "$RESULT_OUTPUT" == "clang-format did not modify any files" ]; then + echo "clang-format-3.8 passed. \o/" + echo "-----" + exit 0 + else + echo "-----" + echo "clang-format failed." + echo "To reproduce it locally please run: " + echo -e "\t1) git checkout ${1:-}" + echo -e "\t2) git clang-format-3.8 --commit $BASE_COMMIT --diff --binary $(which clang-format-3.8)" + echo "To fix the errors automatically please run: " + echo -e "\t1) git checkout ${1:-}" + echo -e "\t2) git clang-format-3.8 --commit $BASE_COMMIT --binary $(which clang-format-3.8)" + echo "-----" + echo "Style errors found:" + echo "$RESULT_OUTPUT" + exit 1 + fi +} + +# Check only if we have enough arguments +if [[ $# -ne 2 ]]; then + echo "Wrong number of parameters supplied!" + echo "Usage: ./check_format.sh " + exit 1 +fi + +# Run the check +check_shogun_style ${1:-} ${2:-} diff --git a/setup.py b/setup.py index ae66c06651e..389f78e3f64 100644 --- a/setup.py +++ b/setup.py @@ -141,7 +141,7 @@ def shogun_cmake(arguments=None): print("Running CMake") if arguments is None: - arguments='-DPythonModular=ON -DENABLE_TESTING=OFF -DCMAKE_INSTALL_PREFIX=install' + arguments='-DINTERFACE_PYTHON=ON -DENABLE_TESTING=OFF -DCMAKE_INSTALL_PREFIX=install' if distutils.spawn.find_executable('cmake') is not None: print('CMake arguments: %s ' % arguments) @@ -243,12 +243,12 @@ def shogun_package_directories(): def shogun_data_files(): data_files = list() libshogun_files = glob.glob(os.path.join(shogun_generated_install, 'lib/libshogun*')) - modshogun_so_destination = os.path.join('lib', python_package_path(shogun_python_packages_location)) - modshogun_so_file = glob.glob(os.path.join(shogun_python_packages_location, '_modshogun.so'))[0] + shogun_so_destination = os.path.join('lib', python_package_path(shogun_python_packages_location)) + shogun_so_file = glob.glob(os.path.join(shogun_python_packages_location, '_shogun.so'))[0] # appending data files data_files.append(('lib', libshogun_files)) - data_files.append((modshogun_so_destination, [modshogun_so_file])) + data_files.append((shogun_so_destination, [shogun_so_file])) if show_debug_information: print('Shogun Python package data files:') @@ -335,7 +335,7 @@ def run(self): # Shogun package content packages = shogun_packages(), package_dir = shogun_package_directories(), - py_modules =['modshogun'], + py_modules =['shogun'], data_files = shogun_data_files(), # Shogun dependencies diff --git a/src/.r-install.sh b/src/.r-install.sh index 2c3f97912bf..887deb61fcd 100755 --- a/src/.r-install.sh +++ b/src/.r-install.sh @@ -129,7 +129,7 @@ else echo "Installing modular shogun interface for R" cat >"$1/$2/NAMESPACE" <"$1/$2/R/$2" < map = 1; +} + +/** + * A mapping from a 64-bit integer + * to a string. + */ +message Int64ToStringMap { + map map = 1; +} + +/** + * A mapping from a string + * to a double-precision floating point number. + */ +message StringToDoubleMap { + map map = 1; +} + +/** + * A mapping from a 64-bit integer + * to a double-precision floating point number. + */ +message Int64ToDoubleMap { + map map = 1; +} + +/** + * A vector of strings. + */ +message StringVector { + repeated string vector = 1; +} + +/** + * A vector of 64-bit integers. + */ +message Int64Vector { + repeated int64 vector = 1; +} + +/** + * A vector of double-precision floating point numbers. + */ +message DoubleVector { + repeated double vector = 1; +} diff --git a/src/interfaces/coreml/DictVectorizer.proto b/src/interfaces/coreml/DictVectorizer.proto new file mode 100644 index 00000000000..662f97b8cec --- /dev/null +++ b/src/interfaces/coreml/DictVectorizer.proto @@ -0,0 +1,36 @@ +// Copyright (c) 2017, Apple Inc. All rights reserved. +// +// Use of this source code is governed by a BSD-3-clause license that can be +// found in LICENSE.txt or at https://opensource.org/licenses/BSD-3-Clause + +syntax = "proto3"; +option optimize_for = LITE_RUNTIME; + +import public "DataStructures.proto"; + +package CoreML.Specification; + +/** + * Uses an index mapping to convert a dictionary to an array. + * + * The output array will be equal in length to the index mapping vector parameter. + * All keys in the input dictionary must be present in the index mapping vector. + * + * For each item in the input dictionary, insert its value in the ouput array. + * The position of the insertion is determined by the position of the item's key + * in the index mapping. Any keys not present in the input dictionary, will be + * zero in the output array. + * + * For example: if the ``stringToIndex`` paramter is set to ``["a", "c", "b", "z"]``, + * then an input of ``{"a": 4, "c": 8}`` will produce an output of ``[4, 8, 0, 0]``. + * + */ +message DictVectorizer { + oneof Map { + /// String keys to indexes + StringVector stringToIndex = 1; + + /// Int keys to indexes + Int64Vector int64ToIndex = 2; + } +} diff --git a/src/interfaces/coreml/FeatureTypes.proto b/src/interfaces/coreml/FeatureTypes.proto new file mode 100644 index 00000000000..50bad311020 --- /dev/null +++ b/src/interfaces/coreml/FeatureTypes.proto @@ -0,0 +1,87 @@ +// Copyright (c) 2017, Apple Inc. All rights reserved. +// +// Use of this source code is governed by a BSD-3-clause license that can be +// found in LICENSE.txt or at https://opensource.org/licenses/BSD-3-Clause + +syntax = "proto3"; +option optimize_for = LITE_RUNTIME; + +package CoreML.Specification; + +/** + * The 64-bit integer feature type. + */ +message Int64FeatureType {} + +/** + * The double-precision floating point number feature type. + */ +message DoubleFeatureType {} + +/** + * The string feature type. + */ +message StringFeatureType {} + +/** + * The image feature type. + */ +message ImageFeatureType { + // Assumes raw (decompressed) format + enum ColorSpace { + INVALID_COLOR_SPACE = 0; + GRAYSCALE = 10; // 8 bit pixel: 0=black, 255=white + RGB = 20; // 32 bit pixel: RGBA with A channel ignored + BGR = 30; // 32 bit pixel: BGRA with A channel ignored + } + + uint64 width = 1; + uint64 height = 2; + ColorSpace colorSpace = 3; +} + +/** + * The array feature type. + */ +message ArrayFeatureType { + enum ArrayDataType { + INVALID_ARRAY_DATA_TYPE = 0; + FLOAT32 = 65568; // 0x10000 | 32 + DOUBLE = 65600; // 0x10000 | 64 + INT32 = 131104; // 0x20000 | 32 + } + + repeated uint64 shape = 1; /// For neural networks, must be of length 1 or 3, representing input shape [C] or [C,H,W], respectively. + ArrayDataType dataType = 2; +} + +/** + * The dictionary feature type. + */ +message DictionaryFeatureType { + /** + * Key/value type tags, with the following restrictions: + * - ``keyType`` must be a hashable type + * - ``valueType`` is assumed to be a ``double`` + */ + oneof KeyType { + Int64FeatureType int64KeyType = 1; + StringFeatureType stringKeyType = 2; + } +} + +/** + * A feature, which may be optional. + */ +message FeatureType { + oneof Type { + Int64FeatureType int64Type = 1; + DoubleFeatureType doubleType = 2; + StringFeatureType stringType = 3; + ImageFeatureType imageType = 4; + ArrayFeatureType multiArrayType = 5; + DictionaryFeatureType dictionaryType = 6; + } + + bool isOptional = 1000; +} diff --git a/src/interfaces/coreml/FeatureVectorizer.proto b/src/interfaces/coreml/FeatureVectorizer.proto new file mode 100644 index 00000000000..75eaf14b536 --- /dev/null +++ b/src/interfaces/coreml/FeatureVectorizer.proto @@ -0,0 +1,26 @@ +// Copyright (c) 2017, Apple Inc. All rights reserved. +// +// Use of this source code is governed by a BSD-3-clause license that can be +// found in LICENSE.txt or at https://opensource.org/licenses/BSD-3-Clause + +syntax = "proto3"; +option optimize_for = LITE_RUNTIME; + +package CoreML.Specification; + +/** + * A FeatureVectorizer puts one or more features into a single array. + * + * The ordering of features in the output array is determined by + * ``inputList``. + * + * ``inputDimensions`` is a zero based index. + */ +message FeatureVectorizer { + message InputColumn { + string inputColumn = 1; + uint64 inputDimensions = 2; + } + + repeated InputColumn inputList = 1; +} diff --git a/src/interfaces/coreml/GLMClassifier.proto b/src/interfaces/coreml/GLMClassifier.proto new file mode 100644 index 00000000000..47f6f4a3c7b --- /dev/null +++ b/src/interfaces/coreml/GLMClassifier.proto @@ -0,0 +1,43 @@ +// Copyright (c) 2017, Apple Inc. All rights reserved. +// +// Use of this source code is governed by a BSD-3-clause license that can be +// found in LICENSE.txt or at https://opensource.org/licenses/BSD-3-Clause + +syntax = "proto3"; +option optimize_for = LITE_RUNTIME; + +import public "DataStructures.proto"; + +package CoreML.Specification; + +/** + * A generalized linear model classifier. + */ +message GLMClassifier { + message DoubleArray { + repeated double value = 1; + } + + enum PostEvaluationTransform { + Logit = 0; + Probit = 1; /// Only binary classification is supported for probit + } + + enum ClassEncoding { + ReferenceClass = 0; /// First class is the reference class + OneVsRest = 1; /// Also called One vs All + } + + repeated DoubleArray weights = 1; + repeated double offset = 2; + PostEvaluationTransform postEvaluationTransform = 3; + ClassEncoding classEncoding = 4; + + /** + * Required class label mapping. + */ + oneof ClassLabels { + StringVector stringClassLabels = 100; + Int64Vector int64ClassLabels = 101; + } +} diff --git a/src/interfaces/coreml/GLMRegressor.proto b/src/interfaces/coreml/GLMRegressor.proto new file mode 100644 index 00000000000..64093c4f156 --- /dev/null +++ b/src/interfaces/coreml/GLMRegressor.proto @@ -0,0 +1,28 @@ +// Copyright (c) 2017, Apple Inc. All rights reserved. +// +// Use of this source code is governed by a BSD-3-clause license that can be +// found in LICENSE.txt or at https://opensource.org/licenses/BSD-3-Clause + +syntax = "proto3"; +option optimize_for = LITE_RUNTIME; + +package CoreML.Specification; + +/** + * A generalized linear model regressor. + */ +message GLMRegressor { + message DoubleArray { + repeated double value = 1; + } + + enum PostEvaluationTransform { + NoTransform = 0; + Logit = 1; + Probit = 2; + } + + repeated DoubleArray weights = 1; + repeated double offset = 2; + PostEvaluationTransform postEvaluationTransform = 3; +} diff --git a/src/interfaces/coreml/Identity.proto b/src/interfaces/coreml/Identity.proto new file mode 100644 index 00000000000..123a15e5915 --- /dev/null +++ b/src/interfaces/coreml/Identity.proto @@ -0,0 +1,18 @@ +// Copyright (c) 2017, Apple Inc. All rights reserved. +// +// Use of this source code is governed by a BSD-3-clause license that can be +// found in LICENSE.txt or at https://opensource.org/licenses/BSD-3-Clause + +syntax = "proto3"; +option optimize_for = LITE_RUNTIME; + +package CoreML.Specification; + +/** + * An identity model. + * + * This model returns given inputs as outputs, unchanged. + * Intended to be used for testing purposes. + */ +message Identity { +} diff --git a/src/interfaces/coreml/Imputer.proto b/src/interfaces/coreml/Imputer.proto new file mode 100644 index 00000000000..3de280b2f16 --- /dev/null +++ b/src/interfaces/coreml/Imputer.proto @@ -0,0 +1,43 @@ +// Copyright (c) 2017, Apple Inc. All rights reserved. +// +// Use of this source code is governed by a BSD-3-clause license that can be +// found in LICENSE.txt or at https://opensource.org/licenses/BSD-3-Clause + +syntax = "proto3"; +option optimize_for = LITE_RUNTIME; + +import public "DataStructures.proto"; + +package CoreML.Specification; + +/** + * A transformer that replaces missing values with a default value, + * such as a statistically-derived value. + * + * If ``ReplaceValue`` is set, then missing values of that type are + * replaced with the corresponding value. + * + * For example: if ``replaceDoubleValue`` is set to ``NaN`` + * and a single ``NaN`` double value is provided as input, + * then it is replaced by ``imputedDoubleValue``. However + * if the input is an array of doubles, then any instances + * of ``NaN`` in the array is replaced with the corresponding + * value in ``imputedDoubleArray``. + */ +message Imputer { + oneof ImputedValue { + double imputedDoubleValue = 1; + int64 imputedInt64Value = 2; + string imputedStringValue = 3; + DoubleVector imputedDoubleArray = 4; + Int64Vector imputedInt64Array = 5; + StringToDoubleMap imputedStringDictionary = 6; + Int64ToDoubleMap imputedInt64Dictionary = 7; + } + + oneof ReplaceValue { + double replaceDoubleValue = 11; + int64 replaceInt64Value = 12; + string replaceStringValue = 13; + } +} diff --git a/src/interfaces/coreml/LICENSE.txt b/src/interfaces/coreml/LICENSE.txt new file mode 100644 index 00000000000..bbcdc9ef8f3 --- /dev/null +++ b/src/interfaces/coreml/LICENSE.txt @@ -0,0 +1,11 @@ +Copyright (c) 2017, Apple Inc. All rights reserved. + +Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions are met: + +1. Redistributions of source code must retain the above copyright notice, this list of conditions and the following disclaimer. + +2. Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. + +3. Neither the name of the copyright holder(s) nor the names of any contributors may be used to endorse or promote products derived from this software without specific prior written permission. + +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. diff --git a/src/interfaces/coreml/Model.proto b/src/interfaces/coreml/Model.proto new file mode 100644 index 00000000000..b903d983379 --- /dev/null +++ b/src/interfaces/coreml/Model.proto @@ -0,0 +1,190 @@ +// Copyright (c) 2017, Apple Inc. All rights reserved. +// +// Use of this source code is governed by a BSD-3-clause license that can be +// found in LICENSE.txt or at https://opensource.org/licenses/BSD-3-Clause + +/** + * A Core ML model consists of a specification version + * and a model description, + * and can be any one of the following types: + * + * Pipelines + * - ``PipelineClassifier`` + * - ``PipelineRegressor`` + * - ``Pipeline`` + * + * Regressors + * - ``GLMRegressor`` + * - ``SupportVectorRegressor`` + * - ``TreeEnsembleRegressor`` + * - ``NeuralNetworkRegressor`` + * + * Classifiers + * - ``GLMClassifier`` + * - ``SupportVectorClassifier`` + * - ``TreeEnsembleClassifier`` + * - ``NeuralNetworkClassifier`` + * + * Neural Networks + * - ``NeuralNetwork`` + * + * Feature Engineering + * - ``OneHotEncoder`` + * - ``Imputer`` + * - ``FeatureVectorizer`` + * - ``DictVectorizer`` + * - ``Scaler`` + * - ``CategoricalMapping`` + * - ``Normalizer`` + * - ``ArrayFeatureExtractor`` + * + * Simple Mathematical Functions + * - ``Identity`` + */ + +syntax = "proto3"; +option optimize_for = LITE_RUNTIME; + +import public "ArrayFeatureExtractor.proto"; +import public "CategoricalMapping.proto"; +import public "DictVectorizer.proto"; +import public "FeatureTypes.proto"; +import public "FeatureVectorizer.proto"; +import public "GLMRegressor.proto"; +import public "GLMClassifier.proto"; +import public "Identity.proto"; +import public "Imputer.proto"; +import public "NeuralNetwork.proto"; +import public "Normalizer.proto"; +import public "OneHotEncoder.proto"; +import public "Scaler.proto"; +import public "SVM.proto"; +import public "TreeEnsemble.proto"; + +package CoreML.Specification; + +/** + * A pipeline consisting of one or more models. + */ +message Pipeline { + repeated Model models = 1; +} + +/** + * A classifier pipeline. + */ +message PipelineClassifier { + Pipeline pipeline = 1; + + // Required class label mapping + oneof ClassLabels { + StringVector stringClassLabels = 100; + Int64Vector int64ClassLabels = 101; + } +} + +/** + * A regressor pipeline. + */ +message PipelineRegressor { + Pipeline pipeline = 1; +} + +/** + * A feature description, + * consisting of a name, short description, and type. + */ +message FeatureDescription { + string name = 1; + string shortDescription = 2; + FeatureType type = 3; +} + +/** + * Model metadata, + * consisting of a short description, a version string, + * an author, a license, and any other user defined + * key/value meta data. + */ +message Metadata { + string shortDescription = 1; + string versionString = 2; + string author = 3; + string license = 4; + map userDefined = 100; +} + +/** + * A description of a model, + * consisting of descriptions of its input and output features. + * Both regressor and classifier models require the name of the + * primary predicted output feature (``predictedFeatureName``). + * Classifier models can specify the output feature containing + * probabilities for the predicted classes + * (``predictedProbabilitiesName``). + */ +message ModelDescription { + repeated FeatureDescription input = 1; + repeated FeatureDescription output = 10; + + string predictedFeatureName = 11; + string predictedProbabilitiesName = 12; + + Metadata metadata = 100; +} + +/** + * A Core ML model, + * consisting of a specification version, + * a model description, and a model type. + * + * Core ML model compatibility is indicated by + * a monotonically increasing specification version number, + * which is incremented anytime a backward-incompatible change is made + * (this is functionally equivalent to the MAJOR version number + * described by `Semantic Versioning 2.0.0 `_). + * The Core ML framework in macOS currently supports specification version `1`. + */ +message Model { + int32 specificationVersion = 1; + ModelDescription description = 2; + + // start at 200 here + // model specific parameters: + oneof Type { + // pipeline starts at 200 + PipelineClassifier pipelineClassifier = 200; + PipelineRegressor pipelineRegressor = 201; + Pipeline pipeline = 202; + + // regressors start at 300 + GLMRegressor glmRegressor = 300; + SupportVectorRegressor supportVectorRegressor = 301; + TreeEnsembleRegressor treeEnsembleRegressor = 302; + NeuralNetworkRegressor neuralNetworkRegressor = 303; + + // classifiers start at 400 + GLMClassifier glmClassifier = 400; + SupportVectorClassifier supportVectorClassifier = 401; + TreeEnsembleClassifier treeEnsembleClassifier = 402; + NeuralNetworkClassifier neuralNetworkClassifier = 403; + + // generic models start at 500 + NeuralNetwork neuralNetwork = 500; + + // feature engineering starts at 600 + OneHotEncoder oneHotEncoder = 600; + Imputer imputer = 601; + FeatureVectorizer featureVectorizer = 602; + DictVectorizer dictVectorizer = 603; + Scaler scaler = 604; + CategoricalMapping categoricalMapping = 606; + Normalizer normalizer = 607; + ArrayFeatureExtractor arrayFeatureExtractor = 609; + + // simple mathematical functions used for testing start at 900 + Identity identity = 900; + + // reserved until 1000 + } +} diff --git a/src/interfaces/coreml/NeuralNetwork.proto b/src/interfaces/coreml/NeuralNetwork.proto new file mode 100644 index 00000000000..fd367f16a58 --- /dev/null +++ b/src/interfaces/coreml/NeuralNetwork.proto @@ -0,0 +1,2032 @@ +// Copyright (c) 2017, Apple Inc. All rights reserved. +// +// Use of this source code is governed by a BSD-3-clause license that can be +// found in LICENSE.txt or at https://opensource.org/licenses/BSD-3-Clause + +syntax = "proto3"; +option optimize_for = LITE_RUNTIME; + +import public "DataStructures.proto"; + +package CoreML.Specification; + +/** + * A neural network is defined through a collection of layers + * and represents a directed acyclic graph (DAG). + * Each layer has a name, a layer type, + * a list of input names, a list of output names, + * and a collection of parameters specific to the layer type. + * + * The graph structure and connectivity of the neural network + * is inferred from the input and output names. + * A neural network starts with the layer + * whose input name is equal to the value specified in + * ``Model.description.input.name``, + * and ends with the layer + * whose output name is equal to the value specified in + * ``Model.description.output.name``. + * Layers must have unique input and output names, + * and a layer may not have input or output names that + * refer to layers that are not yet defined. + * + * CoreML supports sequential data that can be 1- or 3-dimensional. + * 3-dimensional data typically represents an image feature map, + * whose shape is denoted by ``[C, H, W]``, + * which corresponds to the channel, height, and width, respectively. + * 1-dimensional data is a set of features + * whose shape is denoted by ``[C]``, + * and is equivalent to 3-dimensional data + * with the shape ``[C, 1, 1]``. + * + * For the purposes of this specification, + * batch dimension is ignored. + * Thus, a sequence of 3-dimensional data + * is to be understood as a 4-dimensional array, + * whose shape is denoted by ``[Seq_length, C, H, W]``, + * and a sequence of 1-dimensional data + * is to be understood as a 2-dimensional array, + * whose shape is denoted by ``[Seq_length, C]``, + * which is equivalent to a 4-dimensional array + * with the shape ``[Seq_length, C, 1, 1]``. This axes order is important to + * remember while setting parameters for layers such as "reshape" and "permute". + * + * + * At runtime, all data blobs are internally represented + * as 5-dimensional blobs + * with the shape ``[Seq_length, Batch, C, H, W]``. + * + * A layer may process input data differently if operating over a sequence; + * details of this behavior is documented in the layer's message. + * Otherwise, sequential data is processed like a batch --- + * that is, the sequence of inputs are processed independently and in parallel. + * + * The network input shape specified by ``Model.description.input.type`` + * must be compatible with the expected input shape + * of the network input layer, i.e. the last dimension is the fastest moving one. + * + * All data blobs, as well as weight parameters, + * are stored using row-major ordering, i.e. the last dimension is the fastest moving one. + */ +message NeuralNetwork { + repeated NeuralNetworkLayer layers = 1; + repeated NeuralNetworkPreprocessing preprocessing = 2; +} + +/// Preprocessing +/// ------------- + +/** + * A neural network preprocessor that + * performs a scalar multiplication of an image + * followed by addition of scalar biases to the channels. + * + * Input: X + * An image in BGR or RGB format with shape ``[3, H, W]`` + * or in grayscale format with shape ``[1, H, W]``. + * Output: Y + * An image with format and shape corresponding to the input. + * + * If the input image is in BGR format: + * :: + * Y[0, :, :] = channelScale * X[0, :, :] + blueBias + * Y[1, :, :] = channelScale * X[1, :, :] + greenBias + * Y[2, :, :] = channelScale * X[2, :, :] + redBias + * + * If the input image is in RGB format: + * :: + * Y[0, :, :] = channelScale * X[0, :, :] + redBias + * Y[1, :, :] = channelScale * X[1, :, :] + greenBias + * Y[2, :, :] = channelScale * X[2, :, :] + blueBias + * + * If the input image is in grayscale format: + * :: + * Y[0, :, :] = channelScale * X[0, :, :] + grayBias + */ +message NeuralNetworkImageScaler { + float channelScale = 10; ///Scalar to be multiplied. + float blueBias = 20; ///Scalar blue bias to be added. + float greenBias = 21; ///Scalar green bias to be added. + float redBias = 22; ///Scalar red bias to be added. + float grayBias = 30; ///Scalar bias to be added for grayscale images. +} + +/** + * A neural network preprocessor that + * subtracts the provided mean image from the input image. + * The mean image is subtracted from the input named + * ``NeuralNetworkPreprocessing.featureName``. + */ +message NeuralNetworkMeanImage { + /** + * Mean image stored as a flattened array of floats, + * representing shape [Channel,Height,Width]. + */ + repeated float meanImage = 1; +} + +/// Preprocessing parameters for image inputs. +message NeuralNetworkPreprocessing { + string featureName = 1; + oneof preprocessor { + NeuralNetworkImageScaler scaler = 10; + NeuralNetworkMeanImage meanImage = 11; + } +} + +/// Activation Functions +/// -------------------- + +/** + * A rectified linear unit (ReLU) activation function. + * + * This function has the following formula: + * + * .. math:: + * f(x) = \text{max}(0, x) + */ +message ActivationReLU { +} + +/** + * A leaky rectified linear unit (ReLU) activation function. + * + * This function has the following formula: + * + * .. math:: + * f(x) = \begin{cases} + * x & \text{if } x \geq 0 \\ + * \alpha x & \text{if } x < 0 + * \end{cases} + */ +message ActivationLeakyReLU { + float alpha = 1; //negative slope value for leakyReLU +} + +/** + * A hyperbolic tangent activation function. + * + * This function has the following formula: + * + * .. math:: + * f(x) = \dfrac{1 - e^{-2x}}{1 + e^{-2x}} + */ +message ActivationTanh { +} + +/** + * A scaled hyperbolic tangent activation function. + * + * This function has the following formula: + * + * .. math:: + * f(x) = \alpha \tanh(\beta x) + */ +message ActivationScaledTanh { + float alpha = 1; + float beta = 2; +} + +/** + * A sigmoid activation function. + * + * This function has the following formula: + * + * .. math:: + * f(x) = \dfrac{1}{1 + e^{-x}} + */ +message ActivationSigmoid { +} + +/** + * A linear activation function. + * + * This function has the following formula: + * + * .. math:: + * f(x) = \alpha x + \beta + */ +message ActivationLinear { + float alpha = 1; + float beta = 2; +} + +/** + * A hard sigmoid activation function. + * + * This function has the following formula: + * + * .. math:: + * f(x) = \text{min}(\text{max}(\alpha x + \beta, 0), 1) + */ +message ActivationSigmoidHard { + float alpha = 1; + float beta = 2; +} + +/** + * A parameterized rectified linear unit (PReLU) activation function, + * which takes ``[C]`` or ``[C,H,W]`` as an input and + * applies different parameters in each channel dimension + * (shared across the ``H`` and ``W`` components). + * + * This function has the following formula: + * + * .. math:: + * f(x_i) = \begin{cases} + * x_i & \text{if } x_i \geq 0 \\ + * \alpha_i x_i & \text{if } x_i < 0 + * \end{cases} \;,\;i=1,...,C + */ +message ActivationPReLU { + // parameter of length C or 1. + // If length is 1, same value is used for all channels + WeightParams alpha = 1; +} + +/** + * An exponential linear unit (ELU) activation function. + * + * This function has the following formula: + * + * .. math:: + * f(x) = \begin{cases} + * x & \text{if } x \geq 0 \\ + * \alpha (e^x - 1) & \text{if } x < 0 + * \end{cases} + */ +message ActivationELU { + float alpha = 1; +} + +/** + * A thresholded rectified linear unit (ReLU) activation function. + * + * This function has the following formula: + * + * .. math:: + * f(x) = \begin{cases} + * x & \text{if } x \geq \alpha \\ + * 0 & \text{if } x < \alpha + * \end{cases} + */ +message ActivationThresholdedReLU { + float alpha = 1; +} + +/** + * A softsign activation function. + * + * This function has the following formula: + * + * .. math:: + * f(x) = \dfrac{x}{1 + |x|} + */ +message ActivationSoftsign { +} + +/** + * A softplus activation function. + * + * This function has the following formula: + * + * .. math:: + * f(x) = \text{log}(1 + e^x) + */ +message ActivationSoftplus { +} + +/** + * A parametric softplus activation function, + * which takes ``[C]`` or ``[C,H,W]`` as an input and + * applies different parameters in each channel dimension + * (shared across the ``H`` and ``W`` components). + * + * This function has the following formula: + * + * .. math:: + * f(x_i) = \alpha_i \text{log}(1 + e^{\beta_i x_i}) \;,\;i=1,...,C + */ +message ActivationParametricSoftplus { + // If length is 1, same value is used for all channels + WeightParams alpha = 1; //parameter of length C or 1 + WeightParams beta = 2; //parameter of length C or 1 +} + +message ActivationParams { + oneof NonlinearityType { + ActivationLinear linear = 5; + + ActivationReLU ReLU = 10; + ActivationLeakyReLU leakyReLU = 15; + ActivationThresholdedReLU thresholdedReLU = 20; + ActivationPReLU PReLU = 25; + + ActivationTanh tanh = 30; + ActivationScaledTanh scaledTanh = 31; + + ActivationSigmoid sigmoid = 40; + ActivationSigmoidHard sigmoidHard = 41; + + ActivationELU ELU = 50; + + ActivationSoftsign softsign = 60; + ActivationSoftplus softplus = 70; + ActivationParametricSoftplus parametricSoftplus = 71; + } +} + +/** + * A single neural network layer. + */ +message NeuralNetworkLayer { + string name = 1; //descriptive name of the layer + repeated string input = 2; + repeated string output = 3; + + oneof layer { + // start at 100 here + ConvolutionLayerParams convolution = 100; + + PoolingLayerParams pooling = 120; + + ActivationParams activation = 130; + + InnerProductLayerParams innerProduct = 140; + EmbeddingLayerParams embedding = 150; + + //normalization related layers + BatchnormLayerParams batchnorm = 160; + MeanVarianceNormalizeLayerParams mvn = 165; + L2NormalizeLayerParams l2normalize = 170; + SoftmaxLayerParams softmax = 175; + LRNLayerParams lrn = 180; + + CropLayerParams crop = 190; + PaddingLayerParams padding = 200; + UpsampleLayerParams upsample = 210; + + UnaryFunctionLayerParams unary = 220; + + //elementwise operations + AddLayerParams add = 230; + MultiplyLayerParams multiply = 231; + + AverageLayerParams average = 240; + ScaleLayerParams scale = 245; + + BiasLayerParams bias = 250; + MaxLayerParams max = 260; + MinLayerParams min = 261; + + DotProductLayerParams dot = 270; + ReduceLayerParams reduce = 280; + LoadConstantLayerParams loadConstant = 290; + + //data reorganization + ReshapeLayerParams reshape = 300; + FlattenLayerParams flatten = 301; + PermuteLayerParams permute = 310; + ConcatLayerParams concat = 320; + SplitLayerParams split = 330; + SequenceRepeatLayerParams sequenceRepeat = 340; + + //Recurrent Layers + SimpleRecurrentLayerParams simpleRecurrent = 400; + GRULayerParams gru = 410; + UniDirectionalLSTMLayerParams uniDirectionalLSTM = 420; + BiDirectionalLSTMLayerParams biDirectionalLSTM = 430; + } +} + +/// Border Amounts +/// -------------- + +/** + * Specifies the amount of spatial border to be either padded or cropped. + * + * For padding: + * :: + * H_out = borderAmounts[0].startEdgeSize + H_in + borderAmounts[0].endEdgeSize + * W_out = borderAmounts[1].startEdgeSize + W_in + borderAmounts[1].endEdgeSize + * + * topPaddingAmount == Height startEdgeSize + * bottomPaddingAmount == Height endEdgeSize + * leftPaddingAmount == Width startEdgeSize + * rightPaddingAmount == Width endEdgeSize + * + * For cropping: + * :: + * H_out = (-borderAmounts[0].startEdgeSize) + H_in + (-borderAmounts[0].endEdgeSize) + * W_out = (-borderAmounts[1].startEdgeSize) + W_in + (-borderAmounts[1].endEdgeSize) + * + * topCropAmount == Height startEdgeSize + * bottomCropAmount == Height endEdgeSize + * leftCropAmount == Width startEdgeSize + * rightCropAmount == Width endEdgeSize + */ +message BorderAmounts { + message EdgeSizes { + /** + * The amount to be padded or cropped from the beginning. + */ + uint64 startEdgeSize = 1; + + /** + * The amount to be padded or cropped from the end. + */ + uint64 endEdgeSize = 2; + } + + /** + * The border amounts. + * This must be length 2 in the order ``[H, W]``. + */ + repeated EdgeSizes borderAmounts = 10; +} + +/** + * Specifies the type of padding to be used with Convolution/Deconvolution and Pooling layers. + * After padding, input spatial shape: ``[H_in, W_in]``, gets modified to the + * output spatial shape ``[H_out, W_out]``. + * :: + * topPaddingAmount == Height startEdgeSize == borderAmounts[0].startEdgeSize + * bottomPaddingAmount == Height endEdgeSize == borderAmounts[0].endEdgeSize + * leftPaddingAmount == Width startEdgeSize == borderAmounts[1].startEdgeSize + * rightPaddingAmount == Width endEdgeSize == borderAmounts[1].endEdgeSize + * + * With Convolution or Pooling: + * :: + * H_out = int_division_round_down((H_in + topPaddingAmount + bottomPaddingAmount - KernelSize[0]),stride[0]) + 1 + * + * which is same as: + * :: + * H_out = int_division_round_up((H_in + topPaddingAmount + bottomPaddingAmount - KernelSize[0] + 1),stride[0]) + * + * With Deconvolution: + * :: + * H_out = (H_in-1) * stride[0] + kernelSize[0] - (topPaddingAmount + bottomPaddingAmount) + * + * + * The equivalent expressions hold true for ``W_out`` as well. + * + * + * By default, the values of ``paddingAmounts`` are set to ``0``, + * which results in a "true" valid padding. + * If non-zero values are provided for ``paddingAmounts``, + * "valid" convolution/pooling is performed within the spatially expanded input. + * + */ +message ValidPadding { + BorderAmounts paddingAmounts = 1; +} + +/** + * Specifies the type of padding to be used with Convolution/Deconvolution and pooling layers. + * After padding, input spatial shape: ``[H_in, W_in]``, gets modified to the + * output spatial shape ``[H_out, W_out]``. + * With Convolution or pooling: + * :: + * H_out = int_division_round_up(H_in,stride[0]) + * W_out = int_division_round_up(W_in,stride[1]) + * + * This is achieved by using the following padding amounts: + * :: + * totalPaddingHeight = max(0,(H_out-1) * stride[0] + KernelSize[0] - Hin) + * totalPaddingWidth = max(0,(W_out-1) * stride[1] + KernelSize[1] - Win) + * + * There are two modes of asymmetry: + * ``BOTTOM_RIGHT_HEAVY``, and ``TOP_LEFT_HEAVY``. + * + * If the mode is ``BOTTOM_RIGHT_HEAVY``: + * :: + * topPaddingAmount = floor(totalPaddingHeight / 2) + * bottomPaddingAmount = totalPaddingHeight - topPaddingAmount + * leftPaddingAmount = floor(totalPaddingWidth / 2) + * rightPaddingAmount = totalPaddingWidth - leftPaddingAmount + * + * If the mode is ``TOP_LEFT_HEAVY``: + * :: + * bottomPaddingAmount = floor(totalPaddingHeight / 2) + * topPaddingAmount = totalPaddingHeight - bottomPaddingAmount + * rightPaddingAmount = floor(totalPaddingWidth / 2) + * leftPaddingAmount = totalPaddingWidth - rightPaddingAmount + * + * + * With Deconvolution: + * :: + * H_out = H_in * stride[0] + * W_out = W_in * stride[1] + */ +message SamePadding { + enum SamePaddingMode { + BOTTOM_RIGHT_HEAVY = 0; + TOP_LEFT_HEAVY = 1; + } + SamePaddingMode asymmetryMode = 1; +} + +/** + * Weights for layer parameters. + * Weights are stored as repeated floating point numbers + * using row-major ordering + * and can represent 1-, 2-, 3-, or 4-dimensional data. + */ +message WeightParams { + /** + * Values specified in single / float / FP32 precision. + */ + repeated float floatValue = 1; +} + +/// Layers +/// ------ + +/** + * A layer that performs spatial convolution or deconvolution. + * :: + * y = ConvolutionLayer(x) + * + * Requires 1 input and produces 1 output. + * + * Input + * A blob with shape ``[inputChannels,inputHeight,inputWidth]`` or ``[C_in, H_in, W_in]``. + * + * Output + * A blob with shape ``[outputChannels,outputHeight,outputWidth]`` or ``[C_out, H_out, W_out]``. + * + * + * If ``dilationFactor`` is not 1 and ``isDeconvolution`` is not True, effective kernel size is + * modified as follows: + * :: + * KernelSize[0] <-- (kernelSize[0]-1) * dilationFactor[0] + 1 + * KernelSize[1] <-- (kernelSize[1]-1) * dilationFactor[1] + 1 + * + * Type of padding can be ``valid`` or ``same``. Output spatial dimensions depend on the + * the type of padding. For details, refer to the descriptions of the messages "ValidPadding" + * and "SamePadding". Padded values are all zeros. + * + * For Deconvolution, ``ConvolutionPaddingType`` (``valid`` or ``same``) is ignored when ``outputShape`` is set. + * + * + */ +message ConvolutionLayerParams { + /** + * The number of kernels. + * Same as ``C_out`` used in the layer description. + */ + uint64 outputChannels = 1; + + /** + * Channel dimension of the kernels. + * Must be equal to ``inputChannels / nGroups``. + */ + uint64 kernelChannels = 2; + + /** + * Group convolution as used in AlexNet, + * i.e. weight reuse along channel axis. + * Kernel channels * nGroups = inputChannels. + * If not set or 0, it is set to the default value 1. + */ + uint64 nGroups = 10; + + /** + * Must be length 2 in the order ``[H, W]``. + * If not set, default value ``[3, 3]`` is used. + */ + repeated uint64 kernelSize = 20; + + /** + * Must be length 2 in the order ``[H, W]``. + * If not set, default value ``[1, 1]`` is used. + */ + repeated uint64 stride = 30; + + /** + * Must be length 2 in order ``[H, W]``. + * If not set, default value ``[1, 1]`` is used. + * It is ignored if ``isDeconvolution == true``. + */ + repeated uint64 dilationFactor = 40; + + /** + * The type of padding. + */ + oneof ConvolutionPaddingType { + ValidPadding valid = 50; + SamePadding same = 51; + } + + /** + * Flag to specify whether it is a deconvolution layer. + */ + bool isDeconvolution = 60; + + /** + * Flag to specify whether a bias is to be added or not. + */ + bool hasBias = 70; + + /** + * Weights associated with this layer. + * If convolution (``isDeconvolution == false``), weights have the shape + * ``[outputChannels, kernelChannels, kernelHeight, kernelWidth]``. + * If deconvolution (``isDeconvolution == true``) weights have the shape + * ``[kernelChannels, outputChannels, kernelHeight, kernelWidth]``. + */ + WeightParams weights = 90; + WeightParams bias = 91; /// Must be of size [outputChannels]. + + /** + * The output shape, which has length 2 ``[H_out, W_out]``. + * This is used only for deconvolution (``isDeconvolution == true``). + * If not set, the deconvolution output shape is calculated + * based on ``ConvolutionPaddingType``. + */ + repeated uint64 outputShape = 100; +} + +/** + * A layer that performs a matrix vector product. + * This is equivalent to a fully-connected, or dense layer. + * :: + * y = InnerProductLayer(x) + * + * Requires 1 input and produces 1 output. + * + * Input + * A blob with shape ``[C_in]`` or ``[C_in, 1, 1]``, where ``C_in`` is equal to ``inputChannels``. + * + * Output + * A blob with shape ``[C_out]``, where ``C_out`` is equal to ``outputChannels``. + */ +message InnerProductLayerParams { + uint64 inputChannels = 1; /// Input size: C_in. + uint64 outputChannels = 2; /// Output size: C_out. + + bool hasBias = 10; /// Whether a bias is added or not. + + WeightParams weights = 20; /// Weight matrix [C_out, C_in]. + WeightParams bias = 21; /// Bias vector [C_out]. +} + +/** + * A layer that performs a matrix lookup and optionally adds a bias. + * :: + * y = EmbeddingLayer(x) + * + * Requires 1 input and produces 1 output. + * + * Input + * A sequence of integers with shape ``[1]`` or ``[1, 1, 1]``, (equivalent to ``[Seq_length, 1, 1, 1]``). + * Input values must be in the range ``[0, inputDim - 1]``. + * + * Output + * A sequence of 1-dimensional features of size ``outputChannels`` + * (equivalent to ``[Seq_length, outputChannels, 1, 1]``). + */ +message EmbeddingLayerParams { + uint64 inputDim = 1; /// Size of the input dictionary. + uint64 outputChannels = 2; /// Size of the output vectors. + + bool hasBias = 10; /// Whether a bias is added or not. + + WeightParams weights = 20; /// 2-D weights of dimensions [outputChannels, inputDim]. + WeightParams bias = 21; /// Bias of size [outputChannels]. +} + +/** + * A layer that performs batch normalization, + * which is performed along the channel axis, + * and repeated along the other axes, if present. + * :: + * y = BatchnormLayer(x) + * Requires 1 input and produces 1 output. + * + * This operation is described by the following formula: + * + * .. math:: + * y_i = \gamma_i \dfrac{ (x_i - \mu_i)}{\sqrt{\sigma_i^2 + \epsilon}} + \beta_i \;,\;i=1,....,C + * + * Input + * A blob with shape ``[C]`` or ``[C, H, W]``. + * + * Output + * A blob with the same shape as the input. + */ +message BatchnormLayerParams { + uint64 channels = 1; /// Size of the channel dimension in the input. + + /** + * If ``computeMeanVar == true``, + * the mean and variance are calculated from either + * the single input instance, if ``instanceNormalization == true``, + * or the whole batch, if ``instanceNormalization = false``. + * and the values provided in parameters "mean" and "variance" are ignored. + */ + bool computeMeanVar = 5; + bool instanceNormalization = 6; + + /** + * A small constant to avoid division by 0 while normalizing by variance. + * Defaults to ``1e-5`` if not set or set to ``0``. + */ + float epsilon = 10; + + WeightParams gamma=15; /// Parameter of length [channels] + WeightParams beta=16; /// Parameter of length [channels] + WeightParams mean=17; /// Parameter of length [channels] + WeightParams variance=18; /// Parameter of length [channels] +} + +/** + * A spatial pooling layer. + * :: + * y = PoolingLayer(x) + * Requires 1 input and produces 1 output. + * + * Input + * A blob with shape ``[C, H_in, W_in]``. + * Output + * A blob with shape ``[C, H_out, W_out]``. + * + * Padding options are similar to ``ConvolutionLayerParams`` + * with the additional option of ``ValidCompletePadding`` (``includeLastPixel``), + * which ensures that the last application of the kernel + * always includes the last pixel of the input image. + * :: + * H_out = int_division_round_up((H_in + 2 * paddingAmounts[0] - kernelSize[0]),Stride[0]) + 1) + * if ((H_out - 1) * Stride >= H_in + paddingAmounts[0]) { + * H_out = H_out - 1 + * } + * + * The equivalent expressions hold true for ``W_out`` as well. + * Only symmetric padding is supported with this option. + */ +message PoolingLayerParams { + enum PoolingType{ + MAX = 0; + AVERAGE = 1; + L2 = 2; + } + PoolingType type = 1; /// Type of pooling operation. + + /** + * Must be length 2 in the order ``[H, W]``. + * If not set, default value ``[3, 3]`` is used. + */ + repeated uint64 kernelSize = 10; + + /** + * Must be length 2 in the order ``[H, W]``. + * If not set, default value ``[1, 1]`` is used. + */ + repeated uint64 stride = 20; + + message ValidCompletePadding { + /** + * Must be length 2 in order ``[H, W]``. + * If not set, value ``[0, 0]`` is used. + */ + repeated uint64 paddingAmounts = 10; + } + + oneof PoolingPaddingType { + ValidPadding valid = 30; + SamePadding same = 31; + ValidCompletePadding includeLastPixel = 32; + } + + /** + * If true, padded values are excluded from the count (denominator) + * when computing average pooling. + */ + bool avgPoolExcludePadding = 50; + + /** + * If true, global pooling is performed. + * Kernel size is inferred from the input data spatial dimensions. + */ + bool globalPooling = 60; +} + +/** + * A layer that performs padding along spatial dimensions. + * :: + * y = PaddingLayer(x) + * Requires 1 input and produces 1 output. + * + * Input + * A blob with shape ``[C, H_in, W_in]``. + * + * Output + * A blob with shape ``[C, H_out, W_out]``. + * Output dimensions are calculated as follows: + * :: + * H_out = H_in + topPaddingAmount + bottomPaddingAmount + * W_out = W_in + leftPaddingAmount + rightPaddingAmount + * + * topPaddingAmount == Height startEdgeSize == borderAmounts[0].startEdgeSize + * bottomPaddingAmount == Height endEdgeSize == borderAmounts[0].endEdgeSize + * leftPaddingAmount == Width startEdgeSize == borderAmounts[1].startEdgeSize + * rightPaddingAmount == Width endEdgeSize == borderAmounts[1].endEdgeSize + * + * There are three types of padding: + * + * - ``PaddingConstant``, which fills a constant value at the border. + * - ``PaddingReflection``, which reflects the values at the border. + * - ``PaddingReplication``, which replicates the values at the border. + * + * Given the following input: + * :: + * [1, 3, 4] : 1 2 3 4 + * 5 6 7 8 + * 9 10 11 12 + * + * Here is the output of applying the padding + * ``(top=2, left=2, bottom=0, right=0)`` + * with each of the supported types: + * + * - ``PaddingConstant`` (``value = 0``): + * :: + * [1, 5, 6] : 0 0 0 0 0 0 + * 0 0 0 0 0 0 + * 0 0 1 2 3 4 + * 0 0 5 6 7 8 + * 0 0 9 10 11 12 + * + * - ``PaddingReflection``: + * :: + * [1, 5, 6] : 11 10 9 10 11 12 + * 7 6 5 6 7 8 + * 3 2 1 2 3 4 + * 7 6 5 6 7 8 + * 11 10 9 10 11 12 + * + * - ``PaddingReplication``: + * :: + * [1, 5, 6] : 1 1 1 2 3 4 + * 1 1 1 2 3 4 + * 1 1 1 2 3 4 + * 5 5 5 6 7 8 + * 9 9 9 10 11 12 + */ +message PaddingLayerParams { + /** + * Fill a constant value in the padded region. + */ + message PaddingConstant { + float value = 1; + } + + /** + * Reflect the values at the border for padding. + */ + message PaddingReflection { + } + + /** + * Replicate the values at the border for padding. + */ + message PaddingReplication { + } + + oneof PaddingType { + PaddingConstant constant = 1; + PaddingReflection reflection = 2; + PaddingReplication replication = 3; + } + + BorderAmounts paddingAmounts = 10; /// Amounts to be padded to the input. +} + +/** + * A layer that concatenates along the channel axis (default) or sequence axis. + * :: + * y = ConcatLayer(x1,x2,....) + * Requires more than 1 input and produces 1 output. + * + * The input and output formats are dependent on ``sequenceConcat``. + * + * If ``sequenceConcat == true``: + * + * Input + * Sequences of length ``Seq_i`` of blobs with shape ``[C, H, W]``. + * Output + * A Sequence of length ``summation(Seq_i)`` of blobs with shape ``[C, H, W]``. + * + * If ``sequenceConcat == false``: + * + * Input + * A blob with shape ``[C_i, H, W]``, where ``i = 1, 2, ...``. + * Output + * A blob with shape ``[summation(C_i), H, W]``. + */ +message ConcatLayerParams { + /** + * If true, concatenate along the sequence axis instead of the channel axis. + */ + bool sequenceConcat = 100; +} + +/** + * A layer that performs local response normalization (LRN). + * :: + * y = LRNLayer(x) + * Requires 1 input and produces 1 output. + * + * Input + * A blob with shape ``[C, H, W]`` + * Output + * A blob with the same shape as the input. + * + * This layer is described by the following formula: + * + * .. math:: + * x_i \leftarrow \dfrac{x_i}{\left ( k + \dfrac{\alpha}{C} \sum_j x_j^2 \right )^\beta} + * + * where the summation is done over a ``(localSize, 1, 1)`` neighborhood --- + * that is, over a window "across" channels in 1x1 spatial neighborhoods. + */ +message LRNLayerParams { + float alpha = 1; + float beta = 2; + uint64 localSize = 3; /// Number of channels in the normalization window. + float k = 4; /// Defaults to 1 if not set or 0. Must be strictly positive. +} + +/** + * Softmax Normalization Layer + * + * A layer that performs softmax normalization. + * Normalization is done along the channel axis. + * :: + * y = SoftmaxLayer(x) + * Requires 1 input and produces 1 output. + * + * Input + * A blob with shape ``[C]`` or ``[C, H, W]``. + * Output + * A blob with the same shape as the input. + * + * This layer is described by the following formula: + * + * .. math:: + * x_i \leftarrow \dfrac{e^{x_i}}{\sum_i{e^{x_i}}} + */ +message SoftmaxLayerParams { +} + +/** + * A layer that uniformly splits across the channel dimension + * to produce a specified number of outputs. + * :: + * (y1,y2,...yN) = SplitLayer(x), where N = nOutputs + * Requires 1 input and produces multiple outputs. + * + * Input + * A blob with shape ``[C]`` or ``[C, H, W]`` + * Output + * ``nOutputs`` blobs with shapes + * ``[C/nOutputs]`` or ``[C/nOutputs, H, W]`` + */ +message SplitLayerParams { + uint64 nOutputs=1; /// The number of outputs. +} + +/** + * A layer that performs elementwise addition. + * :: + * y = AddLayer(x1,x2,...) + * Requires 1 or more than 1 input and produces 1 output. + * + * Input + * One or more blobs with shape ``[C, H, W]``. + * Output + * A blob with shape equal to the input blob. + * + * If only one input is provided, scalar addition is performed: + * + * .. math:: + * y = x + \alpha + * + * If multiple inputs are provided, + * each input is broadcasted to the shape of the first input and added. + * The shapes of inputs after the first must be one of the following: + * ``[1]``, ``[C]``, ``[1, H, W]``, or ``[C, H, W]``. + */ +message AddLayerParams { + /** + * Scalar to be added to the input. + * Only used if there is a single input. + */ + float alpha = 1; +} + +/** + * A layer that performs elementwise multiplication. + * :: + * y = MultiplyLayer(x1,x2,...) + * Requires 1 or more than 1 input and produces 1 output. + * + * Input + * One or more blobs with shape ``[C, H, W]``. + * Output + * A blob with shape equal to the first input blob. + * + * If only one input is provided, scalar multiplication is performed: + * + * .. math:: + * y = \alpha x + * + * If multiple inputs are provided, + * each input is broadcasted to the shape of the first input and multiplied. + * The shapes of inputs after the first must be one of the following: + * ``[1]``, ``[C]``, ``[1, H, W]``, or ``[C, H, W]``. + */ +message MultiplyLayerParams { + /** + * Scalar to be multiplied with the input. + * Only used if there is a single input. + */ + float alpha = 1; +} + +/** + * A layer that applies a unary function. + * :: + * y = UnaryFunctionLayer(x) + * Requires 1 input and produces 1 output. + * + * Input + * A blob with shape ``[C]`` or ``[C, H, W]``. + * Output + * A blob with the same shape as the input. + * + * The input is first modified by shifting and scaling: + * + * .. math:: + * x \leftarrow \text{scale} \cdot x + \text{shift} + */ +message UnaryFunctionLayerParams { + /** + * A unary operator. + * + * The following functions are supported: + * + * ``SQRT`` + * .. math:: f(x) = \sqrt{x} + * + * ``RSQRT`` + * .. math:: f(x) = \dfrac{1}{\sqrt{x + \epsilon}} + * + * ``INVERSE`` + * .. math:: f(x) = \dfrac{1}{x + \epsilon} + * + * ``POWER`` + * .. math:: f(x) = x^\alpha + * + * ``EXP`` + * .. math:: f(x) = e^x + * + * ``LOG`` + * .. math:: f(x) = \log x + * + * ``ABS`` + * .. math:: f(x) = |x| + * + * ``THRESHOLD`` + * .. math:: f(x) = \text{max}(\alpha, x) + */ + enum Operation{ + SQRT = 0; + RSQRT = 1; + INVERSE = 2; + POWER = 3; + EXP = 4; + LOG = 5; + ABS = 6; + THRESHOLD = 7; + } + Operation type = 1; /// The type of unary function. + + /** + * A constant used in ``POWER`` and ``THRESHOLD`` functions. + */ + float alpha = 2; + + /** + * A small constant to avoid division by 0 while normalizing variance. + * Defaults to ``1e-6`` if not set or set to ``0``. + */ + float epsilon = 3; + + /** + * Input is shifted by this amount + * before the unary function is applied. + * Defaults to ``0.0`` if not set. + */ + float shift = 4; + + /** + * Input is scaled by this amount + * before the unary function is applied. + * Defaults to ``1.0`` if not set or set to ``0``. + */ + float scale = 5; +} + +/** + * A layer that scales up spatial dimensions + * using nearest neighbor interpolation. + * :: + * y = UpsampleLayer(x) + * Requires 1 input and produces 1 output. + * + * Input + * A blob with shape ``[C, H, W]``. + * Output + * A blob with shape ``[C, scalingFactor[0] * H, scalingFactor[1] * W]`` + */ +message UpsampleLayerParams { + /** + * Scaling Factor. + * Must be length 2 in order ``[H, W]``. + * If not set, default value ``[1, 1]`` is used. + */ + repeated uint64 scalingFactor = 1; +} + +/** + * A layer that performs elementwise addition of a bias, + * which is broadcasted to match the input shape. + * :: + * y = BiasLayer(x) + * Requires 1 input and produces 1 output. + * + * Input + * A blob with shape ``[C, H, W]``. + * Output + * A blob with the same shape as the input. + */ +message BiasLayerParams { + /** + * The shape of the bias. + * Must be one of the following: + * ``[1]``, ``[C]``, ``[1, H, W]`` or ``[C, H, W]``. + */ + repeated uint64 shape = 1; + + /** + * The bias values. + * The size must be equal to the product of the ``shape`` dimensions. + */ + WeightParams bias = 2; +} + +/** + * A layer that performs elmentwise multiplication by a scale factor + * and optionally adds a bias; + * both the scale and bias are broadcasted to match the input shape. + * :: + * y = ScaleLayer(x) + * Requires 1 input and produces 1 output. + * + * Input + * A blob with shape ``[C, H, W]``. + * Output + * A blob with the same shape as the input. + */ +message ScaleLayerParams { + /** + * The shape of the scale. + * Must be one of the following: + * ``[1]``, ``[C]``, ``[1, H, W]`` or ``[C, H, W]``. + */ + repeated uint64 shapeScale = 1; + + /** + * The scale values. + * The size must be equal to the product of the ``shape`` dimensions. + */ + WeightParams scale = 2; /// Scale values. Size must be equal to the product of dimensions specified in shapeScale. + + bool hasBias = 3; /// If true, a bias is added after scaling. + + /** + * The shape of the bias. + * Must be one of the following: + * ``[1]``, ``[C]``, ``[1, H, W]`` or ``[C, H, W]``. + */ + repeated uint64 shapeBias = 4; + + /** + * The bias values. + * The size must be equal to the product of the ``shape`` dimensions. + */ + WeightParams bias = 5; +} + +/** + * A layer that loads data as a parameter and provides it as an output. + * :: + * y = LoadConstantLayer() + * Takes no input. Produces 1 output. + * + * Input + * None + * Output: + * A blob with shape ``[C, H, W]`` + */ +message LoadConstantLayerParams { + /** + * The shape of the constant to be loaded, + * which must be``[C, H, W]``. + */ + repeated uint64 shape = 1; + + /** + * The data values, + * of size ``C * H * W``. + */ + WeightParams data = 2; +} + +/** + * A layer that performs L2 normalization, i.e. divides by the + * the square root of the sum of squares of all elements of input. + * :: + * y = L2NormalizeLayer(x) + * Requires 1 input and produces 1 output. + * + * Input + * A blob with shape ``[C]`` or ``[C, H, W]``. + * Output + * A blob with the same shape as the input. + * + * This layer is described by the following formula: + * + * .. math:: + * x_i \leftarrow \dfrac{x_i}{\sqrt{\sum{x_i^2} + \epsilon}} + */ +message L2NormalizeLayerParams { + /** + * A small constant to avoid division by 0 while normalizing variance. + * Defaults to ``1e-6`` if not set or set to ``0``. + */ + float epsilon = 1; +} + +/// Data Reorganization Layers +/// -------------------------- + +/** + * A layer that flattens the input. + * :: + * y = FlattenLayer(x) + * Requires 1 input and produces 1 output. + * + * Input + * A blob with shape ``[C, H, W]``. + * Output + * A blob with shape ``[C * H * W, 1, 1]`` + * + * There are two flatten orders: ``CHANNEL_FIRST`` and ``CHANNEL_LAST``. + * ``CHANNEL_FIRST`` does not require data to be rearranged, + * because row major ordering is used by internal storage. + * ``CHANNEL_LAST`` requires data to be rearranged. + */ +message FlattenLayerParams { + enum FlattenOrder { + CHANNEL_FIRST = 0; + CHANNEL_LAST = 1; + } + FlattenOrder mode = 1; +} + +/** + * A layer that recasts the input into a new shape. + * :: + * y = ReshapeLayer(x) + * Requires 1 input and produces 1 output. + * + * Input + * A blob with shape ``[C, H, W]`` or ``[Seq, C, H, W]``. + * Output + * A blob with shape ``[C_out, H_out, W_out]`` + * or ``[Seq_out, C_out, H_out, W_out]``. + * + * There are two reshape orders: ``CHANNEL_FIRST`` and ``CHANNEL_LAST``. + * ``CHANNEL_FIRST`` is equivalent to + * flattening the input to ``[C * H * W, 1, 1]`` in channel first order + * and then reshaping it to the target shape; + * no data rearrangement is required. + * ``CHANNEL_LAST`` is equivalent to + * flattening the input to ``[H * W * C, 1, 1]`` in channel last order, + * reshaping it to ``[H_out, W_out, C_out]`` (it is now in "H_out-major"" order), + * and then permuting it to ``[C_out, H_out, W_out]``; + * both the flattening and permuting requires the data to be rearranged. + */ +message ReshapeLayerParams { + /** + * The shape of the output. + * Must be of length 3 or 4. + * If set to 3, ``targetShape`` is interpreted as + * ``[C_out, H_out, W_out]``, and sequence length of the input is preserved. + * If set to 4, ``targetShape`` is interpreted as + * ``[Seq_out, C_out, H_out, W_out]``, + * where ``Seq_out`` is the new sequence length. + */ + repeated int64 targetShape = 1; + + enum ReshapeOrder { + CHANNEL_FIRST = 0; + CHANNEL_LAST = 1; + } + ReshapeOrder mode = 2; +} + +/** + * A layer that rearranges the dimensions and data of an input. + * :: + * y = PermuteLayer(x) + * Requires 1 input and produces 1 output. + * + * Input + * A sequence of 3-dimensional blobs. ``InputShape = [Seq, C, H, W]``. + * Output + * A sequence of a different length of 3-dimensional blobs. + * Shape: ``[InputShape[axis[0]], InputShape[axis[1]], + * InputShape[axis[2]], InputShape[axis[3]]]``. Hence output is a sequence of length ``InputShape[axis[0]]``. + * + * Examples: + * + * - If ``axis`` is set to ``[0, 3, 1, 2]``, + * then the output has shape ``[W,C,H]`` + * and has the same sequence length that of the input. + * - If ``axis`` is set to ``[3, 1, 2, 0]``, + * and the input is a sequence of data + * with length ``Seq`` and shape ``[C, 1, 1]``, + * then the output is a unit sequence of data with shape ``[C, 1, Seq]``. + * - If ``axis`` is set to ``[0, 3, 2, 1]``, + * the output is a reverse of the input: ``[C, H, W] -> [W, H, C]``. + * - If ``axis`` is not set, or is set to ``[0, 1, 2, 3]``, + * the output is the same as the input. + */ +message PermuteLayerParams { + /** + The order in which to permute the dimensions. + Must have length 4 and a permutation of ``[0, 1, 2, 3]``. + */ + repeated uint64 axis = 1; +} + +/** + * A layer that reduces the input to a scalar value using a specified operation. + * :: + * y = ReduceLayer(x) + * Requires 1 input and produces 1 output. + * + * Input + * A blob with shape ``[C, H, W]``. + * Output + * A scalar value. + */ +message ReduceLayerParams { + /* + * The following reduction operations are supported + * and are applied on flattened input array: + * + * ``SUM`` + * Sum of all elements + * + * .. math:: \sum{x_i} + * + * ``AVG`` + * Sum of all elements divided by the number of elements + * + * .. math:: \dfrac{\sum^n{x_i}}{n} + * + * ``PROD`` + * Product of all elements + * + * .. math:: \prod{x_i} + * + * ``LOGSUM`` + * Sum of the natural logarithm of all elements + * + * .. math:: \sum{\ln{(x_i + \epsilon)}} + * + * ``SUMSQUARE`` + * Sum of squares of all elements + * + * .. math:: \sum{x^2} + * + * ``L1`` + * L1 normalization of all elements + * + * .. math:: ||x||_1 = \sum{|x_i|} + * + * ``L2`` + * L2 normalization of all elements + * + * .. math:: ||x||_2 = \sqrt{\sum{x_i^2}} + * + */ + enum ReduceOperation { + SUM = 0; + AVG = 1; + PROD = 2; + LOGSUM = 3; + SUMSQUARE = 4; + L1 = 5; + L2 = 6; + } + ReduceOperation mode = 1; /// Specifies function used to reduce. + + /** + * Used if mode is ``LOGSUM``. + * Defaults to ``1e-6`` if not set or is set to ``0``. + */ + float epsilon = 2; +} + +/** + * A layer that crops the spatial dimensions of an input. + * If two inputs are provided, the shape of the second input is used as the reference shape. + * :: + * y = CropLayer(x1) or y = CropLayer(x1,x2) + * Requires 1 or 2 inputs and produces 1 output. + * + * Input + * - 1 input case: A blob with shape ``[C, H_in, W_in]``. + * - 2 input case: 1st blob with shape ``[C, H_in, W_in]``, 2nd blob with shape ``[C, H_out, W_out]``. + * + * Output + * A blob with shape ``[C, H_out, W_out]``. + * + * If one input is used, output is computed as follows: + * :: + * y = x1[:, topCropAmount:H_in - bottomCropAmount, leftCropAmount:W_in - rightCropAmount] + * + * topCropAmount == Height startEdgeSize == borderAmounts[0].startEdgeSize + * bottomCropAmount == Height endEdgeSize == borderAmounts[0].endEdgeSize + * leftCropAmount == Width startEdgeSize == borderAmounts[1].startEdgeSize + * rightCropAmount == Width endEdgeSize == borderAmounts[1].endEdgeSize + * + * H_out = H_in - topCropAmount - bottomCropAmount + * W_out = W_in - leftCropAmount - rightCropAmount + * + * If two inputs are used, output is computed as follows: + * :: + * y = x1[:, offset[0]:offset[0] + H_out, offset[1]:offset[1] + W_out] + */ +message CropLayerParams { + /** + * The amounts to be cropped from the input. + * Used only if a single input is provided. + */ + BorderAmounts cropAmounts = 1; + + /** + * The offset amounts. + * Used only if two inputs are provided. + * Must be of length 2, in order ``[H, W]``. + */ + repeated uint64 offset = 5; +} + +/** + * A layer that computes the elementwise average of the inputs. + * :: + * y = AverageLayer(x1,x2,...) + * Requires multiple inputs and produces 1 output. + * + * Input + * Multiple blobs, each with shape ``[C]`` or ``[C, H, W]``. + * Output + * A blob with the same shape as each input. + */ +message AverageLayerParams { +} + +/** + * A layer that computes the elementwise maximum over the inputs. + * :: + * y = MaxLayer(x1,x2,...) + * Requires multiple inputs and produces 1 output. + * + * Input + * Multiple blobs, each with shape ``[C]`` or ``[C, H, W]``. + * Output + * A blob with the same shape as each input. + */ +message MaxLayerParams { +} + +/** + * A layer that computes the elementwise minimum over the inputs. + * :: + * y = MinLayer(x1,x2,...) + * Requires multiple inputs and produces 1 output. + * + * Input + * Multiple blobs, each with shape ``[C]`` or ``[C, H, W]``. + * Output + * A blob with the same shape as each input. + */ +message MinLayerParams { +} + +/** + * A layer that computes the dot product of two vectors. + * :: + * y = DotProductLayer(x1,x2) + * Requires 2 inputs and produces 1 output. + * + * Input + * Two blobs with shape ``[C]``. + * Output + * A scalar. + */ +message DotProductLayerParams { + /** + * If true, inputs are normalized first, + * thereby computing the cosine similarity. + */ + bool cosineSimilarity = 1; +} + +/** + * A layer that performs mean variance normalization. + * :: + * y = MeanVarianceNormalizeLayer(x) + * Requires 1 input and produces 1 output. + * + * Input + * A blob with shape ``[C]`` or ``[C, H, W]``. + * Output + * A blob with the same shape as the input. + * + * If ``acrossChannels == true`` + * normalization is performed on flattened input. + * + * If ``acrossChannels == false`` + * normalization is performed within a channel, + * across spatial dimensions. + */ +message MeanVarianceNormalizeLayerParams { + /** + * If true, mean and variance are computed across channels. + */ + bool acrossChannels = 1; + + /** + * If false, only mean is subtracted. + */ + bool normalizeVariance = 2; + + /** + * A small constant to avoid division by 0 while normalizing variance. + * Defaults to ``1e-6`` if not set or set to ``0``. + */ + float epsilon = 3; +} + +/** + * A layer that repeats a sequence. + * :: + * y = SequenceRepeatLayer(x) + * Requires 1 input and produces 1 output. + * + * Input + * A sequence of blobs, i.e. shape is either ``[Seq, C]`` or ``[Seq, C, H, W]``. + * Output + * A sequence of length ``nRepetitions * Seq`` + * with shape corresponding to the input, + * i.e. shape is either ``[nRepetitions * Seq, C]`` or ``[nRepetitions * Seq, C, H, W]``. + */ +message SequenceRepeatLayerParams { + /** + * Number of repetitions. + * Defaults to ``1`` if not set or set to ``0``. + */ + uint64 nRepetitions = 1; +} + +/// Recurrent Layers +/// ---------------- + +/* + * The following activations are supported with recurrent layers: + * - Linear + * - Sigmoid + * - Tanh + * - ReLU + * - Scaled Hyperbolic Tangent: alpha * tanh(beta * x), currently only supported for alpha = 1.7159, beta = 2/3 + * - Hard Sigmoid: min(max(alpha * x + beta, 0), 1), currently only supported for alpha = 0.2, beta = 0.5 + */ + +/** + * A simple recurrent layer. + * :: + * y_t = SimpleRecurrentLayer(x_t, y_{t-1}) + * + * Input + * A sequence of vectors of size ``inputVectorSize`` + * with shape ``[Seq, inputVectorSize]``. + * Output + * A vector of size ``outputVectorSize``. It is either the final output or a sequence of outputs at all time steps. + * - Output Shape: ``[1,outputVectorSize]`` , if ``sequenceOutput == false`` + * - Output Shape: ``[Seq,outputVectorSize]`` , if ``sequenceOutput == true`` + * + * This layer is described by the following equation: + * + * .. math:: + * \boldsymbol{y_t} = f(\mathrm{clip}(W \boldsymbol{x_t} + \ + * R \boldsymbol{y_{t-1}} + b)) + * + * - ``W`` is a 2-dimensional weight matrix + * (``[outputVectorSize, inputVectorSize]``, row-major) + * - ``R`` is a 2-dimensional recursion matrix + * (``[outputVectorSize, outputVectorSize]``, row-major) + * - ``b`` is a 1-dimensional bias vector (``[outputVectorSize]``) + * - ``f()`` is an activation + * - ``clip()`` is a function that constrains values between ``[-50.0, 50.0]`` + */ +message SimpleRecurrentLayerParams { + uint64 inputVectorSize = 1; /// The size of the input vectors. + uint64 outputVectorSize = 2; /// The size of the output vectors. + + /** + * Activations supported are Linear, Sigmoid, Tanh, ReLU, Scaled Tanh (alpha = 1.71, beta = 2/3), Hard sigmoid (alpha = 0.2, beta = 0.5) + */ + ActivationParams activation = 10; /// The activation function. + + /** + If false output is just the result after final state update. + If true, output is a sequence, containing outputs at all time steps. + */ + bool sequenceOutput = 15; + + bool hasBiasVector = 20; /// If false, no bias is added. + + WeightParams weightMatrix = 30; /// Weight matrix W. + WeightParams recursionMatrix = 31; /// Recursion Weight matrix R. + WeightParams biasVector = 32; /// Bias vector b. + + bool reverseInput = 100; + // If true, then the node processes the input sequence from right to left +} + +/** + * Gated-Recurrent Unit (GRU) Layer + * :: + * y_t = GRULayer(x_t, y_{t-1}) + * + * Input + * A sequence of vectors of size ``inputVectorSize`` + * with shape ``[Seq, inputVectorSize]``. + * Output + * A vector of size ``outputVectorSize``. It is either the final output or a sequence of outputs at all time steps. + * - Output Shape: ``[1,outputVectorSize]`` , if ``sequenceOutput == false`` + * - Output Shape: ``[Seq,outputVectorSize]`` , if ``sequenceOutput == true`` + * + * This layer is described by the following equations: + * + * Update Gate + * .. math:: + * \boldsymbol{z_t} = \ + * f(\mathrm{clip}(W_z \boldsymbol{x_t} + \ + * R_z \boldsymbol{y_{t-1}} + b_z) + * + * Reset Gate + * .. math:: + * \boldsymbol{r_t} = \ + * f(\mathrm{clip}(W_r \boldsymbol{x_t} + \ + * R_r \boldsymbol{y_{t-1}} + b_r)) + * + * Cell Memory State + * .. math:: + * \boldsymbol{c_t} = \ + * \boldsymbol{y_{t-1}} \odot \boldsymbol{r_t} + * + * Output Gate + * .. math:: + * \boldsymbol{o_t} = \ + * g(\mathrm{clip}(W_o \boldsymbol{x_t} + \ + * R_o \boldsymbol{c_t} + b_o)) + * + * Output + * .. math:: + * \boldsymbol{y_t} = \ + * (1 - \boldsymbol{z_t}) \odot \boldsymbol{o_t} + \ + * \boldsymbol{z_t} \odot \boldsymbol{y_{t-1}} + * + * - ``W_z``, ``W_r``, ``W_o`` are 2-dimensional input weight matrices + * (``[outputVectorSize, inputVectorSize]``, row-major) + * - ``R_z``, ``R_r``, ``R_o`` are 2-dimensional recursion matrices + * (``[outputVectorSize, outputVectorSize]``, row-major) + * - ``b_z``, ``b_r``, ``b_o`` are 1-dimensional bias vectors + * (``[outputVectorSize]``) + * - ``f()``, ``g()`` are activations + * - ``clip()`` is a function that constrains values between ``[-50.0, 50.0]`` + * - ``⊙`` denotes the elementwise product of matrices + */ +message GRULayerParams { + uint64 inputVectorSize = 1; /// Size of the input vectors. + uint64 outputVectorSize = 2; /// Size of the output vectors. + + /** + * 2 element array representing activations [f(), g()] in that order. + * Typical values used = [sigmoid, tanh]. + * Activations supported are Linear, Sigmoid, Tanh, ReLU, Scaled Tanh (alpha = 1.71, beta = 2/3), Hard sigmoid (alpha = 0.2, beta = 0.5) + */ + repeated ActivationParams activations = 10; + + /** + * If false output is just the result after final state update. + * If true, output is a sequence, containing outputs at all time steps. + */ + bool sequenceOutput = 15; + + /** + * If false, no biases (``b_z``, ``b_r``, ``b_o``) are added. + */ + bool hasBiasVectors = 20; + + WeightParams updateGateWeightMatrix = 30; /// Weight Matrix W_z. + WeightParams resetGateWeightMatrix = 31; /// Weight Matrix W_r. + WeightParams outputGateWeightMatrix = 32; /// Weight Matrix W_o. + + WeightParams updateGateRecursionMatrix = 50; /// Recursion Weight Matrix R_z. + WeightParams resetGateRecursionMatrix = 51; /// Recursion Weight Matrix R_r. + WeightParams outputGateRecursionMatrix = 52; /// Recursion Weight Matrix R_o. + + WeightParams updateGateBiasVector = 70; /// Bias vector b_z. + WeightParams resetGateBiasVector = 71; /// Bias vector b_r. + WeightParams outputGateBiasVector = 72; /// Bias vector b_o. + + /// If true, then the node processes the input sequence from right to left + bool reverseInput = 100; +} + +/** + * Long short-term memory (LSTM) parameters. + * + * This is described by the following equations: + * + * Input Gate + * .. math:: + * \boldsymbol{i_t} = \ + * f(\mathrm{clip}(W_i \boldsymbol{x_t} + \ + * R_i \boldsymbol{y_{t-1}} + \ + * p_i \odot c_{t-1} + b_i)) + * + * Forget Gate + * .. math:: + * \boldsymbol{f_t} = \ + * f(\mathrm{clip}(W_f \boldsymbol{x_t} + \ + * R_f \boldsymbol{y_{t-1}} + \ + * p_f \odot c_{t-1} + b_f)) + * + * Block Input + * .. math:: + * \boldsymbol{z_t} = \ + * g(\mathrm{clip}(W_z \boldsymbol{x_t} + \ + * R_z \boldsymbol{y_{t-1}} + b_z)) + * + * Cell Memory State + * .. math:: + * \boldsymbol{c_t} = \ + * \boldsymbol{c_{t-1}} \odot \boldsymbol{f_t} + \ + * \boldsymbol{i_t} \odot \boldsymbol{z_t} + * + * Output Gate + * .. math:: + * \boldsymbol{o_t} = \ + * f(\mathrm{clip}(W_o \boldsymbol{x_t} + \ + * R_o \boldsymbol{y_{t-1}} + \ + * p_o \odot c_t + b_o)) + * + * Output + * .. math:: + * \boldsymbol{y_t} = \ + * h(\boldsymbol{c_t}) \odot \boldsymbol{o_t} + * + * - ``W_i``, ``W_f``, ``W_z``, ``W_o`` are 2-dimensional input weight matrices + * (``[outputVectorSize, inputVectorSize]``, row-major) + * - ``R_i``, ``R_f``, ``R_z``, ``R_o`` are 2-dimensional recursion matrices + * (``[outputVectorSize, outputVectorSize]``, row-major) + * - ``b_i``, ``b_f``, ``b_z``, ``b_o`` are 1-dimensional bias vectors + * (``[outputVectorSize]``) + * - ``p_``, ``p_f``, ``p_o`` are 1-dimensional peephole vectors + * (``[outputVectorSize]``) + * - ``f()``, ``g()``, ``h()`` are activations + * - ``clip()`` is a function that constrains values between ``[-50.0, 50.0]`` + * - ``⊙`` denotes the elementwise product of matrices + */ +message LSTMParams { + /** + * If true, output is a sequence, containing outputs at all time steps. + * If false, output is just the result after final state update. + */ + bool sequenceOutput = 10; + + /** + * If false, no biases (``b_i``, ``b_f``, ``b_z``, ``b_o``) are added. + */ + bool hasBiasVectors = 20; + + /** + * If true, a vector of ``1`` values is added to ``b_f``. + */ + bool forgetBias = 30; + + /** + * If true, peephole vectors are included. + */ + bool hasPeepholeVectors = 40; + + /** + * If the coupled Input and Forget flag is on, the behaviour of + * ``c_t`` is changed to the following (i.e. forget gate is not used): + * + * .. math:: + * \boldsymbol{c_t} = \ + * \boldsymbol{c_{t-1}} \odot (1 - \boldsymbol{i_t}) + \ + * \boldsymbol{i_t} \odot \boldsymbol{z_t} + * + */ + bool coupledInputAndForgetGate = 50; + + /** + * Places a limit on the maximum and minimum values of ``c_t``. + * c_t = min(c_t, cellClipThreshold) + * c_t = max(c_t, -cellClipThreshold) + * If 0, it is set to its default value = 50.0. + */ + float cellClipThreshold = 60; +} + +/** + * Weights for long short-term memory (LSTM) layers + */ +message LSTMWeightParams { + WeightParams inputGateWeightMatrix = 1; /// Weight Matrix W_i. + WeightParams forgetGateWeightMatrix = 2; /// Weight Matrix W_f. + WeightParams blockInputWeightMatrix = 3; /// Weight Matrix W_z. + WeightParams outputGateWeightMatrix = 4; /// Weight Matrix W_o. + + WeightParams inputGateRecursionMatrix = 20; /// Recursion Weight Matrix R_i. + WeightParams forgetGateRecursionMatrix = 21; /// Recursion Weight Matrix R_f. + WeightParams blockInputRecursionMatrix = 22; /// Recursion Weight Matrix R_z. + WeightParams outputGateRecursionMatrix = 23; /// Recursion Weight Matrix R_o. + + //biases: + WeightParams inputGateBiasVector = 40; /// Bias vector b_i. + WeightParams forgetGateBiasVector = 41; /// Bias vector b_f. + WeightParams blockInputBiasVector = 42; /// Bias vector b_z. + WeightParams outputGateBiasVector = 43; /// Bias vector b_o. + + //peepholes: + WeightParams inputGatePeepholeVector = 60; /// Peephole vector p_i. + WeightParams forgetGatePeepholeVector = 61; /// Peephole vector p_f. + WeightParams outputGatePeepholeVector = 62; /// Peephole vector p_o. +} + +/** + * A unidirectional long short-term memory (LSTM) layer. + * :: + * (y_t, c_t) = UniDirectionalLSTMLayer(x_t, y_{t-1}, c_{t-1}) + * + * Input + * A sequence of vectors of size ``inputVectorSize`` + * with shape ``[Seq, inputVectorSize]``. + * Output + * A vector of size ``outputVectorSize``. It is either the final output or a sequence of outputs at all time steps. + * - Output Shape: ``[1,outputVectorSize]`` , if ``sequenceOutput == false`` + * - Output Shape: ``[Seq,outputVectorSize]`` , if ``sequenceOutput == true`` + * + */ +message UniDirectionalLSTMLayerParams { + uint64 inputVectorSize = 1; /// Size of the input vectors. + uint64 outputVectorSize = 2; /// Size of the output vectors. + + /** + * 3 element array representing activations [f(),g(),h()] in that order. + * Typical values used = [sigmoid, tanh, tanh]. + * Activations supported are Linear, Sigmoid, Tanh, ReLU, Scaled Tanh (alpha = 1.71, beta = 2/3), Hard sigmoid (alpha = 0.2, beta = 0.5) + */ + repeated ActivationParams activations = 10; + + LSTMParams params = 15; + + LSTMWeightParams weightParams = 20; /// Weights, biases and peepholes. + + /// If true, then the node processes the input sequence from right to left + bool reverseInput = 100; +} + +/** + * Bidirectional long short-term memory (LSTM) layer + * :: + * (y_t, c_t, y_t_reverse, c_t_reverse) = BiDirectionalLSTMLayer(x_t, y_{t-1}, c_{t-1}, y_{t-1}_reverse, c_{t-1}_reverse) + * + * Input + * A sequence of vectors of size ``inputVectorSize`` + * with shape ``[Seq, inputVectorSize]``. + * Output + * A vector of size ``2 * outputVectorSize``. It is either the final output or a sequence of outputs at all time steps. + * - Output Shape: ``[1, 2 * outputVectorSize]`` , if ``sequenceOutput == false`` + * - Output Shape: ``[Seq, 2 * outputVectorSize]`` , if ``sequenceOutput == true`` + * + * The first LSTM operates on the input sequence in the forward direction. + * The second LSTM operates on the input sequence in the reverse direction. + * + * Example: given the input sequence ``[x_1, x_2, x_3]``, + * where ``x_i`` are vectors at time index ``i``: + * + * The forward LSTM output is ``[yf_1, yf_2, yf_3]``, + * + * where ``yf_i`` are vectors of size ``outputVectorSize``: + * + * - ``yf_1`` is the output at the end of sequence {``x_1``} + * - ``yf_2`` is the output at the end of sequence {``x_1``, ``x_2``} + * - ``yf_3`` is the output at the end of sequence {``x_1``, ``x_2``, ``x_3``} + * + * The backward LSTM output: ``[yb_1, yb_2, yb_3]``, + * + * where ``yb_i`` are vectors of size ``outputVectorSize``: + * + * - ``yb_1`` is the output at the end of sequence {``x_3``} + * - ``yb_2`` is the output at the end of sequence {``x_3``, ``x_2``} + * - ``yb_3`` is the output at the end of sequence {``x_3``, ``x_2``, ``x_1``} + * + * Output of the bi-dir layer: + * + * - if ``sequenceOutput = True`` : { ``[yf_1, yb_3]``, ``[yf_2, yb_2]``, ``[yf_3, yb_1]`` } + * - if ``sequenceOutput = False`` : { ``[yf_3, yb_3]`` } + */ +message BiDirectionalLSTMLayerParams { + /** + * Size of the input vectors. + */ + uint64 inputVectorSize = 1; + /** + * Size of the outputs vectors. + * It is same for both forward and backward LSTMs. + */ + uint64 outputVectorSize = 2; + + /** + * 3 element array representing activations [f(),g(),h()] in that order. + * Typical values used = [sigmoid, tanh, tanh]. + * Activations supported are Linear, Sigmoid, Tanh, ReLU, Scaled Tanh (alpha = 1.71, beta = 2/3), Hard sigmoid (alpha = 0.2, beta = 0.5) + */ + repeated ActivationParams activationsForwardLSTM = 10; + /** + * Currently, backward LSTM activations + * must be same as the ones for the forward LSTM. + */ + repeated ActivationParams activationsBackwardLSTM = 11; + + /** + * Common parameters shared by the forward and backward LSTMs. + */ + LSTMParams params = 15; + + /** + * Weights and biases. + * Must be a length 2 message, + * for the forward and backward LSTM respectively. + */ + repeated LSTMWeightParams weightParams = 20; +} + +/// Neural Network Specializations +/// ------------------------------ + +/** + * A neural network specialized as a classifier. + */ +message NeuralNetworkClassifier { + repeated NeuralNetworkLayer layers = 1; + repeated NeuralNetworkPreprocessing preprocessing = 2; + + /** + * Mapping from indexed vector of probabilities to class label + */ + oneof ClassLabels { + StringVector stringClassLabels = 100; + Int64Vector int64ClassLabels = 101; + } +} + +/** + * A neural network specialized as a regressor. + */ +message NeuralNetworkRegressor { + repeated NeuralNetworkLayer layers = 1; + repeated NeuralNetworkPreprocessing preprocessing = 2; +} diff --git a/src/interfaces/coreml/Normalizer.proto b/src/interfaces/coreml/Normalizer.proto new file mode 100644 index 00000000000..627f7e2e3af --- /dev/null +++ b/src/interfaces/coreml/Normalizer.proto @@ -0,0 +1,38 @@ +// Copyright (c) 2017, Apple Inc. All rights reserved. +// +// Use of this source code is governed by a BSD-3-clause license that can be +// found in LICENSE.txt or at https://opensource.org/licenses/BSD-3-Clause + +syntax = "proto3"; +option optimize_for = LITE_RUNTIME; + +package CoreML.Specification; + +/** + * A normalization preprocessor. + */ +message Normalizer { + /** + * There are three normalization modes, + * which have the corresponding formulas: + * + * Max + * .. math:: + * max(x_i) + * + * L1 + * .. math:: + * z = ||x||_1 = \sum_{i=1}^{n} |x_i| + * + * L2 + * .. math:: + * z = ||x||_2 = \sqrt{\sum_{i=1}^{n} x_i^2} + */ + enum NormType { + LMax = 0; + L1 = 1; + L2 = 2; + } + + NormType normType = 1; +} diff --git a/src/interfaces/coreml/OneHotEncoder.proto b/src/interfaces/coreml/OneHotEncoder.proto new file mode 100644 index 00000000000..f47cf281662 --- /dev/null +++ b/src/interfaces/coreml/OneHotEncoder.proto @@ -0,0 +1,41 @@ +// Copyright (c) 2017, Apple Inc. All rights reserved. +// +// Use of this source code is governed by a BSD-3-clause license that can be +// found in LICENSE.txt or at https://opensource.org/licenses/BSD-3-Clause + +syntax = "proto3"; +option optimize_for = LITE_RUNTIME; + +import public "DataStructures.proto"; + +package CoreML.Specification; + +/** + * Transforms a categorical feature into an array. The array will be all + * zeros expect a single entry of one. + * + * Each categorical value will map to an index, this mapping is given by + * either the ``stringCategories`` parameter or the ``int64Categories`` + * parameter. + */ +message OneHotEncoder { + enum HandleUnknown { + ErrorOnUnknown = 0; + IgnoreUnknown = 1; // Output will be all zeros for unknown values. + } + + /** + * Mapping to be used for the encoding. The position of the category in + * the below vector determines where the single one entry will be in the + * output. + */ + oneof CategoryType { + StringVector stringCategories = 1; + Int64Vector int64Categories = 2; + } + + // Output can be a dictionary with only one entry, instead of an array. + bool outputSparse = 10; + + HandleUnknown handleUnknown = 11; +} diff --git a/src/interfaces/coreml/README.rst b/src/interfaces/coreml/README.rst new file mode 100644 index 00000000000..97ea0b1dac2 --- /dev/null +++ b/src/interfaces/coreml/README.rst @@ -0,0 +1,10 @@ +================================== +Core ML Model Format Specification +================================== + +This directory contains the protobuf message definitions +that comprise the Core ML model document (``.mlmodel``) format. + +The top-level message is ``Model``, which is defined in ``Model.proto``. +Other message types describe data structures, feature types, +feature engineering model types, and predictive model types. \ No newline at end of file diff --git a/src/interfaces/coreml/SVM.proto b/src/interfaces/coreml/SVM.proto new file mode 100644 index 00000000000..932a4ec2166 --- /dev/null +++ b/src/interfaces/coreml/SVM.proto @@ -0,0 +1,195 @@ +// Copyright (c) 2017, Apple Inc. All rights reserved. +// +// Use of this source code is governed by a BSD-3-clause license that can be +// found in LICENSE.txt or at https://opensource.org/licenses/BSD-3-Clause + +syntax = "proto3"; +option optimize_for = LITE_RUNTIME; + +import public "DataStructures.proto"; + +package CoreML.Specification; + +/// Kernel Definitions +/// ------------------ + +/** + * A linear kernel. + * + * This function has the following formula: + * + * .. math:: + * K(\boldsymbol{x}, \boldsymbol{x'}) = \boldsymbol{x}^T \boldsymbol{x'} + */ +message LinearKernel { +} + +/** + * A Gaussian radial basis function (RBF) kernel. + * + * This function has the following formula: + * + * .. math:: + * K(\boldsymbol{x}, \boldsymbol{x'}) = \ + * \exp(-\gamma || \boldsymbol{x} - \boldsymbol{x'} ||^2 ) + * + */ +message RBFKernel { + double gamma = 1; +} + +/** + * A polynomial kernel. + * + * This function has the following formula: + * + * .. math:: + * K(\boldsymbol{x}, \boldsymbol{x'}) = \ + * (\gamma \boldsymbol{x}^T \boldsymbol{x'} + c)^{degree} + */ +message PolyKernel { + int32 degree = 1; + double c = 2; + double gamma = 3; +} + +/** + * A sigmoid kernel. + * + * This function has the following formula: + * + * .. math:: + * K(\boldsymbol{x}, \boldsymbol{x'}) = \ + * \tanh(\gamma \boldsymbol{x}^T \boldsymbol{x'} + c) + */ +message SigmoidKernel { + double gamma = 1; + double c = 2; +} + +/** + * A kernel. + */ +message Kernel { + oneof kernel { + LinearKernel linearKernel = 1; + RBFKernel rbfKernel = 2; + PolyKernel polyKernel = 3; + SigmoidKernel sigmoidKernel = 4; + } +} + + +/// Support Vector Definitions +/// -------------------------- + +/** + * A sparse node. + */ +message SparseNode { + int32 index = 1; // 1-based indexes, like libsvm + double value = 2; +} + +/** + * A sparse vector. + */ +message SparseVector { + repeated SparseNode nodes = 1; +} + +/** + * One or more sparse support vectors. + */ +message SparseSupportVectors { + repeated SparseVector vectors = 1; +} + +/** + * A dense vector. + */ +message DenseVector { + repeated double values = 1; +} + +/** + * One or more dense support vectors. + */ +message DenseSupportVectors { + repeated DenseVector vectors = 1; +} + +/** + * One or more coefficients. + */ +message Coefficients { + repeated double alpha = 1; +} + +/** + * A support vector regressor. + */ +message SupportVectorRegressor { + Kernel kernel = 1; + + // Support vectors, either sparse or dense format + oneof supportVectors { + SparseSupportVectors sparseSupportVectors = 2; + DenseSupportVectors denseSupportVectors = 3; + } + + // Coefficients, one for each support vector + Coefficients coefficients = 4; + + double rho = 5; +} + +/** + * A support vector classifier + */ +message SupportVectorClassifier { + Kernel kernel = 1; + + /** + * The number of support vectors for each class. + */ + repeated int32 numberOfSupportVectorsPerClass = 2; + + /** + * The support vectors, in either sparse or dense format. + */ + oneof supportVectors { + SparseSupportVectors sparseSupportVectors = 3; + DenseSupportVectors denseSupportVectors = 4; + } + + /** + * The coefficients, essentially a two dimensional array of + * size: (numberOfClasses-1) by (total number of support vectors) + */ + repeated Coefficients coefficients = 5; + + /** + * Constants for decision function, + * with K*(K-1) / 2 elements, + * where K is the number of classes. + */ + repeated double rho = 6; + + /** + * Pairwise probability information for A vs B classifier. + * Total of K*(K-1)/2 elements where K is the number of classes. + * These fields are optional, + * and only required if you want probabilities or multi class predictions. + */ + repeated double probA = 7; + repeated double probB = 8; + + /** + * Class label mapping. + */ + oneof ClassLabels { + StringVector stringClassLabels = 100; + Int64Vector int64ClassLabels = 101; + } +} diff --git a/src/interfaces/coreml/Scaler.proto b/src/interfaces/coreml/Scaler.proto new file mode 100644 index 00000000000..f0e13d54be2 --- /dev/null +++ b/src/interfaces/coreml/Scaler.proto @@ -0,0 +1,34 @@ +// Copyright (c) 2017, Apple Inc. All rights reserved. +// +// Use of this source code is governed by a BSD-3-clause license that can be +// found in LICENSE.txt or at https://opensource.org/licenses/BSD-3-Clause + +syntax = "proto3"; +option optimize_for = LITE_RUNTIME; + +package CoreML.Specification; + +/** + * A scaling operation. + * + * This function has the following formula: + * + * .. math:: + * f(x) = scaleValue \cdot (x + shiftValue) + * + * If the ``scaleValue`` is not given, the default value 1 is used. + * If the ``shiftValue`` is not given, the default value 0 is used. + * + * If ``scaleValue`` and ``shiftValue`` are each a single value + * and the input is an array, then the scale and shift are applied + * to each element of the array. + * + * If the input is an integer, then it is converted to a double to + * perform the scaling operation. If the output type is an integer, + * then it is cast to an integer. If that cast is lossy, then an + * error is generated. + */ +message Scaler { + repeated double shiftValue = 1; + repeated double scaleValue = 2; +} diff --git a/src/interfaces/coreml/TreeEnsemble.proto b/src/interfaces/coreml/TreeEnsemble.proto new file mode 100644 index 00000000000..defebee9885 --- /dev/null +++ b/src/interfaces/coreml/TreeEnsemble.proto @@ -0,0 +1,161 @@ +// Copyright (c) 2017, Apple Inc. All rights reserved. +// +// Use of this source code is governed by a BSD-3-clause license that can be +// found in LICENSE.txt or at https://opensource.org/licenses/BSD-3-Clause + +/** + * Each tree is a collection of nodes, + * each of which is identified by a unique identifier. + * + * Each node is either a branch or a leaf node. + * A branch node evaluates a value according to a behavior; + * if true, the node identified by ``true_child_node_id`` is evaluated next, + * if false, the node identified by ``false_child_node_id`` is evaluated next. + * A leaf node adds the evaluation value to the base prediction value + * to get the final prediction. + * + * A tree must have exactly one root node, + * which has no parent node. + * A tree must not terminate on a branch node. + * All leaf nodes must be accessible + * by evaluating one or more branch nodes in sequence, + * starting from the root node. + */ + +syntax = "proto3"; +option optimize_for = LITE_RUNTIME; + +import public "DataStructures.proto"; + +package CoreML.Specification; + +/** + * A tree ensemble post-evaluation transform. + */ +enum TreeEnsemblePostEvaluationTransform { + NoTransform = 0; + Classification_SoftMax = 1; + Regression_Logistic = 2; + Classification_SoftMaxWithZeroClassReference = 3; +} + +/** + * Tree ensemble parameters. + */ +message TreeEnsembleParameters { + message TreeNode { + uint64 treeId = 1; + uint64 nodeId = 2; + + enum TreeNodeBehavior { + BranchOnValueLessThanEqual = 0; + BranchOnValueLessThan = 1; + BranchOnValueGreaterThanEqual = 2; + BranchOnValueGreaterThan = 3; + BranchOnValueEqual = 4; + BranchOnValueNotEqual = 5; + LeafNode = 6; + } + + /** + * The branch mode parameters. + * + * If branch is false, + * then the parameters in this section must be filled in + * to determine how the branching functions. + */ + TreeNodeBehavior nodeBehavior = 3; + + /** + * If the node behavior mode is a branch mode, + * then these values must be filled in. + */ + uint64 branchFeatureIndex = 10; + double branchFeatureValue = 11; + uint64 trueChildNodeId = 12; + uint64 falseChildNodeId = 13; + bool missingValueTracksTrueChild = 14; + + /** + * The leaf mode. + * + * If ``nodeBahavior`` == ``LeafNode``, + * then the evaluationValue is added to the base prediction value + * in order to get the final prediction. + * To support multiclass classification + * as well as regression and binary classification, + * the evaluation value is encoded here as a sparse vector, + * with evaluationIndex being the index of the base vector + * that evaluation value is added to. + * In the single class case, + * it is expected that evaluationIndex is exactly 0. + */ + message EvaluationInfo { + uint64 evaluationIndex = 1; + double evaluationValue = 2; + } + + repeated EvaluationInfo evaluationInfo = 20; + + /** + * The relative hit rate of a node for optimization purposes. + * + * This value has no effect on the accuracy of the result; + * it allows the tree to optimize for frequent branches. + * The value is relative, + * compared to the hit rates of other branch nodes. + * + * You typically use a proportion of training samples + * that reached this node + * or some similar metric to derive this value. + */ + double relativeHitRate = 30; + } + + repeated TreeNode nodes = 1; + + /** + * The number of prediction dimensions or classes in the model. + * + * All instances of ``evaluationIndex`` in a leaf node + * must be less than this value, + * and the number of values in the ``basePredictionValue`` field + * must be equal to this value. + * + * For regression, + * this is the dimension of the prediction. + * For classification, + * this is the number of classes. + */ + uint64 numPredictionDimensions = 2; + + /** + * The base prediction value. + * + * The number of values in this must match + * the default values of the tree model. + */ + repeated double basePredictionValue = 3; +} + +/** + * A tree ensemble classifier. + */ +message TreeEnsembleClassifier { + TreeEnsembleParameters treeEnsemble = 1; + TreeEnsemblePostEvaluationTransform postEvaluationTransform = 2; + + // Required class label mapping + oneof ClassLabels { + StringVector stringClassLabels = 100; + Int64Vector int64ClassLabels = 101; + } +} + +/** + * A tree ensemble regressor. + */ +message TreeEnsembleRegressor { + TreeEnsembleParameters treeEnsemble = 1; + TreeEnsemblePostEvaluationTransform postEvaluationTransform = 2; +} diff --git a/src/interfaces/csharp/CMakeLists.txt b/src/interfaces/csharp/CMakeLists.txt new file mode 100644 index 00000000000..2384ed5a5d4 --- /dev/null +++ b/src/interfaces/csharp/CMakeLists.txt @@ -0,0 +1,23 @@ +FIND_PACKAGE(CSharp REQUIRED) +UNSET(TARGET_SWIGFLAGS) + +include(ShogunInterfaces) + +GENERATE_INTERFACE_TARGET(csharp ${CMAKE_CURRENT_SOURCE_DIR} "") + +ADD_CUSTOM_COMMAND(TARGET interface_csharp + POST_BUILD + COMMAND "${CSHARP_COMPILER}" /t:library *.cs /out:shogun.dll + COMMENT "Creating shogun.dll") +IF(NOT CSHARP_INSTALL_DIR) + SET(CSHARP_INSTALL_DIR "lib/cli/shogun") +ENDIF(NOT CSHARP_INSTALL_DIR) + +SET(INTERFACE_CSHARP_BUILD_DIR ${CMAKE_CURRENT_BINARY_DIR} PARENT_SCOPE) +INSTALL(TARGETS interface_csharp + LIBRARY DESTINATION ${CSHARP_INSTALL_DIR} + COMPONENT csharp) + +INSTALL(FILES ${CMAKE_CURRENT_BINARY_DIR}/shogun.dll + DESTINATION ${CSHARP_INSTALL_DIR} + COMPONENT csharp) diff --git a/src/interfaces/java_modular/sg_print_functions.cpp b/src/interfaces/csharp/sg_print_functions.cpp similarity index 81% rename from src/interfaces/java_modular/sg_print_functions.cpp rename to src/interfaces/csharp/sg_print_functions.cpp index ea4d1d854a2..b9637930f62 100644 --- a/src/interfaces/java_modular/sg_print_functions.cpp +++ b/src/interfaces/csharp/sg_print_functions.cpp @@ -15,7 +15,3 @@ void sg_global_print_error(FILE* target, const char* str) { fprintf(target, "%s", str); } - -void sg_global_cancel_computations(bool &delayed, bool &immediately) -{ -} diff --git a/src/interfaces/csharp_modular/swig_typemaps.i b/src/interfaces/csharp/swig_typemaps.i similarity index 100% rename from src/interfaces/csharp_modular/swig_typemaps.i rename to src/interfaces/csharp/swig_typemaps.i diff --git a/src/interfaces/csharp_modular/CMakeLists.txt b/src/interfaces/csharp_modular/CMakeLists.txt deleted file mode 100644 index 439a078ce1b..00000000000 --- a/src/interfaces/csharp_modular/CMakeLists.txt +++ /dev/null @@ -1,23 +0,0 @@ -FIND_PACKAGE(CSharp REQUIRED) -UNSET(TARGET_SWIGFLAGS) - -include(CommonModularInterface) - -GENERATE_MODULAR_TARGET(csharp ${CMAKE_CURRENT_SOURCE_DIR} "") - -ADD_CUSTOM_COMMAND(TARGET csharp_modular - POST_BUILD - COMMAND "${CSHARP_COMPILER}" /t:library *.cs /out:modshogun.dll - COMMENT "Creating modshogun.dll") -IF(NOT CSHARP_INSTALL_DIR) - SET(CSHARP_INSTALL_DIR "lib/cli/shogun") -ENDIF(NOT CSHARP_INSTALL_DIR) - -SET(CSHARP_MODULAR_BUILD_DIR ${CMAKE_CURRENT_BINARY_DIR} PARENT_SCOPE) -INSTALL(TARGETS csharp_modular - LIBRARY DESTINATION ${CSHARP_INSTALL_DIR} - COMPONENT csharp) - -INSTALL(FILES ${CMAKE_CURRENT_BINARY_DIR}/modshogun.dll - DESTINATION ${CSHARP_INSTALL_DIR} - COMPONENT csharp) diff --git a/src/interfaces/java_modular/CMakeLists.txt b/src/interfaces/java/CMakeLists.txt similarity index 80% rename from src/interfaces/java_modular/CMakeLists.txt rename to src/interfaces/java/CMakeLists.txt index 89e61ac56ab..6cd45381232 100644 --- a/src/interfaces/java_modular/CMakeLists.txt +++ b/src/interfaces/java/CMakeLists.txt @@ -10,20 +10,20 @@ IF(JBLAS) SET(HAVE_JBLAS 1) ELSE() MESSAGE(FATAL_ERROR "JBlas could not be found!\n" - "It is required for java modular interface!!!") + "It is required for java interface!") ENDIF() SET(TARGET_SWIGFLAGS "-package\;org.shogun") -include(CommonModularInterface) +include(ShogunInterfaces) include_directories(${JNI_INCLUDE_DIRS}) -GENERATE_MODULAR_TARGET(java ${CMAKE_CURRENT_SOURCE_DIR} "") +GENERATE_INTERFACE_TARGET(java ${CMAKE_CURRENT_SOURCE_DIR} "") set(CMAKE_JAVA_TARGET_OUTPUT_NAME shogun) SET(CMAKE_JAVA_INCLUDE_PATH ${JBLAS}) -ADD_CUSTOM_COMMAND(TARGET java_modular POST_BUILD +ADD_CUSTOM_COMMAND(TARGET interface_java POST_BUILD COMMAND cmake -E echo "Compiling Java files..." COMMAND cmake -E make_directory ${CMAKE_CURRENT_BINARY_DIR}/shogun_jar COMMAND ${Java_JAVAC_EXECUTABLE} *.java @@ -36,7 +36,7 @@ ADD_CUSTOM_COMMAND(TARGET java_modular POST_BUILD #ADD_JAR(shogun_jar "${CMAKE_CURRENT_BINARY_DIR}/*.java") -#add_dependencies(shogun_jar java_modular) +#add_dependencies(shogun_jar interface_java) IF(NOT JAVA_INSTALL_DIR) SET(JAVA_INSTALL_DIR "share/java") @@ -46,8 +46,8 @@ IF(NOT JNI_INSTALL_DIR) SET(JNI_INSTALL_DIR "lib/cli/shogun") ENDIF(NOT JNI_INSTALL_DIR) -SET(JAVA_MODULAR_BUILD_DIR ${CMAKE_CURRENT_BINARY_DIR} PARENT_SCOPE) -INSTALL(TARGETS java_modular +SET(INTERFACE_JAVA_BUILD_DIR ${CMAKE_CURRENT_BINARY_DIR} PARENT_SCOPE) +INSTALL(TARGETS interface_java LIBRARY DESTINATION ${JNI_INSTALL_DIR} COMPONENT java) diff --git a/src/interfaces/java_modular/build.xml b/src/interfaces/java/build.xml similarity index 100% rename from src/interfaces/java_modular/build.xml rename to src/interfaces/java/build.xml diff --git a/src/interfaces/ruby_modular/sg_print_functions.cpp b/src/interfaces/java/sg_print_functions.cpp similarity index 81% rename from src/interfaces/ruby_modular/sg_print_functions.cpp rename to src/interfaces/java/sg_print_functions.cpp index ea4d1d854a2..b9637930f62 100644 --- a/src/interfaces/ruby_modular/sg_print_functions.cpp +++ b/src/interfaces/java/sg_print_functions.cpp @@ -15,7 +15,3 @@ void sg_global_print_error(FILE* target, const char* str) { fprintf(target, "%s", str); } - -void sg_global_cancel_computations(bool &delayed, bool &immediately) -{ -} diff --git a/src/interfaces/java_modular/swig_typemaps.i b/src/interfaces/java/swig_typemaps.i similarity index 100% rename from src/interfaces/java_modular/swig_typemaps.i rename to src/interfaces/java/swig_typemaps.i diff --git a/src/interfaces/lua_modular/CMakeLists.txt b/src/interfaces/lua/CMakeLists.txt similarity index 64% rename from src/interfaces/lua_modular/CMakeLists.txt rename to src/interfaces/lua/CMakeLists.txt index 37ab57d903a..cdad9e1e927 100644 --- a/src/interfaces/lua_modular/CMakeLists.txt +++ b/src/interfaces/lua/CMakeLists.txt @@ -2,16 +2,16 @@ FIND_PACKAGE(PkgConfig) pkg_search_module(LUA REQUIRED lua lua5.1) IF (NOT LUA_FOUND) MESSAGE(FATAL_ERROR "Lua could not be found!\n" - "Lua is required for compiling lua modular interface!") + "Lua is required for compiling lua interface!") ENDIF () SET(LUA_EXECUTABLE lua) UNSET(TARGET_SWIGFLAGS) -include(CommonModularInterface) +include(ShogunInterfaces) include_directories(${LUA_INCLUDE_DIRS}) -SET(LUA_MODULAR_BUILD_DIR ${CMAKE_CURRENT_BINARY_DIR} PARENT_SCOPE) -GENERATE_MODULAR_TARGET(lua ${CMAKE_CURRENT_SOURCE_DIR} "${LUA_LDFLAGS}") +SET(INTERFACE_LUA_BUILD_DIR ${CMAKE_CURRENT_BINARY_DIR} PARENT_SCOPE) +GENERATE_INTERFACE_TARGET(lua ${CMAKE_CURRENT_SOURCE_DIR} "${LUA_LDFLAGS}") IF (NOT LIB_INSTALL_DIR) SET(LIB_INSTALL_DIR lib) @@ -19,6 +19,6 @@ ENDIF() STRING(REGEX REPLACE "([0-9]*.[0-9]*).[0-9]*" "${LIB_INSTALL_DIR}/lua/\\1" LUA_EXTENSION_DIR ${LUA_VERSION}) -INSTALL(TARGETS lua_modular +INSTALL(TARGETS interface_lua LIBRARY DESTINATION ${LUA_EXTENSION_DIR} COMPONENT lua) diff --git a/src/interfaces/csharp_modular/sg_print_functions.cpp b/src/interfaces/lua/sg_print_functions.cpp similarity index 81% rename from src/interfaces/csharp_modular/sg_print_functions.cpp rename to src/interfaces/lua/sg_print_functions.cpp index ea4d1d854a2..b9637930f62 100644 --- a/src/interfaces/csharp_modular/sg_print_functions.cpp +++ b/src/interfaces/lua/sg_print_functions.cpp @@ -15,7 +15,3 @@ void sg_global_print_error(FILE* target, const char* str) { fprintf(target, "%s", str); } - -void sg_global_cancel_computations(bool &delayed, bool &immediately) -{ -} diff --git a/src/interfaces/lua_modular/swig_typemaps.i b/src/interfaces/lua/swig_typemaps.i similarity index 100% rename from src/interfaces/lua_modular/swig_typemaps.i rename to src/interfaces/lua/swig_typemaps.i diff --git a/src/interfaces/octave/CMakeLists.txt b/src/interfaces/octave/CMakeLists.txt new file mode 100644 index 00000000000..bbabf4af893 --- /dev/null +++ b/src/interfaces/octave/CMakeLists.txt @@ -0,0 +1,16 @@ +FIND_PACKAGE(Octave 3.6 REQUIRED) +SET(OCTAVE_APIVERSION ${OCTAVE_API_VERSION}) +UNSET(TARGET_SWIGFLAGS) + +include(ShogunInterfaces) +include_directories(${OCTAVE_INCLUDE_DIRS}) + +GENERATE_INTERFACE_TARGET(octave ${CMAKE_CURRENT_SOURCE_DIR} ${OCTAVE_LIBRARY}) +set_target_properties(interface_octave PROPERTIES PREFIX "") +set_target_properties(interface_octave PROPERTIES SUFFIX ".oct") +target_compile_definitions(interface_octave PRIVATE "OCTAVE_APIVERSION=${OCTAVE_API_VERSION}") + +SET(INTERFACE_OCTAVE_BINARY_DIR ${CMAKE_CURRENT_BINARY_DIR} PARENT_SCOPE) +install (TARGETS interface_octave + DESTINATION ${OCTAVE_OCT_LOCAL_API_FILE_DIR}/shogun + COMPONENT octave) diff --git a/src/interfaces/octave_modular/sg_print_functions.cpp b/src/interfaces/octave/sg_print_functions.cpp similarity index 86% rename from src/interfaces/octave_modular/sg_print_functions.cpp rename to src/interfaces/octave/sg_print_functions.cpp index c696a81e0d8..3c692039a26 100644 --- a/src/interfaces/octave_modular/sg_print_functions.cpp +++ b/src/interfaces/octave/sg_print_functions.cpp @@ -1,13 +1,13 @@ #include -#include +#include #include #include #include +#include #include #include #include -#include #include #include @@ -19,7 +19,7 @@ void sg_global_print_message(FILE* target, const char* str) void sg_global_print_warning(FILE* target, const char* str) { - if (target==stdout) + if (target == stdout) ::warning(str); else fprintf(target, "%s", str); @@ -29,7 +29,3 @@ void sg_global_print_error(FILE* target, const char* str) { fprintf(target, "%s", str); } - -void sg_global_cancel_computations(bool &delayed, bool &immediately) -{ -} diff --git a/src/interfaces/octave_modular/swig_typemaps.i b/src/interfaces/octave/swig_typemaps.i similarity index 100% rename from src/interfaces/octave_modular/swig_typemaps.i rename to src/interfaces/octave/swig_typemaps.i diff --git a/src/interfaces/octave_modular/CMakeLists.txt b/src/interfaces/octave_modular/CMakeLists.txt deleted file mode 100644 index 536c2b37b3d..00000000000 --- a/src/interfaces/octave_modular/CMakeLists.txt +++ /dev/null @@ -1,16 +0,0 @@ -FIND_PACKAGE(Octave 3.6 REQUIRED) -SET(OCTAVE_APIVERSION ${OCTAVE_API_VERSION}) -UNSET(TARGET_SWIGFLAGS) - -include(CommonModularInterface) -include_directories(${OCTAVE_INCLUDE_DIRS}) - -GENERATE_MODULAR_TARGET(octave ${CMAKE_CURRENT_SOURCE_DIR} ${OCTAVE_LIBRARY}) -set_target_properties(octave_modular PROPERTIES PREFIX "") -set_target_properties(octave_modular PROPERTIES SUFFIX ".oct") -target_compile_definitions(octave_modular PRIVATE "OCTAVE_APIVERSION=${OCTAVE_API_VERSION}") - -SET(OCTAVE_MODULAR_BINARY_DIR ${CMAKE_CURRENT_BINARY_DIR} PARENT_SCOPE) -install (TARGETS octave_modular - DESTINATION ${OCTAVE_OCT_LOCAL_API_FILE_DIR}/shogun - COMPONENT octave) diff --git a/src/interfaces/perl_modular/pdl.i b/src/interfaces/perl/pdl.i similarity index 100% rename from src/interfaces/perl_modular/pdl.i rename to src/interfaces/perl/pdl.i diff --git a/src/interfaces/perl/sg_print_functions.cpp b/src/interfaces/perl/sg_print_functions.cpp new file mode 100644 index 00000000000..a9e7089b519 --- /dev/null +++ b/src/interfaces/perl/sg_print_functions.cpp @@ -0,0 +1,42 @@ +/* For co-existence with stdio only */ + +#ifdef __cplusplus +extern "C" { +#endif + +#define PERLIO_NOT_STDIO 0 +#include + +#ifdef __cplusplus +} +#endif + +#include + +void sg_global_print_message(FILE* target, const char* str) +{ + fprintf(target, "%s", str); +} + +void sg_global_print_warning(FILE* target, const char* str) +{ + SV* err = get_sv("@", GV_ADD); + if (target == stdout) + { + if (sv_isobject(err)) + pdl_warn(0); + else + croak("%s", SvPV_nolen(err)); + } + else + fprintf(target, "%s", str); +} + +void sg_global_print_error(FILE* target, const char* str) +{ + if (target == stdout) // PerlIO_stdout()) //"ERRSV" ($@) + croak(str); + // SWIG_croak(str); + else + fprintf(target, "%s", str); +} diff --git a/src/interfaces/perl_modular/swig_typemaps.i b/src/interfaces/perl/swig_typemaps.i similarity index 100% rename from src/interfaces/perl_modular/swig_typemaps.i rename to src/interfaces/perl/swig_typemaps.i diff --git a/src/interfaces/perl_modular/sg_print_functions.cpp b/src/interfaces/perl_modular/sg_print_functions.cpp deleted file mode 100644 index 7cb0e25b896..00000000000 --- a/src/interfaces/perl_modular/sg_print_functions.cpp +++ /dev/null @@ -1,72 +0,0 @@ -/* For co-existence with stdio only */ - -#ifdef __cplusplus -extern "C" { -#endif - -#define PERLIO_NOT_STDIO 0 -#include - -#ifdef __cplusplus -} -#endif - - -#include - -void sg_global_print_message(FILE* target, const char* str) -{ - fprintf(target, "%s", str); -} - -void sg_global_print_warning(FILE* target, const char* str) -{ - SV *err = get_sv("@", GV_ADD); - if (target == stdout) - { - if (sv_isobject(err)) - pdl_warn(0); - else - croak("%s", SvPV_nolen(err)); - } - else - fprintf(target, "%s", str); -} - -void sg_global_print_error(FILE* target, const char* str) -{ - if (target == stdout) // PerlIO_stdout()) //"ERRSV" ($@) - croak(str); - //SWIG_croak(str); - else - fprintf(target, "%s", str); -} - -//PTZ121009 used in threads...so cannot access stdin like this... -// why not checking kill stuff??? -void sg_global_cancel_computations(bool &delayed, bool &immediately) -{ -#if 0 - using namespace shogun; - - dTHX; /* fetch context */ - //PerlIO_init(pTHX); - // PerlIO *f = PerlIO_stdin(); crashes in Perl_csighandler () from /usr/lib/libperl.so.5.14 - - if(!f) {return;} - if(PerlIO_flush(f)) //check signal - { - SG_SPRINT("\nImmediately return to matlab prompt / Prematurely finish computations / Do nothing (I/P/D)? "); - char answer= PerlIO_getc(f); - if (answer == 'I') - immediately=true; - else if (answer == 'P') - { - PerlIO_clearerr(f); - delayed=true; - } - else - SG_SPRINT("\n"); - } -#endif -} diff --git a/src/interfaces/python_modular/CMakeLists.txt b/src/interfaces/python/CMakeLists.txt similarity index 67% rename from src/interfaces/python_modular/CMakeLists.txt rename to src/interfaces/python/CMakeLists.txt index 09c58132ddc..a9f0ce62e40 100644 --- a/src/interfaces/python_modular/CMakeLists.txt +++ b/src/interfaces/python/CMakeLists.txt @@ -5,7 +5,7 @@ FIND_PACKAGE(NumPy REQUIRED) SET(HAVE_PYTHON 1) -#custom swig flags for python modular interface +#custom swig flags for python interface IF(${PYTHON_VERSION_MAJOR} VERSION_EQUAL 3) SET(TARGET_SWIGFLAGS "-builtin\;-modern\;-modernargs\;-threads\;-py3") SET(PYTHON3 1) @@ -19,11 +19,11 @@ ENDIF() IF ((NOT "${PYTHON_VERSION_STRING}" VERSION_LESS "3.5") AND ("${SWIG_VERSION}" VERSION_LESS "3.0.8")) MESSAGE(FATAL_ERROR - "Building Python_modular interface for Python >= 3.5 requires SWIG >= 3.0.8.") + "Building Python interface for Python >= 3.5 requires SWIG >= 3.0.8.") ENDIF ((NOT "${PYTHON_VERSION_STRING}" VERSION_LESS "3.5") AND ("${SWIG_VERSION}" VERSION_LESS "3.0.8")) -include(CommonModularInterface) +include(ShogunInterfaces) include_directories(${PYTHON_INCLUDE_PATH} ${NUMPY_INCLUDE_DIRS}) #TODO: check for SWIG version where this bug has been applied already: @@ -31,7 +31,7 @@ include_directories(${PYTHON_INCLUDE_PATH} ${NUMPY_INCLUDE_DIRS}) SET(SWIG_CXX_COMPILER_FLAGS "${SWIG_CXX_COMPILER_FLAGS} -Wno-c++11-narrowing") -GENERATE_MODULAR_TARGET(python ${CMAKE_CURRENT_SOURCE_DIR} ${PYTHON_LIBRARIES}) +GENERATE_INTERFACE_TARGET(python ${CMAKE_CURRENT_SOURCE_DIR} ${PYTHON_LIBRARIES}) # pacakges directory if(CMAKE_HOST_UNIX) @@ -64,31 +64,10 @@ elseif(CMAKE_HOST_WIN32) endif() SET(PYTHON_PACKAGES_PATH "${_PYTHON_PACKAGES_PATH}" CACHE PATH "Where to install the python packages.") -INSTALL(TARGETS _python_modular +INSTALL(TARGETS _interface_python DESTINATION ${PYTHON_PACKAGES_PATH} COMPONENT python) -INSTALL(FILES ${CMAKE_CURRENT_BINARY_DIR}/modshogun.py +INSTALL(FILES ${CMAKE_CURRENT_BINARY_DIR}/shogun.py DESTINATION ${PYTHON_PACKAGES_PATH} - COMPONENT python) - -# create the __init.py__ files for modules and the main -file(WRITE ${CMAKE_CURRENT_BINARY_DIR}/__init__.py.in "from modshogun import *") -file(WRITE ${CMAKE_CURRENT_BINARY_DIR}/main_init.py.in "__all__= [\n") -configure_file(${CMAKE_CURRENT_BINARY_DIR}/__init__.py.in __init__.py COPYONLY) -SET(MODULES Kernel Distance Features Classifier Regression Converter - Loss Clustering Evaluation IO Library Mathematics - ModelSelection Preprocessor Structure Distribution Statistics Latent) - -FOREACH(module ${MODULES}) - INSTALL(FILES ${CMAKE_CURRENT_BINARY_DIR}/__init__.py - DESTINATION ${PYTHON_PACKAGES_PATH}/shogun/${module} - COMPONENT python) - file(APPEND ${CMAKE_CURRENT_BINARY_DIR}/main_init.py.in "\"${module}\",\n") -ENDFOREACH() - -file(APPEND ${CMAKE_CURRENT_BINARY_DIR}/main_init.py.in "]") -configure_file(${CMAKE_CURRENT_BINARY_DIR}/main_init.py.in main_init.py COPYONLY) -INSTALL(FILES ${CMAKE_CURRENT_BINARY_DIR}/main_init.py - DESTINATION ${PYTHON_PACKAGES_PATH}/shogun - COMPONENT python - RENAME __init__.py) + COMPONENT python +) diff --git a/src/interfaces/python_modular/CustomKernel_protocols.i b/src/interfaces/python/CustomKernel_protocols.i similarity index 100% rename from src/interfaces/python_modular/CustomKernel_protocols.i rename to src/interfaces/python/CustomKernel_protocols.i diff --git a/src/interfaces/python_modular/DenseFeatures_protocols.i b/src/interfaces/python/DenseFeatures_protocols.i similarity index 100% rename from src/interfaces/python_modular/DenseFeatures_protocols.i rename to src/interfaces/python/DenseFeatures_protocols.i diff --git a/src/interfaces/python_modular/DenseLabels_protocols.i b/src/interfaces/python/DenseLabels_protocols.i similarity index 100% rename from src/interfaces/python_modular/DenseLabels_protocols.i rename to src/interfaces/python/DenseLabels_protocols.i diff --git a/src/interfaces/python_modular/SGVector_protocols.i b/src/interfaces/python/SGVector_protocols.i similarity index 100% rename from src/interfaces/python_modular/SGVector_protocols.i rename to src/interfaces/python/SGVector_protocols.i diff --git a/src/interfaces/python_modular/protocols_helper.i b/src/interfaces/python/protocols_helper.i similarity index 100% rename from src/interfaces/python_modular/protocols_helper.i rename to src/interfaces/python/protocols_helper.i diff --git a/src/interfaces/python_modular/sg_print_functions.cpp b/src/interfaces/python/sg_print_functions.cpp similarity index 54% rename from src/interfaces/python_modular/sg_print_functions.cpp rename to src/interfaces/python/sg_print_functions.cpp index 4b86b8d8a5c..7f3f16dfc75 100644 --- a/src/interfaces/python_modular/sg_print_functions.cpp +++ b/src/interfaces/python/sg_print_functions.cpp @@ -13,7 +13,7 @@ void sg_global_print_message(FILE* target, const char* str) void sg_global_print_warning(FILE* target, const char* str) { - if (target==stdout) + if (target == stdout) { PyGILState_STATE gil = PyGILState_Ensure(); PyErr_Warn(NULL, str); @@ -25,7 +25,7 @@ void sg_global_print_warning(FILE* target, const char* str) void sg_global_print_error(FILE* target, const char* str) { - if (target==stdout) + if (target == stdout) { PyGILState_STATE gil = PyGILState_Ensure(); PyErr_SetString(PyExc_RuntimeError, str); @@ -34,26 +34,3 @@ void sg_global_print_error(FILE* target, const char* str) else fprintf(target, "%s", str); } - -void sg_global_cancel_computations(bool &delayed, bool &immediately) -{ - using namespace shogun; - - PyGILState_STATE gil = PyGILState_Ensure(); - if (PyErr_CheckSignals()) - { - SG_SPRINT("\nImmediately return to matlab prompt / Prematurely finish computations / Do nothing (I/P/D)? "); - char answer=fgetc(stdin); - - if (answer == 'I') - immediately=true; - else if (answer == 'P') - { - PyErr_Clear(); - delayed=true; - } - else - SG_SPRINT("\n"); - } - PyGILState_Release(gil); -} diff --git a/src/interfaces/python_modular/swig_typemaps.i b/src/interfaces/python/swig_typemaps.i similarity index 100% rename from src/interfaces/python_modular/swig_typemaps.i rename to src/interfaces/python/swig_typemaps.i diff --git a/src/interfaces/r_modular/CMakeLists.txt b/src/interfaces/r/CMakeLists.txt similarity index 57% rename from src/interfaces/r_modular/CMakeLists.txt rename to src/interfaces/r/CMakeLists.txt index 33cbf5725d8..7b596acb2b1 100644 --- a/src/interfaces/r_modular/CMakeLists.txt +++ b/src/interfaces/r/CMakeLists.txt @@ -2,27 +2,27 @@ FIND_PACKAGE(R REQUIRED) UNSET(TARGET_SWIGFLAGS) -include(CommonModularInterface) +include(ShogunInterfaces) SET(SWIG_CXX_COMPILER_FLAGS "${SWIG_CXX_COMPILER_FLAGS} ${R_CXX_FLAGS}") -IF(APPLE) +IF("${CMAKE_CXX_COMPILER_ID}" STREQUAL "Clang") SET(SWIG_CXX_COMPILER_FLAGS "${SWIG_CXX_COMPILER_FLAGS} -DDO_NOT_USE_CXX_HEADERS") ENDIF() -GENERATE_MODULAR_TARGET(r ${CMAKE_CURRENT_SOURCE_DIR} ${R_LIBRARIES}) -set_target_properties(r_modular PROPERTIES PREFIX "") +GENERATE_INTERFACE_TARGET(r ${CMAKE_CURRENT_SOURCE_DIR} ${R_LIBRARIES}) +set_target_properties(interface_r PROPERTIES PREFIX "") -ADD_CUSTOM_COMMAND(TARGET r_modular +ADD_CUSTOM_COMMAND(TARGET interface_r POST_BUILD - COMMAND echo 'wd="${CMAKE_CURRENT_BINARY_DIR}" \; setwd( wd ) \; f="modshogun.R" \; fdata="modshogun.RData" \; source( f ) \; save( list=ls( all=TRUE ) , file=fdata , compress=TRUE ) \; q( save="no" ) \;' | ${R_EXECUTABLE} --silent --no-save + COMMAND echo 'wd="${CMAKE_CURRENT_BINARY_DIR}" \; setwd( wd ) \; f="shogun.R" \; fdata="shogun.RData" \; source( f ) \; save( list=ls( all=TRUE ) , file=fdata , compress=TRUE ) \; q( save="no" ) \;' | ${R_EXECUTABLE} --silent --no-save WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR} - COMMENT "Generating modshogun.RData" + COMMENT "Generating shogun.RData" ) -install(FILES ${CMAKE_CURRENT_BINARY_DIR}/modshogun.so +install(FILES ${CMAKE_CURRENT_BINARY_DIR}/shogun.so DESTINATION ${R_COMPONENT_LIB_PATH}/shogun/libs COMPONENT r) -install(FILES ${CMAKE_CURRENT_BINARY_DIR}/modshogun.RData +install(FILES ${CMAKE_CURRENT_BINARY_DIR}/shogun.RData DESTINATION ${R_COMPONENT_LIB_PATH}/shogun/R COMPONENT r) diff --git a/src/interfaces/r_modular/sg_print_functions.cpp b/src/interfaces/r/sg_print_functions.cpp similarity index 64% rename from src/interfaces/r_modular/sg_print_functions.cpp rename to src/interfaces/r/sg_print_functions.cpp index 7a8c7c7bfdc..985b43ace97 100644 --- a/src/interfaces/r_modular/sg_print_functions.cpp +++ b/src/interfaces/r/sg_print_functions.cpp @@ -1,39 +1,34 @@ #include -#include -#include +#include +#include #include +#include #include #include -#include -#include +#include #include void sg_global_print_message(FILE* target, const char* str) { - if (target==stdout) - Rprintf((char*) "%s", str); + if (target == stdout) + Rprintf((char*)"%s", str); else fprintf(target, "%s", str); } void sg_global_print_warning(FILE* target, const char* str) { - if (target==stdout) - Rprintf((char*) "%s", str); + if (target == stdout) + Rprintf((char*)"%s", str); else fprintf(target, "%s", str); } void sg_global_print_error(FILE* target, const char* str) { - if (target==stdout) - Rprintf((char*) "%s", str); + if (target == stdout) + Rprintf((char*)"%s", str); else fprintf(target, "%s", str); } - -void sg_global_cancel_computations(bool &delayed, bool &immediately) -{ - /* R_Suicide((char*) "sg stopped by SIGINT\n"); */ -} diff --git a/src/interfaces/r_modular/swig_typemaps.i b/src/interfaces/r/swig_typemaps.i similarity index 100% rename from src/interfaces/r_modular/swig_typemaps.i rename to src/interfaces/r/swig_typemaps.i diff --git a/src/interfaces/ruby/CMakeLists.txt b/src/interfaces/ruby/CMakeLists.txt new file mode 100644 index 00000000000..e806eab5995 --- /dev/null +++ b/src/interfaces/ruby/CMakeLists.txt @@ -0,0 +1,29 @@ +FIND_PACKAGE(Ruby REQUIRED) +FIND_PACKAGE(RubyNArray REQUIRED) +SET(NARRAY_LIB ${RUBY_NARRAY_LIBRARY} PARENT_SCOPE) +UNSET(TARGET_SWIGFLAGS) + +include(ShogunInterfaces) +INCLUDE_DIRECTORIES(${RUBY_INCLUDE_DIRS} ${RUBY_NARRAY_INCLUDE_DIR}) + +CONFIGURE_FILE("${CMAKE_CURRENT_SOURCE_DIR}/shogun.rb.in" + "${CMAKE_CURRENT_BINARY_DIR}/shogun.rb" @ONLY) + +GENERATE_INTERFACE_TARGET(ruby ${CMAKE_CURRENT_SOURCE_DIR} ${RUBY_LIBRARY}) +set_target_properties(interface_ruby PROPERTIES PREFIX "") +target_compile_definitions(interface_ruby PRIVATE NARRAY_LIB="${RUBY_NARRAY_LIBRARY}") + +SET(INTERFACE_RUBY_BINARY_DIR ${CMAKE_CURRENT_BINARY_DIR} PARENT_SCOPE) + +EXECUTE_PROCESS(COMMAND ${RUBY_EXECUTABLE} -r rbconfig -e "print RbConfig::CONFIG['sitearchdir']" + OUTPUT_VARIABLE RUBY_SITE_ARCH_DIR + ERROR_QUIET) + +INSTALL(FILES ${CMAKE_CURRENT_BINARY_DIR}/shogun.rb + DESTINATION ${RUBY_SITE_ARCH_DIR} + COMPONENT ruby) + +INSTALL(FILES ${CMAKE_CURRENT_BINARY_DIR}/shogun.so + DESTINATION ${RUBY_SITE_ARCH_DIR} + COMPONENT ruby + RENAME shogun${EXT_LIB_SWIG_RUBY}) diff --git a/src/interfaces/lua_modular/sg_print_functions.cpp b/src/interfaces/ruby/sg_print_functions.cpp similarity index 73% rename from src/interfaces/lua_modular/sg_print_functions.cpp rename to src/interfaces/ruby/sg_print_functions.cpp index 72afe5ab1f4..b9637930f62 100644 --- a/src/interfaces/lua_modular/sg_print_functions.cpp +++ b/src/interfaces/ruby/sg_print_functions.cpp @@ -8,14 +8,10 @@ void sg_global_print_message(FILE* target, const char* str) void sg_global_print_warning(FILE* target, const char* str) { - fprintf(target, "%s", str); + fprintf(target, "%s", str); } void sg_global_print_error(FILE* target, const char* str) { fprintf(target, "%s", str); } - -void sg_global_cancel_computations(bool &delayed, bool &immediately) -{ -} diff --git a/src/interfaces/ruby/shogun.rb.in b/src/interfaces/ruby/shogun.rb.in new file mode 100644 index 00000000000..c60d3beb55f --- /dev/null +++ b/src/interfaces/ruby/shogun.rb.in @@ -0,0 +1,3 @@ +require 'rubygems' +require 'narray' +require 'shogun@EXT_LIB_SWIG_RUBY@' diff --git a/src/interfaces/ruby_modular/swig_typemaps.i b/src/interfaces/ruby/swig_typemaps.i similarity index 98% rename from src/interfaces/ruby_modular/swig_typemaps.i rename to src/interfaces/ruby/swig_typemaps.i index 0e52d2fa36a..d330adad8ae 100644 --- a/src/interfaces/ruby_modular/swig_typemaps.i +++ b/src/interfaces/ruby/swig_typemaps.i @@ -114,7 +114,7 @@ TYPEMAP_SGVECTOR(float64_t, NUM2DBL, rb_float_new) array = SG_MALLOC(SGTYPE, rows * cols); } for (j = 0; j < cols; j++) { - array[i * cols + j] = R2SG(rb_ary_entry(vec, j)); + array[j * rows + i] = R2SG(rb_ary_entry(vec, j)); } } } @@ -136,7 +136,7 @@ TYPEMAP_SGVECTOR(float64_t, NUM2DBL, rb_float_new) for (i = 0; i < rows; i++) { VALUE vec = rb_ary_new2(cols); for (j = 0; j < cols; j++) { - rb_ary_push(vec, SG2R($1.matrix[i * cols + j])); + rb_ary_push(vec, SG2R($1.matrix[j * rows + i])); } rb_ary_push(arr, vec); } diff --git a/src/interfaces/ruby_modular/CMakeLists.txt b/src/interfaces/ruby_modular/CMakeLists.txt deleted file mode 100644 index c2585d410fc..00000000000 --- a/src/interfaces/ruby_modular/CMakeLists.txt +++ /dev/null @@ -1,29 +0,0 @@ -FIND_PACKAGE(Ruby REQUIRED) -FIND_PACKAGE(RubyNArray REQUIRED) -SET(NARRAY_LIB ${RUBY_NARRAY_LIBRARY} PARENT_SCOPE) -UNSET(TARGET_SWIGFLAGS) - -include(CommonModularInterface) -INCLUDE_DIRECTORIES(${RUBY_INCLUDE_DIRS} ${RUBY_NARRAY_INCLUDE_DIR}) - -CONFIGURE_FILE("${CMAKE_CURRENT_SOURCE_DIR}/modshogun.rb.in" - "${CMAKE_CURRENT_BINARY_DIR}/modshogun.rb" @ONLY) - -GENERATE_MODULAR_TARGET(ruby ${CMAKE_CURRENT_SOURCE_DIR} ${RUBY_LIBRARY}) -set_target_properties(ruby_modular PROPERTIES PREFIX "") -target_compile_definitions(ruby_modular PRIVATE NARRAY_LIB="${RUBY_NARRAY_LIBRARY}") - -SET(RUBY_MODULAR_BINARY_DIR ${CMAKE_CURRENT_BINARY_DIR} PARENT_SCOPE) - -EXECUTE_PROCESS(COMMAND ${RUBY_EXECUTABLE} -r rbconfig -e "print RbConfig::CONFIG['sitearchdir']" - OUTPUT_VARIABLE RUBY_SITE_ARCH_DIR - ERROR_QUIET) - -INSTALL(FILES ${CMAKE_CURRENT_BINARY_DIR}/modshogun.rb - DESTINATION ${RUBY_SITE_ARCH_DIR} - COMPONENT ruby) - -INSTALL(FILES ${CMAKE_CURRENT_BINARY_DIR}/modshogun.so - DESTINATION ${RUBY_SITE_ARCH_DIR} - COMPONENT ruby - RENAME modshogun${EXT_LIB_SWIG_RUBY_MODULAR}) diff --git a/src/interfaces/ruby_modular/modshogun.rb.in b/src/interfaces/ruby_modular/modshogun.rb.in deleted file mode 100644 index 966a414ade2..00000000000 --- a/src/interfaces/ruby_modular/modshogun.rb.in +++ /dev/null @@ -1,3 +0,0 @@ -require 'rubygems' -require 'narray' -require 'modshogun@EXT_LIB_SWIG_RUBY_MODULAR@' diff --git a/src/interfaces/modular/Boost.i b/src/interfaces/swig/Boost.i similarity index 100% rename from src/interfaces/modular/Boost.i rename to src/interfaces/swig/Boost.i diff --git a/src/interfaces/modular/Boost_includes.i b/src/interfaces/swig/Boost_includes.i similarity index 100% rename from src/interfaces/modular/Boost_includes.i rename to src/interfaces/swig/Boost_includes.i diff --git a/src/interfaces/modular/Classifier.i b/src/interfaces/swig/Classifier.i similarity index 98% rename from src/interfaces/modular/Classifier.i rename to src/interfaces/swig/Classifier.i index da11cef942b..dd0a9a8adcd 100644 --- a/src/interfaces/modular/Classifier.i +++ b/src/interfaces/swig/Classifier.i @@ -53,8 +53,8 @@ #ifndef HAVE_PYTHON %rename(SVM) CSVM; #endif -%rename(SVMLin) CSVMLin; #ifdef USE_GPL_SHOGUN +%rename(SVMLin) CSVMLin; %rename(SVMOcas) CSVMOcas; #endif //USE_GPL_SHOGUN %rename(SVMSGD) CSVMSGD; @@ -67,7 +67,9 @@ %rename(MKLClassification) CMKLClassification; %rename(MKLOneClass) CMKLOneClass; %rename(VowpalWabbit) CVowpalWabbit; +#ifdef USE_GPL_SHOGUN %rename(FeatureBlockLogisticRegression) CFeatureBlockLogisticRegression; +#endif //USE_GPL_SHOGUN %rename(DirectorLinearMachine) CDirectorLinearMachine; %rename(DirectorKernelMachine) CDirectorKernelMachine; %rename(BaggingMachine) CBaggingMachine; @@ -115,8 +117,8 @@ %include %include %include -%include #ifdef USE_GPL_SHOGUN +%include %include #endif //USE_GPL_SHOGUN %include @@ -130,7 +132,9 @@ %include %include %include +#ifdef USE_GPL_SHOGUN %include +#endif //USE_GPL_SHOGUN %include %include %include diff --git a/src/interfaces/modular/Classifier_includes.i b/src/interfaces/swig/Classifier_includes.i similarity index 97% rename from src/interfaces/modular/Classifier_includes.i rename to src/interfaces/swig/Classifier_includes.i index 741e9151ec6..3dc5ddc37f1 100644 --- a/src/interfaces/modular/Classifier_includes.i +++ b/src/interfaces/swig/Classifier_includes.i @@ -1,3 +1,4 @@ + %{ #include #include @@ -23,9 +24,9 @@ #include #include #include - #include #include #ifdef USE_GPL_SHOGUN + #include #include #endif //USE_GPL_SHOGUN #include @@ -39,7 +40,9 @@ #include #include #include +#ifdef USE_GPL_SHOGUN #include +#endif //USE_GPL_SHOGUN #include #include #include diff --git a/src/interfaces/modular/Clustering.i b/src/interfaces/swig/Clustering.i similarity index 100% rename from src/interfaces/modular/Clustering.i rename to src/interfaces/swig/Clustering.i diff --git a/src/interfaces/modular/Clustering_includes.i b/src/interfaces/swig/Clustering_includes.i similarity index 100% rename from src/interfaces/modular/Clustering_includes.i rename to src/interfaces/swig/Clustering_includes.i diff --git a/src/interfaces/modular/Converter.i b/src/interfaces/swig/Converter.i similarity index 100% rename from src/interfaces/modular/Converter.i rename to src/interfaces/swig/Converter.i diff --git a/src/interfaces/modular/Converter_includes.i b/src/interfaces/swig/Converter_includes.i similarity index 100% rename from src/interfaces/modular/Converter_includes.i rename to src/interfaces/swig/Converter_includes.i diff --git a/src/interfaces/modular/Distance.i b/src/interfaces/swig/Distance.i similarity index 100% rename from src/interfaces/modular/Distance.i rename to src/interfaces/swig/Distance.i diff --git a/src/interfaces/modular/Distance_includes.i b/src/interfaces/swig/Distance_includes.i similarity index 100% rename from src/interfaces/modular/Distance_includes.i rename to src/interfaces/swig/Distance_includes.i diff --git a/src/interfaces/modular/Distribution.i b/src/interfaces/swig/Distribution.i similarity index 100% rename from src/interfaces/modular/Distribution.i rename to src/interfaces/swig/Distribution.i diff --git a/src/interfaces/modular/Distribution_includes.i b/src/interfaces/swig/Distribution_includes.i similarity index 100% rename from src/interfaces/modular/Distribution_includes.i rename to src/interfaces/swig/Distribution_includes.i diff --git a/src/interfaces/modular/Ensemble.i b/src/interfaces/swig/Ensemble.i similarity index 100% rename from src/interfaces/modular/Ensemble.i rename to src/interfaces/swig/Ensemble.i diff --git a/src/interfaces/modular/Ensemble_includes.i b/src/interfaces/swig/Ensemble_includes.i similarity index 100% rename from src/interfaces/modular/Ensemble_includes.i rename to src/interfaces/swig/Ensemble_includes.i diff --git a/src/interfaces/modular/Evaluation.i b/src/interfaces/swig/Evaluation.i similarity index 87% rename from src/interfaces/modular/Evaluation.i rename to src/interfaces/swig/Evaluation.i index ca70a132c91..c67263a13a4 100644 --- a/src/interfaces/modular/Evaluation.i +++ b/src/interfaces/swig/Evaluation.i @@ -47,10 +47,6 @@ %rename(GradientEvaluation) CGradientEvaluation; %rename(MulticlassOVREvaluation) CMulticlassOVREvaluation; %rename(CrossValidationResult) CCrossValidationResult; -%rename(CrossValidationOutput) CCrossValidationOutput; -%rename(CrossValidationPrintOutput) CCrossValidationPrintOutput; -%rename(CrossValidationMKLStorage) CCrossValidationMKLStorage; -%rename(CrossValidationMulticlassStorage) CCrossValidationMulticlassStorage; %rename(StructuredAccuracy) CStructuredAccuracy; %rename(DirectorContingencyTableEvaluation) CDirectorContingencyTableEvaluation; @@ -70,6 +66,7 @@ %include %include %include +%include %include %include %include @@ -78,9 +75,5 @@ %include %include %include -%include -%include -%include -%include %include -%include +%include \ No newline at end of file diff --git a/src/interfaces/modular/Evaluation_includes.i b/src/interfaces/swig/Evaluation_includes.i similarity index 84% rename from src/interfaces/modular/Evaluation_includes.i rename to src/interfaces/swig/Evaluation_includes.i index 6950aebbfe8..d96f226d22c 100644 --- a/src/interfaces/modular/Evaluation_includes.i +++ b/src/interfaces/swig/Evaluation_includes.i @@ -14,6 +14,7 @@ #include #include #include + #include #include #include #include @@ -22,10 +23,6 @@ #include #include #include - #include - #include - #include - #include #include #include %} diff --git a/src/interfaces/modular/Features.i b/src/interfaces/swig/Features.i similarity index 100% rename from src/interfaces/modular/Features.i rename to src/interfaces/swig/Features.i diff --git a/src/interfaces/modular/Features_includes.i b/src/interfaces/swig/Features_includes.i similarity index 100% rename from src/interfaces/modular/Features_includes.i rename to src/interfaces/swig/Features_includes.i diff --git a/src/interfaces/modular/GaussianProcess.i b/src/interfaces/swig/GaussianProcess.i similarity index 100% rename from src/interfaces/modular/GaussianProcess.i rename to src/interfaces/swig/GaussianProcess.i diff --git a/src/interfaces/modular/GaussianProcess_includes.i b/src/interfaces/swig/GaussianProcess_includes.i similarity index 100% rename from src/interfaces/modular/GaussianProcess_includes.i rename to src/interfaces/swig/GaussianProcess_includes.i diff --git a/src/interfaces/modular/IO.i b/src/interfaces/swig/IO.i similarity index 100% rename from src/interfaces/modular/IO.i rename to src/interfaces/swig/IO.i diff --git a/src/interfaces/modular/IO_includes.i b/src/interfaces/swig/IO_includes.i similarity index 100% rename from src/interfaces/modular/IO_includes.i rename to src/interfaces/swig/IO_includes.i diff --git a/src/interfaces/modular/Kernel.i b/src/interfaces/swig/Kernel.i similarity index 99% rename from src/interfaces/modular/Kernel.i rename to src/interfaces/swig/Kernel.i index 3a4ce6f1617..a4b73db1dcf 100644 --- a/src/interfaces/modular/Kernel.i +++ b/src/interfaces/swig/Kernel.i @@ -49,7 +49,9 @@ PROTOCOLS_CUSTOMKERNEL(CustomKernel, float32_t, "f\0", NPY_FLOAT32) %rename(CustomKernel) CCustomKernel; %rename(DiagKernel) CDiagKernel; +#ifdef USE_GPL_SHOGUN %rename(DistantSegmentsKernel) CDistantSegmentsKernel; +#endif //USE_GPL_SHOGUN %rename(WaveKernel) CWaveKernel; %rename(CauchyKernel) CCauchyKernel; %rename(DiceKernelNormalizer) CDiceKernelNormalizer; @@ -182,7 +184,9 @@ namespace shogun %include %include %include +#ifdef USE_GPL_SHOGUN %include +#endif //USE_GPL_SHOGUN %include %include %include diff --git a/src/interfaces/modular/Kernel_includes.i b/src/interfaces/swig/Kernel_includes.i similarity index 98% rename from src/interfaces/modular/Kernel_includes.i rename to src/interfaces/swig/Kernel_includes.i index a38a8a920d2..ad6533e4d48 100644 --- a/src/interfaces/modular/Kernel_includes.i +++ b/src/interfaces/swig/Kernel_includes.i @@ -19,7 +19,9 @@ #include #include #include +#ifdef USE_GPL_SHOGUN #include +#endif //USE_GPL_SHOGUN #include #include #include diff --git a/src/interfaces/modular/Latent.i b/src/interfaces/swig/Latent.i similarity index 95% rename from src/interfaces/modular/Latent.i rename to src/interfaces/swig/Latent.i index c6a7317f260..6b57e20c873 100644 --- a/src/interfaces/modular/Latent.i +++ b/src/interfaces/swig/Latent.i @@ -29,4 +29,6 @@ %include +#ifdef USE_GPL_SHOGUN %include +#endif //USE_GPL_SHOGUN diff --git a/src/interfaces/modular/Latent_includes.i b/src/interfaces/swig/Latent_includes.i similarity index 80% rename from src/interfaces/modular/Latent_includes.i rename to src/interfaces/swig/Latent_includes.i index 84270b52634..237cfc16420 100644 --- a/src/interfaces/modular/Latent_includes.i +++ b/src/interfaces/swig/Latent_includes.i @@ -5,6 +5,8 @@ #include +#ifdef USE_GPL_SHOGUN #include +#endif //USE_GPL_SHOGUN %} diff --git a/src/interfaces/modular/Library.i b/src/interfaces/swig/Library.i similarity index 100% rename from src/interfaces/modular/Library.i rename to src/interfaces/swig/Library.i diff --git a/src/interfaces/modular/Library_includes.i b/src/interfaces/swig/Library_includes.i similarity index 100% rename from src/interfaces/modular/Library_includes.i rename to src/interfaces/swig/Library_includes.i diff --git a/src/interfaces/modular/Loss.i b/src/interfaces/swig/Loss.i similarity index 100% rename from src/interfaces/modular/Loss.i rename to src/interfaces/swig/Loss.i diff --git a/src/interfaces/modular/Loss_includes.i b/src/interfaces/swig/Loss_includes.i similarity index 100% rename from src/interfaces/modular/Loss_includes.i rename to src/interfaces/swig/Loss_includes.i diff --git a/src/interfaces/modular/Machine.i b/src/interfaces/swig/Machine.i similarity index 100% rename from src/interfaces/modular/Machine.i rename to src/interfaces/swig/Machine.i diff --git a/src/interfaces/modular/Mathematics.i b/src/interfaces/swig/Mathematics.i similarity index 89% rename from src/interfaces/modular/Mathematics.i rename to src/interfaces/swig/Mathematics.i index 6bb00a75145..9fc4993afe0 100644 --- a/src/interfaces/modular/Mathematics.i +++ b/src/interfaces/swig/Mathematics.i @@ -12,59 +12,39 @@ /* Remove C Prefix */ %rename(Math) CMath; %rename(Statistics) CStatistics; +#ifdef USE_GPL_SHOGUN %rename(SparseInverseCovariance) CSparseInverseCovariance; +#endif //USE_GPL_SHOGUN // fix overloaded methods in Math #if defined(SWIGLUA) || defined(SWIGR) namespace shogun { -#ifdef USE_INT8 -%rename(dot_int8) CMath::dot(int8_t const *,int8_t const *,int32_t); -#endif - -#ifdef USE_UINT8 -%rename(dot_uint8) CMath::dot(uint8_t const *,uint8_t const *,int32_t); -#endif - -#ifdef USE_INT16 -%rename(dot_int16) CMath::dot(int16_t const *,int16_t const *,int32_t); -#endif - -#ifdef USE_UINT16 -%rename(dot_uint16) CMath::dot(uint16_t const *,uint16_t const *,int32_t); -#endif - #ifdef USE_INT32 -%rename(dot_int32) CMath::dot(int32_t const *,int32_t const *,int32_t); %rename(pow_int32) CMath::pow(int32_t,int32_t); %rename(random_int32) CMath::random(int32_t,int32_t); #endif #ifdef USE_UINT32 -%rename(dot_uint32) CMath::dot(uint32_t const *,uint32_t const *,int32_t); %rename(random_uint32) CMath::random(uint32_t,uint32_t); #endif #ifdef USE_INT64 -%rename(dot_int64) CMath::dot(int64_t const *,int64_t const *,int32_t); %rename(random_int64) CMath::random(int64_t,int64_t); #endif #ifdef USE_UINT64 -%rename(dot_uint64) CMath::dot(uint64_t const *,uint64_t const *,int32_t); %rename(random_uint64) CMath::random(uint64_t,uint64_t); #endif #ifdef USE_FLOAT32 -%rename(dot_float32) CMath::dot(float32_t const *,float32_t const *,int32_t); %rename(normal_random_float32) CMath::normal_random(float32_t,float32_t); %rename(random_float32) CMath::random(float32_t,float32_t); %rename(sqrt_float32) CMath::sqrt(float32_t); #endif #ifdef USE_FLOAT64 -%rename(dot_float64) CMath::dot(float64_t const *,float64_t const *,int32_t); %rename(normal_random_float64) CMath::normal_random(float64_t,float64_t); %rename(pow_float64_int32) CMath::pow(float64_t,int32_t); %rename(pow_float64_float64) CMath::pow(float64_t,float64_t); @@ -194,7 +174,9 @@ namespace shogun /* Include Class Headers to make them visible from within the target language */ %include %include +#ifdef USE_GPL_SHOGUN %include +#endif //USE_GPL_SHOGUN /* Log-det framework */ %include diff --git a/src/interfaces/modular/Mathematics_includes.i b/src/interfaces/swig/Mathematics_includes.i similarity index 97% rename from src/interfaces/modular/Mathematics_includes.i rename to src/interfaces/swig/Mathematics_includes.i index 1654ae085f1..11b6e1c3d77 100644 --- a/src/interfaces/modular/Mathematics_includes.i +++ b/src/interfaces/swig/Mathematics_includes.i @@ -1,7 +1,9 @@ %{ #include #include +#ifdef USE_GPL_SHOGUN #include +#endif //USE_GPL_SHOGUN /* Log-det framework */ #include diff --git a/src/interfaces/modular/Metric.i b/src/interfaces/swig/Metric.i similarity index 100% rename from src/interfaces/modular/Metric.i rename to src/interfaces/swig/Metric.i diff --git a/src/interfaces/modular/Metric_includes.i b/src/interfaces/swig/Metric_includes.i similarity index 100% rename from src/interfaces/modular/Metric_includes.i rename to src/interfaces/swig/Metric_includes.i diff --git a/src/interfaces/modular/Minimizer.i b/src/interfaces/swig/Minimizer.i similarity index 100% rename from src/interfaces/modular/Minimizer.i rename to src/interfaces/swig/Minimizer.i diff --git a/src/interfaces/modular/Minimizer_includes.i b/src/interfaces/swig/Minimizer_includes.i similarity index 100% rename from src/interfaces/modular/Minimizer_includes.i rename to src/interfaces/swig/Minimizer_includes.i diff --git a/src/interfaces/modular/ModelSelection.i b/src/interfaces/swig/ModelSelection.i similarity index 100% rename from src/interfaces/modular/ModelSelection.i rename to src/interfaces/swig/ModelSelection.i diff --git a/src/interfaces/modular/ModelSelection_includes.i b/src/interfaces/swig/ModelSelection_includes.i similarity index 100% rename from src/interfaces/modular/ModelSelection_includes.i rename to src/interfaces/swig/ModelSelection_includes.i diff --git a/src/interfaces/modular/Multiclass.i b/src/interfaces/swig/Multiclass.i similarity index 98% rename from src/interfaces/modular/Multiclass.i rename to src/interfaces/swig/Multiclass.i index 8317a9178c8..21f1b9b1a18 100644 --- a/src/interfaces/modular/Multiclass.i +++ b/src/interfaces/swig/Multiclass.i @@ -52,10 +52,12 @@ %rename(ECOCAEDDecoder) CECOCAEDDecoder; %rename(ECOCLLBDecoder) CECOCLLBDecoder; +#ifdef USE_GPL_SHOGUN %rename(MulticlassTreeGuidedLogisticRegression) CMulticlassTreeGuidedLogisticRegression; %rename(MulticlassLogisticRegression) CMulticlassLogisticRegression; -%rename(MulticlassLibLinear) CMulticlassLibLinear; %rename(MulticlassOCAS) CMulticlassOCAS; +#endif //USE_GPL_SHOGUN +%rename(MulticlassLibLinear) CMulticlassLibLinear; %rename(MulticlassSVM) CMulticlassSVM; %rename(MulticlassLibSVM) CMulticlassLibSVM; %rename(LaRank) CLaRank; @@ -121,10 +123,12 @@ namespace shogun %include %include +#ifdef USE_GPL_SHOGUN %include %include -%include %include +#endif // USE_GPL_SHOGUN +%include %include %include %include diff --git a/src/interfaces/modular/Multiclass_includes.i b/src/interfaces/swig/Multiclass_includes.i similarity index 98% rename from src/interfaces/modular/Multiclass_includes.i rename to src/interfaces/swig/Multiclass_includes.i index 4057458cf77..689ee8662fa 100644 --- a/src/interfaces/modular/Multiclass_includes.i +++ b/src/interfaces/swig/Multiclass_includes.i @@ -46,10 +46,12 @@ #include #include +#ifdef USE_GPL_SHOGUN #include #include - #include #include +#endif //USE_GPL_SHOGUN + #include #include #include #include diff --git a/src/interfaces/modular/NeuralNets.i b/src/interfaces/swig/NeuralNets.i similarity index 100% rename from src/interfaces/modular/NeuralNets.i rename to src/interfaces/swig/NeuralNets.i diff --git a/src/interfaces/modular/NeuralNets_includes.i b/src/interfaces/swig/NeuralNets_includes.i similarity index 100% rename from src/interfaces/modular/NeuralNets_includes.i rename to src/interfaces/swig/NeuralNets_includes.i diff --git a/src/interfaces/swig/ParameterObserver.i b/src/interfaces/swig/ParameterObserver.i new file mode 100644 index 00000000000..439c3dbf786 --- /dev/null +++ b/src/interfaces/swig/ParameterObserver.i @@ -0,0 +1,23 @@ +%include "std_vector.i" +%include "std_string.i" +%template(ParameterList) std::vector; + +%rename(ParameterObserverCV) CParameterObserverCV; + +%{ + #include + #include +#ifdef HAVE_TFLOGGER + #include + #include + #include +#endif // HAVE_TFLOGGER +%} + +%include +%include +#ifdef HAVE_TFLOGGER +%include +%include +%include +#endif // HAVE_TFLOGGER diff --git a/src/interfaces/modular/Preprocessor.i b/src/interfaces/swig/Preprocessor.i similarity index 100% rename from src/interfaces/modular/Preprocessor.i rename to src/interfaces/swig/Preprocessor.i diff --git a/src/interfaces/modular/Preprocessor_includes.i b/src/interfaces/swig/Preprocessor_includes.i similarity index 100% rename from src/interfaces/modular/Preprocessor_includes.i rename to src/interfaces/swig/Preprocessor_includes.i diff --git a/src/interfaces/modular/Regression.i b/src/interfaces/swig/Regression.i similarity index 100% rename from src/interfaces/modular/Regression.i rename to src/interfaces/swig/Regression.i diff --git a/src/interfaces/modular/Regression_includes.i b/src/interfaces/swig/Regression_includes.i similarity index 100% rename from src/interfaces/modular/Regression_includes.i rename to src/interfaces/swig/Regression_includes.i diff --git a/src/interfaces/modular/SGBase.i b/src/interfaces/swig/SGBase.i similarity index 97% rename from src/interfaces/modular/SGBase.i rename to src/interfaces/swig/SGBase.i index 54e52841824..368889f587e 100644 --- a/src/interfaces/modular/SGBase.i +++ b/src/interfaces/swig/SGBase.i @@ -111,9 +111,6 @@ public void readExternal(java.io.ObjectInput in) throws java.io.IOException, jav extern void sg_global_print_message(FILE* target, const char* str); extern void sg_global_print_warning(FILE* target, const char* str); extern void sg_global_print_error(FILE* target, const char* str); -#ifndef DISABLE_CANCEL_CALLBACK - extern void sg_global_cancel_computations(bool &delayed, bool &immediately); -#endif #ifdef SWIGR #include @@ -162,7 +159,7 @@ public void readExternal(java.io.ObjectInput in) throws java.io.IOException, jav #if !defined(SWIGJAVA) && !defined(SWIGCSHARP) #ifndef DISABLE_CANCEL_CALLBACK shogun::init_shogun(&sg_global_print_message, &sg_global_print_warning, - &sg_global_print_error, &sg_global_cancel_computations); + &sg_global_print_error); #else shogun::init_shogun(&sg_global_print_message, &sg_global_print_warning, &sg_global_print_error); @@ -448,9 +445,9 @@ except ImportError: import copyreg as copy_reg def _sg_reconstructor(cls, base, state): try: - if isinstance(cls, str) and cls.startswith('modshogun.'): + if isinstance(cls, str) and cls.startswith('shogun.'): if base is object: - import modshogun + import shogun return eval(cls+'()') else: base.__new__(cls, state) @@ -479,7 +476,7 @@ def _sg_reduce_ex(self, proto): base = object state = None - args = ('modshogun.' + self.get_name(), base, state) + args = ('shogun.' + self.get_name(), base, state) try: diff --git a/src/interfaces/modular/Statistics.i b/src/interfaces/swig/Statistics.i similarity index 100% rename from src/interfaces/modular/Statistics.i rename to src/interfaces/swig/Statistics.i diff --git a/src/interfaces/modular/Statistics_includes.i b/src/interfaces/swig/Statistics_includes.i similarity index 100% rename from src/interfaces/modular/Statistics_includes.i rename to src/interfaces/swig/Statistics_includes.i diff --git a/src/interfaces/modular/Structure.i b/src/interfaces/swig/Structure.i similarity index 96% rename from src/interfaces/modular/Structure.i rename to src/interfaces/swig/Structure.i index e8264b21c11..48795aa1dc8 100644 --- a/src/interfaces/modular/Structure.i +++ b/src/interfaces/swig/Structure.i @@ -56,11 +56,13 @@ %rename(StructuredOutputMachine) CStructuredOutputMachine; %rename(LinearStructuredOutputMachine) CLinearStructuredOutputMachine; %rename(KernelStructuredOutputMachine) CKernelStructuredOutputMachine; -%rename(DualLibQPBMSOSVM) CDualLibQPBMSOSVM; +#ifdef USE_GPL_SHOGUN +%rename(DualLibQPBMSOSVM) CDualLibQPBMSOSVM; #ifdef USE_MOSEK %rename(PrimalMosekSOSVM) CPrimalMosekSOSVM; #endif /* USE_MOSEK */ +#endif //USE_GPL_SHOGUN %rename(StochasticSOSVM) CStochasticSOSVM; %rename(FWSOSVM) CFWSOSVM; @@ -74,7 +76,9 @@ %include %include +#ifdef USE_GPL_SHOGUN %include +#endif //USE_GPL_SHOGUN %include %include %include @@ -105,11 +109,13 @@ %include %include +#ifdef USE_GPL_SHOGUN %include #ifdef USE_MOSEK %include #endif /* USE_MOSEK */ +#endif //USE_GPL_SHOGUN %include %include diff --git a/src/interfaces/modular/Structure_includes.i b/src/interfaces/swig/Structure_includes.i similarity index 92% rename from src/interfaces/modular/Structure_includes.i rename to src/interfaces/swig/Structure_includes.i index 29a3203df23..8faddc64711 100644 --- a/src/interfaces/modular/Structure_includes.i +++ b/src/interfaces/swig/Structure_includes.i @@ -6,8 +6,10 @@ #include #include #include - + +#ifdef USE_GPL_SHOGUN #include +#endif //USE_GPL_SHOGUN #include #include #include @@ -38,11 +40,12 @@ #include #include - #include - +#ifdef USE_GPL_SHOGUN +#include #ifdef USE_MOSEK #include #endif /* USE_MOSEK */ +#endif //USE_GPL_SHOGUN #include #include diff --git a/src/interfaces/modular/Transfer.i b/src/interfaces/swig/Transfer.i similarity index 95% rename from src/interfaces/modular/Transfer.i rename to src/interfaces/swig/Transfer.i index df81c915673..841736587de 100644 --- a/src/interfaces/modular/Transfer.i +++ b/src/interfaces/swig/Transfer.i @@ -20,12 +20,14 @@ %rename(TaskRelationBase) CTaskRelation; %rename(TaskTree) CTaskTree; %rename(TaskGroup) CTaskGroup; +#ifdef USE_GPL_SHOGUN %rename(MultitaskLinearMachineBase) CMultitaskLinearMachine; %rename(MultitaskLeastSquaresRegression) CMultitaskLeastSquaresRegression; %rename(MultitaskLogisticRegression) CMultitaskLogisticRegression; %rename(MultitaskL12LogisticRegression) CMultitaskL12LogisticRegression; %rename(MultitaskTraceLogisticRegression) CMultitaskTraceLogisticRegression; %rename(MultitaskClusteredLogisticRegression) CMultitaskClusteredLogisticRegression; +#endif //USE_GPL_SHOGUN %rename(MultitaskROCEvaluation) CMultitaskROCEvaluation; @@ -50,16 +52,21 @@ %include %include %include +#ifdef USE_GPL_SHOGUN %include %include %include %include %include -%include +#endif //USE_GPL_SHOGUN %include %include +#ifdef USE_GPL_SHOGUN +%include +#endif // USE_GPL_SHOGUN + /* Domain adaptation includes */ #ifdef USE_SVMLIGHT %include diff --git a/src/interfaces/modular/Transfer_includes.i b/src/interfaces/swig/Transfer_includes.i similarity index 93% rename from src/interfaces/modular/Transfer_includes.i rename to src/interfaces/swig/Transfer_includes.i index e354b4a389b..5c699c0d736 100644 --- a/src/interfaces/modular/Transfer_includes.i +++ b/src/interfaces/swig/Transfer_includes.i @@ -11,15 +11,20 @@ #include #include #include +#ifdef USE_GPL_SHOGUN #include #include #include #include #include - #include +#endif //USE_GPL_SHOGUN #include +#ifdef USE_GPL_SHOGUN + #include +#endif /* USE_GPL_SHOGUN */ + #ifdef USE_SVMLIGHT #include #endif /* USE_SVMLIGHT */ diff --git a/src/interfaces/modular/abstract_types_extension.i b/src/interfaces/swig/abstract_types_extension.i similarity index 98% rename from src/interfaces/modular/abstract_types_extension.i rename to src/interfaces/swig/abstract_types_extension.i index cfb435c3df2..f78065f10a4 100644 --- a/src/interfaces/modular/abstract_types_extension.i +++ b/src/interfaces/swig/abstract_types_extension.i @@ -18,7 +18,7 @@ //PTZ121108 example of classifier in examples/undocumented/libshogun/classifier_latent_svm.cpp //extention to make use of CData,CLatentModel //TODO:PTZ121108 put it in another file like classifier_latent_svm.i or %include examples/undocumented/libshogun/classifier_latent_svm.cpp - //or find a clever way to wrap CLatenModel, CData instanciation, bless({}, modshogun::LatentModel) + //or find a clever way to wrap CLatenModel, CData instanciation, bless({}, shogun::LatentModel) // is not enough and would need a new wrapper, but yet new CLatentModel() is not working, // (with error: "cannot allocate an object of abstract type") ? %inline %{ diff --git a/src/interfaces/modular/bagging.i b/src/interfaces/swig/bagging.i similarity index 100% rename from src/interfaces/modular/bagging.i rename to src/interfaces/swig/bagging.i diff --git a/src/interfaces/modular/bagging_includes.i b/src/interfaces/swig/bagging_includes.i similarity index 100% rename from src/interfaces/modular/bagging_includes.i rename to src/interfaces/swig/bagging_includes.i diff --git a/src/interfaces/modular/modshogun.doxy.in b/src/interfaces/swig/shogun.doxy.in similarity index 86% rename from src/interfaces/modular/modshogun.doxy.in rename to src/interfaces/swig/shogun.doxy.in index 8e2521e8f40..900ac13d1bf 100644 --- a/src/interfaces/modular/modshogun.doxy.in +++ b/src/interfaces/swig/shogun.doxy.in @@ -1,6 +1,6 @@ INPUT = @LIBSHOGUN_SRC_DIR@ -OUTPUT_DIRECTORY = modshogun -XML_OUTPUT = doxygen_xml +OUTPUT_DIRECTORY = shogun_doxygen +XML_OUTPUT = xml CREATE_SUBDIRS = NO OUTPUT_LANGUAGE = English FILE_PATTERNS = *.h diff --git a/src/interfaces/modular/modshogun.i b/src/interfaces/swig/shogun.i similarity index 92% rename from src/interfaces/modular/modshogun.i rename to src/interfaces/swig/shogun.i index 7dea9e39d40..f25432c21e9 100644 --- a/src/interfaces/modular/modshogun.i +++ b/src/interfaces/swig/shogun.i @@ -20,13 +20,13 @@ %include "swig_config.h" %define DOCSTR -"The `modshogun` module gathers all modules available in the SHOGUN toolkit." +"The `shogun` module gathers all modules available in the SHOGUN toolkit." %enddef #if defined(USE_SWIG_DIRECTORS) && defined(SWIGPYTHON) -%module(directors="1", docstring=DOCSTR) modshogun +%module(directors="1", docstring=DOCSTR) shogun #else -%module(docstring=DOCSTR) modshogun +%module(docstring=DOCSTR) shogun #endif #undef DOCSTR @@ -50,11 +50,11 @@ #ifdef HAVE_DOXYGEN #ifndef SWIGRUBY -%include "modshogun_doxygen.i" +%include "shogun_doxygen.i" #endif #endif -%include "modshogun_ignores.i" +%include "shogun_ignores.i" %include "Classifier_includes.i" %include "Clustering_includes.i" @@ -114,6 +114,8 @@ %include "bagging.i" %include "Boost.i" +%include "ParameterObserver.i" + #if defined(SWIGPERL) %include "abstract_types_extension.i" #endif diff --git a/src/interfaces/modular/modshogun_ignores.i b/src/interfaces/swig/shogun_ignores.i similarity index 99% rename from src/interfaces/modular/modshogun_ignores.i rename to src/interfaces/swig/shogun_ignores.i index a5f4feff389..86d35a8466a 100644 --- a/src/interfaces/modular/modshogun_ignores.i +++ b/src/interfaces/swig/shogun_ignores.i @@ -222,3 +222,5 @@ %ignore shogun::CMAPInference::CMAPInference(); %ignore shogun::CGraphCut::CGraphCut(); %ignore shogun::CFactorGraphModel::CFactorGraphModel(); + +%ignore shogun::Range; diff --git a/src/interfaces/modular/swig_config.h.in b/src/interfaces/swig/swig_config.h.in similarity index 100% rename from src/interfaces/modular/swig_config.h.in rename to src/interfaces/swig/swig_config.h.in diff --git a/src/shogun/CMakeLists.txt b/src/shogun/CMakeLists.txt index 367f9a2cf9c..71d42b2209c 100644 --- a/src/shogun/CMakeLists.txt +++ b/src/shogun/CMakeLists.txt @@ -29,6 +29,32 @@ if (MSVC AND (BUILD_EXAMPLES OR BUILD_META_EXAMPLES)) CACHE BOOL "Build libshogun static library" FORCE) endif() +# Allow to hide non-bsd compatible codes +OPTION(LICENSE_GPL_SHOGUN "Include GPL codes of Shogun (non-BSD compatible) in build" ON) +SET(USE_GPL_SHOGUN 0) +IF (LICENSE_GPL_SHOGUN) + SET(SHOGUN_GPL_INCLUDE_DIR ${CMAKE_SOURCE_DIR}/src/gpl) + IF(EXISTS "${SHOGUN_GPL_INCLUDE_DIR}/shogun") + SET(USE_GPL_SHOGUN 1) + + FILE(GLOB_RECURSE GPL_LIBSHOGUN_SRC ${SHOGUN_GPL_INCLUDE_DIR}/*.${EXT_SRC_CPP} ${SHOGUN_GPL_INCLUDE_DIR}/*.${EXT_SRC_C}) + FILE(GLOB_RECURSE GPL_LIBSHOGUN_HEADERS ${SHOGUN_GPL_INCLUDE_DIR}/*.${EXT_SRC_HEADER}) + + LIST(APPEND LIBSHOGUN_SRC ${GPL_LIBSHOGUN_SRC}) + LIST(APPEND LIBSHOGUN_HEADERS ${GPL_LIBSHOGUN_HEADERS}) + INCLUDE_DIRECTORIES(${SHOGUN_GPL_INCLUDE_DIR}) + + SET(USE_GPL_SHOGUN 1) + ELSE() + MESSAGE(FATAL_ERROR "Shogun can only be built with GPL codes if the source files are in ${SHOGUN_GPL_INCLUDE_DIR}. Please download or disable.\n") + ENDIF() +ENDIF() + +OPTION(USE_SVMLIGHT "SVMLight" ON) +IF(USE_SVMLIGHT AND NOT USE_GPL_SHOGUN) + MESSAGE(FATAL_ERROR "Can only use SVMLight when GPL codes are included") +ENDIF() + # add target to compile the libshogun sources add_library(libshogun OBJECT ${LIBSHOGUN_SRC} ${LIBSHOGUN_HEADERS} ${CMAKE_CURRENT_BINARY_DIR}/lib/config.h) set_property(TARGET libshogun PROPERTY POSITION_INDEPENDENT_CODE TRUE) @@ -56,27 +82,17 @@ FOREACH(template ${LIBSHOGUN_SRC_TMP}) STRING(REGEX REPLACE ${EXT_CPP_TMP} ${EXT_CPP_PY} generator_script "${template}") STRING(REGEX REPLACE ".*/(.*).${EXT_CPP_TMP}" "\\1" generated_target "${template}") - IF(MSVC) - FOREACH(h ${LIBSHOGUN_HEADERS}) - FILE(APPEND ${CMAKE_CURRENT_BINARY_DIR}/headers_list.txt "${h}\n") - ENDFOREACH() - - ADD_CUSTOM_COMMAND(OUTPUT ${generated_cpp} - COMMAND ${CMAKE_COMMAND} -E copy_if_different ${template} ${generated_cpp} - COMMAND ${PYTHON_EXECUTABLE} ${generator_script} ${generated_cpp} -in ${CMAKE_CURRENT_BINARY_DIR}/headers_list.txt - DEPENDS ${template} ${generator_script} version ${LIBSHOGUN_HEADERS} ${CMAKE_CURRENT_BINARY_DIR}/headers_list.txt - WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR} - COMMENT "Generating ${generated_cpp}" - ) - ELSE() - ADD_CUSTOM_COMMAND(OUTPUT ${generated_cpp} - COMMAND ${CMAKE_COMMAND} -E copy_if_different ${template} ${generated_cpp} - COMMAND ${PYTHON_EXECUTABLE} ${generator_script} ${generated_cpp} ${LIBSHOGUN_HEADERS} - DEPENDS ${template} ${generator_script} version ${LIBSHOGUN_HEADERS} - WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR} - COMMENT "Generating ${generated_cpp}" - ) - ENDIF() + FOREACH(h ${LIBSHOGUN_HEADERS}) + FILE(APPEND ${CMAKE_CURRENT_BINARY_DIR}/headers_list.txt "${h}\n") + ENDFOREACH() + + ADD_CUSTOM_COMMAND(OUTPUT ${generated_cpp} + COMMAND ${CMAKE_COMMAND} -E copy_if_different ${template} ${generated_cpp} + COMMAND ${PYTHON_EXECUTABLE} ${generator_script} ${generated_cpp} -in ${CMAKE_CURRENT_BINARY_DIR}/headers_list.txt + DEPENDS ${template} ${generator_script} version ${LIBSHOGUN_HEADERS} ${CMAKE_CURRENT_BINARY_DIR}/headers_list.txt + WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR} + COMMENT "Generating ${generated_cpp}" + ) ADD_CUSTOM_TARGET(${generated_target} DEPENDS ${generated_cpp}) add_dependencies(libshogun ${generated_target}) @@ -97,6 +113,14 @@ target_include_directories(shogun PUBLIC $ $ ) +IF (${USE_GPL_SHOGUN}) + target_include_directories(shogun PUBLIC $) +ENDIF() + +get_property(dirs DIRECTORY ${CMAKE_CURRENT_SOURCE_DIR} PROPERTY INCLUDE_DIRECTORIES) +foreach(dir ${dirs}) + message(STATUS "dir='${dir}'") +endforeach() # create shogun_deps meta target to store all the # dependencies of shogun itself, this is used in unit test @@ -106,6 +130,9 @@ target_include_directories(shogun_deps INTERFACE $ $ ) +IF (${USE_GPL_SHOGUN}) + target_include_directories(shogun_deps INTERFACE $) +ENDIF() # add bundled libraries to dependency if(SHOGUN_DEPENDS) @@ -116,6 +143,15 @@ endif() if (LIBSHOGUN_BUILD_STATIC) add_library(shogun-static STATIC $ ${CMAKE_CURRENT_BINARY_DIR}/lib/config.h) set_property(TARGET shogun-static PROPERTY OUTPUT_NAME shogun) + target_include_directories(shogun-static PUBLIC + $ + $ + $ + ) + IF (${USE_GPL_SHOGUN}) + target_include_directories(shogun-static PUBLIC $) + ENDIF() + if(MSVC) target_link_libraries(shogun-static PUBLIC winmm.lib) endif() @@ -157,11 +193,6 @@ if (OPENMP_FOUND) endif() endif() -FIND_PACKAGE(Spinlock) -IF (SPINLOCK_FOUND) - SET(USE_SPINLOCKS 1) -ENDIF() - FIND_PACKAGE(CxaDemangle) # check SSE and SSE2 intrinsics header @@ -172,8 +203,6 @@ ENDIF((NOT CYGWIN) AND (NOT DISABLE_SSE)) FIND_PACKAGE(CxaDemangle) ############################ std lib functions -CHECK_INCLUDE_FILE_CXX("unordered_map" HAVE_STD_UNORDERED_MAP) - include (CheckCXXSymbolExists) CHECK_CXX_SYMBOL_EXISTS(isfinite "cmath" HAVE_DECL_ISFINITE) CHECK_CXX_SYMBOL_EXISTS(isinf "cmath" HAVE_DECL_ISINF) @@ -231,64 +260,30 @@ SHOGUN_DEPENDENCIES( VERSION ${VIENNACL_VERSION_MINIMUM} CONFIG_FLAG HAVE_VIENNACL) -# Linear algebra default global backend setups -SET_LINALG_BACKEND(LINALG_DEFAULT_BACKEND GLOBAL) - -# Linear algebra default module specific backend setup -# Linear solver module -SET_LINALG_BACKEND(LINALG_LINEAR_SOLVER_LIB LINSLV) - -# Eigen solver module -SET_LINALG_BACKEND(LINALG_EIGENSOLVER_LIB EIGSLV) - ####### /LINALG -# prefer original LAPACK, if needed -OPTION(USE_ORIGINAL_LAPACK "Original LAPACK" OFF) -FIND_PACKAGE(LAPACK) -IF (LAPACK_FOUND) - # find out whether it is Accelerate.framework we found for LaPack/BLAS - IF("${LAPACK_LIBRARIES}" MATCHES ".*/Accelerate.framework$") - SET(HAVE_MVEC 1) - SET(HAVE_CATLAS 1) - SET(HAVE_LAPACK 1) - SHOGUN_LINK_LIBS(${LAPACK_LIBRARIES}) - ELSEIF("${LAPACK_LIBRARIES}" MATCHES ".*/mkl_.*") - SET(HAVE_LAPACK 1) - SET(HAVE_MKL 1) - SHOGUN_LINK_LIBS(${LAPACK_LIBRARIES}) - ELSE() - FIND_PACKAGE(Atlas) - IF(Atlas_FOUND) - SET(HAVE_ATLAS 1) - SET(HAVE_LAPACK 1) - SHOGUN_INCLUDE_DIRS(SCOPE PUBLIC ${Atlas_INCLUDE_DIRS}) - IF(USE_ORIGINAL_LAPACK) - SET(ATLAS_CLAPACK_LIBRARY) - FOREACH(ITR ${Atlas_LIBRARIES}) - IF(ITR MATCHES ".*atlas.*lapack.*" OR ITR MATCHES ".*lapack.*atlas.*") - STRING(REGEX REPLACE "lapack" "clapack" ITR ${ITR}) - LIST(APPEND ATLAS_CLAPACK_LIBRARY ${ITR}) - ENDIF(ITR MATCHES ".*atlas.*lapack.*" OR ITR MATCHES ".*lapack.*atlas.*") - ENDFOREACH(ITR ${ATLAS_LIBRARIES}) - MESSAGE(STATUS "using ATLAS-CLAPACK from: ${ATLAS_CLAPACK_LIBRARY}") - SHOGUN_LINK_LIBS(${LAPACK_LIBRARIES} ${ATLAS_CLAPACK_LIBRARY} ${Atlas_LIBRARIES}) - ELSE(USE_ORIGINAL_LAPACK) - SHOGUN_LINK_LIBS(${LAPACK_LIBRARIES} ${Atlas_LIBRARIES}) - ENDIF(USE_ORIGINAL_LAPACK) - ELSE() - FIND_PACKAGE(CBLAS) - if(CBLAS_LIBRARY) - SET(HAVE_LAPACK 1) - SHOGUN_LINK_LIBS(${LAPACK_LIBRARIES} ${CBLAS_LIBRARY}) - else() - SET(HAVE_LAPACK 1) - SHOGUN_LINK_LIBS(${LAPACK_LIBRARIES}) - endif() - ENDIF() - ENDIF() +FIND_PACKAGE(rxcpp) +IF(NOT rxcpp_FOUND) + include(external/rxcpp) + SHOGUN_INCLUDE_DIRS(SCOPE PUBLIC SYSTEM + $ + $ + ) +ELSE() + SHOGUN_INCLUDE_DIRS(SCOPE PUBLIC SYSTEM ${rxcpp_INCLUDE_DIR}) +ENDIF() + +# TFLogger package +FIND_PACKAGE(TFLogger 0.1.0 CONFIG) +IF (TFLogger_FOUND) + SET(HAVE_TFLOGGER 1) + SHOGUN_INCLUDE_DIRS(SCOPE PRIVATE SYSTEM ${TFLogger_INCLUDE_DIR}) + target_link_libraries(shogun PRIVATE tflogger::tflogger) ENDIF() +#### LAPACK +include(ShogunFindLAPACK) + SHOGUN_DEPENDENCIES( LIBRARY GLPK SCOPE PRIVATE @@ -358,10 +353,14 @@ SHOGUN_DEPENDENCIES( SCOPE PUBLIC CONFIG_FLAG HAVE_XML) -SHOGUN_DEPENDENCIES( - LIBRARY HDF5 - SCOPE PUBLIC - CONFIG_FLAG HAVE_HDF5) +if (NOT WIN32) + # FIXME: HDF5 linking on WIN32 is broken. + # at least with the hdf5 supplied in anaconda + SHOGUN_DEPENDENCIES( + LIBRARY HDF5 + SCOPE PUBLIC + CONFIG_FLAG HAVE_HDF5) +endif () SHOGUN_DEPENDENCIES( LIBRARY CURL @@ -434,9 +433,6 @@ SHOGUN_DEPENDENCIES( SCOPE PRIVATE CONFIG_FLAG HAVE_COLPACK) -IF(USE_ARPREC AND NOT LICENSE_GPL_SHOGUN) - MESSAGE(FATAL_ERROR "ARPREC can only be used if USE_GPL_SHOGUN is enabled ") -ENDIF() SHOGUN_DEPENDENCIES( LIBRARY ARPREC SCOPE PRIVATE @@ -482,8 +478,10 @@ IF (CTAGS_FOUND) ADD_CUSTOM_COMMAND(OUTPUT ${CTAGS_FILE} COMMAND ${CTAGS_EXECUTABLE} -f ${CTAGS_FILE} # functions, classes, macroses, enumerations, enumerators, typedefs - --c++-kinds=fcdget - -R ${CMAKE_CURRENT_SOURCE_DIR}) + --c++-kinds=fcdgetp + --fields=+im + -R ${CMAKE_SOURCE_DIR}) + ADD_CUSTOM_TARGET(ctags DEPENDS ${CTAGS_FILE}) SET_SOURCE_FILES_PROPERTIES(${CTAGS_FILE} PROPERTIES GENERATED 1) ENDIF() diff --git a/src/shogun/base/Parameter.cpp b/src/shogun/base/Parameter.cpp index 9e572ab78d4..815ab27afb1 100644 --- a/src/shogun/base/Parameter.cpp +++ b/src/shogun/base/Parameter.cpp @@ -2195,7 +2195,7 @@ TParameter::new_sgserial(CSGObject** param, if (*param != NULL) SG_UNREF(*param); - *param = new_sgserializable(sgserializable_name, generic); + *param = create(sgserializable_name, generic); if (*param == NULL) { string_t buf = {'\0'}; diff --git a/src/shogun/base/SGObject.cpp b/src/shogun/base/SGObject.cpp index 394ed0a1da5..684fd703758 100644 --- a/src/shogun/base/SGObject.cpp +++ b/src/shogun/base/SGObject.cpp @@ -14,33 +14,32 @@ #include #include +#include +#include #include #include -#include -#include +#include #include -#include #include -#include +#include +#include #include #include #include -#ifdef HAVE_CXX11 +#include +#include + #include -#else -#include -#endif namespace shogun { -#ifdef HAVE_CXX11 - typedef std::unordered_map ParametersMap; -#else typedef std::map ParametersMap; -#endif + typedef std::unordered_map> + ObsParamsList; class CSGObject::Self { @@ -150,7 +149,7 @@ namespace shogun using namespace shogun; -CSGObject::CSGObject() : self() +CSGObject::CSGObject() : self(), param_obs_list() { init(); set_global_objects(); @@ -160,7 +159,8 @@ CSGObject::CSGObject() : self() } CSGObject::CSGObject(const CSGObject& orig) -: self(), io(orig.io), parallel(orig.parallel), version(orig.version) + : self(), param_obs_list(), io(orig.io), parallel(orig.parallel), + version(orig.version) { init(); set_global_objects(); @@ -178,9 +178,11 @@ CSGObject::~CSGObject() delete m_model_selection_parameters; delete m_gradient_parameters; delete m_refcount; + delete m_subject_params; + delete m_observable_params; + delete m_subscriber_params; } -#ifdef USE_REFERENCE_COUNTING int32_t CSGObject::ref() { int32_t count = m_refcount->ref(); @@ -210,7 +212,6 @@ int32_t CSGObject::unref() return m_refcount->ref_count(); } } -#endif //USE_REFERENCE_COUNTING #ifdef TRACE_MEMORY_ALLOCS #include @@ -502,6 +503,10 @@ void CSGObject::init() m_save_pre_called = false; m_save_post_called = false; m_hash = 0; + + m_subject_params = new SGSubject(); + m_observable_params = new SGObservable(m_subject_params->get_observable()); + m_subscriber_params = new SGSubscriber(m_subject_params->get_subscriber()); } void CSGObject::print_modsel_params() @@ -729,7 +734,7 @@ bool CSGObject::equals(CSGObject* other, float64_t accuracy, bool tolerant) CSGObject* CSGObject::clone() { SG_DEBUG("Constructing an empty instance of %s\n", get_name()); - CSGObject* copy=new_sgserializable(get_name(), this->m_generic); + CSGObject* copy = create(get_name(), this->m_generic); SG_REF(copy); @@ -755,7 +760,7 @@ bool CSGObject::clone_parameters(CSGObject* other) { REQUIRE(other, "Provided instance must be non-empty.\n"); index_t num_parameters = m_parameters->get_num_parameters(); - + REQUIRE(other->m_parameters->get_num_parameters() == num_parameters, "Number of parameters of provided instance (%d) must match this instance (%d).\n", other->m_parameters->get_num_parameters(), num_parameters); @@ -802,3 +807,83 @@ bool CSGObject::type_erased_has(const BaseTag& _tag) const { return self->has(_tag); } + +void CSGObject::subscribe_to_parameters(ParameterObserverInterface* obs) +{ + auto sub = rxcpp::make_subscriber( + [obs](TimedObservedValue e) { obs->on_next(e); }, + [obs](std::exception_ptr ep) { obs->on_error(ep); }, + [obs]() { obs->on_complete(); }); + + // Create an observable which emits values only if they are about + // parameters selected by the observable. + auto subscription = m_observable_params + ->filter([obs](ObservedValue v) { + return obs->filter(v.get_name()); + }) + .timestamp() + .subscribe(sub); +} + +void CSGObject::observe(const ObservedValue value) +{ + m_subscriber_params->on_next(value); +} + +class CSGObject::ParameterObserverList +{ +public: + void register_param( + const std::string& name, const SG_OBS_VALUE_TYPE type, + const std::string& description) + { + m_list_obs_params[name] = std::make_pair(type, description); + } + + std::string type_name(SG_OBS_VALUE_TYPE type) + { + std::string value; + switch (type) + { + case TENSORBOARD: + value = std::string("Tensorboard"); + break; + case CROSSVALIDATION: + value = std::string("CrossValidation"); + break; + default: + value = std::string("Unknown"); + break; + } + return value; + } + + ObsParamsList get_list() const + { + return m_list_obs_params; + } + +private: + /** List of observable parameters (name, description) */ + ObsParamsList m_list_obs_params; +}; + +void CSGObject::register_observable_param( + const std::string& name, const SG_OBS_VALUE_TYPE type, + const std::string& description) +{ + param_obs_list->register_param(name, type, description); +} + +void CSGObject::list_observable_parameters() +{ + SG_INFO("List of observable parameters of object %s\n", get_name()); + SG_PRINT("------"); + for (auto const& x : param_obs_list->get_list()) + { + SG_PRINT( + "%s [%s]: %s\n", x.first.c_str(), + param_obs_list->type_name(x.second.first).c_str(), + x.second.second.c_str()); + } +} diff --git a/src/shogun/base/SGObject.h b/src/shogun/base/SGObject.h index 5af48151343..08bb98f97a7 100644 --- a/src/shogun/base/SGObject.h +++ b/src/shogun/base/SGObject.h @@ -13,15 +13,19 @@ #ifndef __SGOBJECT_H__ #define __SGOBJECT_H__ -#include -#include -#include -#include #include #include #include -#include +#include +#include +#include #include +#include +#include +#include +#include + +#include /** \namespace shogun * @brief all of classes and functions are contained in the shogun namespace @@ -33,6 +37,7 @@ class SGIO; class Parallel; class Parameter; class CSerializableFile; +class ParameterObserverInterface; template class CMap; @@ -44,15 +49,9 @@ template class SGStringList; * define reference counter macros ******************************************************************************/ -#ifdef USE_REFERENCE_COUNTING #define SG_REF(x) { if (x) (x)->ref(); } #define SG_UNREF(x) { if (x) { if ((x)->unref()==0) (x)=NULL; } } #define SG_UNREF_NO_NULL(x) { if (x) { (x)->unref(); } } -#else -#define SG_REF(x) -#define SG_UNREF(x) -#define SG_UNREF_NO_NULL(x) -#endif /******************************************************************************* * Macros for registering parameters/model selection parameters @@ -125,6 +124,14 @@ enum EGradientAvailability class CSGObject { public: + typedef rxcpp::subjects::subject SGSubject; + typedef rxcpp::observable> + SGObservable; + typedef rxcpp::subscriber< + ObservedValue, rxcpp::observer> + SGSubscriber; + /** default constructor */ CSGObject(); @@ -134,7 +141,6 @@ class CSGObject /** destructor */ virtual ~CSGObject(); -#ifdef USE_REFERENCE_COUNTING /** increase reference counter * * @return reference count @@ -153,7 +159,6 @@ class CSGObject * @return reference count */ int32_t unref(); -#endif //USE_REFERENCE_COUNTING #ifdef TRACE_MEMORY_ALLOCS static void list_memory_allocs(); @@ -402,6 +407,23 @@ class CSGObject return get(tag); } +#ifndef SWIG + /** + * Get parameters observable + * @return RxCpp observable + */ + SGObservable* get_parameters_observable() + { + return m_observable_params; + }; +#endif + + /** Subscribe a parameter observer to watch over params */ + void subscribe_to_parameters(ParameterObserverInterface* obs); + + /** Print to stdout a list of observable parameters */ + void list_observable_parameters(); + protected: /** Can (optionally) be overridden to pre-initialize some member * variables which are not PARAMETER::ADD'ed. Make sure that at @@ -553,6 +575,26 @@ class CSGObject class Self; Unique self; + class ParameterObserverList; + Unique param_obs_list; + +protected: + /** + * Observe a parameter value and emit them to observer. + * @param value Observed parameter's value + */ + void observe(const ObservedValue value); + + /** + * Register which params this object can emit. + * @param name the param name + * @param type the param type + * @param description a user oriented description + */ + void register_observable_param( + const std::string& name, const SG_OBS_VALUE_TYPE type, + const std::string& description); + public: /** io */ SGIO* io; @@ -584,6 +626,15 @@ class CSGObject bool m_save_post_called; RefCount* m_refcount; + + /** Subject used to create the params observer */ + SGSubject* m_subject_params; + + /** Parameter Observable */ + SGObservable* m_observable_params; + + /** Subscriber used to call onNext, onComplete etc.*/ + SGSubscriber* m_subscriber_params; }; } #endif // __SGOBJECT_H__ diff --git a/src/shogun/base/Version.cpp b/src/shogun/base/Version.cpp index 7a56f533e52..28a17c7eadb 100644 --- a/src/shogun/base/Version.cpp +++ b/src/shogun/base/Version.cpp @@ -19,7 +19,7 @@ using namespace shogun; namespace shogun { -const int32_t Version::version_revision = VERSION_REVISION; +const int64_t Version::version_revision = VERSION_REVISION; const int32_t Version::version_year = VERSION_YEAR; const int32_t Version::version_month = VERSION_MONTH; const int32_t Version::version_day = VERSION_DAY; @@ -43,7 +43,7 @@ Version::~Version() void Version::print_version() { - SG_SPRINT("libshogun (%s/%s%d)\n\n", MACHINE, VERSION_RELEASE, version_revision) + SG_SPRINT("libshogun (%s/%s%" PRId64 ")\n\n", MACHINE, VERSION_RELEASE, version_revision) SG_SPRINT("Copyright (C) 1999-2009 Fraunhofer Institute FIRST\n") SG_SPRINT("Copyright (C) 1999-2011 Max Planck Society\n") SG_SPRINT("Copyright (C) 2009-2011 Berlin Institute of Technology\n") @@ -65,7 +65,7 @@ const char* Version::get_version_release() return version_release; } -int32_t Version::get_version_revision() +int64_t Version::get_version_revision() { return version_revision; } diff --git a/src/shogun/base/Version.h b/src/shogun/base/Version.h index f4595a4cca2..edf3b522819 100644 --- a/src/shogun/base/Version.h +++ b/src/shogun/base/Version.h @@ -43,7 +43,7 @@ class Version static const char* get_version_release(); /** get version revision */ - static int32_t get_version_revision(); + static int64_t get_version_revision(); /** get version year */ static int32_t get_version_year(); @@ -88,7 +88,7 @@ class Version static const char version_extra[128]; /** version revision */ - static const int32_t version_revision; + static const int64_t version_revision; /** version year */ static const int32_t version_year; /** version month */ diff --git a/src/shogun/base/class_list.cpp.py b/src/shogun/base/class_list.cpp.py index 46f9887d1be..8ca0a4d9018 100644 --- a/src/shogun/base/class_list.cpp.py +++ b/src/shogun/base/class_list.cpp.py @@ -1,12 +1,36 @@ #!/usr/bin/env python -# This program is free software; you can redistribute it and/or modify -# it under the terms of the GNU General Public License as published by -# the Free Software Foundation; either version 3 of the License, or -# (at your option) any later version. +# Copyright (c) The Shogun Machine Learning Toolbox +# Copyright (c) 2008-2009 Fraunhofer Institute FIRST and Max-Planck-Society +# Written (w) 2008-2009 Soeren Sonnenburg +# Written (w) 2016 - 2017 Heiko Strathmann +# All rights reserved. # -# Written (W) 2008-2009 Soeren Sonnenburg -# Copyright (C) 2008-2009 Fraunhofer Institute FIRST and Max Planck Society +# Redistribution and use in source and binary forms, with or without +# modification, are permitted provided that the following conditions are met: +# +# 1. Redistributions of source code must retain the above copyright notice, this +# list of conditions and the following disclaimer. +# 2. Redistributions in binary form must reproduce the above copyright notice, +# this list of conditions and the following disclaimer in the documentation +# and/or other materials provided with the distribution. +# +# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND +# ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED +# WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE +# DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR +# ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES +# (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; +# LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND +# ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT +# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS +# SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +# +# The views and conclusions contained in the software and documentation are those +# of the authors and should not be interpreted as representing official policies, +# either expressed or implied, of the Shogun Development Team. + +import os class_str = 'class' types = ["BOOL", "CHAR", "INT8", "UINT8", "INT16", "UINT16", "INT32", "UINT32", @@ -81,24 +105,31 @@ def extract_class_name(lines, line_nr, line, blacklist): return c[1:] -def get_includes(classes, basedir="."): - class_headers = [] - for c, t in classes: - class_headers.append(c+".h") - - import os - result = [] - for root, dirs, files in os.walk(basedir): - for f in files: - if f in class_headers: - result.append(os.path.join(os.path.relpath(root, basedir), f)) - +def get_includes(classes, headers_absolute_fnames): includes = [] - result.sort() - for o in result: - includes.append('#include ' % o.strip().lstrip('./')) - return includes + for c in classes: + for h in headers_absolute_fnames: + class_from_header = os.path.splitext(os.path.basename(h))[0] + + # build relative include path from absolute header filename + if class_from_header in c: + # find *last* occurence of "shogun" dir in header + shogun_dir = "shogun" + assert shogun_dir in h + tails = [] + head, tail = os.path.split(h) + while tail != shogun_dir and len(head)>0: + tails += [tail] + head, tail = os.path.split(head) + + # construct include path from collected tails + tails.reverse() + include = os.path.join(*([shogun_dir] + tails)) + + # thats your include header + includes.append("#include <%s>" % include) + return includes def get_definitions(classes): definitions = [] @@ -287,12 +318,6 @@ def get_blacklist(): blacklist[cfg] = 1 return blacklist - -def get_base_src_dir(headers): - import os.path - return os.path.commonprefix(headers) - - if __name__ == '__main__': import sys TEMPL_FILE = sys.argv[1] @@ -307,11 +332,10 @@ def get_base_src_dir(headers): blacklist = get_blacklist() - base_src_dir = get_base_src_dir(HEADERS) classes = extract_classes(HEADERS, False, blacklist, False) template_classes = extract_classes(HEADERS, True, blacklist, False) complex_template_classes = extract_classes(HEADERS, True, blacklist, True) - includes = get_includes(classes+template_classes+complex_template_classes, basedir=base_src_dir) + includes = get_includes(classes+template_classes+complex_template_classes, HEADERS) definitions = get_definitions(classes) template_definitions = get_template_definitions(template_classes, False) complex_template_definitions = get_template_definitions(complex_template_classes, True) diff --git a/src/shogun/base/class_list.cpp.templ b/src/shogun/base/class_list.cpp.templ index 00e076f2604..2afbf2c3d7d 100644 --- a/src/shogun/base/class_list.cpp.templ +++ b/src/shogun/base/class_list.cpp.templ @@ -1,11 +1,33 @@ /* - * This program is free software; you can redistribute it and/or modify - * it under the terms of the GNU General Public License as published by - * the Free Software Foundation; either version 3 of the License, or - * (at your option) any later version. + * Copyright (c) The Shogun Machine Learning Toolbox + * Copyright (c) 2009 Fraunhofer Institute FIRST and Max-Planck-Society + * Written (w) 2009 Soeren Sonnenburg + * Written (w) 2016 - 2017 Heiko Strathmann + * All rights reserved. * - * Written (W) 2009 Soeren Sonnenburg - * Copyright (C) 2009 Fraunhofer Institute FIRST and Max-Planck-Society + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions are met: + * + * 1. Redistributions of source code must retain the above copyright notice, this + * list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright notice, + * this list of conditions and the following disclaimer in the documentation + * and/or other materials provided with the distribution. + * + * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED + * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE + * DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR + * ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES + * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; + * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND + * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + * + * The views and conclusions contained in the software and documentation are those + * of the authors and should not be interpreted as representing official policies, + * either expressed or implied, of the Shogun Development Team. */ #include @@ -22,12 +44,12 @@ REPLACE template_definitions THIS REPLACE complex_template_definitions THIS -typedef CSGObject* (*new_sgserializable_t)(EPrimitiveType generic); +typedef CSGObject* (*create_function)(EPrimitiveType generic); #ifndef DOXYGEN_SHOULD_SKIP_THIS typedef struct { const char* m_class_name; - new_sgserializable_t m_new_sgserializable; + create_function m_create_function; } class_list_entry_t; #endif @@ -36,14 +58,13 @@ REPLACE struct THIS {NULL, NULL} }; -CSGObject* shogun::new_sgserializable(const char* sgserializable_name, - EPrimitiveType generic) +CSGObject* shogun::create(const char* classname, EPrimitiveType generic) { for (class_list_entry_t* i=class_list; i->m_class_name != NULL; i++) { - if (strncmp(i->m_class_name, sgserializable_name, STRING_LEN) == 0) - return i->m_new_sgserializable(generic); + if (strncmp(i->m_class_name, classname, STRING_LEN) == 0) + return i->m_create_function(generic); } return NULL; diff --git a/src/shogun/base/class_list.h b/src/shogun/base/class_list.h index ed1ed6617f1..99b495c3507 100644 --- a/src/shogun/base/class_list.h +++ b/src/shogun/base/class_list.h @@ -18,12 +18,11 @@ namespace shogun { class CSGObject; - /** new shogun serializable + /** new shogun instance * @param sgserializable_name * @param generic */ - CSGObject* new_sgserializable(const char* sgserializable_name, - EPrimitiveType generic); + CSGObject* create(const char* sgserializable_name, EPrimitiveType generic); } #endif /* __SG_CLASS_LIST_H__ */ diff --git a/src/shogun/base/init.cpp b/src/shogun/base/init.cpp index 104636e925d..fe7e46421e4 100644 --- a/src/shogun/base/init.cpp +++ b/src/shogun/base/init.cpp @@ -12,17 +12,23 @@ #include #include +#include +#include +#include +#include +#include + +#include + #include #include #include -#include -#include -#include -#include -#include +#include +#include #include #include +#include #ifdef TRACE_MEMORY_ALLOCS #include shogun::CMap* sg_mallocs=NULL; @@ -39,25 +45,22 @@ namespace shogun Version* sg_version=NULL; CMath* sg_math=NULL; CRandom* sg_rand=NULL; + std::unique_ptr sg_signal(nullptr); std::unique_ptr sg_linalg(nullptr); /// function called to print normal messages - void (*sg_print_message)(FILE* target, const char* str) = NULL; + std::function sg_print_message(nullptr); /// function called to print warning messages - void (*sg_print_warning)(FILE* target, const char* str) = NULL; + std::function sg_print_warning(nullptr); /// function called to print error messages - void (*sg_print_error)(FILE* target, const char* str) = NULL; - - /// function called to cancel things - void (*sg_cancel_computations)(bool &delayed, bool &immediately)=NULL; - + std::function sg_print_error(nullptr); - void init_shogun(void (*print_message)(FILE* target, const char* str), - void (*print_warning)(FILE* target, const char* str), - void (*print_error)(FILE* target, const char* str), - void (*cancel_computations)(bool &delayed, bool &immediately)) + void init_shogun( + const std::function print_message, + const std::function print_warning, + const std::function print_error) { if (!sg_io) sg_io = new shogun::SGIO(); @@ -71,6 +74,8 @@ namespace shogun sg_rand = new shogun::CRandom(); if (!sg_linalg) sg_linalg = std::unique_ptr(new shogun::SGLinalg()); + if (!sg_signal) + sg_signal = std::unique_ptr(new shogun::CSignal()); #ifdef TRACE_MEMORY_ALLOCS if (!sg_mallocs) @@ -87,7 +92,9 @@ namespace shogun sg_print_message=print_message; sg_print_warning=print_warning; sg_print_error=print_error; - sg_cancel_computations=cancel_computations; + + // Set up signal handler + std::signal(SIGINT, sg_signal->handler); init_from_env(); } @@ -111,10 +118,6 @@ namespace shogun sg_mallocs=NULL; SG_UNREF(mallocs); #endif - sg_print_message=NULL; - sg_print_warning=NULL; - sg_print_error=NULL; - sg_cancel_computations=NULL; SG_UNREF(sg_rand); SG_UNREF(sg_math); @@ -122,6 +125,10 @@ namespace shogun SG_UNREF(sg_parallel); SG_UNREF(sg_io); + delete CSignal::m_subscriber; + delete CSignal::m_observable; + delete CSignal::m_subject; + #ifdef HAVE_PROTOBUF ::google::protobuf::ShutdownProtobufLibrary(); #endif @@ -191,12 +198,19 @@ namespace shogun SG_REF(sg_rand); return sg_rand; } + + CSignal* get_global_signal() + { + return sg_signal.get(); + } + #ifndef SWIG // SWIG should skip this part SGLinalg* get_global_linalg() { return sg_linalg.get(); } #endif + void init_from_env() { char* env_log_val = NULL; diff --git a/src/shogun/base/init.h b/src/shogun/base/init.h index 7f60097496a..53b8455a6fb 100644 --- a/src/shogun/base/init.h +++ b/src/shogun/base/init.h @@ -13,6 +13,7 @@ #include +#include #include namespace shogun @@ -23,91 +24,95 @@ namespace shogun class Parallel; class CRandom; class SGLinalg; - -/** This function must be called before libshogun is used. Usually shogun does - * not provide any output messages (neither debugging nor error; apart from - * exceptions). This function allows one to specify customized output - * callback functions and a callback function to check for exceptions: - * - * @param print_message function pointer to print a message - * @param print_warning function pointer to print a warning message - * @param print_error function pointer to print an error message (this will be - * printed before shogun throws an exception) - * - * @param cancel_computations function pointer to check for exception - * - */ -void init_shogun(void (*print_message)(FILE* target, const char* str) = NULL, - void (*print_warning)(FILE* target, const char* str) = NULL, - void (*print_error)(FILE* target, const char* str) = NULL, - void (*cancel_computations)(bool &delayed, bool &immediately)=NULL); - -/** init shogun with defaults */ -void init_shogun_with_defaults(); - -/** This function must be called when one stops using libshogun. It will - * perform a number of cleanups */ -void exit_shogun(); - -/** set the global io object - * - * @param io io object to use - */ -void set_global_io(SGIO* io); - -/** get the global io object - * - * @return io object - */ -SGIO* get_global_io(); - -/** set the global parallel object - * - * @param parallel parallel object to use - */ -void set_global_parallel(Parallel* parallel); - -/** get the global parallel object - * - * @return parallel object - */ -Parallel* get_global_parallel(); - -/** set the global version object - * - * @param version version object to use - */ -void set_global_version(Version* version); - -/** get the global version object - * - * @return version object - */ -Version* get_global_version(); - -/** set the global math object - * - * @param math math object to use - */ -void set_global_math(CMath* math); - -/** get the global math object - * - * @return math object - */ -CMath* get_global_math(); - -/** set the global random object - * - * @param rand random object to use - */ -void set_global_rand(CRandom* rand); - -/** get the global random object - * - * @return random object - */ -CRandom* get_global_rand(); + class CSignal; + + /** This function must be called before libshogun is used. Usually shogun + * does + * not provide any output messages (neither debugging nor error; apart from + * exceptions). This function allows one to specify customized output + * callback functions and a callback function to check for exceptions: + * + * @param print_message function pointer to print a message + * @param print_warning function pointer to print a warning message + * @param print_error function pointer to print an error message (this will + * be + * printed before shogun throws an + * exception) + * + * @param cancel_computations function pointer to check for exception + * + */ + void init_shogun( + const std::function print_message = nullptr, + const std::function print_warning = nullptr, + const std::function print_error = nullptr); + + /** init shogun with defaults */ + void init_shogun_with_defaults(); + + /** This function must be called when one stops using libshogun. It will + * perform a number of cleanups */ + void exit_shogun(); + + /** set the global io object + * + * @param io io object to use + */ + void set_global_io(SGIO* io); + + /** get the global io object + * + * @return io object + */ + SGIO* get_global_io(); + + /** set the global parallel object + * + * @param parallel parallel object to use + */ + void set_global_parallel(Parallel* parallel); + + /** get the global parallel object + * + * @return parallel object + */ + Parallel* get_global_parallel(); + + /** set the global version object + * + * @param version version object to use + */ + void set_global_version(Version* version); + + /** get the global version object + * + * @return version object + */ + Version* get_global_version(); + + /** set the global math object + * + * @param math math object to use + */ + void set_global_math(CMath* math); + + /** get the global math object + * + * @return math object + */ + CMath* get_global_math(); + + /** set the global random object + * + * @param rand random object to use + */ + void set_global_rand(CRandom* rand); + + /** get the global random object + * + * @return random object + */ + CRandom* get_global_rand(); #ifndef SWIG // SWIG should skip this part /** get the global linalg library object @@ -117,20 +122,23 @@ CRandom* get_global_rand(); SGLinalg* get_global_linalg(); #endif +/** get the global singnal handler object + * + * @return linalg object + */ +CSignal* get_global_signal(); + /** Checks environment variables and modifies global objects */ void init_from_env(); /// function called to print normal messages -extern void (*sg_print_message)(FILE* target, const char* str); +extern std::function sg_print_message; /// function called to print warning messages -extern void (*sg_print_warning)(FILE* target, const char* str); +extern std::function sg_print_warning; /// function called to print error messages -extern void (*sg_print_error)(FILE* target, const char* str); - -/// function called to cancel things -extern void (*sg_cancel_computations)(bool &delayed, bool &immediately); +extern std::function sg_print_error; } #endif //__SG_INIT__ diff --git a/src/shogun/base/progress.h b/src/shogun/base/progress.h new file mode 100644 index 00000000000..1c13095617b --- /dev/null +++ b/src/shogun/base/progress.h @@ -0,0 +1,741 @@ +/* +* BSD 3-Clause License +* +* Copyright (c) 2017, Shogun-Toolbox e.V. +* All rights reserved. +* +* Redistribution and use in source and binary forms, with or without +* modification, are permitted provided that the following conditions are met: +* +* * Redistributions of source code must retain the above copyright notice, this +* list of conditions and the following disclaimer. +* +* * Redistributions in binary form must reproduce the above copyright notice, +* this list of conditions and the following disclaimer in the documentation +* and/or other materials provided with the distribution. +* +* * Neither the name of the copyright holder nor the names of its +* contributors may be used to endorse or promote products derived from +* this software without specific prior written permission. +* +* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +* AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE +* DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE +* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL +* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR +* SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER +* CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, +* OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +* OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +* +* Written (W) 2017 Giovanni De Toni +* +*/ + +#ifndef __SG_PROGRESS_H__ +#define __SG_PROGRESS_H__ + +#include +#include +#include +#include + +#include +#include +#include +#include +#include +#include + +#ifdef WIN32 +#include +#else +#include +#include + +#endif + +namespace shogun +{ + + /** Possible print modes */ + enum SG_PRG_MODE + { + ASCII, + UTF8 + }; + + /** + * @class Printer class that displays the progress bar. + */ + class ProgressPrinter + { + public: + /** + * Creates a @ref ProgressPrinter instance. + * @param io SGIO object which will be used to print the progress bar. + * @param max_value interval maximum value. + * @param min_value interval minimum value. + * @param prefix string which will be printed before the progress bar. + * @param mode char mode (UTF8, ASCII etc.). + */ + ProgressPrinter( + const SGIO& io, float64_t max_value, float64_t min_value, + const std::string& prefix, const SG_PRG_MODE mode) + : m_io(io), m_max_value(max_value), m_min_value(min_value), + m_prefix(prefix), m_mode(mode), m_last_progress(0), + m_last_progress_time(0), + m_progress_start_time(CTime::get_curtime()), + m_current_value(min_value) + { + } + ~ProgressPrinter() + { + } + + /** + * Increment and print the progress bar. + * Everything is locked to prevent race conditions + * or characters overlapping (especially within + * multi threaded environments). + */ + void print_progress() const + { + lock.lock(); + if (m_current_value.load() - m_min_value > + m_max_value - m_min_value) + { + increment(); + lock.unlock(); + return; + } + print_progress_impl(); + if (m_current_value.load() - m_min_value == + m_max_value - m_min_value) + { + print_end(); + increment(); + lock.unlock(); + return; + } + increment(); + lock.unlock(); + } + + void print_progress_absolute( + float64_t current_val, float64_t val, float64_t min_val, + float64_t max_val) + { + lock.lock(); + if (val - m_min_value > m_max_value - m_min_value) + { + lock.unlock(); + return; + } + print_progress_absolute_impl(current_val, val, min_val, max_val); + if (val - m_min_value == m_max_value - m_min_value) + { + print_end(); + lock.unlock(); + return; + } + lock.unlock(); + } + + /** + * Manually increment to max size the current value + * to print a complete progress bar. + */ + void premature_end() + { + if (m_current_value.load() < m_max_value - 1) + m_current_value.store(m_max_value); + } + + /** @return last progress as a percentage. */ + inline float64_t get_current_progress() const + { + return m_current_value.load(); + } + + private: + /** + * Logic implementation of the progress bar. + */ + void print_progress_impl() const + { + + // Check if the progress was enabled + if (!m_io.get_show_progress()) + return; + + if (m_max_value <= m_min_value) + return; + + // Check for terminal dimension. This is for provide + // a minimal resize functionality. + set_screen_size(); + + float64_t difference = m_max_value - m_min_value, v = -1, + estimate = 0, total_estimate = 0; + float64_t size_chunk = -1; + + // Check if we have enough space to show the progress bar + // Use only a fraction of it to account for the size of the + // time displayed (decimals and integer). + int32_t progress_bar_space = + (m_columns_num - 50 - m_prefix.length()) * 0.9; + + // TODO: this guy here brokes testing + // REQUIRE( + // progress_bar_space > 0, + // "Not enough terminal space to show the progress bar!\n") + + char str[1000]; + float64_t runtime = CTime::get_curtime(); + + if (difference > 0.0) + v = 100 * (m_current_value.load() - m_min_value) / + (m_max_value - m_min_value); + + // Set up chunk size + size_chunk = difference / (float64_t)progress_bar_space; + + if (m_last_progress == 0) + { + m_last_progress_time = runtime; + m_last_progress = v; + } + else + { + m_last_progress = v - 1e-6; + + if ((v != 100.0) && (runtime - m_last_progress_time < 0.5)) + return; + + m_last_progress_time = runtime; + estimate = (1 - v / 100) * + (m_last_progress_time - m_progress_start_time) / + (v / 100); + total_estimate = + (m_last_progress_time - m_progress_start_time) / (v / 100); + } + + /** Print the actual progress bar to screen **/ + m_io.message(MSG_MESSAGEONLY, "", "", -1, "%s |", m_prefix.c_str()); + for (index_t i = 1; i < progress_bar_space; i++) + { + if (m_current_value.load() - m_min_value > i * size_chunk) + { + m_io.message( + MSG_MESSAGEONLY, "", "", -1, "%s", + get_pb_char().c_str()); + } + else + { + m_io.message(MSG_MESSAGEONLY, "", "", -1, " "); + } + } + m_io.message(MSG_MESSAGEONLY, "", "", -1, "| %.2f\%", v); + + if (estimate > 120) + { + snprintf( + str, sizeof(str), + " %%1.1f minutes remaining %%1.1f minutes total\r"); + m_io.message( + MSG_MESSAGEONLY, "", "", -1, str, estimate / 60, + total_estimate / 60); + } + else + { + snprintf( + str, sizeof(str), + " %%1.1f seconds remaining %%1.1f seconds total\r"); + m_io.message( + MSG_MESSAGEONLY, "", "", -1, str, estimate, total_estimate); + } + } + + /** + * Logic implementation fo the absolute progress bar. + */ + void print_progress_absolute_impl( + float64_t current_val, float64_t val, float64_t min_value, + float64_t max_value) const + { + // Check if the progress was enabled + if (!m_io.get_show_progress()) + return; + + m_current_value.store(current_val); + + if (max_value <= min_value) + return; + + // Check for terminal dimension. This is for provide + // a minimal resize functionality. + set_screen_size(); + + float64_t difference = max_value - min_value, v = -1, estimate = 0, + total_estimate = 0; + float64_t size_chunk = -1; + + // Check if we have enough space to show the progress bar + // Use only a fraction of it to account for the size of the + // time displayed (decimals and integer). + int32_t progress_bar_space = + (m_columns_num - 50 - m_prefix.length()) * 0.9; + + // TODO: this guy here brokes testing + // REQUIRE( + // progress_bar_space > 0, + // "Not enough terminal space to show the progress bar!\n") + + char str[1000]; + float64_t runtime = CTime::get_curtime(); + + if (difference > 0.0) + v = 100 * (val - min_value) / (max_value - min_value); + + // Set up chunk size + size_chunk = difference / (float64_t)progress_bar_space; + + if (m_last_progress == 0) + { + m_last_progress_time = runtime; + m_last_progress = v; + } + else + { + m_last_progress = v - 1e-6; + + if ((v != 100.0) && (runtime - m_last_progress_time < 0.5)) + return; + + m_last_progress_time = runtime; + estimate = (1 - v / 100) * + (m_last_progress_time - m_progress_start_time) / + (v / 100); + total_estimate = + (m_last_progress_time - m_progress_start_time) / (v / 100); + } + + /** Print the actual progress bar to screen **/ + m_io.message(MSG_MESSAGEONLY, "", "", -1, "%s |", m_prefix.c_str()); + for (index_t i = 1; i < progress_bar_space; i++) + { + if (m_current_value.load() - min_value > i * size_chunk) + { + m_io.message( + MSG_MESSAGEONLY, "", "", -1, "%s", + get_pb_char().c_str()); + } + else + { + m_io.message(MSG_MESSAGEONLY, "", "", -1, " "); + } + } + m_io.message(MSG_MESSAGEONLY, "", "", -1, "| %.2f\%", current_val); + + if (estimate > 120) + { + snprintf( + str, sizeof(str), + " %%1.1f minutes remaining %%1.1f minutes total\r"); + m_io.message( + MSG_MESSAGEONLY, "", "", -1, str, estimate / 60, + total_estimate / 60); + } + else + { + snprintf( + str, sizeof(str), + " %%1.1f seconds remaining %%1.1f seconds total\r"); + m_io.message( + MSG_MESSAGEONLY, "", "", -1, str, estimate, total_estimate); + } + } + + /** Print the progress bar end. */ + void print_end() const + { + // Check if the progress was enabled + if (!m_io.get_show_progress()) + return; + + m_io.message(MSG_MESSAGEONLY, "", "", -1, "\n"); + } + + /** + * Return the char which will be used to print the progress. + * @return UTF8/ASCII string + */ + std::string get_pb_char() const + { + switch (m_mode) + { + case ASCII: + return m_ascii_char; + case UTF8: + return m_utf8_char; + default: + return m_ascii_char; + } + } + + /** + * Get the terminal's screen size (Windows and Unix). + */ + void set_screen_size() const + { +#ifdef WIN32 + CONSOLE_SCREEN_BUFFER_INFO csbi; + GetConsoleScreenBufferInfo(GetStdHandle(STD_OUTPUT_HANDLE), &csbi); + m_columns_num = csbi.srWindow.Right - csbi.srWindow.Left + 1; + m_rows_num = csbi.srWindow.Bottom - csbi.srWindow.Top + 1; +#else + struct winsize wind; + ioctl(STDOUT_FILENO, TIOCGWINSZ, &wind); + m_columns_num = wind.ws_col; + m_rows_num = wind.ws_row; +#endif + } + + /* Increment the current value (atomically) */ + void increment() const + { + m_current_value++; + } + + /** IO object */ + SGIO m_io; + /** Maxmimum value */ + float64_t m_max_value; + /** Minimum value */ + float64_t m_min_value; + /** Prefix which will be printed before the progress bar */ + std::string m_prefix; + /** Progres bar's char mode */ + SG_PRG_MODE m_mode; + /** ASCII char */ + std::string m_ascii_char = "#"; + /** UTF8 char */ + std::string m_utf8_char = "\u2588"; + /* Screen column number*/ + mutable int32_t m_columns_num; + /* Screen row number*/ + mutable int32_t m_rows_num; + /** Last progress */ + mutable float64_t m_last_progress; + /** Last progress time */ + mutable float64_t m_last_progress_time; + /** Progress start time */ + mutable float64_t m_progress_start_time; + /** Current value */ + mutable std::atomic m_current_value; + /** Lock for multithreaded operations **/ + mutable CLock lock; + }; + + /** @class Helper class to show a progress bar given a range. + * + * @code + * for (auto i : PRange(Range(1, 10), io)) { ... } + * @endcode + */ + template + class PRange + { + public: + /** + * Constructor, initialize the progress bar manager. + * + * @param range the range to loop over + * @param io the SGIO object which will be used to print the progress + * bar + * @param prefix the string prefix which will be printed before the + * progress bar + * @param mode the char mode used to print the progress bar (ASCII, UTF8 + * etc.) + * @param condition premature stop condition for the loop + */ + PRange( + Range range, const SGIO& io, const std::string prefix, + const SG_PRG_MODE mode, std::function condition) + : m_range(range), m_condition(condition) + { + set_up_range(); + m_printer = std::make_shared( + io, m_end_range, m_begin_range, prefix, mode); + } + + /** @class Wrapper for Range::Iterator spawned by @ref PRange. */ + class PIterator : public std::iterator + { + public: + /** + * Initialize the PIterator object. + * @param value the @ref Range:Iterator object. + * @param shrd_ptr the @ref ProgressPrinter object. + * @param condition premature stop condition for the loop. + */ + PIterator( + typename Range::Iterator value, + std::shared_ptr shrd_ptr, + std::function condition) + : m_value(value), m_printer(shrd_ptr), m_condition(condition) + { + } + PIterator(const PIterator& other) + : m_value(other.m_value), m_printer(other.m_printer), + m_condition(other.m_condition) + { + } + PIterator(PIterator&& other) + : m_value(other.m_value), m_printer(other.m_printer), + m_condition(other.m_condition) + { + } + PIterator& operator=(const PIterator&) = delete; + PIterator& operator++() + { + // Every time we update the iterator we print + // also the updated progress bar + m_printer->print_progress(); + m_value++; + return *this; + } + PIterator operator++(int) + { + PIterator tmp(*this); + ++*this; + return tmp; + } + T operator*() + { + // Since PIterator is a wrapper we have + // to return the actual value of the + // wrapped iterator + return *m_value; + } + bool operator!=(const PIterator& other) + { + if (!(this->m_value != other.m_value)) + { + m_printer->premature_end(); + m_printer->print_progress(); + return false; + } + bool result = evaluate_condition(); + return (this->m_value != other.m_value) && result; + } + + private: + /** + * Evaluate the premature stop condition. + * @return return value of the condition. + */ + bool evaluate_condition() + { + if (!m_condition()) + { + m_printer->premature_end(); + m_printer->print_progress(); + } + return m_condition(); + } + + /* The wrapped range */ + typename Range::Iterator m_value; + /* The ProgressPrinter object which will be used to show the + * progress bar*/ + std::shared_ptr m_printer; + /* The function which will contain the custom condition + * to premature stop the loop */ + std::function m_condition; + }; + + /** Create the iterator that corresponds to the start of the range. + * Used within the range-based loop version of the progress bar. + * + * @code + * for (auto i: progress(range(0, 10), io, ASCII)) + * { + * //Do stuff + * } + * @endcode + * + * @return @ref PIterator that represents the start of the range + */ + PIterator begin() const + { + return PIterator(m_range.begin(), m_printer, m_condition); + } + + /** Create the iterator that corresponds to the end of the range. + * Used within the range-based loop version of the progress bar. + * + * @code + * for (auto i: progress(range(0, 10), io, ASCII)) + * { + * //Do stuff + * } + * @endcode + * + * @return @ref PIterator that represent the end of the range. + */ + PIterator end() const + { + return PIterator(m_range.end(), m_printer, m_condition); + } + + /** + * Return the current progress bar value. + * Used for testing purposes. + * @return current progress bar value. + */ + inline float64_t get_current_progress() const + { + return m_printer->get_current_progress(); + } + + /** + * Print the progress bar. This method must be called + * each time we want the progress bar to be updated. + * @code + * auto pr = progress(range(0,10), ASCII); + * for (int i=0; i<10; i++) + * { + * // Do stuff + * pr.print_progress(); + * } + * pr.complete(); + * @endcode + */ + void print_progress() const + { + m_printer->print_progress(); + } + + /** + * Print the absolute progress bar. This method must be called + * each time we want the progress bar to be updated. + * + * @param current_val current value + * @param val value + * @param min_val minimum value + * @param max_val maximum value + */ + void print_absolute( + float64_t current_val, float64_t val, float64_t min_value, + float64_t max_value) const + { + m_printer->print_progress_absolute( + current_val, val, min_value, max_value); + } + + /** + * Print the progress bar end. This method must be called + * one time, after the loop. + * @code + * auto pr = progress(range(0,10), ASCII); + * for (int i=0; i<10; i++) + * { + * // Do stuff + * pr.print_progress(); + * } + * pr.complete(); + * @endcode + */ + void complete() const + { + m_printer->premature_end(); + m_printer->print_progress(); + } + + /** + * Print the progress bar end. This method must be called + * one time, after the loop. + * @code + * auto pr = progress(range(0,10), ASCII); + * for (int i=0; i<10; i++) + * { + * // Do stuff + * pr.print_absolute(); + * } + * pr.complete_absolute(); + * @endcode + */ + void complete_absolute() const + { + m_printer->print_progress_absolute(100, 100, 0, 100); + } + + private: + /** + * Set up progress range. + */ + void set_up_range() + { + m_begin_range = *(m_range.begin()); + m_end_range = *(m_range.end()); + } + + /** Range we iterate over */ + Range m_range; + /** Observer that will print the actual progress bar */ + std::shared_ptr m_printer; + /* Start of the range */ + float64_t m_begin_range; + /* End of the range */ + float64_t m_end_range; + /* Function which store the premature stop condition */ + std::function m_condition = []() { return true; }; + }; + + /** Creates @ref PRange given a range. + * + * @code + * for (auto i : progress(range(0, 100), io)) { ... } + * @endcode + * + * @param range range used + * @param io SGIO object + * @param mode char printing mode (default: UTF8) + * @param prefix string which will be printed before the progress bar + * (default: PROGRESS: ) + * @param condition premature stopping condition + */ + template + inline PRange progress( + Range range, const SGIO& io, std::string prefix = "PROGRESS: ", + SG_PRG_MODE mode = UTF8, + std::function condition = []() { return true; }) + { + return PRange(range, io, prefix, mode, condition); + } + + /** Creates @ref PRange given a range that uses the global SGIO + * + * @code + * for (auto i : progress( range(0, 100) ) ) { ... } + * @endcode + * + * @param range range used + * @param mode char printing mode (default: UTF8) + * @param prefix string which will be printed before the progress bar + * (default: PROGRESS: ) + * @param condition premature stopping condition + */ + template + inline PRange progress( + Range range, std::string prefix = "PROGRESS: ", + SG_PRG_MODE mode = UTF8, + std::function condition = []() { return true; }) + { + return PRange(range, *sg_io, prefix, mode, condition); + } +}; +#endif /* __SG_PROGRESS_H__ */ diff --git a/src/shogun/base/range.h b/src/shogun/base/range.h index dfb1e213257..27d31d19b43 100644 --- a/src/shogun/base/range.h +++ b/src/shogun/base/range.h @@ -1,128 +1,157 @@ +/* +* BSD 3-Clause License +* +* Copyright (c) 2017, Shogun-Toolbox e.V. +* All rights reserved. +* +* Redistribution and use in source and binary forms, with or without +* modification, are permitted provided that the following conditions are met: +* +* * Redistributions of source code must retain the above copyright notice, this +* list of conditions and the following disclaimer. +* +* * Redistributions in binary form must reproduce the above copyright notice, +* this list of conditions and the following disclaimer in the documentation +* and/or other materials provided with the distribution. +* +* * Neither the name of the copyright holder nor the names of its +* contributors may be used to endorse or promote products derived from +* this software without specific prior written permission. +* +* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +* AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE +* DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE +* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL +* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR +* SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER +* CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, +* OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +* OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +* +* Written (W) 2016 Sergey Lisitsyn +* +*/ + #ifndef __SG_RANGE_H__ #define __SG_RANGE_H__ -#include #include -#ifdef HAVE_CXX11 namespace shogun { - /** @class Helper class to spawn range iterator. - * - * Useful for C++11-style for loops: - * - * @code - * for (auto i : Range(3, 10)) { ... } - * @endcode - */ - template - class Range - { - public: - /** Creates range with specified bounds. - * Assumes rbegin < rend. - * - * @param rbegin lower bound of range - * @param rend upper bound of range (excluding) - */ - Range(T rbegin, T rend) : m_begin(rbegin), m_end(rend) - { - } + /** @class Helper class to spawn range iterator. + * + * Useful for C++11-style for loops: + * + * @code + * for (auto i : Range(3, 10)) { ... } + * @endcode + */ + template + class Range + { + public: + /** Creates range with specified bounds. + * Assumes rbegin < rend. + * + * @param rbegin lower bound of range + * @param rend upper bound of range (excluding) + */ + Range(T rbegin, T rend) : m_begin(rbegin), m_end(rend) + { + } + + /** @class Iterator spawned by @ref Range. */ + class Iterator : public std::iterator + { + public: + Iterator(T value) : m_value(value) + { + } + Iterator(const Iterator& other) : m_value(other.m_value) + { + } + Iterator(Iterator&& other) : m_value(other.m_value) + { + } + Iterator& operator=(const Iterator&) = delete; + Iterator& operator++() + { + m_value++; + return *this; + } + Iterator operator++(int) + { + Iterator tmp(*this); + ++*this; + return tmp; + } + T operator*() + { + return m_value; + } + bool operator!=(const Iterator& other) + { + return this->m_value != other.m_value; + } - /** @class Iterator spawned by @ref Range. */ - class Iterator : public std::iterator - { - public: - Iterator(T value) : m_value(value) - { - } - Iterator(const Iterator& other) : m_value(other.m_value) - { - } - Iterator(Iterator&& other) : m_value(other.m_value) - { - } - Iterator& operator=(const Iterator&) = delete; - Iterator& operator++() - { - m_value++; - return *this; - } - Iterator& operator++(int) - { - Iterator tmp(*this); - tmp++; - return tmp; - } - T operator*() - { - return m_value; - } - bool operator!=(const Iterator& other) - { - return this->m_value != other.m_value; - } - bool operator==(const Iterator& other) - { - return this->m_value == other.m_value; - } - private: - T m_value; - }; - /** Create iterator that corresponds to the start of range. - * - * Usually called through for-loop syntax. - */ - Iterator begin() const - { - return Iterator(m_begin); - } - /** Create iterator that corresponds to the end of range. - * - * Usually called through for-loop syntax. - */ - Iterator end() const - { - return Iterator(m_end); - } - private: - /** begin of range */ - T m_begin; - /** end of range */ - T m_end; - }; + private: + T m_value; + }; + /** Create iterator that corresponds to the start of range. + * + * Usually called through for-loop syntax. + */ + Iterator begin() const + { + return Iterator(m_begin); + } + /** Create iterator that corresponds to the end of range. + * + * Usually called through for-loop syntax. + */ + Iterator end() const + { + return Iterator(m_end); + } - /** Creates @ref Range with specified upper bound. - * - * @code - * for (auto i : range(100)) { ... } - * @endcode - * - * @param rend upper bound of range (excluding) - */ - template - inline Range range(T rend) - { - return Range(0, rend); - } + private: + /** begin of range */ + T m_begin; + /** end of range */ + T m_end; + }; - /** Creates @ref Range with specified bounds. - * - * @code - * for (auto i : range(0, 100)) { ... } - * @endcode - * - * @param rbegin lower bound of range - * @param rend upper bound of range (excluding) - */ - template - inline Range range(T rbegin, T rend) - { - return Range(rbegin, rend); - } + /** Creates @ref Range with specified upper bound. + * + * @code + * for (auto i : range(100)) { ... } + * @endcode + * + * @param rend upper bound of range (excluding) + */ + template + inline Range range(T rend) + { + return Range(0, rend); + } + /** Creates @ref Range with specified bounds. + * + * @code + * for (auto i : range(0, 100)) { ... } + * @endcode + * + * @param rbegin lower bound of range + * @param rend upper bound of range (excluding) + */ + template + inline Range range(T rbegin, T rend) + { + return Range(rbegin, rend); + } } -#endif /* HAVE_CXX */ #endif /* __SG_RANGE_H__ */ diff --git a/src/shogun/base/some.h b/src/shogun/base/some.h index eb8094e4965..464c60949f5 100644 --- a/src/shogun/base/some.h +++ b/src/shogun/base/some.h @@ -1,148 +1,169 @@ #ifndef __SG_SOME_H__ #define __SG_SOME_H__ -#include - -#ifdef HAVE_CXX11 #include -#include - namespace shogun { - /** @class Shogun synonym for the std::shared_ptr. Employs - * exactly the same strategy for reference counting - * as std::shared_ptr: any operation involving copying increases - * the count and once deleted this wrapper decreases the counter. - * - */ - template - class Some - { - public: - Some(const Some& other); - explicit Some(T* other); - Some& operator=(T* other); - ~Some(); - - static Some from_raw(T* raw); - - /** Casts the underlying object back to raw pointer - * - * Be careful to SG_REF obtained pointer if you start to own it. - * - * @return raw pointer (without SG_REF) - */ - operator T*(); - /** Call member function or access member of T - * - * @return raw pointer (without SG_REF) - */ - T* operator->(); - private: - Some(); - void unref(); - void ref(); - private: - T* raw; - }; - - template - Some::Some() - : raw(nullptr) - { - } - template - Some::Some(const Some& other) - : raw(other.raw) - { - ref(); - } - template - Some::Some(T* other) - : raw(other) - { - ref(); - } - template - Some& Some::operator=(T* other) - { - if (raw != other) { - unref(); - raw = other; - ref(); - } - return *this; - } - template - Some::~Some() - { - unref(); - } - template - Some::operator T*() - { - return raw; - } - template - T* Some::operator->() - { - return raw; - } - template - void Some::ref() - { - SG_REF(raw); - } - template - void Some::unref() - { - SG_UNREF(raw); - } - template - Some Some::from_raw(T* raw) - { - Some result(raw); - return result; - } - - /** Creates an instance of any class - * that is wrapped with a shared pointer like - * structure @ref Some - * - * @param args arguments to construct instance of T with (T should - * have compatible constructor) - * - * @return a shared pointer that holds created instance of @ref T - * - */ - template - Some some(Args&&... args) - { - T* ptr = new T(args...); - return Some::from_raw(ptr); - } - - template - inline T wrap(const T& value) - { - return value; - } - - template - inline Some wrap(T* ptr) - { - return Some::from_raw(ptr); - } - - template - inline Some wrap(const Some& other) - { - return other; - } + /** @class Shogun synonym for the std::shared_ptr. Employs + * exactly the same strategy for reference counting + * as std::shared_ptr: any operation involving copying increases + * the count and once deleted this wrapper decreases the counter. + * + */ + template + class Some + { + public: + Some(const Some& other); + template + Some(const Some& other); + explicit Some(T* other); + + Some& operator=(T* other); + ~Some(); + + static Some from_raw(T* raw); + + /** Casts the underlying object back to raw pointer + * + * Be careful to SG_REF obtained pointer if you start to own it. + * + * @return raw pointer (without SG_REF) + */ + operator T*() const; + /** Call member function or access member of T + * + * @return raw pointer (without SG_REF) + */ + T* operator->() const; + + /** + * Get the raw pointer + * + * @return raw pointer (without SG_REF) + */ + T* get() const; + + private: + Some(); + void unref(); + void ref(); + + private: + T* raw = nullptr; + }; + + template + Some::Some() : raw(nullptr) + { + } + template + Some::Some(const Some& other) : raw(other.raw) + { + ref(); + } + template + Some::Some(T* other) : raw(other) + { + ref(); + } + template + template + Some::Some(const Some& other) + { + raw = dynamic_cast(other.get()); + ref(); + } + template + Some& Some::operator=(T* other) + { + if (raw != other) + { + unref(); + raw = other; + ref(); + } + return *this; + } + + template + Some::~Some() + { + unref(); + } + template + Some::operator T*() const + { + return raw; + } + template + T* Some::operator->() const + { + return raw; + } + template + T* Some::get() const + { + return raw; + } + template + void Some::ref() + { + if (raw) + (raw)->ref(); + } + template + void Some::unref() + { + if (raw) + { + if ((raw)->unref() == 0) + (raw) = NULL; + }; + } + template + Some Some::from_raw(T* raw) + { + Some result(raw); + return result; + } + + /** Creates an instance of any class + * that is wrapped with a shared pointer like + * structure @ref Some + * + * @param args arguments to construct instance of T with (T should + * have compatible constructor) + * + * @return a shared pointer that holds created instance of @ref T + * + */ + template + Some some(Args&&... args) + { + T* ptr = new T(args...); + return Some::from_raw(ptr); + } + + template + inline T wrap(const T& value) + { + return value; + } + + template + inline Some wrap(T* ptr) + { + return Some::from_raw(ptr); + } + template + inline Some wrap(const Some& other) + { + return other; + } }; -#endif /* HAVE_CXX11 */ #endif /* __SG_SOME_H__ */ diff --git a/src/shogun/classifier/AveragedPerceptron.cpp b/src/shogun/classifier/AveragedPerceptron.cpp index 35d89daa710..cf9ac008294 100644 --- a/src/shogun/classifier/AveragedPerceptron.cpp +++ b/src/shogun/classifier/AveragedPerceptron.cpp @@ -52,6 +52,7 @@ bool CAveragedPerceptron::train_machine(CFeatures* data) ASSERT(num_vec==train_labels.vlen) SGVector w(num_feat); float64_t* tmp_w=SG_MALLOC(float64_t, num_feat); + memset(tmp_w, 0, sizeof(float64_t)*num_feat); float64_t* output=SG_MALLOC(float64_t, num_vec); //start with uniform w, bias=0, tmp_bias=0 @@ -60,15 +61,13 @@ bool CAveragedPerceptron::train_machine(CFeatures* data) for (int32_t i=0; idense_dot(i, w.vector, w.vlen) + bias; diff --git a/src/shogun/classifier/FeatureBlockLogisticRegression.cpp b/src/shogun/classifier/FeatureBlockLogisticRegression.cpp deleted file mode 100644 index dd3aa8cdd22..00000000000 --- a/src/shogun/classifier/FeatureBlockLogisticRegression.cpp +++ /dev/null @@ -1,266 +0,0 @@ -/* - * This program is free software; you can redistribute it and/or modify - * it under the terms of the GNU General Public License as published by - * the Free Software Foundation; either version 3 of the License, or - * (at your option) any later version. - * - * Copyright (C) 2012 Sergey Lisitsyn - */ - - -#include -#ifdef USE_GPL_SHOGUN -#include -#include - -#include -#include - -namespace shogun -{ - -CFeatureBlockLogisticRegression::CFeatureBlockLogisticRegression() : - CLinearMachine() -{ - init(); - register_parameters(); -} - -CFeatureBlockLogisticRegression::CFeatureBlockLogisticRegression( - float64_t z, CDotFeatures* train_features, - CBinaryLabels* train_labels, CIndexBlockRelation* feature_relation) : - CLinearMachine() -{ - init(); - set_feature_relation(feature_relation); - set_z(z); - set_features(train_features); - set_labels(train_labels); - register_parameters(); -} - -void CFeatureBlockLogisticRegression::init() -{ - m_feature_relation=NULL; - m_z=0.0; - m_q=2.0; - m_termination=0; - m_regularization=0; - m_tolerance=1e-3; - m_max_iter=1000; -} - -CFeatureBlockLogisticRegression::~CFeatureBlockLogisticRegression() -{ - SG_UNREF(m_feature_relation); -} - -void CFeatureBlockLogisticRegression::register_parameters() -{ - SG_ADD((CSGObject**)&m_feature_relation, "feature_relation", "feature relation", MS_NOT_AVAILABLE); - SG_ADD(&m_z, "z", "regularization coefficient", MS_AVAILABLE); - SG_ADD(&m_q, "q", "q of L1/Lq", MS_AVAILABLE); - SG_ADD(&m_termination, "termination", "termination", MS_NOT_AVAILABLE); - SG_ADD(&m_regularization, "regularization", "regularization", MS_NOT_AVAILABLE); - SG_ADD(&m_tolerance, "tolerance", "tolerance", MS_NOT_AVAILABLE); - SG_ADD(&m_max_iter, "max_iter", "maximum number of iterations", MS_NOT_AVAILABLE); -} - -CIndexBlockRelation* CFeatureBlockLogisticRegression::get_feature_relation() const -{ - SG_REF(m_feature_relation); - return m_feature_relation; -} - -void CFeatureBlockLogisticRegression::set_feature_relation(CIndexBlockRelation* feature_relation) -{ - SG_REF(feature_relation); - SG_UNREF(m_feature_relation); - m_feature_relation = feature_relation; -} - -int32_t CFeatureBlockLogisticRegression::get_max_iter() const -{ - return m_max_iter; -} - -int32_t CFeatureBlockLogisticRegression::get_regularization() const -{ - return m_regularization; -} - -int32_t CFeatureBlockLogisticRegression::get_termination() const -{ - return m_termination; -} - -float64_t CFeatureBlockLogisticRegression::get_tolerance() const -{ - return m_tolerance; -} - -float64_t CFeatureBlockLogisticRegression::get_z() const -{ - return m_z; -} - -float64_t CFeatureBlockLogisticRegression::get_q() const -{ - return m_q; -} - -void CFeatureBlockLogisticRegression::set_max_iter(int32_t max_iter) -{ - ASSERT(max_iter>=0) - m_max_iter = max_iter; -} - -void CFeatureBlockLogisticRegression::set_regularization(int32_t regularization) -{ - ASSERT(regularization==0 || regularization==1) - m_regularization = regularization; -} - -void CFeatureBlockLogisticRegression::set_termination(int32_t termination) -{ - ASSERT(termination>=0 && termination<=4) - m_termination = termination; -} - -void CFeatureBlockLogisticRegression::set_tolerance(float64_t tolerance) -{ - ASSERT(tolerance>0.0) - m_tolerance = tolerance; -} - -void CFeatureBlockLogisticRegression::set_z(float64_t z) -{ - m_z = z; -} - -void CFeatureBlockLogisticRegression::set_q(float64_t q) -{ - m_q = q; -} - -bool CFeatureBlockLogisticRegression::train_machine(CFeatures* data) -{ - if (data && (CDotFeatures*)data) - set_features((CDotFeatures*)data); - - ASSERT(features) - ASSERT(m_labels) - - int32_t n_vecs = m_labels->get_num_labels(); - SGVector y(n_vecs); - for (int32_t i=0; iget_label(i); - - slep_options options = slep_options::default_options(); - options.q = m_q; - options.regularization = m_regularization; - options.termination = m_termination; - options.tolerance = m_tolerance; - options.max_iter = m_max_iter; - options.loss = LOGISTIC; - - EIndexBlockRelationType relation_type = m_feature_relation->get_relation_type(); - switch (relation_type) - { - case GROUP: - { - CIndexBlockGroup* feature_group = (CIndexBlockGroup*)m_feature_relation; - SGVector ind = feature_group->get_SLEP_ind(); - options.ind = ind.vector; - options.n_feature_blocks = ind.vlen-1; - if (ind[ind.vlen-1] > features->get_dim_feature_space()) - SG_ERROR("Group of features covers more features than available\n") - - options.gWeight = SG_MALLOC(double, options.n_feature_blocks); - for (int32_t i=0; iget_dim_feature_space(); - SGVector new_w(n_feats); - for (int i=0; i ind_t = feature_tree->get_SLEP_ind_t(); - SGVector G; - if (feature_tree->is_general()) - { - G = feature_tree->get_SLEP_G(); - options.general = true; - } - options.ind_t = ind_t.vector; - options.G = G.vector; - options.n_nodes = ind_t.vlen/3; - options.n_feature_blocks = ind_t.vlen/3; - options.mode = FEATURE_TREE; - options.loss = LOGISTIC; - - slep_result_t result = slep_solver(features, y.vector, m_z, options); - - int32_t n_feats = features->get_dim_feature_space(); - SGVector new_w(n_feats); - for (int i=0; i w = get_w(); - return CMath::exp(-(features->dense_dot(vec_idx, w.vector, w.vlen) + bias)); -} - -SGVector CFeatureBlockLogisticRegression::apply_get_outputs(CFeatures* data) -{ - if (data) - { - if (!data->has_property(FP_DOT)) - SG_ERROR("Specified features are not of type CDotFeatures\n") - - set_features((CDotFeatures*) data); - } - - if (!features) - return SGVector(); - - int32_t num=features->get_num_vectors(); - SGVector w = get_w(); - ASSERT(num>0) - ASSERT(w.vlen==features->get_dim_feature_space()) - - float64_t* out=SG_MALLOC(float64_t, num); - features->dense_dot_range(out, 0, num, NULL, w.vector, w.vlen, bias); - for (int32_t i=0; i(out,num); -} - -} -#endif //USE_GPL_SHOGUN diff --git a/src/shogun/classifier/FeatureBlockLogisticRegression.h b/src/shogun/classifier/FeatureBlockLogisticRegression.h deleted file mode 100644 index 89cafbdeb38..00000000000 --- a/src/shogun/classifier/FeatureBlockLogisticRegression.h +++ /dev/null @@ -1,141 +0,0 @@ -/* - * This program is free software; you can redistribute it and/or modify - * it under the terms of the GNU General Public License as published by - * the Free Software Foundation; either version 3 of the License, or - * (at your option) any later version. - * - * Copyright (C) 2012 Sergey Lisitsyn - */ - -#ifndef FEATUREBLOCKLOGISTICREGRESSION_H_ -#define FEATUREBLOCKLOGISTICREGRESSION_H_ - -#include -#ifdef USE_GPL_SHOGUN -#include -#include - -namespace shogun -{ -/** @brief class FeatureBlockLogisticRegression, a linear - * binary logistic loss classifier for problems with complex feature relations. - * Currently two feature relations are supported - feature group - * (done via CIndexBlockGroup) and feature tree (done via CIndexTree). - * Handling of feature relations is done via L1/Lq (for groups) and L1/L2 - * (for trees) regularization. - * - * The underlying solver is based on the SLEP library. - * - * @see CIndexBlock - * @see CIndexBlockGroup - * @see CIndexBlockTree - */ -class CFeatureBlockLogisticRegression : public CLinearMachine -{ - - public: - MACHINE_PROBLEM_TYPE(PT_BINARY) - - /** default constructor */ - CFeatureBlockLogisticRegression(); - - /** constructor - * - * @param z regularization coefficient - * @param training_data training features - * @param training_labels training labels - * @param task_relation task relation - */ - CFeatureBlockLogisticRegression( - float64_t z, CDotFeatures* training_data, - CBinaryLabels* training_labels, CIndexBlockRelation* task_relation); - - /** destructor */ - virtual ~CFeatureBlockLogisticRegression(); - - /** get name */ - virtual const char* get_name() const - { - return "FeatureBlockLogisticRegression"; - } - - /** getter for feature relation - * @return feature relation - */ - CIndexBlockRelation* get_feature_relation() const; - - /** setter for feature relation - * @param feature_relation feature relation - */ - void set_feature_relation(CIndexBlockRelation* feature_relation); - - virtual float64_t apply_one(int32_t vec_idx); - - /** get max iter */ - int32_t get_max_iter() const; - /** get q */ - float64_t get_q() const; - /** get regularization */ - int32_t get_regularization() const; - /** get termination */ - int32_t get_termination() const; - /** get tolerance */ - float64_t get_tolerance() const; - /** get z */ - float64_t get_z() const; - - /** set max iter */ - void set_max_iter(int32_t max_iter); - /** set q */ - void set_q(float64_t q); - /** set regularization */ - void set_regularization(int32_t regularization); - /** set termination */ - void set_termination(int32_t termination); - /** set tolerance */ - void set_tolerance(float64_t tolerance); - /** set z */ - void set_z(float64_t z); - - protected: - - virtual SGVector apply_get_outputs(CFeatures* data); - - /** train machine */ - virtual bool train_machine(CFeatures* data=NULL); - - private: - - /** register parameters */ - void register_parameters(); - - /** Initializes Parameters to std values */ - void init(); - - protected: - - /** feature tree */ - CIndexBlockRelation* m_feature_relation; - - /** regularization type */ - int32_t m_regularization; - - /** termination criteria */ - int32_t m_termination; - - /** max iteration */ - int32_t m_max_iter; - - /** tolerance */ - float64_t m_tolerance; - - /** q of L1/Lq */ - float64_t m_q; - - /** regularization coefficient */ - float64_t m_z; - -}; -} -#endif //USE_GPL_SHOGUN -#endif diff --git a/src/shogun/classifier/GaussianProcessClassification.cpp b/src/shogun/classifier/GaussianProcessClassification.cpp index 468905f84b6..9ef7cb9f591 100644 --- a/src/shogun/classifier/GaussianProcessClassification.cpp +++ b/src/shogun/classifier/GaussianProcessClassification.cpp @@ -39,7 +39,9 @@ #include #include #include +#ifdef USE_GPL_SHOGUN #include +#endif //USE_GPL_SHOGUN using namespace shogun; @@ -117,10 +119,14 @@ CBinaryLabels* CGaussianProcessClassification::apply_binary( { if (m_method->get_inference_type()== INF_FITC_LAPLACE_SINGLE) { +#ifdef USE_GPL_SHOGUN CSingleFITCLaplaceInferenceMethod* fitc_method= CSingleFITCLaplaceInferenceMethod::obtain_from_generic(m_method); data=fitc_method->get_inducing_features(); SG_UNREF(fitc_method); +#else + SG_GPL_ONLY +#endif //USE_GPL_SHOGUN } else data=m_method->get_features(); @@ -149,10 +155,14 @@ bool CGaussianProcessClassification::train_machine(CFeatures* data) // set inducing features for FITC inference method if (m_method->get_inference_type()==INF_FITC_LAPLACE_SINGLE) { +#ifdef USE_GPL_SHOGUN CSingleFITCLaplaceInferenceMethod* fitc_method= CSingleFITCLaplaceInferenceMethod::obtain_from_generic(m_method); fitc_method->set_inducing_features(data); SG_UNREF(fitc_method); +#else + SG_ERROR("Single FITC Laplace inference only supported under GPL.\n") +#endif //USE_GPL_SHOGUN } else m_method->set_features(data); diff --git a/src/shogun/classifier/LDA.cpp b/src/shogun/classifier/LDA.cpp index 6c2969ee1c5..51085187aa5 100644 --- a/src/shogun/classifier/LDA.cpp +++ b/src/shogun/classifier/LDA.cpp @@ -10,45 +10,51 @@ */ #include -#include -#include -#include #include -#include -#include -#include #include +#include +#include +#include +#include +#include using namespace Eigen; using namespace shogun; -CLDA::CLDA(float64_t gamma, ELDAMethod method) - :CLinearMachine() +CLDA::CLDA(float64_t gamma, ELDAMethod method, bool bdc_svd) + : CLinearMachine(false) { init(); m_method=method; m_gamma=gamma; + m_bdc_svd = bdc_svd; } -CLDA::CLDA(float64_t gamma, CDenseFeatures *traindat, - CLabels *trainlab, ELDAMethod method) - :CLinearMachine(), m_gamma(gamma) +CLDA::CLDA( + float64_t gamma, CDenseFeatures* traindat, CLabels* trainlab, + ELDAMethod method, bool bdc_svd) + : CLinearMachine(false), m_gamma(gamma) { init(); set_features(traindat); set_labels(trainlab); m_method=method; m_gamma=gamma; + m_bdc_svd = bdc_svd; } void CLDA::init() { m_method=AUTO_LDA; m_gamma=0; - SG_ADD((machine_int_t*) &m_method, "m_method", - "Method used for LDA calculation", MS_NOT_AVAILABLE); - SG_ADD((machine_int_t*) &m_gamma, "m_gamma", - "Regularization parameter", MS_NOT_AVAILABLE); + m_bdc_svd = true; + SG_ADD( + (machine_int_t*)&m_method, "m_method", + "Method used for LDA calculation", MS_NOT_AVAILABLE); + SG_ADD( + (machine_int_t*)&m_gamma, "m_gamma", "Regularization parameter", + MS_NOT_AVAILABLE); + SG_ADD(&m_bdc_svd, "m_bdc_svd", "Use BDC-SVD algorithm", MS_NOT_AVAILABLE); } CLDA::~CLDA() @@ -58,8 +64,10 @@ CLDA::~CLDA() bool CLDA::train_machine(CFeatures *data) { REQUIRE(m_labels, "Labels for the given features are not specified!\n") - REQUIRE(m_labels->get_label_type()==LT_BINARY, "The labels should of type" - " CBinaryLabels! you provided %s \n",m_labels->get_name()) + REQUIRE( + m_labels->get_label_type() == LT_BINARY, + "The labels should of type CBinaryLabels! Provided type is %s \n", + m_labels->get_name()) if(data) { @@ -73,192 +81,117 @@ bool CLDA::train_machine(CFeatures *data) REQUIRE(data, "Features have not been provided.\n") } - SGVectortrain_labels=((CBinaryLabels *)m_labels)->get_int_labels(); - REQUIRE(train_labels.vector,"Provided Labels are empty!\n") + REQUIRE( + data->get_num_vectors() == m_labels->get_num_labels(), + "Number of training examples(%d) should be equal to number of labels " + "(%d)!\n", + data->get_num_vectors(), m_labels->get_num_labels()); - REQUIRE(data->get_num_vectors() == train_labels.vlen,"Number of training examples(%d) should be " - "equal to number of labels (%d)!\n", data->get_num_vectors(), train_labels.vlen); + REQUIRE( + features->get_feature_class() == C_DENSE, + "LDA only works with dense features") if(data->get_feature_type() == F_SHORTREAL) - return CLDA::train_machine_templated(train_labels, data); + return CLDA::train_machine_templated(); else if(data->get_feature_type() == F_DREAL) - return CLDA::train_machine_templated(train_labels, data); + return CLDA::train_machine_templated(); else if(data->get_feature_type() == F_LONGREAL) - return CLDA::train_machine_templated(train_labels, data); + return CLDA::train_machine_templated(); return false; } template -bool CLDA::train_machine_templated(SGVector train_labels, CFeatures *data) -{ - SGMatrixfeature_matrix=((CDenseFeatures*)features) - ->get_feature_matrix(); - int32_t num_feat=feature_matrix.num_rows; - int32_t num_vec=feature_matrix.num_cols; - - SGVector classidx_neg(num_vec); - SGVector classidx_pos(num_vec); - - int32_t i=0; - int32_t num_neg=0; - int32_t num_pos=0; - - for(i=0; i w_st(num_feat); - w_st.zero(); - typename SGMatrix::EigenMatrixXt fmatrix=typename SGMatrix::EigenMatrixXtMap(feature_matrix.matrix, num_feat, num_vec); - typename SGVector::EigenVectorXt mean_neg(num_feat); - mean_neg.setZero(); - typename SGVector::EigenVectorXt mean_pos(num_feat); - mean_pos.setZero(); - - //mean neg - for(i=0; iscatter_matrix(num_feat, num_feat); - typename SGMatrix::EigenMatrixXtMap scatter(scatter_matrix.matrix, num_feat, num_feat); - - if (m_method == FLD_LDA || (m_method==AUTO_LDA && num_vec>num_feat)) - { - // covariance matrix. - typename SGMatrix::EigenMatrixXt cov_mat(num_feat, num_feat); - cov_mat=fmatrix*fmatrix.transpose(); - scatter=cov_mat/(num_vec-1); - ST trace=scatter.trace(); - ST s=1.0-((ST) m_gamma); - scatter*=s; - scatter.diagonal()+=Eigen::DenseBase::EigenVectorXt>::Constant(num_feat, trace*((ST)m_gamma)/num_feat); - - // the usual way - // we need to find a Basic Linear Solution of A.x=b for 'x'. - // Instead of crudely Inverting A, we go for solve() using Decompositions. - // where: - // MatrixXd A=scatter; - // VectorXd b=mean_pos-mean_neg; - // VectorXd x=w; - typename SGVector::EigenVectorXtMap x(w_st.vector, num_feat); - LLT::EigenMatrixXt> decomposition(scatter); - x=decomposition.solve(mean_pos-mean_neg); +bool CLDA::train_machine_templated() +{ + index_t num_feat = ((CDenseFeatures*)features)->get_num_features(); + index_t num_vec = features->get_num_vectors(); + ; - // get the weights w_neg(for -ve class) and w_pos(for +ve class) - typename SGVector::EigenVectorXt w_neg=decomposition.solve(mean_neg); - typename SGVector::EigenVectorXt w_pos=decomposition.solve(mean_pos); + bool lda_more_efficient = (m_method == AUTO_LDA && num_vec <= num_feat); - // get the bias. - bias=((float64_t)(0.5*(w_neg.dot(mean_neg)-w_pos.dot(mean_pos)))); - } + if (m_method == SVD_LDA || lda_more_efficient) + return solver_svd(); else - { - //for algorithmic detail, please refer to section 16.3.1. of Bayesian - //Reasoning and Machine Learning by David Barber. - - //we will perform SVD here. - typename SGMatrix::EigenMatrixXtMap fmatrix1(feature_matrix.matrix, num_feat, num_vec); - - // to hold the centered positive and negative class data - typename SGMatrix::EigenMatrixXt cen_pos(num_feat,num_pos); - typename SGMatrix::EigenMatrixXt cen_neg(num_feat,num_neg); - - for(i=0; i::EigenMatrixXt Sw= num_pos*cen_pos+num_neg*cen_neg; - ST trace=Sw.trace(); - ST s=1.0-((ST)m_gamma); - Sw *=s; - Sw.diagonal()+=Eigen::DenseBase::EigenVectorXt>::Constant(num_feat, trace*((ST)m_gamma)/num_feat); - - //total mean - typename SGVector::EigenVectorXt mean_total=(num_pos*mean_pos+num_neg*mean_neg)/(ST)num_vec; - - //between class matrix - typename SGMatrix::EigenMatrixXt Sb(num_feat,2); - Sb.col(0)=sqrt(num_pos)*(mean_pos-mean_total); - Sb.col(1)=sqrt(num_neg)*(mean_neg-mean_total); + return solver_classic(); +} - JacobiSVD::EigenMatrixXt> svd(fmatrix1, ComputeThinU); +template +bool CLDA::solver_svd() +{ + auto dense_feat = static_cast*>(features); - // basis to represent the solution - typename SGMatrix::EigenMatrixXt Q=svd.matrixU(); - // modified between class scatter - Sb=Q.transpose()*(Sb*(Sb.transpose()))*Q; + // keep just one dimension to do binary classification + const index_t projection_dim = 1; + auto solver = std::unique_ptr>( + new LDACanVarSolver( + dense_feat, + new CMulticlassLabels(static_cast(m_labels)), + projection_dim, m_gamma, m_bdc_svd)); - // modified within class scatter - Sw=Q.transpose()*Sw*Q; + SGVector w_st(solver->get_eigenvectors()); - // to find SVD((inverse(Chol(Sw)))' * Sb * (inverse(Chol(Sw)))) - //1.get Cw=Chol(Sw) - //find the decomposition of Cw'. - HouseholderQR::EigenMatrixXt> decomposition(Sw.llt().matrixU().transpose()); + auto class_mean = solver->get_class_mean(); + ST m_neg = linalg::dot(w_st, class_mean[0]); + ST m_pos = linalg::dot(w_st, class_mean[1]); - //2.get P=inv(Cw')*Sb_new - //MatrixXd P=decomposition.solve(Sb); - //3. final value to be put in SVD will be therefore: - // final_ output =(inv(Cw')*(P'))' - JacobiSVD::EigenMatrixXt> svd2(decomposition.solve((decomposition.solve(Sb)) - .transpose()).transpose(), ComputeThinU); + // change the sign of w if needed to get the correct labels + float64_t sign = (m_pos > m_neg) ? 1 : -1; - // Since this is a linear classifier, with only binary classes, - // we need to keep only the 1st eigenvector. - Map::EigenVectorXt> x(w_st.vector, num_feat); - x=Q*(svd2.matrixU().col(0)); - // get the bias - bias=((float64_t)(x.transpose()*mean_total)); - bias=bias*(-1); - } + SGVector w(dense_feat->get_num_features()); + // copy w_st into w + for (index_t i = 0; i < w.size(); ++i) + w[i] = sign * w_st[i]; + set_w(w); - SGVector w(num_feat); - w.zero(); + set_bias(-0.5 * sign * (m_neg + m_pos)); - //copy w_st into w - for(i = 0; i < w.size(); ++i) - w[i] = (float64_t) w_st[i]; + return true; +} +template +bool CLDA::solver_classic() +{ + auto dense_feat = static_cast*>(features); + index_t num_feat = dense_feat->get_num_features(); + + auto solver = std::unique_ptr>( + new LDASolver( + dense_feat, + new CMulticlassLabels(static_cast(m_labels)), + m_gamma)); + + auto class_mean = solver->get_class_mean(); + auto class_count = solver->get_class_count(); + SGMatrix scatter_matrix = solver->get_within_cov(); + + // the usual way + // we need to find a Basic Linear Solution of A.x=b for 'x'. + // Instead of crudely Inverting A, we go for solve() using Decompositions. + // where: + // MatrixXd A=scatter; + // VectorXd b=mean_pos-mean_neg; + // VectorXd x=w; + auto decomposition = linalg::cholesky_factor(scatter_matrix); + SGVector w_st = linalg::cholesky_solver( + decomposition, + linalg::add(class_mean[1], class_mean[0], (ST)1, (ST)-1)); + + // get the weights w_neg(for -ve class) and w_pos(for +ve class) + auto w_neg = linalg::cholesky_solver(decomposition, class_mean[0]); + auto w_pos = linalg::cholesky_solver(decomposition, class_mean[1]); + + SGVector w(num_feat); + // copy w_st into w + for (index_t i = 0; i < w.size(); ++i) + w[i] = (float64_t)w_st[i]; set_w(w); + // get the bias. + set_bias( + (float64_t)( + 0.5 * (linalg::dot(w_neg, class_mean[0]) - + linalg::dot(w_pos, class_mean[1])))); + return true; } diff --git a/src/shogun/classifier/LDA.h b/src/shogun/classifier/LDA.h index 06f43d56af3..debbd08e1b7 100644 --- a/src/shogun/classifier/LDA.h +++ b/src/shogun/classifier/LDA.h @@ -32,7 +32,7 @@ enum ELDAMethod /** Singular Value Decomposition based LDA. */ SVD_LDA = 20, - /** Fisher two class discrimiant based LDA. + /** Fisher two class discriminant based LDA. */ FLD_LDA = 30 }; @@ -55,7 +55,8 @@ template class CDenseFeatures; * is maximized, where * \f[S_b := ({\bf m_{+1}} - {\bf m_{-1}})({\bf m_{+1}} - {\bf m_{-1}})^T \f] * is the between class scatter matrix and - * \f[S_w := \sum_{c\in\{-1,+1\}}\sum_{{\bf x}\in X_{c}}({\bf x} - {\bf m_c})({\bf x} - {\bf m_c})^T \f] + * \f[S_w := \sum_{c\in\{-1,+1\}}\sum_{{\bf x}\in X_{c}}({\bf x} - {\bf + * m_c})({\bf x} - {\bf m_c})^T \f] * is the within class scatter matrix with mean \f${\bf m_c} := * \frac{1}{N}\sum_{j=1}^N {\bf x_j^c}\f$ and \f$X_c:=\{x_1^c, \dots, x_N^c\}\f$ * the set of examples of class c. @@ -66,10 +67,14 @@ template class CDenseFeatures; * * ::SVD_LDA : Singular Valued decomposition method. * The above derivation of Fisher's LDA requires the invertibility of the within - * class matrix. However, this condition gets void when there are fewer data-points - * than dimensions. A solution is to require that \f${\bf W}\f$ lies only in the subspace - * spanned by the data. A basis of the data \f${\bf X}\f$ is found using the thin-SVD technique - * which returns an orthonormal non-square basis matrix \f${\bf Q}\f$. We then require the + * class matrix. However, this condition gets void when there are fewer + * data-points + * than dimensions. A solution is to require that \f${\bf W}\f$ lies only in the + * subspace + * spanned by the data. A basis of the data \f${\bf X}\f$ is found using the + * thin-SVD technique + * which returns an orthonormal non-square basis matrix \f${\bf Q}\f$. We then + * require the * solution \f${\bf w}\f$ to be expressed in this basis. * \f[{\bf W} := {\bf Q} {\bf{W^\prime}}\f] * The between class Matrix is replaced with: @@ -80,8 +85,13 @@ template class CDenseFeatures; * been projected down to the basis that spans the data. * see: Bayesian Reasoning and Machine Learning, section 16.3.1. * - * ::AUTO_LDA : This mode automagically chooses one of the above modes for - * the users based on whether N > D (chooses ::FLD_LDA) or N < D(chooses ::SVD_LDA) + * ::AUTO_LDA : This mode automagically chooses one of the above modes + * for + * the users based on whether N > D (chooses ::FLD_LDA) or N < D(chooses + * ::SVD_LDA) + * Note that even if N > D FLD_LDA may fail being the covariance matrix not + * invertible, + * in such case one should use SVD_LDA. * \sa CLinearMachine * \sa http://en.wikipedia.org/wiki/Linear_discriminant_analysis */ @@ -94,19 +104,33 @@ class CLDA : public CLinearMachine /** constructor * * @param gamma gamma - * @param method LDA using Fisher's algorithm or Singular Value Decomposition : ::FLD_LDA/::SVD_LDA/::AUTO_LDA[default] + * @param method LDA using Fisher's algorithm or Singular Value + * Decomposition : ::FLD_LDA/::SVD_LDA/::AUTO_LDA[default] + * @param bdc_svd when using SVD solver switch between + * Bidiagonal Divide and Conquer algorithm (BDC) and + * Jacobi's algorithm, for the differences @see linalg::SVDAlgorithm. + * [default = BDC-SVD] */ - CLDA(float64_t gamma=0, ELDAMethod method=AUTO_LDA); + CLDA( + float64_t gamma = 0, ELDAMethod method = AUTO_LDA, + bool bdc_svd = true); /** constructor * * @param gamma gamma * @param traindat training features * @param trainlab labels for training features - * @param method LDA using Fisher's algorithm or Singular Value Decomposition : ::FLD_LDA/::SVD_LDA/::AUTO_LDA[default] - + * @param method LDA using Fisher's algorithm or Singular Value + * Decomposition : ::FLD_LDA/::SVD_LDA/::AUTO_LDA[default] + * @param bdc_svd when using SVD solver switch between + * Bidiagonal Divide and Conquer algorithm (BDC-SVD) and + * Jacobi's algorithm, for the differences @see linalg::SVDAlgorithm. + * [default = BDC-SVD] */ - CLDA(float64_t gamma, CDenseFeatures* traindat, CLabels* trainlab, ELDAMethod method=AUTO_LDA); + CLDA( + float64_t gamma, CDenseFeatures* traindat, + CLabels* trainlab, ELDAMethod method = AUTO_LDA, + bool bdc_svd = true); virtual ~CLDA(); /** set gamma @@ -170,7 +194,24 @@ class CLDA : public CLinearMachine * @see train_machine */ template - bool train_machine_templated(SGVector train_labels, CFeatures * features); + bool train_machine_templated(); + + /** + * Train the machine with the svd-based solver (@see CFisherLDA). + * @param features training data + * @param labels labels for training data + */ + template + bool solver_svd(); + + /** + * Train the machine with the classic method based on the cholesky + * decomposition of the covariance matrix. + * @param features training data + * @param labels labels for training data + */ + template + bool solver_classic(); protected: @@ -180,6 +221,8 @@ class CLDA : public CLinearMachine float64_t m_gamma; /** LDA mode */ ELDAMethod m_method; + /** use bdc-svd algorithm */ + bool m_bdc_svd; }; } #endif//ifndef diff --git a/src/shogun/classifier/LPBoost.cpp b/src/shogun/classifier/LPBoost.cpp index b1b9bd5618c..1765439d6fe 100644 --- a/src/shogun/classifier/LPBoost.cpp +++ b/src/shogun/classifier/LPBoost.cpp @@ -124,9 +124,8 @@ bool CLPBoost::train_machine(CFeatures* data) int32_t num_hypothesis=0; CTime time; - CSignal::clear_cancel(); - while (!(CSignal::cancel_computations())) + while (!(cancel_computation())) { int32_t max_dim=0; float64_t violator=find_max_violator(max_dim); diff --git a/src/shogun/classifier/Perceptron.cpp b/src/shogun/classifier/Perceptron.cpp index 07cc289e5ab..7e1a3866a52 100644 --- a/src/shogun/classifier/Perceptron.cpp +++ b/src/shogun/classifier/Perceptron.cpp @@ -13,6 +13,7 @@ #include #include #include +#include using namespace shogun; @@ -64,21 +65,23 @@ bool CPerceptron::train_machine(CFeatures* data) w.vector[i]=1.0/num_feat; } - CSignal::clear_cancel(); //loop till we either get everything classified right or reach max_iter - while (!(CSignal::cancel_computations()) && (!converged && iterindex_iterator()) { - output[i] = features->dense_dot(i, w.vector, w.vlen) + bias; + const auto predicted_label = features->dense_dot(example_idx, w.vector, w.vlen) + bias; + const auto true_label = train_labels[example_idx]; + output[example_idx] = predicted_label; - if (CMath::sign(output[i]) != train_labels.vector[i]) + if (CMath::sign(predicted_label) != true_label) { - converged=false; - bias+=learn_rate*train_labels.vector[i]; - features->add_to_dense_vec(learn_rate*train_labels.vector[i], i, w.vector, w.vlen); + converged = false; + const auto gradient = learn_rate * train_labels[example_idx]; + bias += gradient; + features->add_to_dense_vec(gradient, example_idx, w.vector, w.vlen); } } diff --git a/src/shogun/classifier/mkl/MKL.cpp b/src/shogun/classifier/mkl/MKL.cpp index df758630d63..bbb96b05cd3 100644 --- a/src/shogun/classifier/mkl/MKL.cpp +++ b/src/shogun/classifier/mkl/MKL.cpp @@ -427,7 +427,6 @@ bool CMKL::train_machine(CFeatures* data) #endif mkl_iterations = 0; - CSignal::clear_cancel(); training_time_clock.start(); @@ -449,17 +448,13 @@ bool CMKL::train_machine(CFeatures* data) //but if we don't actually unref() the object we might leak memory... //So as a workaround we only unref when the reference count was >1 //before. -#ifdef USE_REFERENCE_COUNTING int32_t refs=this->ref(); -#endif svm->set_callback_function(this, perform_mkl_step_helper); svm->train(); SG_DONE() svm->set_callback_function(NULL, NULL); -#ifdef USE_REFERENCE_COUNTING if (refs>1) this->unref(); -#endif } else { @@ -482,7 +477,7 @@ bool CMKL::train_machine(CFeatures* data) mkl_iterations++; - if (perform_mkl_step(sumw, suma) || CSignal::cancel_computations()) + if (perform_mkl_step(sumw, suma) || cancel_computation()) break; } diff --git a/src/shogun/classifier/mkl/MKLClassification.cpp b/src/shogun/classifier/mkl/MKLClassification.cpp index 2b7833f9d6b..e80564b5ab4 100644 --- a/src/shogun/classifier/mkl/MKLClassification.cpp +++ b/src/shogun/classifier/mkl/MKLClassification.cpp @@ -38,3 +38,16 @@ void CMKLClassification::init_training() REQUIRE(m_labels->get_num_labels(), "Number of labels is zero.\n"); REQUIRE(m_labels->get_label_type() == LT_BINARY, "Labels must be binary.\n"); } + +CMKLClassification* CMKLClassification::obtain_from_generic(CMachine* machine) +{ + if (machine == NULL) + return NULL; + + if (machine->get_classifier_type() != CT_MKLCLASSIFICATION) + SG_SERROR("Provided machine is not of type CMKLClassification!") + + CMKLClassification* casted = dynamic_cast(machine); + SG_REF(casted) + return casted; +} \ No newline at end of file diff --git a/src/shogun/classifier/mkl/MKLClassification.h b/src/shogun/classifier/mkl/MKLClassification.h index af666d5f83a..c868c9cc744 100644 --- a/src/shogun/classifier/mkl/MKLClassification.h +++ b/src/shogun/classifier/mkl/MKLClassification.h @@ -41,6 +41,13 @@ class CMKLClassification : public CMKL */ virtual float64_t compute_sum_alpha(); + /** + * Helper method used to specialize a base class instance. + * @param machine the machine we want to cast + * @return a MKLClassification machine (already SGREF'ed) + */ + static CMKLClassification* obtain_from_generic(CMachine* machine); + /** @return object name */ virtual const char* get_name() const { return "MKLClassification"; } diff --git a/src/shogun/classifier/mkl/MKLMulticlass.cpp b/src/shogun/classifier/mkl/MKLMulticlass.cpp index 8ea604c60a9..d3b00c108e5 100644 --- a/src/shogun/classifier/mkl/MKLMulticlass.cpp +++ b/src/shogun/classifier/mkl/MKLMulticlass.cpp @@ -370,9 +370,8 @@ bool CMKLMulticlass::train_machine(CFeatures* data) int32_t numberofsilpiterations=0; bool final=false; - CSignal::clear_cancel(); - while (!(CSignal::cancel_computations()) && !final) + while (!(cancel_computation()) && !final) { //curweights.clear(); diff --git a/src/shogun/classifier/mkl/MKLMulticlassGLPK.cpp b/src/shogun/classifier/mkl/MKLMulticlassGLPK.cpp index 8b692d7ee79..6bba60e1fd4 100644 --- a/src/shogun/classifier/mkl/MKLMulticlassGLPK.cpp +++ b/src/shogun/classifier/mkl/MKLMulticlassGLPK.cpp @@ -13,6 +13,8 @@ #include #include + +#include #include #ifdef USE_GLPK diff --git a/src/shogun/classifier/svm/GPBTSVM.cpp b/src/shogun/classifier/svm/GPBTSVM.cpp deleted file mode 100644 index b787807786d..00000000000 --- a/src/shogun/classifier/svm/GPBTSVM.cpp +++ /dev/null @@ -1,135 +0,0 @@ -/* - * This program is free software; you can redistribute it and/or modify - * it under the terms of the GNU General Public License as published by - * the Free Software Foundation; either version 3 of the License, or - * (at your option) any later version. - * - * Written (W) 1999-2009 Soeren Sonnenburg - * Copyright (C) 1999-2009 Fraunhofer Institute FIRST and Max-Planck-Society - */ - - -#include -#ifdef USE_GPL_SHOGUN -#include -#include -#include -#include - -using namespace shogun; - -CGPBTSVM::CGPBTSVM() -: CSVM(), model(NULL) -{ -} - -CGPBTSVM::CGPBTSVM(float64_t C, CKernel* k, CLabels* lab) -: CSVM(C, k, lab), model(NULL) -{ -} - -CGPBTSVM::~CGPBTSVM() -{ - SG_FREE(model); -} - -bool CGPBTSVM::train_machine(CFeatures* data) -{ - float64_t* solution; /* store the solution found */ - QPproblem prob; /* object containing the solvers */ - - ASSERT(kernel) - ASSERT(m_labels && m_labels->get_num_labels()) - ASSERT(m_labels->get_label_type() == LT_BINARY) - if (data) - { - if (m_labels->get_num_labels() != data->get_num_vectors()) - SG_ERROR("Number of training vectors does not match number of labels\n") - kernel->init(data, data); - } - - SGVector lab=((CBinaryLabels*) m_labels)->get_int_labels(); - prob.KER=new sKernel(kernel, lab.vlen); - prob.y=lab.vector; - prob.ell=lab.vlen; - SG_INFO("%d trainlabels\n", prob.ell) - - // /*** set options defaults ***/ - prob.delta = epsilon; - prob.maxmw = kernel->get_cache_size(); - prob.verbosity = 0; - prob.preprocess_size = -1; - prob.projection_projector = -1; - prob.c_const = get_C1(); - prob.chunk_size = get_qpsize(); - prob.linadd = get_linadd_enabled(); - - if (prob.chunk_size < 2) prob.chunk_size = 2; - if (prob.q <= 0) prob.q = prob.chunk_size / 3; - if (prob.q < 2) prob.q = 2; - if (prob.q > prob.chunk_size) prob.q = prob.chunk_size; - prob.q = prob.q & (~1); - if (prob.maxmw < 5) - prob.maxmw = 5; - - /*** set the problem description for final report ***/ - SG_INFO("\nTRAINING PARAMETERS:\n") - SG_INFO("\tNumber of training documents: %d\n", prob.ell) - SG_INFO("\tq: %d\n", prob.chunk_size) - SG_INFO("\tn: %d\n", prob.q) - SG_INFO("\tC: %lf\n", prob.c_const) - SG_INFO("\tkernel type: %d\n", prob.ker_type) - SG_INFO("\tcache size: %dMb\n", prob.maxmw) - SG_INFO("\tStopping tolerance: %lf\n", prob.delta) - - // /*** compute the number of cache rows up to maxmw Mb. ***/ - if (prob.preprocess_size == -1) - prob.preprocess_size = (int32_t) ( (float64_t)prob.chunk_size * 1.5 ); - - if (prob.projection_projector == -1) - { - if (prob.chunk_size <= 20) prob.projection_projector = 0; - else prob.projection_projector = 1; - } - - /*** compute the problem solution *******************************************/ - solution = SG_MALLOC(float64_t, prob.ell); - prob.gpdtsolve(solution); - /****************************************************************************/ - - CSVM::set_objective(prob.objective_value); - - int32_t num_sv=0; - int32_t bsv=0; - int32_t i=0; - int32_t k=0; - - for (i = 0; i < prob.ell; i++) - { - if (solution[i] > prob.DELTAsv) - { - num_sv++; - if (solution[i] > (prob.c_const - prob.DELTAsv)) bsv++; - } - } - - create_new_model(num_sv); - set_bias(prob.bee); - - SG_INFO("SV: %d BSV = %d\n", num_sv, bsv) - - for (i = 0; i < prob.ell; i++) - { - if (solution[i] > prob.DELTAsv) - { - set_support_vector(k, i); - set_alpha(k++, solution[i]*((CBinaryLabels*) m_labels)->get_label(i)); - } - } - - delete prob.KER; - SG_FREE(solution); - - return true; -} -#endif //USE_GPL_SHOGUN diff --git a/src/shogun/classifier/svm/GPBTSVM.h b/src/shogun/classifier/svm/GPBTSVM.h deleted file mode 100644 index 810df77d5fd..00000000000 --- a/src/shogun/classifier/svm/GPBTSVM.h +++ /dev/null @@ -1,65 +0,0 @@ -/* - * This program is free software; you can redistribute it and/or modify - * it under the terms of the GNU General Public License as published by - * the Free Software Foundation; either version 3 of the License, or - * (at your option) any later version. - * - * Written (W) 1999-2009 Soeren Sonnenburg - * Copyright (C) 1999-2009 Fraunhofer Institute FIRST and Max-Planck-Society - */ - -#ifndef _GPBTSVM_H___ -#define _GPBTSVM_H___ - -#include -#ifdef USE_GPL_SHOGUN -#include -#include -#include - - -namespace shogun -{ -/** @brief class GPBTSVM */ -class CGPBTSVM : public CSVM -{ - public: - /** default constructor */ - CGPBTSVM(); - - /** constructor - * - * @param C constant C - * @param k kernel - * @param lab labels - */ - CGPBTSVM(float64_t C, CKernel* k, CLabels* lab); - virtual ~CGPBTSVM(); - - /** get classifier type - * - * @return classifier type GPBT - */ - virtual EMachineType get_classifier_type() { return CT_GPBT; } - - /** @return object name */ - virtual const char* get_name() const { return "GPBTSVM"; } - - protected: - /** train SVM classifier - * - * @param data training data (parameter can be avoided if distance or - * kernel-based classifiers are used and distance/kernels are - * initialized with train data) - * - * @return whether training was successful - */ - virtual bool train_machine(CFeatures* data=NULL); - - protected: - /** SVM model */ - struct svm_model* model; -}; -} -#endif //USE_GPL_SHOGUN -#endif diff --git a/src/shogun/classifier/svm/LibLinear.cpp b/src/shogun/classifier/svm/LibLinear.cpp index d81989f62d6..3940dabba75 100644 --- a/src/shogun/classifier/svm/LibLinear.cpp +++ b/src/shogun/classifier/svm/LibLinear.cpp @@ -10,14 +10,15 @@ */ #include -#include -#include -#include #include +#include #include -#include #include +#include #include +#include +#include +#include using namespace shogun; @@ -77,7 +78,7 @@ CLibLinear::~CLibLinear() bool CLibLinear::train_machine(CFeatures* data) { - CSignal::clear_cancel(); + ASSERT(m_labels) ASSERT(m_labels->get_label_type() == LT_BINARY) @@ -313,9 +314,9 @@ void CLibLinear::solve_l2r_l1l2_svc( index[i] = i; } - + auto pb = progress(range(10)); CTime start_time; - while (iter < max_iterations && !CSignal::cancel_computations()) + while (iter < max_iterations && !cancel_computation()) { if (m_max_train_time > 0 && start_time.cur_time_diff() > m_max_train_time) break; @@ -392,7 +393,8 @@ void CLibLinear::solve_l2r_l1l2_svc( iter++; float64_t gap=PGmax_new - PGmin_new; - SG_SABS_PROGRESS(gap, -CMath::log10(gap), -CMath::log10(1), -CMath::log10(eps), 6) + pb.print_absolute( + gap, -CMath::log10(gap), -CMath::log10(1), -CMath::log10(eps)); if(gap <= eps) { @@ -414,7 +416,7 @@ void CLibLinear::solve_l2r_l1l2_svc( PGmin_old = -CMath::INFTY; } - SG_DONE() + pb.complete_absolute(); SG_INFO("optimization finished, #iter = %d\n",iter) if (iter >= max_iterations) { @@ -522,9 +524,9 @@ void CLibLinear::solve_l1r_l2_svc( } } - + auto pb = progress(range(10)); CTime start_time; - while (iter < max_iterations && !CSignal::cancel_computations()) + while (iter < max_iterations && !cancel_computation()) { if (m_max_train_time > 0 && start_time.cur_time_diff() > m_max_train_time) break; @@ -742,8 +744,9 @@ void CLibLinear::solve_l1r_l2_svc( Gmax_init = Gmax_new; iter++; - SG_SABS_PROGRESS(Gmax_new, -CMath::log10(Gmax_new), - -CMath::log10(Gmax_init), -CMath::log10(eps*Gmax_init), 6); + pb.print_absolute( + Gmax_new, -CMath::log10(Gmax_new), -CMath::log10(Gmax_init), + -CMath::log10(eps * Gmax_init)); if(Gmax_new <= eps*Gmax_init) { @@ -760,7 +763,7 @@ void CLibLinear::solve_l1r_l2_svc( Gmax_old = Gmax_new; } - SG_DONE() + pb.complete_absolute(); SG_INFO("optimization finished, #iter = %d\n", iter) if(iter >= max_iterations) SG_WARNING("\nWARNING: reaching max number of iterations\n") @@ -893,8 +896,9 @@ void CLibLinear::solve_l1r_lr( } } + auto pb = progress(range(10)); CTime start_time; - while (iter < max_iterations && !CSignal::cancel_computations()) + while (iter < max_iterations && !cancel_computation()) { if (m_max_train_time > 0 && start_time.cur_time_diff() > m_max_train_time) break; @@ -1105,7 +1109,9 @@ void CLibLinear::solve_l1r_lr( if(iter == 0) Gmax_init = Gmax_new; iter++; - SG_SABS_PROGRESS(Gmax_new, -CMath::log10(Gmax_new), -CMath::log10(Gmax_init), -CMath::log10(eps*Gmax_init), 6) + pb.print_absolute( + Gmax_new, -CMath::log10(Gmax_new), -CMath::log10(Gmax_init), + -CMath::log10(eps * Gmax_init)); if(Gmax_new <= eps*Gmax_init) { @@ -1122,7 +1128,7 @@ void CLibLinear::solve_l1r_lr( Gmax_old = Gmax_new; } - SG_DONE() + pb.complete_absolute(); SG_INFO("optimization finished, #iter = %d\n", iter) if(iter >= max_iterations) SG_WARNING("\nWARNING: reaching max number of iterations\n") @@ -1230,6 +1236,7 @@ void CLibLinear::solve_l2r_lr_dual(SGVector& w, const liblinear_probl index[i] = i; } + auto pb = progress(range(10)); while (iter < max_iter) { for (i=0; i& w, const liblinear_probl Gmax_init = Gmax; iter++; - SG_SABS_PROGRESS(Gmax, -CMath::log10(Gmax), -CMath::log10(Gmax_init), -CMath::log10(eps*Gmax_init), 6) + pb.print_absolute( + Gmax, -CMath::log10(Gmax), -CMath::log10(Gmax_init), + -CMath::log10(eps * Gmax_init)); if(Gmax < eps) break; @@ -1314,7 +1323,7 @@ void CLibLinear::solve_l2r_lr_dual(SGVector& w, const liblinear_probl } - SG_DONE() + pb.complete_absolute(); SG_INFO("optimization finished, #iter = %d\n",iter) if (iter >= max_iter) SG_WARNING("reaching max number of iterations\nUsing -s 0 may be faster (also see FAQ)\n\n") diff --git a/src/shogun/classifier/svm/MPDSVM.cpp b/src/shogun/classifier/svm/MPDSVM.cpp index 33c7a60f9cf..89abf942983 100644 --- a/src/shogun/classifier/svm/MPDSVM.cpp +++ b/src/shogun/classifier/svm/MPDSVM.cpp @@ -79,7 +79,6 @@ bool CMPDSVM::train_machine(CFeatures* data) bool primalcool; bool dualcool; - CSignal::clear_cancel(); //if (nustop) //etas[1] = 1; @@ -96,7 +95,7 @@ bool CMPDSVM::train_machine(CFeatures* data) } // go ... - while (niter++ < maxiter && !CSignal::cancel_computations()) + while (niter++ < maxiter && !cancel_computation()) { int32_t maxpidx=-1; float64_t maxpviol = -1; diff --git a/src/shogun/classifier/svm/NewtonSVM.cpp b/src/shogun/classifier/svm/NewtonSVM.cpp index 5431d6e3aa0..4fcfa75c79f 100644 --- a/src/shogun/classifier/svm/NewtonSVM.cpp +++ b/src/shogun/classifier/svm/NewtonSVM.cpp @@ -12,6 +12,7 @@ #ifdef HAVE_LAPACK #include #include +#include #include #include #include @@ -23,19 +24,21 @@ //#define V_NEWTON using namespace shogun; -CNewtonSVM::CNewtonSVM() -: CLinearMachine(), C(1), use_bias(true) +CNewtonSVM::CNewtonSVM() : CLinearMachine(true) { + lambda = 1; + num_iter = 20; + prec = 1e-6; + C = 1; } -CNewtonSVM::CNewtonSVM(float64_t c, CDotFeatures* traindat, CLabels* trainlab, int32_t itr) -: CLinearMachine() +CNewtonSVM::CNewtonSVM( + float64_t c, CDotFeatures* traindat, CLabels* trainlab, int32_t itr) + : CLinearMachine(true) { lambda=1/c; num_iter=itr; prec=1e-6; - num_iter=20; - use_bias=true; C=c; set_features(traindat); set_labels(trainlab); @@ -49,7 +52,7 @@ CNewtonSVM::~CNewtonSVM() bool CNewtonSVM::train_machine(CFeatures* data) { - CSignal::clear_cancel(); + ASSERT(m_labels) ASSERT(m_labels->get_label_type() == LT_BINARY) @@ -80,7 +83,7 @@ bool CNewtonSVM::train_machine(CFeatures* data) float64_t obj, *grad=SG_MALLOC(float64_t, x_d+1); float64_t t; - while(!CSignal::cancel_computations()) + while (!cancel_computation()) { iter++; @@ -232,20 +235,19 @@ void CNewtonSVM::line_search_linear(float64_t* weights, float64_t* d, float64_t* out, float64_t* tx) { SGVector Y=((CBinaryLabels*) m_labels)->get_labels(); - float64_t* outz=SG_MALLOC(float64_t, x_n); - float64_t* temp1=SG_MALLOC(float64_t, x_n); - float64_t* temp1forout=SG_MALLOC(float64_t, x_n); - float64_t* outzsv=SG_MALLOC(float64_t, x_n); - float64_t* Ysv=SG_MALLOC(float64_t, x_n); - float64_t* Xsv=SG_MALLOC(float64_t, x_n); - float64_t* temp2=SG_MALLOC(float64_t, x_d); + SGVector outz(x_n); + SGVector temp1(x_n); + SGVector temp1forout(x_n); + SGVector outzsv(x_n); + SGVector Ysv(x_n); + SGVector Xsv(x_n); float64_t t=0.0; - float64_t* Xd=SG_MALLOC(float64_t, x_n); + SGVector Xd(x_n); for (int32_t i=0; idense_dot(i, d, x_d); - SGVector::add_scalar(d[x_d], Xd, x_n); + linalg::add_scalar(Xd, d[x_d]); #ifdef DEBUG_NEWTON CMath::display_vector(d, x_d+1, "Weight vector"); @@ -259,19 +261,22 @@ void CNewtonSVM::line_search_linear(float64_t* weights, float64_t* d, float64_t* float64_t wd; cblas_dgemm(CblasColMajor, CblasTrans, CblasNoTrans, 1, 1, x_d, lambda, weights, x_d, d, x_d, 0.0, &wd, 1); - float64_t tempg, dd; + float64_t dd; cblas_dgemm(CblasColMajor, CblasTrans, CblasNoTrans, 1, 1, x_d, lambda, d, x_d, d, x_d, 0.0, &dd, 1); float64_t g, h; - int32_t sv_len=0, *sv=SG_MALLOC(int32_t, x_n); + int32_t sv_len=0; + SGVector sv(x_n); + SGVector sg_out(out, x_n, false); do { - SGVector::vector_multiply(temp1, Y.vector, Xd, x_n); - sg_memcpy(temp1forout, temp1, sizeof(float64_t)*x_n); - SGVector::scale_vector(t, temp1forout, x_n); - SGVector::add(outz, 1.0, out, -1.0, temp1forout, x_n); + // FIXME:: port it to linalg:: + SGVector::vector_multiply(temp1.vector, Y.vector, Xd.vector, x_n); + sg_memcpy(temp1forout.vector, temp1.vector, sizeof(float64_t)*x_n); + linalg::scale(temp1forout, temp1forout, t); + linalg::add(sg_out, temp1forout, outz, 1.0, -1.0); // Calculation of sv sv_len=0; @@ -286,19 +291,22 @@ void CNewtonSVM::line_search_linear(float64_t* weights, float64_t* d, float64_t* for (int32_t i=0; i::vector_multiply(temp1, outzsv, Ysv, sv_len); - tempg=CMath::dot(temp1, Xsv, sv_len); + memset(temp1.vector, 0, sizeof(float64_t)*sv_len); + SGVector::vector_multiply(temp1.vector, outzsv.vector, Ysv.vector, sv_len); + // in case sv_len < x_n tempg != dot(temp1, Xsv); + float64_t tempg = 0.0; + for (auto i = 0; i < sv_len; ++i) + tempg += temp1[i]*Xsv[i]; g=wd+(t*dd); g-=tempg; // Calculation of second derivative 'h' cblas_dgemm(CblasColMajor, CblasTrans, CblasNoTrans, 1, 1, sv_len, 1.0, - Xsv, sv_len, Xsv, sv_len, 0.0, &h, 1); + Xsv.vector, sv_len, Xsv.vector, sv_len, 0.0, &h, 1); h+=dd; // Calculation of 1D Newton step 'd' @@ -309,19 +317,8 @@ void CNewtonSVM::line_search_linear(float64_t* weights, float64_t* d, float64_t* } while(1); - for (int32_t i=0; i #include #include +#include #include using namespace shogun; COnlineLibLinear::COnlineLibLinear() - : COnlineLinearMachine() + : COnlineLinearMachine() { init(); } COnlineLibLinear::COnlineLibLinear(float64_t C_reg) + : COnlineLinearMachine() { init(); C1=C_reg; @@ -35,6 +37,7 @@ COnlineLibLinear::COnlineLibLinear(float64_t C_reg) COnlineLibLinear::COnlineLibLinear( float64_t C_reg, CStreamingDotFeatures* traindat) + : COnlineLinearMachine() { init(); C1=C_reg; @@ -45,6 +48,7 @@ COnlineLibLinear::COnlineLibLinear( } COnlineLibLinear::COnlineLibLinear(COnlineLibLinear *mch) + : COnlineLinearMachine() { init(); C1 = mch->C1; @@ -52,17 +56,7 @@ COnlineLibLinear::COnlineLibLinear(COnlineLibLinear *mch) use_bias = mch->use_bias; set_features(mch->features); - - w_dim = mch->w_dim; - if (w_dim > 0) - { - w = SG_MALLOC(float32_t, w_dim); - sg_memcpy(w, mch->w, w_dim*sizeof(float32_t)); - } - else - { - w = NULL; - } + m_w = mch->m_w.clone(); bias = mch->bias; } @@ -129,8 +123,7 @@ void COnlineLibLinear::stop_train() SG_INFO("Optimization finished.\n") // calculate objective value - for (int32_t i=0; i ex, float64_t label) QD = diag[y_current + 1]; // Dot product of vector with itself - QD += CMath::dot(ex.vector, ex.vector, ex.vlen); + QD += linalg::dot(ex, ex); // Dot product of vector with learned weights - G = CMath::dot(ex.vector, w, w_dim); + G = linalg::dot(ex, m_w); if (use_bias) G += bias; @@ -193,9 +186,7 @@ void COnlineLibLinear::train_one(SGVector ex, float64_t label) alpha_current = CMath::min(CMath::max(alpha_current - G/QD, 0.0), C); d = (alpha_current - alpha_old) * y_current; - for (int32_t i=0; i < w_dim; ++i) - w[i] += d*ex[i]; - + linalg::add(m_w, ex, m_w, 1.0f, (float32_t)d); if (use_bias) bias += d; @@ -220,7 +211,7 @@ void COnlineLibLinear::train_one(SGSparseVector ex, float64_t label) QD += SGSparseVector::sparse_dot(ex, ex); // Dot product of vector with learned weights - G = ex.dense_dot(1.0,w,w_dim,0.0); + G = ex.dense_dot(1.0,m_w.vector,m_w.vlen,0.0); if (use_bias) G += bias; @@ -262,7 +253,7 @@ void COnlineLibLinear::train_one(SGSparseVector ex, float64_t label) d = (alpha_current - alpha_old) * y_current; for (int32_t i=0; i < ex.num_feat_entries; i++) - w[ex.features[i].feat_index] += d*ex.features[i].entry; + m_w[ex.features[i].feat_index] += d*ex.features[i].entry; if (use_bias) @@ -276,9 +267,9 @@ void COnlineLibLinear::train_one(SGSparseVector ex, float64_t label) void COnlineLibLinear::train_example(CStreamingDotFeatures *feature, float64_t label) { - features->expand_if_required(w, w_dim); + feature->expand_if_required(m_w.vector, m_w.vlen); - if (features->get_feature_class() == C_STREAMING_DENSE) { + if (feature->get_feature_class() == C_STREAMING_DENSE) { CStreamingDenseFeatures *feat = dynamic_cast *>(feature); if (feat == NULL) @@ -286,7 +277,7 @@ void COnlineLibLinear::train_example(CStreamingDotFeatures *feature, float64_t l train_one(feat->get_vector(), label); } - else if (features->get_feature_class() == C_STREAMING_SPARSE) { + else if (feature->get_feature_class() == C_STREAMING_SPARSE) { CStreamingSparseFeatures *feat = dynamic_cast *>(feature); if (feat == NULL) diff --git a/src/shogun/classifier/svm/OnlineLibLinear.h b/src/shogun/classifier/svm/OnlineLibLinear.h index 7b34af26132..3f0926f6405 100644 --- a/src/shogun/classifier/svm/OnlineLibLinear.h +++ b/src/shogun/classifier/svm/OnlineLibLinear.h @@ -126,21 +126,22 @@ class COnlineLibLinear : public COnlineLinearMachine */ virtual void train_example(CStreamingDotFeatures *feature, float64_t label); +private: + /** Set up parameters */ + void init(); + /** train on one vector * @param ex the example being trained * @param label label of this example */ - virtual void train_one(SGVector ex, float64_t label); + void train_one(SGVector ex, float64_t label); /** train on one *sparse* vector * @param ex the example being trained * @param label label of this example + * */ - virtual void train_one(SGSparseVector ex, float64_t label); - -private: - /** Set up parameters */ - void init(); + void train_one(SGSparseVector ex, float64_t label); private: /// use bias or not diff --git a/src/shogun/classifier/svm/OnlineSVMSGD.cpp b/src/shogun/classifier/svm/OnlineSVMSGD.cpp index 6e7495b97b9..a82884a0f24 100644 --- a/src/shogun/classifier/svm/OnlineSVMSGD.cpp +++ b/src/shogun/classifier/svm/OnlineSVMSGD.cpp @@ -22,6 +22,7 @@ #include #include +#include #include #include #include @@ -79,10 +80,7 @@ bool COnlineSVMSGD::train(CFeatures* data) // allocate memory for w and initialize everyting w and bias with 0 ASSERT(features) ASSERT(features->get_has_labels()) - if (w) - SG_FREE(w); - w_dim=1; - w=new float32_t; + m_w = SGVector(1); bias=0; // Shift t in order to have a @@ -100,15 +98,13 @@ bool COnlineSVMSGD::train(CFeatures* data) if (features->is_seekable()) features->reset_stream(); - CSignal::clear_cancel(); - ELossType loss_type = loss->get_loss_type(); bool is_log_loss = false; if ((loss_type == L_LOGLOSS) || (loss_type == L_LOGLOSSMARGIN)) is_log_loss = true; int32_t vec_count; - for(int32_t e=0; eexpand_if_required(w, w_dim); + features->expand_if_required(m_w.vector, m_w.vlen); float64_t eta = 1.0 / (lambda * t); float64_t y = features->get_label(); - float64_t z = y * (features->dense_dot(w, w_dim) + bias); + float64_t z = y * (features->dense_dot(m_w.vector, m_w.vlen) + bias); if (z < 1 || is_log_loss) { float64_t etd = -eta * loss->first_derivative(z,1); - features->add_to_dense_vec(etd * y / wscale, w, w_dim); + features->add_to_dense_vec(etd * y / wscale, m_w.vector, m_w.vlen); if (use_bias) { @@ -140,7 +136,7 @@ bool COnlineSVMSGD::train(CFeatures* data) float32_t r = 1 - eta * lambda * skip; if (r < 0.8) r = pow(1 - eta * lambda, skip); - SGVector::scale_vector(r, w, w_dim); + linalg::scale(m_w, m_w, r); count = skip; } t++; @@ -158,7 +154,7 @@ bool COnlineSVMSGD::train(CFeatures* data) } features->end_parser(); - float64_t wnorm = CMath::dot(w,w, w_dim); + float64_t wnorm = linalg::dot(m_w, m_w); SG_INFO("Norm: %.6f, Bias: %.6f\n", wnorm, bias) return true; diff --git a/src/shogun/classifier/svm/QPBSVMLib.cpp b/src/shogun/classifier/svm/QPBSVMLib.cpp deleted file mode 100644 index b91a2f1ec24..00000000000 --- a/src/shogun/classifier/svm/QPBSVMLib.cpp +++ /dev/null @@ -1,687 +0,0 @@ -/*----------------------------------------------------------------------- - * - * This program is free software; you can redistribute it and/or modify - * it under the terms of the GNU General Public License as published by - * the Free Software Foundation; either version 3 of the License, or - * (at your option) any later version. - * - * Library for solving QP task required for learning SVM without bias term. - * - * Written (W) 2006-2009 Vojtech Franc, xfrancv@cmp.felk.cvut.cz - * Written (W) 2007 Soeren Sonnenburg - * Copyright (C) 2006-2009 Center for Machine Perception, CTU FEL Prague - * Copyright (C) 2007-2009 Fraunhofer Institute FIRST - * - * - * min 0.5*x'*H*x + f'*x - * - * subject to C >= x(i) >= 0 for all i - * - * H [dim x dim] is symmetric positive semi-definite matrix. - * f [dim x 1] is an arbitrary vector. - * - * The precision of found solution is given by parameters - * tmax, tolabs, tolrel which define the stopping conditions: - * - * t >= tmax -> exit_flag = 0 Number of iterations. - * UB-LB <= tolabs -> exit_flag = 1 Abs. tolerance. - * UB-LB <= UB*tolrel -> exit_flag = 2 Relative tolerance. - * - * UB ... Upper bound on the optimal solution. - * LB ... Lower bound on the optimal solution. - * t ... Number of iterations. - * History ... Value of LB and UB wrt. number of iterations. - * - * 1. Generalized Gauss-Seidel methods - * exitflag = qpbsvm_sca( &get_col, diag_H, f, UB, dim, tmax, - * tolabs, tolrel, x, Nabla, &t, &History, verb ) - * - * 2. Greedy variant - Udpate variable yielding the best improvement. - * exitflag = qpbsvm_scas( &get_col, diag_H, f, UB, dim, tmax, - * tolabs, tolrel, x, Nabla, &t, &History, verb ) - * - * 3. Updates variable which most violates the KKT conditions - * exitflag = qpbsvm_scamv( &get_col, diag_H, f, UB, dim, tmax, - * tolabs, tolrel, tolKKT, x, Nabla, &t, &History, verb ) - * --------------------------------------------------------------------- */ - -#include -#ifdef USE_GPL_SHOGUN -#include -#include - -#include -#include -#include - -#include -#include - -using namespace shogun; - -#define HISTORY_BUF 1000000 - -#define INDEX(ROW,COL,DIM) ((COL*DIM)+ROW) - -CQPBSVMLib::CQPBSVMLib() -{ - SG_UNSTABLE("CQPBSVMLib::CQPBSVMLib()", "\n") - - m_H=0; - m_dim = 0; - m_diag_H = NULL; - - m_f = NULL; - m_UB = 0.0; - m_tmax = INT_MAX; - m_tolabs = 0; - m_tolrel = 1e-6; - m_tolKKT = 0; - m_solver = QPB_SOLVER_SCA; -} - -CQPBSVMLib::CQPBSVMLib( - float64_t* H, int32_t n, float64_t* f, int32_t m, float64_t UB) -: CSGObject() -{ - ASSERT(H && n>0) - m_H=H; - m_dim = n; - m_diag_H=NULL; - - m_f=f; - m_UB=UB; - m_tmax = INT_MAX; - m_tolabs = 0; - m_tolrel = 1e-6; - m_tolKKT = 0; - m_solver = QPB_SOLVER_SCA; -} - -CQPBSVMLib::~CQPBSVMLib() -{ - SG_FREE(m_diag_H); -} - -int32_t CQPBSVMLib::solve_qp(float64_t* result, int32_t len) -{ - int32_t status = -1; - ASSERT(len==m_dim) - float64_t* Nabla=SG_MALLOC(float64_t, m_dim); - for (int32_t i=0; i 0 ) { - SG_PRINT("%d: Q_P=%m_f, Q_D=%m_f, Q_P-Q_D=%m_f, (Q_P-Q_D)/|Q_P|=%m_f \n", - t, Q_P, Q_D, Q_P-Q_D,(Q_P-Q_D)/CMath::abs(Q_P)); - } - - exitflag = -1; - while( exitflag == -1 ) - { - t++; - - for(i = 0; i < m_dim; i++ ) { - if( m_diag_H[i] > 0 ) { - /* variable update */ - x_old = x[i]; - x[i] = CMath::min(m_UB,CMath::max(0.0, x[i] - Nabla[i]/m_diag_H[i])); - - /* update Nabla */ - delta_x = x[i] - x_old; - if( delta_x != 0 ) { - col_H = (float64_t*)get_col(i); - for(j = 0; j < m_dim; j++ ) { - Nabla[j] += col_H[j]*delta_x; - } - } - - } - } - - /* compute Q_P and Q_D */ - xHx = 0; - xf = 0; - xi_sum = 0; - KKTsatisf = 1; - for(i = 0; i < m_dim; i++ ) { - xHx += x[i]*(Nabla[i] - m_f[i]); - xf += x[i]*m_f[i]; - xi_sum += CMath::max(0.0,-Nabla[i]); - - if((x[i] > 0 && x[i] < m_UB && CMath::abs(Nabla[i]) > m_tolKKT) || - (x[i] == 0 && Nabla[i] < -m_tolKKT) || - (x[i] == m_UB && Nabla[i] > m_tolKKT)) KKTsatisf = 0; - } - - Q_P = 0.5*xHx + xf; - Q_D = -0.5*xHx - m_UB*xi_sum; - - /* stopping conditions */ - if(t >= m_tmax) exitflag = 0; - else if(Q_P-Q_D <= m_tolabs) exitflag = 1; - else if(Q_P-Q_D <= CMath::abs(Q_P)*m_tolrel) exitflag = 2; - else if(KKTsatisf == 1) exitflag = 3; - - if( verb > 0 && (t % verb == 0 || t==1)) { - SG_PRINT("%d: Q_P=%m_f, Q_D=%m_f, Q_P-Q_D=%m_f, (Q_P-Q_D)/|Q_P|=%m_f \n", - t, Q_P, Q_D, Q_P-Q_D,(Q_P-Q_D)/CMath::abs(Q_P)); - } - - /* Store m_UB LB to History buffer */ - if( t < History_size ) { - History[INDEX(0,t,2)] = Q_P; - History[INDEX(1,t,2)] = Q_D; - } - else { - tmp_ptr=SG_MALLOC(float64_t, (History_size+HISTORY_BUF)*2); - memset(tmp_ptr, 0, sizeof(float64_t)*(History_size+HISTORY_BUF)*2); - - for( i = 0; i < History_size; i++ ) { - tmp_ptr[INDEX(0,i,2)] = History[INDEX(0,i,2)]; - tmp_ptr[INDEX(1,i,2)] = History[INDEX(1,i,2)]; - } - tmp_ptr[INDEX(0,t,2)] = Q_P; - tmp_ptr[INDEX(1,t,2)] = Q_D; - - History_size += HISTORY_BUF; - SG_FREE(History); - History = tmp_ptr; - } - } - - (*ptr_t) = t; - (*ptr_History) = History; - - SG_PRINT("QP: %f QD: %f\n", Q_P, Q_D) - - return( exitflag ); -} - - -/* -------------------------------------------------------------- - -Usage: exitflag = qpbsvm_scas(m_UB, m_dim, m_tmax, - m_tolabs, m_tolrel, m_tolKKT, x, Nabla, &t, &History, verb ) - --------------------------------------------------------------- */ -int32_t CQPBSVMLib::qpbsvm_scas(float64_t *x, - float64_t *Nabla, - int32_t *ptr_t, - float64_t **ptr_History, - int32_t verb) -{ - float64_t *History; - float64_t *col_H; - float64_t *tmp_ptr; - float64_t x_old; - float64_t x_new; - float64_t delta_x; - float64_t max_x=CMath::INFTY; - float64_t xHx; - float64_t Q_P; - float64_t Q_D; - float64_t xf; - float64_t xi_sum; - float64_t max_update; - float64_t curr_update; - int32_t History_size; - int32_t t; - int32_t i, j; - int32_t max_i=-1; - int32_t exitflag; - int32_t KKTsatisf; - - /* ------------------------------------------------------------ */ - /* Initialization */ - /* ------------------------------------------------------------ */ - - t = 0; - - History_size = (m_tmax < HISTORY_BUF ) ? m_tmax+1 : HISTORY_BUF; - History=SG_MALLOC(float64_t, History_size*2); - memset(History, 0, sizeof(float64_t)*History_size*2); - - /* compute Q_P and Q_D */ - xHx = 0; - xf = 0; - xi_sum = 0; - for(i = 0; i < m_dim; i++ ) { - xHx += x[i]*(Nabla[i] - m_f[i]); - xf += x[i]*m_f[i]; - xi_sum += CMath::max(0.0,-Nabla[i]); - } - - Q_P = 0.5*xHx + xf; - Q_D = -0.5*xHx - m_UB*xi_sum; - History[INDEX(0,t,2)] = Q_P; - History[INDEX(1,t,2)] = Q_D; - - if( verb > 0 ) { - SG_PRINT("%d: Q_P=%m_f, Q_D=%m_f, Q_P-Q_D=%m_f, (Q_P-Q_D)/|Q_P|=%m_f \n", - t, Q_P, Q_D, Q_P-Q_D,(Q_P-Q_D)/CMath::abs(Q_P)); - } - - exitflag = -1; - while( exitflag == -1 ) - { - t++; - - max_update = -CMath::INFTY; - for(i = 0; i < m_dim; i++ ) { - if( m_diag_H[i] > 0 ) { - /* variable update */ - x_old = x[i]; - x_new = CMath::min(m_UB,CMath::max(0.0, x[i] - Nabla[i]/m_diag_H[i])); - - curr_update = -0.5*m_diag_H[i]*(x_new*x_new-x_old*x_old) - - (Nabla[i] - m_diag_H[i]*x_old)*(x_new - x_old); - - if( curr_update > max_update ) { - max_i = i; - max_update = curr_update; - max_x = x_new; - } - } - } - - x_old = x[max_i]; - x[max_i] = max_x; - - /* update Nabla */ - delta_x = max_x - x_old; - if( delta_x != 0 ) { - col_H = (float64_t*)get_col(max_i); - for(j = 0; j < m_dim; j++ ) { - Nabla[j] += col_H[j]*delta_x; - } - } - - /* compute Q_P and Q_D */ - xHx = 0; - xf = 0; - xi_sum = 0; - KKTsatisf = 1; - for(i = 0; i < m_dim; i++ ) { - xHx += x[i]*(Nabla[i] - m_f[i]); - xf += x[i]*m_f[i]; - xi_sum += CMath::max(0.0,-Nabla[i]); - - if((x[i] > 0 && x[i] < m_UB && CMath::abs(Nabla[i]) > m_tolKKT) || - (x[i] == 0 && Nabla[i] < -m_tolKKT) || - (x[i] == m_UB && Nabla[i] > m_tolKKT)) KKTsatisf = 0; - } - - Q_P = 0.5*xHx + xf; - Q_D = -0.5*xHx - m_UB*xi_sum; - - /* stopping conditions */ - if(t >= m_tmax) exitflag = 0; - else if(Q_P-Q_D <= m_tolabs) exitflag = 1; - else if(Q_P-Q_D <= CMath::abs(Q_P)*m_tolrel) exitflag = 2; - else if(KKTsatisf == 1) exitflag = 3; - - if( verb > 0 && (t % verb == 0 || t==1)) { - SG_PRINT("%d: Q_P=%m_f, Q_D=%m_f, Q_P-Q_D=%m_f, (Q_P-Q_D)/|Q_P|=%m_f \n", - t, Q_P, Q_D, Q_P-Q_D,(Q_P-Q_D)/CMath::abs(Q_P)); - } - - /* Store m_UB LB to History buffer */ - if( t < History_size ) { - History[INDEX(0,t,2)] = Q_P; - History[INDEX(1,t,2)] = Q_D; - } - else { - tmp_ptr=SG_MALLOC(float64_t, (History_size+HISTORY_BUF)*2); - memset(tmp_ptr, 0, (History_size+HISTORY_BUF)*2*sizeof(float64_t)); - for( i = 0; i < History_size; i++ ) { - tmp_ptr[INDEX(0,i,2)] = History[INDEX(0,i,2)]; - tmp_ptr[INDEX(1,i,2)] = History[INDEX(1,i,2)]; - } - tmp_ptr[INDEX(0,t,2)] = Q_P; - tmp_ptr[INDEX(1,t,2)] = Q_D; - - History_size += HISTORY_BUF; - SG_FREE(History); - History = tmp_ptr; - } - } - - (*ptr_t) = t; - (*ptr_History) = History; - - return( exitflag ); -} - -/* -------------------------------------------------------------- - -Usage: exitflag = qpbsvm_scamv(m_UB, m_dim, m_tmax, - m_tolabs, m_tolrel, m_tolKKT, x, Nabla, &t, &History, verb ) - --------------------------------------------------------------- */ -int32_t CQPBSVMLib::qpbsvm_scamv(float64_t *x, - float64_t *Nabla, - int32_t *ptr_t, - float64_t **ptr_History, - int32_t verb) -{ - float64_t *History; - float64_t *col_H; - float64_t delta_x; - float64_t x_new; - float64_t max_viol; - float64_t fval; - int32_t t; - int32_t i; - int32_t u=-1; - int32_t exitflag; - - /* ------------------------------------------------------------ */ - /* Initialization */ - /* ------------------------------------------------------------ */ - - t = 0; - exitflag = -1; - while( exitflag == -1 && t <= m_tmax) - { - t++; - - max_viol = 0; - for(i = 0; i < m_dim; i++ ) - { - if( x[i] == 0 ) - { - if( max_viol < -Nabla[i]) { u = i; max_viol = -Nabla[i]; } - } - else if( x[i] > 0 && x[i] < m_UB ) - { - if( max_viol < CMath::abs(Nabla[i]) ) { u = i; max_viol = CMath::abs(Nabla[i]); } - } - else if( max_viol < Nabla[i]) { u = i; max_viol = Nabla[i]; } - } - -/* SG_PRINT("%d: max_viol=%m_f, u=%d\n", t, max_viol, u)*/ - - if( max_viol <= m_tolKKT ) - { - exitflag = 1; - } - else - { - /* update */ - x_new = CMath::min(m_UB,CMath::max(0.0, x[u] - Nabla[u]/m_diag_H[u])); - - delta_x = x_new - x[u]; - x[u] = x_new; - - col_H = (float64_t*)get_col(u); - for(i = 0; i < m_dim; i++ ) { - Nabla[i] += col_H[i]*delta_x; - } - } - } - - History=SG_MALLOC(float64_t, (t+1)*2); - memset(History, 0, sizeof(float64_t)*(t+1)*2); - - fval = 0; - for(fval = 0, i = 0; i < m_dim; i++ ) { - fval += 0.5*x[i]*(Nabla[i]+m_f[i]); - } - - History[INDEX(0,t,2)] = fval; - History[INDEX(1,t,2)] = 0; - - (*ptr_t) = t; - (*ptr_History) = History; - - - - return( exitflag ); -} - -/* -------------------------------------------------------------- - -Usage: exitflag = qpbsvm_prloqo(m_UB, m_dim, m_tmax, - m_tolabs, m_tolrel, m_tolKKT, x, Nabla, &t, &History, verb ) - --------------------------------------------------------------- */ -int32_t CQPBSVMLib::qpbsvm_prloqo(float64_t *x, - float64_t *Nabla, - int32_t *ptr_t, - float64_t **ptr_History, - int32_t verb) -{ - float64_t* lb=SG_MALLOC(float64_t, m_dim); - float64_t* ub=SG_MALLOC(float64_t, m_dim); - float64_t* primal=SG_MALLOC(float64_t, 3*m_dim); - float64_t* dual=SG_MALLOC(float64_t, 1+2*m_dim); - float64_t* a=SG_MALLOC(float64_t, m_dim); - - for (int32_t i=0; i::display_vector(m_f, m_dim, "m_f"); - int32_t result=pr_loqo(m_dim, 1, m_f, m_H, a, &b, lb, ub, primal, dual, - 2, 5, 1, -0.95, 10,0); - - SG_FREE(a); - SG_FREE(lb); - SG_FREE(ub); - SG_FREE(primal); - SG_FREE(dual); - - *ptr_t=0; - *ptr_History=NULL; - return result; -} - -int32_t CQPBSVMLib::qpbsvm_gauss_seidel(float64_t *x, - float64_t *Nabla, - int32_t *ptr_t, - float64_t **ptr_History, - int32_t verb) -{ - for (int32_t i=0; i -#ifdef USE_GPL_SHOGUN -#include -#include -#include -#include - -namespace shogun -{ - -#ifndef DOXYGEN_SHOULD_SKIP_THIS -enum E_QPB_SOLVER -{ - QPB_SOLVER_SCA, // sequential coordinate wise (gaussian seidel based) - QPB_SOLVER_SCAS, // sequential coordinate wise selecting the variable - // gaining 'best' improved - QPB_SOLVER_SCAMV, // sequential coordinate wise selecting variable most violating kkt's - QPB_SOLVER_PRLOQO,// via pr_loqo - QPB_SOLVER_CPLEX, // via cplex - QPB_SOLVER_GS, // gaussian seidel - QPB_SOLVER_GRADDESC // gaussian seidel -}; -#endif - -/** @brief class QPBSVMLib */ -class CQPBSVMLib: public CSGObject -{ - public: - /** default constructor */ - CQPBSVMLib(); - - /** constructor - * - * @param H symmetric matrix of size n x n - * @param n size of H's matrix - * @param f is vector of size m - * @param m size of vector f - * @param UB UB - */ - CQPBSVMLib( - float64_t* H, int32_t n, float64_t* f, int32_t m, float64_t UB=1.0); - - /// result has to be allocated & zeroed - int32_t solve_qp(float64_t* result, int32_t len); - - /** set solver - * - * @param solver new solver - */ - inline void set_solver(E_QPB_SOLVER solver) - { - m_solver=solver; - } - - virtual ~CQPBSVMLib(); - - protected: - /** get col - * - * @param col col to get - * @return col indexed by col - */ - inline float64_t* get_col(int32_t col) - { - return &m_H[m_dim*col]; - } - - /** Usage: exitflag = qpbsvm_sca(UB, dim, tmax, - tolabs, tolrel, tolKKT, x, Nabla, &t, &History, verb ) */ - int32_t qpbsvm_sca( - float64_t *x, float64_t *Nabla, int32_t *ptr_t, - float64_t **ptr_History, int32_t verb); - /** Usage: exitflag = qpbsvm_scas(UB, dim, tmax, - tolabs, tolrel, tolKKT, x, Nabla, &t, &History, verb ) */ - int32_t qpbsvm_scas( - float64_t *x, float64_t *Nabla, int32_t *ptr_t, - float64_t **ptr_History, int32_t verb); - /** Usage: exitflag = qpbsvm_scamv(UB, dim, tmax, - tolabs, tolrel, tolKKT, x, Nabla, &t, &History, verb ) */ - int32_t qpbsvm_scamv( - float64_t *x, float64_t *Nabla, int32_t *ptr_t, - float64_t **ptr_History, int32_t verb); - /** Usage: exitflag = qpbsvm_prloqo(UB, dim, tmax, - tolabs, tolrel, tolKKT, x, Nabla, &t, &History, verb ) */ - int32_t qpbsvm_prloqo( - float64_t *x, float64_t *Nabla, int32_t *ptr_t, - float64_t **ptr_History, int32_t verb); - /** Usage: exitflag = qpbsvm_gauss_seidel(UB, dim, tmax, - tolabs, tolrel, tolKKT, x, Nabla, &t, &History, verb ) */ - int32_t qpbsvm_gauss_seidel( - float64_t *x, float64_t *Nabla, int32_t *ptr_t, - float64_t **ptr_History, int32_t verb); - /** Usage: exitflag = qpbsvm_gradient_descent(UB, dim, tmax, - tolabs, tolrel, tolKKT, x, Nabla, &t, &History, verb ) */ - int32_t qpbsvm_gradient_descent( - float64_t *x, float64_t *Nabla, int32_t *ptr_t, - float64_t **ptr_History, int32_t verb); -#ifdef USE_CPLEX - /** Usage: exitflag = qpbsvm_cplex(UB, dim, tmax, - tolabs, tolrel, tolKKT, x, Nabla, &t, &History, verb ) */ - int32_t qpbsvm_cplex( - float64_t *x, float64_t *Nabla, int32_t *ptr_t, - float64_t **ptr_History, int32_t verb); -#endif - - /** @return object name */ - inline const char* get_name() const { return "QPBSVMLib"; } - - protected: - /** matrix H */ - float64_t* m_H; - /** diagonal of H */ - float64_t* m_diag_H; - /** dim */ - int32_t m_dim; - - /** vector f */ - float64_t* m_f; - - /** UB */ - float64_t m_UB; - - /** tmax */ - int32_t m_tmax; - /** tolabs */ - float64_t m_tolabs; - /** tolrel */ - float64_t m_tolrel; - /** tolKKT */ - float64_t m_tolKKT; - /** solver */ - E_QPB_SOLVER m_solver; -}; -} -#endif //USE_GPL_SHOGUN -#endif //QPBSVMLIB_H__ diff --git a/src/shogun/classifier/svm/SGDQN.cpp b/src/shogun/classifier/svm/SGDQN.cpp index e005485b7cc..aed685fcfad 100644 --- a/src/shogun/classifier/svm/SGDQN.cpp +++ b/src/shogun/classifier/svm/SGDQN.cpp @@ -137,14 +137,13 @@ bool CSGDQN::train(CFeatures* data) calibrate(); SG_INFO("Training on %d vectors\n", num_vec) - CSignal::clear_cancel(); ELossType loss_type = loss->get_loss_type(); bool is_log_loss = false; if ((loss_type == L_LOGLOSS) || (loss_type == L_LOGLOSSMARGIN)) is_log_loss = true; - for(int32_t e=0; e #include #include -#include #include +#include #include #include @@ -52,6 +53,7 @@ #ifdef HAVE_PTHREAD #include + #endif using namespace shogun; @@ -640,12 +642,16 @@ int32_t CSVMLight::optimize_to_convergence(int32_t* docs, int32_t* label, int32_ CTime start_time; mkl_converged=false; - + auto pb = progress(range(10), *this->io); #ifdef CYGWIN for (;((iteration<100 || (!mkl_converged && callback) ) || (retrain && (!terminate))); iteration++){ #else - CSignal::clear_cancel(); - for (;((!CSignal::cancel_computations()) && ((iteration<3 || (!mkl_converged && callback) ) || (retrain && (!terminate)))); iteration++){ + + for (; ((!cancel_computation()) && + ((iteration < 3 || (!mkl_converged && callback)) || + (retrain && (!terminate)))); + iteration++) + { #endif if(use_kernel_cache) @@ -901,7 +907,9 @@ int32_t CSVMLight::optimize_to_convergence(int32_t* docs, int32_t* label, int32_ if (bestmaxdiff>worstmaxdiff) worstmaxdiff=bestmaxdiff; - SG_ABS_PROGRESS(bestmaxdiff, -CMath::log10(bestmaxdiff), -CMath::log10(worstmaxdiff), -CMath::log10(epsilon), 6) + pb.print_absolute( + bestmaxdiff, -CMath::log10(bestmaxdiff), -CMath::log10(worstmaxdiff), + -CMath::log10(epsilon)); /* Terminate loop */ if (m_max_train_time > 0 && start_time.cur_time_diff() > m_max_train_time) { @@ -910,6 +918,7 @@ int32_t CSVMLight::optimize_to_convergence(int32_t* docs, int32_t* label, int32_ } } /* end of loop */ + pb.complete_absolute(); SG_DEBUG("inactive:%d\n", inactivenum) diff --git a/src/shogun/classifier/svm/SVMLin.cpp b/src/shogun/classifier/svm/SVMLin.cpp deleted file mode 100644 index 388cf2f2ae6..00000000000 --- a/src/shogun/classifier/svm/SVMLin.cpp +++ /dev/null @@ -1,109 +0,0 @@ -/* - * This program is free software; you can redistribute it and/or modify - * it under the terms of the GNU General Public License as published by - * the Free Software Foundation; either version 3 of the License, or - * (at your option) any later version. - * - * Written (W) 2006-2009 Soeren Sonnenburg - * Copyright (C) 2006-2009 Fraunhofer Institute FIRST and Max-Planck-Society - */ - -#include -#include -#include -#include -#include -#include -#include -#include - -using namespace shogun; - -CSVMLin::CSVMLin() -: CLinearMachine(), C1(1), C2(1), epsilon(1e-5), use_bias(true) -{ -} - -CSVMLin::CSVMLin( - float64_t C, CDotFeatures* traindat, CLabels* trainlab) -: CLinearMachine(), C1(C), C2(C), epsilon(1e-5), use_bias(true) -{ - set_features(traindat); - set_labels(trainlab); -} - - -CSVMLin::~CSVMLin() -{ -} - -bool CSVMLin::train_machine(CFeatures* data) -{ - ASSERT(m_labels) - - if (data) - { - if (!data->has_property(FP_DOT)) - SG_ERROR("Specified features are not of type CDotFeatures\n") - set_features((CDotFeatures*) data); - } - - ASSERT(features) - - SGVector train_labels=((CBinaryLabels*) m_labels)->get_labels(); - int32_t num_feat=features->get_dim_feature_space(); - int32_t num_vec=features->get_num_vectors(); - - ASSERT(num_vec==train_labels.vlen) - - struct options Options; - struct data Data; - struct vector_double Weights; - struct vector_double Outputs; - - Data.l=num_vec; - Data.m=num_vec; - Data.u=0; - Data.n=num_feat+1; - Data.nz=num_feat+1; - Data.Y=train_labels.vector; - Data.features=features; - Data.C = SG_MALLOC(float64_t, Data.l); - - Options.algo = SVM; - Options.lambda=1/(2*get_C1()); - Options.lambda_u=1/(2*get_C1()); - Options.S=10000; - Options.R=0.5; - Options.epsilon = get_epsilon(); - Options.cgitermax=10000; - Options.mfnitermax=50; - Options.Cp = get_C2()/get_C1(); - Options.Cn = 1; - - if (use_bias) - Options.bias=1.0; - else - Options.bias=0.0; - - for (int32_t i=0;i0) - Data.C[i]=Options.Cp; - else - Data.C[i]=Options.Cn; - } - ssl_train(&Data, &Options, &Weights, &Outputs); - ASSERT(Weights.vec && Weights.d==num_feat+1) - - float64_t sgn=train_labels.vector[0]; - for (int32_t i=0; i(Weights.vec, num_feat)); - set_bias(Weights.vec[num_feat]); - - SG_FREE(Data.C); - SG_FREE(Outputs.vec); - return true; -} diff --git a/src/shogun/classifier/svm/SVMLin.h b/src/shogun/classifier/svm/SVMLin.h deleted file mode 100644 index 950305f1584..00000000000 --- a/src/shogun/classifier/svm/SVMLin.h +++ /dev/null @@ -1,119 +0,0 @@ -/* - * This program is free software; you can redistribute it and/or modify - * it under the terms of the GNU General Public License as published by - * the Free Software Foundation; either version 3 of the License, or - * (at your option) any later version. - * - * Written (W) 2006-2009 Soeren Sonnenburg - * Copyright (C) 2006-2009 Fraunhofer Institute FIRST and Max-Planck-Society - */ - -#ifndef _SVMLIN_H___ -#define _SVMLIN_H___ - -#include -#include -#include -#include - -namespace shogun -{ -/** @brief class SVMLin */ -class CSVMLin : public CLinearMachine -{ - public: - - /** problem type */ - MACHINE_PROBLEM_TYPE(PT_BINARY); - - /** default constructor */ - CSVMLin(); - - /** constructor - * - * @param C constant C - * @param traindat training features - * @param trainlab labels for features - */ - CSVMLin( - float64_t C, CDotFeatures* traindat, - CLabels* trainlab); - virtual ~CSVMLin(); - - /** get classifier type - * - * @return classifier type SVMLIN - */ - virtual EMachineType get_classifier_type() { return CT_SVMLIN; } - - /** set C - * - * @param c_neg new C constant for negatively labeled examples - * @param c_pos new C constant for positively labeled examples - * - */ - inline void set_C(float64_t c_neg, float64_t c_pos) { C1=c_neg; C2=c_pos; } - - /** get C1 - * - * @return C1 - */ - inline float64_t get_C1() { return C1; } - - /** get C2 - * - * @return C2 - */ - inline float64_t get_C2() { return C2; } - - /** set if bias shall be enabled - * - * @param enable_bias if bias shall be enabled - */ - inline void set_bias_enabled(bool enable_bias) { use_bias=enable_bias; } - - /** get if bias is enabled - * - * @return if bias is enabled - */ - inline bool get_bias_enabled() { return use_bias; } - - /** set epsilon - * - * @param eps new epsilon - */ - inline void set_epsilon(float64_t eps) { epsilon=eps; } - - /** get epsilon - * - * @return epsilon - */ - inline float64_t get_epsilon() { return epsilon; } - - /** @return object name */ - virtual const char* get_name() const { return "SVMLin"; } - - protected: - /** train SVM classifier - * - * @param data training data (parameter can be avoided if distance or - * kernel-based classifiers are used and distance/kernels are - * initialized with train data) - * - * @return whether training was successful - */ - virtual bool train_machine(CFeatures* data=NULL); - - protected: - /** C1 */ - float64_t C1; - /** C2 */ - float64_t C2; - /** epsilon */ - float64_t epsilon; - - /** if bias is used */ - bool use_bias; -}; -} -#endif diff --git a/src/shogun/classifier/svm/SVMOcas.cpp b/src/shogun/classifier/svm/SVMOcas.cpp deleted file mode 100644 index ee63169fdbd..00000000000 --- a/src/shogun/classifier/svm/SVMOcas.cpp +++ /dev/null @@ -1,368 +0,0 @@ -/* - * This program is free software; you can redistribute it and/or modify - * it under the terms of the GNU General Public License as published by - * the Free Software Foundation; either version 3 of the License, or - * (at your option) any later version. - * - * Written (W) 2007-2008 Vojtech Franc - * Written (W) 2007-2009 Soeren Sonnenburg - * Copyright (C) 2007-2009 Fraunhofer Institute FIRST and Max-Planck-Society - */ - -#include -#ifdef USE_GPL_SHOGUN - -#include -#include -#include -#include -#include -#include -#include -#include -#include - -using namespace shogun; - -CSVMOcas::CSVMOcas() -: CLinearMachine() -{ - init(); -} - -CSVMOcas::CSVMOcas(E_SVM_TYPE type) -: CLinearMachine() -{ - init(); - method=type; -} - -CSVMOcas::CSVMOcas( - float64_t C, CDotFeatures* traindat, CLabels* trainlab) -: CLinearMachine() -{ - init(); - C1=C; - C2=C; - - set_features(traindat); - set_labels(trainlab); -} - - -CSVMOcas::~CSVMOcas() -{ -} - -bool CSVMOcas::train_machine(CFeatures* data) -{ - SG_INFO("C=%f, epsilon=%f, bufsize=%d\n", get_C1(), get_epsilon(), bufsize) - SG_DEBUG("use_bias = %i\n", get_bias_enabled()) - - ASSERT(m_labels) - ASSERT(m_labels->get_label_type() == LT_BINARY) - if (data) - { - if (!data->has_property(FP_DOT)) - SG_ERROR("Specified features are not of type CDotFeatures\n") - set_features((CDotFeatures*) data); - } - ASSERT(features) - - int32_t num_vec=features->get_num_vectors(); - lab = SGVector(num_vec); - for (int32_t i=0; iget_label(i); - - current_w = SGVector(features->get_dim_feature_space()); - current_w.zero(); - - if (num_vec!=lab.vlen || num_vec<=0) - SG_ERROR("num_vec=%d num_train_labels=%d\n", num_vec, lab.vlen) - - SG_FREE(old_w); - old_w=SG_CALLOC(float64_t, current_w.vlen); - bias=0; - old_bias=0; - - tmp_a_buf=SG_CALLOC(float64_t, current_w.vlen); - cp_value=SG_CALLOC(float64_t*, bufsize); - cp_index=SG_CALLOC(uint32_t*, bufsize); - cp_nz_dims=SG_CALLOC(uint32_t, bufsize); - cp_bias=SG_CALLOC(float64_t, bufsize); - - float64_t TolAbs=0; - float64_t QPBound=0; - int32_t Method=0; - if (method == SVM_OCAS) - Method = 1; - ocas_return_value_T result = svm_ocas_solver( get_C1(), num_vec, get_epsilon(), - TolAbs, QPBound, get_max_train_time(), bufsize, Method, - &CSVMOcas::compute_W, - &CSVMOcas::update_W, - &CSVMOcas::add_new_cut, - &CSVMOcas::compute_output, - &CSVMOcas::sort, - &CSVMOcas::print, - this); - - SG_INFO("Ocas Converged after %d iterations\n" - "==================================\n" - "timing statistics:\n" - "output_time: %f s\n" - "sort_time: %f s\n" - "add_time: %f s\n" - "w_time: %f s\n" - "solver_time %f s\n" - "ocas_time %f s\n\n", result.nIter, result.output_time, result.sort_time, - result.add_time, result.w_time, result.qp_solver_time, result.ocas_time); - - SG_FREE(tmp_a_buf); - - primal_objective = result.Q_P; - - uint32_t num_cut_planes = result.nCutPlanes; - - SG_DEBUG("num_cut_planes=%d\n", num_cut_planes) - for (uint32_t i=0; icurrent_w.vlen; - float64_t* W = o->current_w.vector; - float64_t* oldW=o->old_w; - - for(uint32_t j=0; j bias=o->old_bias*(1-t) + t*o->bias; - sq_norm_W += CMath::sq(o->bias); - - return( sq_norm_W ); -} - -/*---------------------------------------------------------------------------------- - sparse_add_new_cut( new_col_H, new_cut, cut_length, nSel ) does the following: - - new_a = sum(data_X(:,find(new_cut ~=0 )),2); - new_col_H = [sparse_A(:,1:nSel)'*new_a ; new_a'*new_a]; - sparse_A(:,nSel+1) = new_a; - - ---------------------------------------------------------------------------------*/ -int CSVMOcas::add_new_cut( - float64_t *new_col_H, uint32_t *new_cut, uint32_t cut_length, - uint32_t nSel, void* ptr) -{ - CSVMOcas* o = (CSVMOcas*) ptr; - CDotFeatures* f = o->features; - uint32_t nDim=(uint32_t) o->current_w.vlen; - float64_t* y = o->lab.vector; - - float64_t** c_val = o->cp_value; - uint32_t** c_idx = o->cp_index; - uint32_t* c_nzd = o->cp_nz_dims; - float64_t* c_bias = o->cp_bias; - - float64_t sq_norm_a; - uint32_t i, j, nz_dims; - - /* temporary vector */ - float64_t* new_a = o->tmp_a_buf; - memset(new_a, 0, sizeof(float64_t)*nDim); - - for(i=0; i < cut_length; i++) - { - f->add_to_dense_vec(y[new_cut[i]], new_cut[i], new_a, nDim); - - if (o->use_bias) - c_bias[nSel]+=y[new_cut[i]]; - } - - /* compute new_a'*new_a and count number of non-zerou dimensions */ - nz_dims = 0; - sq_norm_a = CMath::sq(c_bias[nSel]); - for(j=0; j < nDim; j++ ) { - if(new_a[j] != 0) { - nz_dims++; - sq_norm_a += new_a[j]*new_a[j]; - } - } - - /* sparsify new_a and insert it to the last column of sparse_A */ - c_nzd[nSel] = nz_dims; - c_idx[nSel]=NULL; - c_val[nSel]=NULL; - - if(nz_dims > 0) - { - c_idx[nSel]=SG_MALLOC(uint32_t, nz_dims); - c_val[nSel]=SG_MALLOC(float64_t, nz_dims); - - uint32_t idx=0; - for(j=0; j < nDim; j++ ) - { - if(new_a[j] != 0) - { - c_idx[nSel][idx] = j; - c_val[nSel][idx++] = new_a[j]; - } - } - } - - new_col_H[nSel] = sq_norm_a; - - for(i=0; i < nSel; i++) - { - float64_t tmp = c_bias[nSel]*c_bias[i]; - for(j=0; j < c_nzd[i]; j++) - tmp += new_a[c_idx[i][j]]*c_val[i][j]; - - new_col_H[i] = tmp; - } - //CMath::display_vector(new_col_H, nSel+1, "new_col_H"); - //CMath::display_vector((int32_t*) c_idx[nSel], (int32_t) nz_dims, "c_idx"); - //CMath::display_vector((float64_t*) c_val[nSel], nz_dims, "c_val"); - return 0; -} - -int CSVMOcas::sort(float64_t* vals, float64_t* data, uint32_t size) -{ - CMath::qsort_index(vals, data, size); - return 0; -} - -/*---------------------------------------------------------------------- - sparse_compute_output( output ) does the follwing: - - output = data_X'*W; - ----------------------------------------------------------------------*/ -int CSVMOcas::compute_output(float64_t *output, void* ptr) -{ - CSVMOcas* o = (CSVMOcas*) ptr; - CDotFeatures* f=o->features; - int32_t nData=f->get_num_vectors(); - - float64_t* y = o->lab.vector; - - f->dense_dot_range(output, 0, nData, y, o->current_w.vector, o->current_w.vlen, 0.0); - - for (int32_t i=0; ibias; - return 0; -} - -/*---------------------------------------------------------------------- - sq_norm_W = compute_W( alpha, nSel ) does the following: - - oldW = W; - W = sparse_A(:,1:nSel)'*alpha; - sq_norm_W = W'*W; - dp_WoldW = W'*oldW'; - - ----------------------------------------------------------------------*/ -void CSVMOcas::compute_W( - float64_t *sq_norm_W, float64_t *dp_WoldW, float64_t *alpha, uint32_t nSel, - void* ptr ) -{ - CSVMOcas* o = (CSVMOcas*) ptr; - uint32_t nDim= (uint32_t) o->current_w.vlen; - CMath::swap(o->current_w.vector, o->old_w); - float64_t* W=o->current_w.vector; - float64_t* oldW=o->old_w; - memset(W, 0, sizeof(float64_t)*nDim); - float64_t old_bias=o->bias; - float64_t bias=0; - - float64_t** c_val = o->cp_value; - uint32_t** c_idx = o->cp_index; - uint32_t* c_nzd = o->cp_nz_dims; - float64_t* c_bias = o->cp_bias; - - for(uint32_t i=0; i 0 && alpha[i] > 0) - { - for(uint32_t j=0; j < nz_dims; j++) - W[c_idx[i][j]] += alpha[i]*c_val[i][j]; - } - bias += c_bias[i]*alpha[i]; - } - - *sq_norm_W = CMath::dot(W,W, nDim) + CMath::sq(bias); - *dp_WoldW = CMath::dot(W,oldW, nDim) + bias*old_bias; - //SG_PRINT("nSel=%d sq_norm_W=%f dp_WoldW=%f\n", nSel, *sq_norm_W, *dp_WoldW) - - o->bias = bias; - o->old_bias = old_bias; -} - -void CSVMOcas::init() -{ - use_bias=true; - bufsize=3000; - C1=1; - C2=1; - - epsilon=1e-3; - method=SVM_OCAS; - old_w=NULL; - tmp_a_buf=NULL; - cp_value=NULL; - cp_index=NULL; - cp_nz_dims=NULL; - cp_bias=NULL; - - primal_objective = 0.0; - - m_parameters->add(&C1, "C1", "Cost constant 1."); - m_parameters->add(&C2, "C2", "Cost constant 2."); - m_parameters->add(&use_bias, "use_bias", - "Indicates if bias is used."); - m_parameters->add(&epsilon, "epsilon", "Convergence precision."); - m_parameters->add(&bufsize, "bufsize", "Maximum number of cutting planes."); - m_parameters->add((machine_int_t*) &method, "method", - "SVMOcas solver type."); -} - -float64_t CSVMOcas::compute_primal_objective() const -{ - return primal_objective; -} - -#endif //USE_GPL_SHOGUN diff --git a/src/shogun/classifier/svm/SVMOcas.h b/src/shogun/classifier/svm/SVMOcas.h deleted file mode 100644 index 838385a9406..00000000000 --- a/src/shogun/classifier/svm/SVMOcas.h +++ /dev/null @@ -1,242 +0,0 @@ -/* - * This program is free software; you can redistribute it and/or modify - * it under the terms of the GNU General Public License as published by - * the Free Software Foundation; either version 3 of the License, or - * (at your option) any later version. - * - * Written (W) 2007-2009 Vojtech Franc - * Written (W) 2007-2009 Soeren Sonnenburg - * Copyright (C) 2007-2009 Fraunhofer Institute FIRST and Max-Planck-Society - */ - - -#ifndef _SVMOCAS_H___ -#define _SVMOCAS_H___ - -#include -#ifdef USE_GPL_SHOGUN - -#include -#include -#include -#include -#include - -namespace shogun -{ -#ifndef DOXYGEN_SHOULD_SKIP_THIS -enum E_SVM_TYPE -{ - SVM_OCAS = 0, - SVM_BMRM = 1 -}; -#endif - -/** @brief class SVMOcas */ -class CSVMOcas : public CLinearMachine -{ - public: - - /** problem type */ - MACHINE_PROBLEM_TYPE(PT_BINARY); - - /** default constructor */ - CSVMOcas(); - - /** constructor - * - * @param type a E_SVM_TYPE - */ - CSVMOcas(E_SVM_TYPE type); - - /** constructor - * - * @param C constant C - * @param traindat training features - * @param trainlab labels for training features - */ - CSVMOcas( - float64_t C, CDotFeatures* traindat, - CLabels* trainlab); - virtual ~CSVMOcas(); - - /** get classifier type - * - * @return classifier type SVMOCAS - */ - virtual EMachineType get_classifier_type() { return CT_SVMOCAS; } - - /** set C - * - * @param c_neg new C constant for negatively labeled examples - * @param c_pos new C constant for positively labeled examples - * - */ - inline void set_C(float64_t c_neg, float64_t c_pos) { C1=c_neg; C2=c_pos; } - - /** get C1 - * - * @return C1 - */ - inline float64_t get_C1() { return C1; } - - /** get C2 - * - * @return C2 - */ - inline float64_t get_C2() { return C2; } - - /** set epsilon - * - * @param eps new epsilon - */ - inline void set_epsilon(float64_t eps) { epsilon=eps; } - - /** get epsilon - * - * @return epsilon - */ - inline float64_t get_epsilon() { return epsilon; } - - /** set if bias shall be enabled - * - * @param enable_bias if bias shall be enabled - */ - inline void set_bias_enabled(bool enable_bias) { use_bias=enable_bias; } - - /** check if bias is enabled - * - * @return if bias is enabled - */ - inline bool get_bias_enabled() { return use_bias; } - - /** set buffer size - * - * @param sz buffer size - */ - inline void set_bufsize(int32_t sz) { bufsize=sz; } - - /** get buffer size - * - * @return buffer size - */ - inline int32_t get_bufsize() { return bufsize; } - - /** compute the primal objective value - * - * @return the primal objective - */ - virtual float64_t compute_primal_objective() const; - - protected: - /** compute W - * - * @param sq_norm_W square normed W - * @param dp_WoldW dp W old W - * @param alpha alpha - * @param nSel nSel - * @param ptr ptr - */ - static void compute_W( - float64_t *sq_norm_W, float64_t *dp_WoldW, float64_t *alpha, - uint32_t nSel, void* ptr); - - /** update W - * - * @param t t - * @param ptr ptr - * @return something floaty - */ - static float64_t update_W(float64_t t, void* ptr ); - - /** add new cut - * - * @param new_col_H new col H - * @param new_cut new cut - * @param cut_length length of cut - * @param nSel nSel - * @param ptr ptr - */ - static int add_new_cut( - float64_t *new_col_H, uint32_t *new_cut, uint32_t cut_length, - uint32_t nSel, void* ptr ); - - /** compute output - * - * @param output output - * @param ptr ptr - */ - static int compute_output( float64_t *output, void* ptr ); - - /** sort - * - * @param vals vals - * @param data data - * @param size size - */ - static int sort( float64_t* vals, float64_t* data, uint32_t size); - - /** print nothing */ - static inline void print(ocas_return_value_T value) - { - return; - } - - protected: - /** train SVM classifier - * - * @param data training data (parameter can be avoided if distance or - * kernel-based classifiers are used and distance/kernels are - * initialized with train data) - * - * @return whether training was successful - */ - virtual bool train_machine(CFeatures* data=NULL); - - /** @return object name */ - inline const char* get_name() const { return "SVMOcas"; } - private: - void init(); - - protected: - /** if bias is used */ - bool use_bias; - /** buffer size */ - int32_t bufsize; - /** C1 */ - float64_t C1; - /** C2 */ - float64_t C2; - /** epsilon */ - float64_t epsilon; - /** method */ - E_SVM_TYPE method; - - /** current W */ - SGVector current_w; - /** old W */ - float64_t* old_w; - /** old bias */ - float64_t old_bias; - /** nDim big */ - float64_t* tmp_a_buf; - /** labels */ - SGVector lab; - - /** sparse representation of - * cutting planes */ - float64_t** cp_value; - /** cutting plane index */ - uint32_t** cp_index; - /** cutting plane dimensions */ - uint32_t* cp_nz_dims; - /** bias dimensions */ - float64_t* cp_bias; - - /** primal objective */ - float64_t primal_objective; -}; -} -#endif - -#endif //USE_GPL_SHOGUN diff --git a/src/shogun/classifier/svm/SVMSGD.cpp b/src/shogun/classifier/svm/SVMSGD.cpp index f608ff3eef0..bad90f8b9fc 100644 --- a/src/shogun/classifier/svm/SVMSGD.cpp +++ b/src/shogun/classifier/svm/SVMSGD.cpp @@ -24,6 +24,7 @@ #include #include #include +#include #include #include @@ -109,14 +110,13 @@ bool CSVMSGD::train_machine(CFeatures* data) calibrate(); SG_INFO("Training on %d vectors\n", num_vec) - CSignal::clear_cancel(); ELossType loss_type = loss->get_loss_type(); bool is_log_loss = false; if ((loss_type == L_LOGLOSS) || (loss_type == L_LOGLOSSMARGIN)) is_log_loss = true; - for(int32_t e=0; e::scale_vector(r, w.vector, w.vlen); + linalg::scale(w, w, r); count = skip; } t++; } } - float64_t wnorm = CMath::dot(w.vector,w.vector, w.vlen); + float64_t wnorm = linalg::dot(w, w); SG_INFO("Norm: %.6f, Bias: %.6f\n", wnorm, bias) set_w(w); diff --git a/src/shogun/classifier/svm/WDSVMOcas.cpp b/src/shogun/classifier/svm/WDSVMOcas.cpp deleted file mode 100644 index 64b5e70fbbc..00000000000 --- a/src/shogun/classifier/svm/WDSVMOcas.cpp +++ /dev/null @@ -1,645 +0,0 @@ -/* - * This program is free software; you can redistribute it and/or modify - * it under the terms of the GNU General Public License as published by - * the Free Software Foundation; either version 3 of the License, or - * (at your option) any later version. - * - * Written (W) 2007-2008 Vojtech Franc - * Written (W) 2007-2009 Soeren Sonnenburg - * Copyright (C) 2007-2009 Fraunhofer Institute FIRST and Max-Planck-Society - */ - -#include -#ifdef USE_GPL_SHOGUN - -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include - -using namespace shogun; - -#ifndef DOXYGEN_SHOULD_SKIP_THIS -struct wdocas_thread_params_output -{ - float32_t* out; - int32_t* val; - float64_t* output; - CWDSVMOcas* wdocas; - int32_t start; - int32_t end; -}; - -struct wdocas_thread_params_add -{ - CWDSVMOcas* wdocas; - float32_t* new_a; - uint32_t* new_cut; - int32_t start; - int32_t end; - uint32_t cut_length; -}; -#endif // DOXYGEN_SHOULD_SKIP_THIS - -CWDSVMOcas::CWDSVMOcas() -: CMachine(), use_bias(false), bufsize(3000), C1(1), C2(1), - epsilon(1e-3), method(SVM_OCAS) -{ - SG_UNSTABLE("CWDSVMOcas::CWDSVMOcas()", "\n") - - w=NULL; - old_w=NULL; - features=NULL; - degree=6; - from_degree=40; - wd_weights=NULL; - w_offsets=NULL; - normalization_const=1.0; -} - -CWDSVMOcas::CWDSVMOcas(E_SVM_TYPE type) -: CMachine(), use_bias(false), bufsize(3000), C1(1), C2(1), - epsilon(1e-3), method(type) -{ - w=NULL; - old_w=NULL; - features=NULL; - degree=6; - from_degree=40; - wd_weights=NULL; - w_offsets=NULL; - normalization_const=1.0; -} - -CWDSVMOcas::CWDSVMOcas( - float64_t C, int32_t d, int32_t from_d, CStringFeatures* traindat, - CLabels* trainlab) -: CMachine(), use_bias(false), bufsize(3000), C1(C), C2(C), epsilon(1e-3), - degree(d), from_degree(from_d) -{ - w=NULL; - old_w=NULL; - method=SVM_OCAS; - features=traindat; - set_labels(trainlab); - wd_weights=NULL; - w_offsets=NULL; - normalization_const=1.0; -} - - -CWDSVMOcas::~CWDSVMOcas() -{ -} - -CBinaryLabels* CWDSVMOcas::apply_binary(CFeatures* data) -{ - SGVector outputs = apply_get_outputs(data); - return new CBinaryLabels(outputs); -} - -CRegressionLabels* CWDSVMOcas::apply_regression(CFeatures* data) -{ - SGVector outputs = apply_get_outputs(data); - return new CRegressionLabels(outputs); -} - -SGVector CWDSVMOcas::apply_get_outputs(CFeatures* data) -{ - if (data) - { - if (data->get_feature_class() != C_STRING || - data->get_feature_type() != F_BYTE) - { - SG_ERROR("Features not of class string type byte\n") - } - - set_features((CStringFeatures*) data); - } - ASSERT(features) - - set_wd_weights(); - set_normalization_const(); - - SGVector outputs; - if (features) - { - int32_t num=features->get_num_vectors(); - ASSERT(num>0) - - outputs = SGVector(num); - - for (int32_t i=0; i0 && degree<=8) - SG_FREE(wd_weights); - wd_weights=SG_MALLOC(float32_t, degree); - SG_FREE(w_offsets); - w_offsets=SG_MALLOC(int32_t, degree); - int32_t w_dim_single_c=0; - - for (int32_t i=0; iget_label_type() == LT_BINARY) - if (data) - { - if (data->get_feature_class() != C_STRING || - data->get_feature_type() != F_BYTE) - { - SG_ERROR("Features not of class string type byte\n") - } - set_features((CStringFeatures*) data); - } - - ASSERT(get_features()) - CAlphabet* alphabet=get_features()->get_alphabet(); - ASSERT(alphabet && alphabet->get_alphabet()==RAWDNA) - - alphabet_size=alphabet->get_num_symbols(); - string_length=features->get_num_vectors(); - SGVector labvec=((CBinaryLabels*) m_labels)->get_labels(); - lab=labvec.vector; - - w_dim_single_char=set_wd_weights(); - //CMath::display_vector(wd_weights, degree, "wd_weights"); - SG_DEBUG("w_dim_single_char=%d\n", w_dim_single_char) - w_dim=string_length*w_dim_single_char; - SG_DEBUG("cutting plane has %d dims\n", w_dim) - num_vec=get_features()->get_max_vector_length(); - - set_normalization_const(); - SG_INFO("num_vec: %d num_lab: %d\n", num_vec, labvec.vlen) - ASSERT(num_vec==labvec.vlen) - ASSERT(num_vec>0) - - SG_FREE(w); - w=SG_MALLOC(float32_t, w_dim); - memset(w, 0, w_dim*sizeof(float32_t)); - - SG_FREE(old_w); - old_w=SG_MALLOC(float32_t, w_dim); - memset(old_w, 0, w_dim*sizeof(float32_t)); - bias=0; - old_bias=0; - - cuts=SG_MALLOC(float32_t*, bufsize); - memset(cuts, 0, sizeof(*cuts)*bufsize); - cp_bias=SG_MALLOC(float64_t, bufsize); - memset(cp_bias, 0, sizeof(float64_t)*bufsize); - -/////speed tests///// - /*float64_t* tmp = SG_MALLOC(float64_t, num_vec); - float64_t start=CTime::get_curtime(); - CMath::random_vector(w, w_dim, (float32_t) 0, (float32_t) 1000); - compute_output(tmp, this); - start=CTime::get_curtime()-start; - SG_PRINT("timing:%f\n", start) - SG_FREE(tmp); - exit(1);*/ -/////speed tests///// - float64_t TolAbs=0; - float64_t QPBound=0; - uint8_t Method=0; - if (method == SVM_OCAS) - Method = 1; - ocas_return_value_T result = svm_ocas_solver( get_C1(), num_vec, get_epsilon(), - TolAbs, QPBound, get_max_train_time(), bufsize, Method, - &CWDSVMOcas::compute_W, - &CWDSVMOcas::update_W, - &CWDSVMOcas::add_new_cut, - &CWDSVMOcas::compute_output, - &CWDSVMOcas::sort, - &CWDSVMOcas::print, - this); - - SG_INFO("Ocas Converged after %d iterations\n" - "==================================\n" - "timing statistics:\n" - "output_time: %f s\n" - "sort_time: %f s\n" - "add_time: %f s\n" - "w_time: %f s\n" - "solver_time %f s\n" - "ocas_time %f s\n\n", result.nIter, result.output_time, result.sort_time, - result.add_time, result.w_time, result.qp_solver_time, result.ocas_time); - - for (int32_t i=bufsize-1; i>=0; i--) - SG_FREE(cuts[i]); - SG_FREE(cuts); - - lab=NULL; - SG_UNREF(alphabet); - - return true; -} - -/*---------------------------------------------------------------------------------- - sq_norm_W = sparse_update_W( t ) does the following: - - W = oldW*(1-t) + t*W; - sq_norm_W = W'*W; - - ---------------------------------------------------------------------------------*/ -float64_t CWDSVMOcas::update_W( float64_t t, void* ptr ) -{ - float64_t sq_norm_W = 0; - CWDSVMOcas* o = (CWDSVMOcas*) ptr; - uint32_t nDim = (uint32_t) o->w_dim; - float32_t* W=o->w; - float32_t* oldW=o->old_w; - float64_t bias=o->bias; - float64_t old_bias=bias; - - for(uint32_t j=0; j bias=bias; - o->old_bias=old_bias; - - return( sq_norm_W ); -} - -/*---------------------------------------------------------------------------------- - sparse_add_new_cut( new_col_H, new_cut, cut_length, nSel ) does the following: - - new_a = sum(data_X(:,find(new_cut ~=0 )),2); - new_col_H = [sparse_A(:,1:nSel)'*new_a ; new_a'*new_a]; - sparse_A(:,nSel+1) = new_a; - - ---------------------------------------------------------------------------------*/ -void* CWDSVMOcas::add_new_cut_helper( void* ptr) -{ - wdocas_thread_params_add* p = (wdocas_thread_params_add*) ptr; - CWDSVMOcas* o = p->wdocas; - int32_t start = p->start; - int32_t end = p->end; - int32_t string_length = o->string_length; - //uint32_t nDim=(uint32_t) o->w_dim; - uint32_t cut_length=p->cut_length; - uint32_t* new_cut=p->new_cut; - int32_t* w_offsets = o->w_offsets; - float64_t* y = o->lab; - int32_t alphabet_size = o->alphabet_size; - float32_t* wd_weights = o->wd_weights; - int32_t degree = o->degree; - CStringFeatures* f = o->features; - float64_t normalization_const = o->normalization_const; - - // temporary vector - float32_t* new_a = p->new_a; - //float32_t* new_a = SG_MALLOC(float32_t, nDim); - //memset(new_a, 0, sizeof(float32_t)*nDim); - - int32_t* val=SG_MALLOC(int32_t, cut_length); - for (int32_t j=start; jw_dim_single_char*j; - memset(val,0,sizeof(int32_t)*cut_length); - int32_t lim=CMath::min(degree, string_length-j); - int32_t len; - - for (int32_t k=0; kget_feature_vector(j+k, len, free_vec); - float32_t wd = wd_weights[k]/normalization_const; - - for(uint32_t i=0; i < cut_length; i++) - { - val[i]=val[i]*alphabet_size + vec[new_cut[i]]; - new_a[offs+val[i]]+=wd * y[new_cut[i]]; - } - offs+=w_offsets[k]; - f->free_feature_vector(vec, j+k, free_vec); - } - } - - //p->new_a=new_a; - SG_FREE(val); - return NULL; -} - -int CWDSVMOcas::add_new_cut( - float64_t *new_col_H, uint32_t *new_cut, uint32_t cut_length, - uint32_t nSel, void* ptr) -{ - CWDSVMOcas* o = (CWDSVMOcas*) ptr; - uint32_t i; - float64_t* c_bias = o->cp_bias; - uint32_t nDim=(uint32_t) o->w_dim; - float32_t** cuts=o->cuts; - float32_t* new_a=SG_MALLOC(float32_t, nDim); - memset(new_a, 0, sizeof(float32_t)*nDim); -#ifdef HAVE_PTHREAD - - wdocas_thread_params_add* params_add=SG_MALLOC(wdocas_thread_params_add, o->parallel->get_num_threads()); - pthread_t* threads=SG_MALLOC(pthread_t, o->parallel->get_num_threads()); - - int32_t string_length = o->string_length; - int32_t t; - int32_t nthreads=o->parallel->get_num_threads()-1; - int32_t step= string_length/o->parallel->get_num_threads(); - - if (step<1) - { - nthreads=string_length-1; - step=1; - } - - for (t=0; tuse_bias) - c_bias[nSel]+=o->lab[new_cut[i]]; - } - - // insert new_a into the last column of sparse_A - for(i=0; i < nSel; i++) - new_col_H[i] = CMath::dot(new_a, cuts[i], nDim) + c_bias[nSel]*c_bias[i]; - new_col_H[nSel] = CMath::dot(new_a, new_a, nDim) + CMath::sq(c_bias[nSel]); - - cuts[nSel]=new_a; - //CMath::display_vector(new_col_H, nSel+1, "new_col_H"); - //CMath::display_vector(cuts[nSel], nDim, "cut[nSel]"); - // - - return 0; -} - -int CWDSVMOcas::sort( float64_t* vals, float64_t* data, uint32_t size) -{ - CMath::qsort_index(vals, data, size); - return 0; -} - -/*---------------------------------------------------------------------- - sparse_compute_output( output ) does the follwing: - - output = data_X'*W; - ----------------------------------------------------------------------*/ -void* CWDSVMOcas::compute_output_helper(void* ptr) -{ - wdocas_thread_params_output* p = (wdocas_thread_params_output*) ptr; - CWDSVMOcas* o = p->wdocas; - int32_t start = p->start; - int32_t end = p->end; - float32_t* out = p->out; - float64_t* output = p->output; - int32_t* val = p->val; - - CStringFeatures* f=o->get_features(); - - int32_t degree = o->degree; - int32_t string_length = o->string_length; - int32_t alphabet_size = o->alphabet_size; - int32_t* w_offsets = o->w_offsets; - float32_t* wd_weights = o->wd_weights; - float32_t* w= o->w; - - float64_t* y = o->lab; - float64_t normalization_const = o->normalization_const; - - - for (int32_t j=0; jw_dim_single_char*j; - for (int32_t i=start ; iget_feature_vector(j+k, len, free_vec); - float32_t wd = wd_weights[k]; - - for (int32_t i=start; i>8)&255); - val[ii+2]=val[ii+2]*alphabet_size + ((x>>16)&255); - val[ii+3]=val[ii+3]*alphabet_size + (x>>24); - out[ii]+=wd*w[offs+val[ii]]; - out[ii+1]+=wd*w[offs+val[ii+1]]; - out[ii+2]+=wd*w[offs+val[ii+2]]; - out[ii+3]+=wd*w[offs+val[ii+3]]; - }*/ - - /*for (int32_t i=0; i>3; i++) // fastest on 64bit: 1.5s - { - uint64_t x=((uint64_t*) vec)[i]; - int32_t ii=i<<3; - val[ii]=val[ii]*alphabet_size + (x&255); - val[ii+1]=val[ii+1]*alphabet_size + ((x>>8)&255); - val[ii+2]=val[ii+2]*alphabet_size + ((x>>16)&255); - val[ii+3]=val[ii+3]*alphabet_size + ((x>>24)&255); - val[ii+4]=val[ii+4]*alphabet_size + ((x>>32)&255); - val[ii+5]=val[ii+5]*alphabet_size + ((x>>40)&255); - val[ii+6]=val[ii+6]*alphabet_size + ((x>>48)&255); - val[ii+7]=val[ii+7]*alphabet_size + (x>>56); - out[ii]+=wd*w[offs+val[ii]]; - out[ii+1]+=wd*w[offs+val[ii+1]]; - out[ii+2]+=wd*w[offs+val[ii+2]]; - out[ii+3]+=wd*w[offs+val[ii+3]]; - out[ii+4]+=wd*w[offs+val[ii+4]]; - out[ii+5]+=wd*w[offs+val[ii+5]]; - out[ii+6]+=wd*w[offs+val[ii+6]]; - out[ii+7]+=wd*w[offs+val[ii+7]]; - }*/ - offs+=w_offsets[k]; - f->free_feature_vector(vec, j+k, free_vec); - } - } - - for (int32_t i=start; ibias + out[i]*y[i]/normalization_const; - - //CMath::display_vector(o->w, o->w_dim, "w"); - //CMath::display_vector(output, nData, "out"); - return NULL; -} - -int CWDSVMOcas::compute_output( float64_t *output, void* ptr ) -{ -#ifdef HAVE_PTHREAD - CWDSVMOcas* o = (CWDSVMOcas*) ptr; - int32_t nData=o->num_vec; - wdocas_thread_params_output* params_output=SG_MALLOC(wdocas_thread_params_output, o->parallel->get_num_threads()); - pthread_t* threads = SG_MALLOC(pthread_t, o->parallel->get_num_threads()); - - float32_t* out=SG_MALLOC(float32_t, nData); - int32_t* val=SG_MALLOC(int32_t, nData); - memset(out, 0, sizeof(float32_t)*nData); - - int32_t t; - int32_t nthreads=o->parallel->get_num_threads()-1; - int32_t step= nData/o->parallel->get_num_threads(); - - if (step<1) - { - nthreads=nData-1; - step=1; - } - - for (t=0; tw_dim; - CMath::swap(o->w, o->old_w); - float32_t* W=o->w; - float32_t* oldW=o->old_w; - float32_t** cuts=o->cuts; - memset(W, 0, sizeof(float32_t)*nDim); - float64_t* c_bias = o->cp_bias; - float64_t old_bias=o->bias; - float64_t bias=0; - - for (uint32_t i=0; i 0) - SGVector::vec1_plus_scalar_times_vec2(W, (float32_t) alpha[i], cuts[i], nDim); - - bias += c_bias[i]*alpha[i]; - } - - *sq_norm_W = CMath::dot(W,W, nDim) +CMath::sq(bias); - *dp_WoldW = CMath::dot(W,oldW, nDim) + bias*old_bias;; - //SG_PRINT("nSel=%d sq_norm_W=%f dp_WoldW=%f\n", nSel, *sq_norm_W, *dp_WoldW) - - o->bias = bias; - o->old_bias = old_bias; -} - -#endif //USE_GPL_SHOGUN diff --git a/src/shogun/classifier/svm/WDSVMOcas.h b/src/shogun/classifier/svm/WDSVMOcas.h deleted file mode 100644 index 85050faf030..00000000000 --- a/src/shogun/classifier/svm/WDSVMOcas.h +++ /dev/null @@ -1,379 +0,0 @@ -/* - * This program is free software; you can redistribute it and/or modify - * it under the terms of the GNU General Public License as published by - * the Free Software Foundation; either version 3 of the License, or - * (at your option) any later version. - * - * Written (W) 2007-2008 Vojtech Franc - * Written (W) 2007-2009 Soeren Sonnenburg - * Copyright (C) 2007-2009 Fraunhofer Institute FIRST and Max-Planck-Society - */ - - -#ifndef _WDSVMOCAS_H___ -#define _WDSVMOCAS_H___ - -#include -#ifdef USE_GPL_SHOGUN - -#include -#include -#include -#include -#include - -namespace shogun -{ -template class CStringFeatures; - -/** @brief class WDSVMOcas */ -class CWDSVMOcas : public CMachine -{ - public: - /** problem type */ - MACHINE_PROBLEM_TYPE(PT_BINARY); - - /** default constructor */ - CWDSVMOcas(); - - /** constructor - * - * @param type type of SVM - */ - CWDSVMOcas(E_SVM_TYPE type); - - /** constructor - * - * @param C constant C - * @param d degree - * @param from_d from degree - * @param traindat training features - * @param trainlab labels for training features - */ - CWDSVMOcas( - float64_t C, int32_t d, int32_t from_d, - CStringFeatures* traindat, CLabels* trainlab); - virtual ~CWDSVMOcas(); - - /** get classifier type - * - * @return classifier type WDSVMOCAS - */ - virtual EMachineType get_classifier_type() { return CT_WDSVMOCAS; } - - /** set C - * - * @param c_neg new C constant for negatively labeled examples - * @param c_pos new C constant for positively labeled examples - * - */ - inline void set_C(float64_t c_neg, float64_t c_pos) { C1=c_neg; C2=c_pos; } - - /** get C1 - * - * @return C1 - */ - inline float64_t get_C1() { return C1; } - - /** get C2 - * - * @return C2 - */ - inline float64_t get_C2() { return C2; } - - /** set epsilon - * - * @param eps new epsilon - */ - inline void set_epsilon(float64_t eps) { epsilon=eps; } - - /** get epsilon - * - * @return epsilon - */ - inline float64_t get_epsilon() { return epsilon; } - - /** set features - * - * @param feat features to set - */ - inline void set_features(CStringFeatures* feat) - { - SG_REF(feat); - SG_UNREF(features); - features=feat; - } - - /** get features - * - * @return features - */ - inline CStringFeatures* get_features() - { - SG_REF(features); - return features; - } - - /** set if bias shall be enabled - * - * @param enable_bias if bias shall be enabled - */ - inline void set_bias_enabled(bool enable_bias) { use_bias=enable_bias; } - - /** check if bias is enabled - * - * @return if bias is enabled - */ - inline bool get_bias_enabled() { return use_bias; } - - /** set buffer size - * - * @param sz buffer size - */ - inline void set_bufsize(int32_t sz) { bufsize=sz; } - - /** get buffer size - * - * @return buffer size - */ - inline int32_t get_bufsize() { return bufsize; } - - /** set degree - * - * @param d degree - * @param from_d from degree - */ - inline void set_degree(int32_t d, int32_t from_d) - { - degree=d; - from_degree=from_d; - } - - /** get degree - * - * @return degree - */ - inline int32_t get_degree() { return degree; } - - /** classify objects - * for binary classification problems - * - * @param data (test)data to be classified - * @return classified labels - */ - virtual CBinaryLabels* apply_binary(CFeatures* data=NULL); - - /** classify objects - * for regression problems - * - * @param data (test)data to be classified - * @return classified labels - */ - virtual CRegressionLabels* apply_regression(CFeatures* data=NULL); - - /** classify one example - * - * @param num number of example to classify - * @return classified result - */ - virtual float64_t apply_one(int32_t num) - { - ASSERT(features) - if (!wd_weights) - set_wd_weights(); - - int32_t len=0; - float64_t sum=0; - bool free_vec; - uint8_t* vec=features->get_feature_vector(num, len, free_vec); - //SG_INFO("len %d, string_length %d\n", len, string_length) - ASSERT(len==string_length) - - for (int32_t j=0; jfree_feature_vector(vec, num, free_vec); - return sum/normalization_const; - } - - /** set normalization const */ - inline void set_normalization_const() - { - ASSERT(features) - normalization_const=0; - for (int32_t i=0; i apply_get_outputs(CFeatures* data); - - /** set wd weights - * - * @return w_dim_single_c - */ - int32_t set_wd_weights(); - - /** compute W - * - * @param sq_norm_W square normed W - * @param dp_WoldW dp W old W - * @param alpha alpha - * @param nSel nSel - * @param ptr ptr - */ - static void compute_W( - float64_t *sq_norm_W, float64_t *dp_WoldW, float64_t *alpha, - uint32_t nSel, void* ptr ); - - /** update W - * - * @param t t - * @param ptr ptr - * @return something floaty - */ - static float64_t update_W(float64_t t, void* ptr ); - - /** helper function for adding a new cut - * - * @param ptr - * @return ptr - */ - static void* add_new_cut_helper(void* ptr); - - /** add new cut - * - * @param new_col_H new col H - * @param new_cut new cut - * @param cut_length length of cut - * @param nSel nSel - * @param ptr ptr - */ - static int add_new_cut( - float64_t *new_col_H, uint32_t *new_cut, uint32_t cut_length, - uint32_t nSel, void* ptr ); - - /** helper function for computing the output - * - * @param ptr - * @return ptr - */ - static void* compute_output_helper(void* ptr); - - /** compute output - * - * @param output output - * @param ptr ptr - */ - static int compute_output( float64_t *output, void* ptr ); - - /** sort - * - * @param vals vals - * @param data data - * @param size size - */ - static int sort( float64_t* vals, float64_t* data, uint32_t size); - - /** print nothing */ - static inline void print(ocas_return_value_T value) - { - return; - } - - - /** @return object name */ - virtual const char* get_name() const { return "WDSVMOcas"; } - - protected: - /** train classifier - * - * @param data training data (parameter can be avoided if distance or - * kernel-based classifiers are used and distance/kernels are - * initialized with train data) - * - * @return whether training was successful - */ - virtual bool train_machine(CFeatures* data=NULL); - - protected: - /** features */ - CStringFeatures* features; - /** if bias shall be used */ - bool use_bias; - /** buffer size */ - int32_t bufsize; - /** C1 */ - float64_t C1; - /** C2 */ - float64_t C2; - /** epsilon */ - float64_t epsilon; - /** method */ - E_SVM_TYPE method; - - /** degree */ - int32_t degree; - /** from degree */ - int32_t from_degree; - /** wd weights */ - float32_t* wd_weights; - /** num vectors */ - int32_t num_vec; - /** length of string in vector */ - int32_t string_length; - /** size of alphabet */ - int32_t alphabet_size; - - /** normalization const */ - float64_t normalization_const; - - /** bias */ - float64_t bias; - /** old_bias */ - float64_t old_bias; - /** w offsets */ - int32_t* w_offsets; - /** w dim */ - int32_t w_dim; - /** w dim of a single char */ - int32_t w_dim_single_char; - /** w */ - float32_t* w; - /** old w*/ - float32_t* old_w; - /** labels */ - float64_t* lab; - - /** cuts */ - float32_t** cuts; - /** bias dimensions */ - float64_t* cp_bias; -}; -} -#endif - -#endif //USE_GPL_SHOGUN diff --git a/src/shogun/classifier/vw/VowpalWabbit.cpp b/src/shogun/classifier/vw/VowpalWabbit.cpp index 6d5928f1762..4b17973dd83 100644 --- a/src/shogun/classifier/vw/VowpalWabbit.cpp +++ b/src/shogun/classifier/vw/VowpalWabbit.cpp @@ -59,10 +59,9 @@ CVowpalWabbit::CVowpalWabbit(CVowpalWabbit *vw) save_predictions = vw->save_predictions; prediction_fd = vw->prediction_fd; - w = reg->weight_vectors[0]; + m_w.vector = reg->weight_vectors[0]; reg->weight_vectors[0] = NULL; - copy(vw->w, vw->w+vw->w_dim, w); - w_dim = vw->w_dim; + copy(vw->m_w.vector, vw->m_w.vector+vw->m_w.vlen, m_w.vector); bias = vw->bias; } @@ -83,7 +82,7 @@ void CVowpalWabbit::reinitialize_weights() } reg->init(env); - w = reg->weight_vectors[0]; + m_w.vector = reg->weight_vectors[0]; reg->weight_vectors[0] = NULL; } @@ -114,9 +113,8 @@ void CVowpalWabbit::set_exact_adaptive_norm(bool exact_adaptive) void CVowpalWabbit::load_regressor(char* file_name) { reg->load_regressor(file_name); - w = reg->weight_vectors[0]; + m_w = SGVector(reg->weight_vectors[0], 1 << env->num_bits); reg->weight_vectors[0] = NULL; - w_dim = 1 << env->num_bits; } void CVowpalWabbit::set_regressor_out(char* file_name, bool is_text) @@ -163,9 +161,8 @@ bool CVowpalWabbit::train_machine(CFeatures* feat) "loss", "last", "counter", "weight", "label", "predict", "features"); } - CSignal::clear_cancel(); features->start_parser(); - while (!(CSignal::cancel_computations()) && (env->passes_complete < env->num_passes)) + while (!(cancel_computation()) && (env->passes_complete < env->num_passes)) { while (features->get_next_example()) { @@ -271,9 +268,8 @@ void CVowpalWabbit::init(CStreamingVwFeatures* feat) save_predictions = false; prediction_fd = -1; - w = reg->weight_vectors[0]; + m_w = SGVector(reg->weight_vectors[0], 1 << env->num_bits); reg->weight_vectors[0] = NULL; - w_dim = 1 << env->num_bits; bias = 0.; } diff --git a/src/shogun/clustering/GMM.cpp b/src/shogun/clustering/GMM.cpp index f2877377871..045637695b6 100644 --- a/src/shogun/clustering/GMM.cpp +++ b/src/shogun/clustering/GMM.cpp @@ -11,15 +11,15 @@ #ifdef HAVE_LAPACK +#include #include #include #include -#include +#include #include #include -#include +#include #include - #include using namespace shogun; @@ -32,8 +32,7 @@ CGMM::CGMM() : CDistribution(), m_components(), m_coefficients() CGMM::CGMM(int32_t n, ECovType cov_type) : CDistribution(), m_components(), m_coefficients() { - m_coefficients.vector=SG_MALLOC(float64_t, n); - m_coefficients.vlen=n; + m_coefficients = SGVector(n); m_components = vector(n); for (int32_t i=0; i components, SGVector coefficients, bool m_components[i]->set_cov_type(components[i]->get_cov_type()); SGVector old_mean=components[i]->get_mean(); - SGVector new_mean(old_mean.vlen); - sg_memcpy(new_mean.vector, old_mean.vector, old_mean.vlen*sizeof(float64_t)); + SGVector new_mean = old_mean.clone(); m_components[i]->set_mean(new_mean); SGVector old_d=components[i]->get_d(); - SGVector new_d(old_d.vlen); - sg_memcpy(new_d.vector, old_d.vector, old_d.vlen*sizeof(float64_t)); + SGVector new_d = old_d.clone(); m_components[i]->set_d(new_d); if (components[i]->get_cov_type()==FULL) { SGMatrix old_u=components[i]->get_u(); - SGMatrix new_u(old_u.num_rows, old_u.num_cols); - sg_memcpy(new_u.matrix, old_u.matrix, old_u.num_rows*old_u.num_cols*sizeof(float64_t)); + SGMatrix new_u = old_u.clone(); m_components[i]->set_u(new_u); } @@ -154,8 +150,8 @@ float64_t CGMM::train_em(float64_t min_cov, int32_t max_iter, float64_t min_chan int32_t iter=0; float64_t log_likelihood_prev=0; float64_t log_likelihood_cur=0; - float64_t* logPxy=SG_MALLOC(float64_t, num_vectors*m_components.size()); - float64_t* logPx=SG_MALLOC(float64_t, num_vectors); + SGVector logPxy(num_vectors * m_components.size()); + SGVector logPx(num_vectors); //float64_t* logPost=SG_MALLOC(float64_t, num_vectors*m_components.vlen); while (iter v=dotdata->get_computed_dot_feature_vector(i); for (int32_t j=0; jcompute_log_PDF(v)+CMath::log(m_coefficients[j]); - logPx[i]+=CMath::exp(logPxy[i*m_components.size()+j]); + logPxy[index_t(i * m_components.size() + j)] = + m_components[j]->compute_log_PDF(v) + + CMath::log(m_coefficients[j]); + logPx[i] += + CMath::exp(logPxy[index_t(i * m_components.size() + j)]); } logPx[i]=CMath::log(logPx[i]); @@ -179,7 +178,8 @@ float64_t CGMM::train_em(float64_t min_cov, int32_t max_iter, float64_t min_chan for (int32_t j=0; j logPxy(num_vectors * m_components.size()); + SGVector logPx(num_vectors); + SGVector logPost(num_vectors * m_components.size()); + SGVector logPostSum(m_components.size()); + SGVector logPostSum2(m_components.size()); + SGVector logPostSumSum( + m_components.size() * (m_components.size() - 1) / 2); + SGVector split_crit(m_components.size()); + SGVector merge_crit( + m_components.size() * (m_components.size() - 1) / 2); + SGVector split_ind(m_components.size()); + SGVector merge_ind( + m_components.size() * (m_components.size() - 1) / 2); while (iter v=dotdata->get_computed_dot_feature_vector(i); for (int32_t j=0; jcompute_log_PDF(v)+CMath::log(m_coefficients[j]); - logPx[i]+=CMath::exp(logPxy[i*m_components.size()+j]); + logPxy[index_t(i * m_components.size() + j)] = + m_components[j]->compute_log_PDF(v) + + CMath::log(m_coefficients[j]); + logPx[i] += + CMath::exp(logPxy[index_t(i * m_components.size() + j)]); } logPx[i]=CMath::log(logPx[i]); for (int32_t j=0; jget_num_vectors(); - float64_t* init_logPxy=SG_MALLOC(float64_t, num_vectors*m_components.size()); - float64_t* init_logPx=SG_MALLOC(float64_t, num_vectors); - float64_t* init_logPx_fix=SG_MALLOC(float64_t, num_vectors); - float64_t* post_add=SG_MALLOC(float64_t, num_vectors); + SGVector init_logPxy(num_vectors * m_components.size()); + SGVector init_logPx(num_vectors); + SGVector init_logPx_fix(num_vectors); + SGVector post_add(num_vectors); for (int32_t i=0; i v=dotdata->get_computed_dot_feature_vector(i); for (int32_t j=0; jcompute_log_PDF(v)+CMath::log(m_coefficients[j]); - init_logPx[i]+=CMath::exp(init_logPxy[i*m_components.size()+j]); + init_logPxy[index_t(i * m_components.size() + j)] = + m_components[j]->compute_log_PDF(v) + + CMath::log(m_coefficients[j]); + init_logPx[i] += + CMath::exp(init_logPxy[index_t(i * m_components.size() + j)]); if (j!=comp1 && j!=comp2 && j!=comp3) { - init_logPx_fix[i]+=CMath::exp(init_logPxy[i*m_components.size()+j]); + init_logPx_fix[i] += CMath::exp( + init_logPxy[index_t(i * m_components.size() + j)]); } } init_logPx[i]=CMath::log(init_logPx[i]); - post_add[i]=CMath::log(CMath::exp(init_logPxy[i*m_components.size()+comp1]-init_logPx[i])+ - CMath::exp(init_logPxy[i*m_components.size()+comp2]-init_logPx[i])+ - CMath::exp(init_logPxy[i*m_components.size()+comp3]-init_logPx[i])); + post_add[i] = CMath::log( + CMath::exp( + init_logPxy[index_t(i * m_components.size() + comp1)] - + init_logPx[i]) + + CMath::exp( + init_logPxy[index_t(i * m_components.size() + comp2)] - + init_logPx[i]) + + CMath::exp( + init_logPxy[index_t(i * m_components.size() + comp3)] - + init_logPx[i])); } vector components(3); @@ -385,8 +401,10 @@ void CGMM::partial_em(int32_t comp1, int32_t comp2, int32_t comp3, float64_t min float64_t noise_mag=SGVector::twonorm(components[0]->get_mean().vector, dim_n)*0.1/ CMath::sqrt((float64_t)dim_n); - SGVector::add(components[1]->get_mean().vector, alpha1, components[1]->get_mean().vector, alpha2, - components[2]->get_mean().vector, dim_n); + auto temp_mean = components[2]->get_mean(); + auto temp_mean_result = components[1]->get_mean(); + linalg::add(temp_mean_result, temp_mean, temp_mean_result, alpha1, alpha2); + components[1]->set_mean(temp_mean_result); for (int32_t i=0; i c1=components[1]->get_cov(); SGMatrix c2=components[2]->get_cov(); - SGVector::add(c1.matrix, alpha1, c1.matrix, alpha2, c2.matrix, dim_n*dim_n); + linalg::add(c1, c2, c1, alpha1, alpha2); components[1]->set_d(SGVector(SGMatrix::compute_eigenvectors(c1.matrix, dim_n, dim_n), dim_n)); components[1]->set_u(c1); @@ -434,8 +452,10 @@ void CGMM::partial_em(int32_t comp1, int32_t comp2, int32_t comp3, float64_t min } else if(components[0]->get_cov_type()==DIAG) { - SGVector::add(components[1]->get_d().vector, alpha1, components[1]->get_d().vector, - alpha2, components[2]->get_d().vector, dim_n); + auto result_d = components[1]->get_d(); + auto temp_d = components[2]->get_d(); + linalg::add(result_d, temp_d, result_d, alpha1, alpha2); + components[1]->set_d(result_d); float64_t new_d=0; for (int32_t i=0; i alpha(num_vectors, 3); - float64_t* logPxy=SG_MALLOC(float64_t, num_vectors*3); - float64_t* logPx=SG_MALLOC(float64_t, num_vectors); + SGVector logPxy(num_vectors * 3); + SGVector logPx(num_vectors); //float64_t* logPost=SG_MALLOC(float64_t, num_vectors*m_components.vlen); while (iter alpha, float64_t min_cov) @@ -523,113 +537,121 @@ void CGMM::max_likelihood(SGMatrix alpha, float64_t min_cov) float64_t alpha_sum; float64_t alpha_sum_sum=0; - float64_t* mean_sum; - float64_t* cov_sum=NULL; for (int32_t i=0; i mean_sum(num_dim); + linalg::zero(mean_sum); for (int32_t j=0; j v=dotdata->get_computed_dot_feature_vector(j); - SGVector::add(mean_sum, alpha.matrix[j*alpha.num_cols+i], v.vector, 1, mean_sum, v.vlen); + linalg::add( + v, mean_sum, mean_sum, alpha.matrix[j * alpha.num_cols + i], + 1.0); } - for (int32_t j=0; jset_mean(SGVector(mean_sum, num_dim)); + m_components[i]->set_mean(mean_sum); - ECovType cov_type=m_components[i]->get_cov_type(); + SGMatrix cov_sum; + ECovType cov_type = m_components[i]->get_cov_type(); if (cov_type==FULL) { - cov_sum=SG_MALLOC(float64_t, num_dim*num_dim); - memset(cov_sum, 0, num_dim*num_dim*sizeof(float64_t)); + cov_sum = SGMatrix(num_dim, num_dim); + linalg::zero(cov_sum); } else if(cov_type==DIAG) { - cov_sum=SG_MALLOC(float64_t, num_dim); - memset(cov_sum, 0, num_dim*sizeof(float64_t)); + cov_sum = SGMatrix(1, num_dim); + linalg::zero(cov_sum); } else if(cov_type==SPHERICAL) { - cov_sum=SG_MALLOC(float64_t, 1); - cov_sum[0]=0; + cov_sum = SGMatrix(1, 1); + linalg::zero(cov_sum); } for (int32_t j=0; j v=dotdata->get_computed_dot_feature_vector(j); - SGVector::add(v.vector, 1, v.vector, -1, mean_sum, v.vlen); + linalg::add(v, mean_sum, v, 1.0, -1.0); switch (cov_type) { case FULL: - cblas_dger(CblasRowMajor, num_dim, num_dim, alpha.matrix[j*alpha.num_cols+i], v.vector, 1, v.vector, - 1, (double*) cov_sum, num_dim); - - break; - case DIAG: - for (int32_t k=0; k(v.vector, 1, v.vlen); + auto temp_result = linalg::matrix_prod( + temp_matrix, temp_matrix, true, false); + cov_sum = temp_result.get_diagonal_vector().clone(); + linalg::scale( + cov_sum, cov_sum, alpha.matrix[j * alpha.num_cols + i]); + } + + break; + case SPHERICAL: + float64_t temp = 0; + + temp = linalg::dot(v, v); + + cov_sum(0, 0) += + temp * alpha.matrix[j * alpha.num_cols + i]; + break; } } switch (cov_type) { case FULL: - for (int32_t j=0; j::compute_eigenvectors(cov_sum, num_dim, num_dim); - for (int32_t j=0; j d0 = + SGMatrix::compute_eigenvectors(cov_sum); + for (int32_t j = 0; j < num_dim; j++) + d0[j] = CMath::max(min_cov, d0[j]); - m_components[i]->set_d(SGVector(d0, num_dim)); - m_components[i]->set_u(SGMatrix(cov_sum, num_dim, num_dim)); + m_components[i]->set_d(d0); + m_components[i]->set_u(cov_sum); - break; - case DIAG: - for (int32_t j=0; jset_d(SGVector(cov_sum, num_dim)); + m_components[i]->set_d(cov_sum.get_row_vector(0)); - break; - case SPHERICAL: - cov_sum[0]/=alpha_sum*num_dim; - cov_sum[0]=CMath::max(min_cov, cov_sum[0]); + break; + case SPHERICAL: + cov_sum[0] /= alpha_sum * num_dim; + cov_sum[0] = CMath::max(min_cov, cov_sum[0]); - m_components[i]->set_d(SGVector(cov_sum, 1)); + m_components[i]->set_d(cov_sum.get_row_vector(0)); - break; + break; } m_coefficients.vector[i]=alpha_sum; alpha_sum_sum+=alpha_sum; } - for (int32_t i=0; i CGMM::alpha_init(SGMatrix init_means) SGVector label_num(init_means.num_cols); for (int32_t i=0; itrain(new CDenseFeatures(init_means)); CMulticlassLabels* init_labels=(CMulticlassLabels*) knn->apply(features); SGMatrix alpha(num_vectors, int32_t(m_components.size())); - memset(alpha.matrix, 0, num_vectors*m_components.size()*sizeof(float64_t)); + linalg::zero(alpha); for (int32_t i=0; iget_int_label(i)]=1; + alpha[i * m_components.size() + init_labels->get_int_label(i)] = 1; SG_UNREF(init_labels); @@ -766,7 +788,7 @@ SGVector CGMM::sample() { REQUIRE(m_components.size()>0, "Number of mixture components is %d but " "must be positive\n", m_components.size()); - float64_t rand_num=CMath::random(float64_t(0), float64_t(1)); + float64_t rand_num = CMath::random(0.0, 1.0); float64_t cum_sum=0; for (int32_t i=0; i +#include #include #include -#include #include +#include #include -#include using namespace shogun; @@ -83,21 +84,21 @@ bool CHierarchical::train_machine(CFeatures* data) float64_t* distances=SG_MALLOC(float64_t, num_pairs); int32_t offs=0; - for (int32_t i=0; iio)) { for (int32_t j=i+1; jdistance(i,j); - index[offs].idx1=i; - index[offs].idx2=j; - offs++; //offs=i*(i+1)/2+j + distances[offs] = distance->distance(i, j); + index[offs].idx1 = i; + index[offs].idx2 = j; + offs++; // offs=i*(i+1)/2+j } - SG_PROGRESS(i, 0, num-1) } CMath::qsort_index(distances, index, (num-1)*num/2); //CMath::display_vector(distances, (num-1)*num/2, "dists"); + auto pb = progress(range(0, num_pairs - 1), *this->io); int32_t k=-1; int32_t l=0; for (; l=merges && k X = ((CDenseFeatures*)features)->get_feature_matrix(); + REQUIRE(X.data(), "Features have not been provided.\n"); int n = X.num_rows; int p = X.num_cols; diff --git a/src/shogun/distance/CustomMahalanobisDistance.cpp b/src/shogun/distance/CustomMahalanobisDistance.cpp index c4f8f84cc9e..734353390c1 100644 --- a/src/shogun/distance/CustomMahalanobisDistance.cpp +++ b/src/shogun/distance/CustomMahalanobisDistance.cpp @@ -10,8 +10,7 @@ #include - -#include +#include using namespace shogun; using namespace Eigen; diff --git a/src/shogun/distance/Distance.cpp b/src/shogun/distance/Distance.cpp index 950f159d427..342936f9130 100644 --- a/src/shogun/distance/Distance.cpp +++ b/src/shogun/distance/Distance.cpp @@ -9,14 +9,15 @@ * Copyright (C) 2006-2009 Fraunhofer Institute FIRST and Max-Planck-Society */ -#include -#include -#include -#include -#include -#include #include #include +#include +#include +#include +#include +#include +#include +#include #include #include @@ -28,6 +29,7 @@ #ifdef HAVE_OPENMP #include + #endif using namespace shogun; @@ -241,15 +243,11 @@ void CDistance::do_precompute_matrix() SG_FREE(precomputed_matrix); precomputed_matrix=SG_MALLOC(float32_t, num*(num+1)/2); - for (int32_t i=0; iio)) { - SG_PROGRESS(i*i,0,num*num) for (int32_t j=0; j<=i; j++) precomputed_matrix[i*(i+1)/2+j] = compute(i,j) ; } - - SG_PROGRESS(num*num,0,num*num) - SG_DONE() } void CDistance::init() @@ -279,8 +277,6 @@ SGMatrix CDistance::get_distance_matrix() int64_t total_num = int64_t(m)*n; int64_t total=0; - int64_t total_start=0; - int64_t total_end=total_num; // if lhs == rhs and sizes match assume k(i,j)=k(j,i) bool symmetric= (lhs && lhs==rhs && m==n); @@ -289,6 +285,8 @@ SGMatrix CDistance::get_distance_matrix() result=SG_MALLOC(T, total_num); + PRange pb = PRange( + range(total_num), *this->io, "PROGRESS: ", UTF8, []() { return true; }); int32_t num_threads; int64_t step; #pragma omp parallel shared(num_threads, step) @@ -333,17 +331,16 @@ SGMatrix CDistance::get_distance_matrix() if (symmetric && i!=j) total++; - if (total%100 == 0) - SG_OBJ_PROGRESS(this, total, total_start, total_end) + pb.print_progress(); - if (CSignal::cancel_computations()) - break; + // TODO: replace with new signal + // if (CSignal::cancel_computations()) + // break; } } } } - - SG_DONE() + pb.complete(); return SGMatrix(result,m,n,true); } diff --git a/src/shogun/distance/MahalanobisDistance.cpp b/src/shogun/distance/MahalanobisDistance.cpp index f05a4248c70..d343d0763a3 100644 --- a/src/shogun/distance/MahalanobisDistance.cpp +++ b/src/shogun/distance/MahalanobisDistance.cpp @@ -50,7 +50,8 @@ bool CMahalanobisDistance::init(CFeatures* l, CFeatures* r) } else { - mean = ((CDenseFeatures*) l)->get_mean((CDotFeatures*) lhs, (CDotFeatures*) rhs); + mean = ((CDenseFeatures*)l) + ->compute_mean((CDotFeatures*)lhs, (CDotFeatures*)rhs); icov = CDotFeatures::compute_cov((CDotFeatures*) lhs, (CDotFeatures*) rhs); } diff --git a/src/shogun/distributions/Gaussian.cpp b/src/shogun/distributions/Gaussian.cpp index 3520649568e..d5c36c6e9e4 100644 --- a/src/shogun/distributions/Gaussian.cpp +++ b/src/shogun/distributions/Gaussian.cpp @@ -10,22 +10,24 @@ */ #include -#ifdef HAVE_LAPACK - +#include #include #include -#include +#include #include +#include using namespace shogun; +using namespace linalg; CGaussian::CGaussian() : CDistribution(), m_constant(0), m_d(), m_u(), m_mean(), m_cov_type(FULL) { register_params(); } -CGaussian::CGaussian(const SGVector mean, SGMatrix cov, ECovType cov_type) - : CDistribution() +CGaussian::CGaussian( + const SGVector mean, SGMatrix cov, ECovType cov_type) + : CDistribution() { ASSERT(mean.vlen==cov.num_rows) ASSERT(cov.num_rows==cov.num_cols) @@ -118,102 +120,115 @@ float64_t CGaussian::get_log_likelihood_example(int32_t num_example) float64_t CGaussian::update_params_em(float64_t* alpha_k, int32_t len) { CDotFeatures* dotdata=dynamic_cast(features); - REQUIRE(dotdata,"dynamic cast from CFeatures to CDotFeatures returned NULL\n") + REQUIRE( + dotdata, "dynamic cast from CFeatures to CDotFeatures returned NULL\n"); int32_t num_dim=dotdata->get_dim_feature_space(); // compute mean float64_t alpha_k_sum=0; SGVector mean(num_dim); - mean.fill_vector(mean.vector,mean.vlen,0); - for (int32_t i=0;i v=dotdata->get_computed_dot_feature_vector(i); - SGVector::add(mean.vector, alpha_k[i], v.vector, 1, mean.vector, v.vlen); + linalg::add(v, mean, mean, alpha_k[i], 1.0); } - for (int32_t i=0; i cov_sum; ECovType cov_type=get_cov_type(); if (cov_type==FULL) { - cov_sum=SG_MALLOC(float64_t, num_dim*num_dim); - memset(cov_sum, 0, num_dim*num_dim*sizeof(float64_t)); + cov_sum = SGMatrix(num_dim, num_dim); + cov_sum.zero(); } else if(cov_type==DIAG) { - cov_sum=SG_MALLOC(float64_t,num_dim); - memset(cov_sum, 0, num_dim*sizeof(float64_t)); + cov_sum = SGMatrix(1, num_dim); + cov_sum.zero(); } else if(cov_type==SPHERICAL) { - cov_sum=SG_MALLOC(float64_t,1); - cov_sum[0]=0; + cov_sum = SGMatrix(1, 1); + cov_sum.zero(); } for (int32_t j=0; j v=dotdata->get_computed_dot_feature_vector(j); - SGVector::add(v.vector, 1, v.vector, -1, mean.vector, v.vlen); + linalg::add(v, mean, v, -1.0, 1.0); switch (cov_type) { - case FULL: - cblas_dger(CblasRowMajor, num_dim, num_dim, alpha_k[j], v.vector, 1, v.vector, - 1, (double*) cov_sum, num_dim); - - break; - case DIAG: - for (int32_t k=0; k(alpha_k[j], v, v, cov_sum); +#endif + break; + case DIAG: + for (int32_t k = 0; k < num_dim; k++) + cov_sum(1, k) += v.vector[k] * v.vector[k] * alpha_k[j]; - break; - case SPHERICAL: - float64_t temp=0; + break; + case SPHERICAL: + float64_t temp = 0; - for (int32_t k=0; k::compute_eigenvectors(cov_sum, num_dim, num_dim); - - set_d(SGVector(d0, num_dim)); - set_u(SGMatrix(cov_sum, num_dim, num_dim)); - - break; + case FULL: + { + linalg::scale(cov_sum, cov_sum, 1 / alpha_k_sum); - case DIAG: - for (int32_t j=0; j d0(num_dim); +#ifdef HAVE_LAPACK + d0.vector = SGMatrix::compute_eigenvectors( + cov_sum.matrix, num_dim, num_dim); +#else + // FIXME use eigenvectors computeation warpper by micmn + typename SGMatrix::EigenMatrixXtMap eig = cov_sum; + typename SGVector::EigenVectorXtMap eigenvalues_eig = d0; + + Eigen::EigenSolver::EigenMatrixXt> solver( + eig); + eigenvalues_eig = solver.eigenvalues().real(); +#endif + + set_d(d0); + set_u(cov_sum); + + break; + } + case DIAG: + linalg::scale(cov_sum, cov_sum, 1 / alpha_k_sum); - set_d(SGVector(cov_sum,num_dim)); + set_d(cov_sum.get_row_vector(0)); - break; + break; - case SPHERICAL: - cov_sum[0]/=alpha_k_sum*num_dim; + case SPHERICAL: + cov_sum[0] /= alpha_k_sum * num_dim; - set_d(SGVector(cov_sum,1)); + set_d(cov_sum.get_row_vector(0)); - break; + break; } return alpha_k_sum; @@ -223,24 +238,26 @@ float64_t CGaussian::compute_log_PDF(SGVector point) { ASSERT(m_mean.vector && m_d.vector) ASSERT(point.vlen == m_mean.vlen) - float64_t* difference=SG_MALLOC(float64_t, m_mean.vlen); - sg_memcpy(difference, point.vector, sizeof(float64_t)*m_mean.vlen); + SGVector difference = point.clone(); - for (int32_t i = 0; i < m_mean.vlen; i++) - difference[i] -= m_mean.vector[i]; + linalg::add(difference, m_mean, difference, -1.0, 1.0); float64_t answer=m_constant; if (m_cov_type==FULL) { - float64_t* temp_holder=SG_MALLOC(float64_t, m_d.vlen); - cblas_dgemv(CblasRowMajor, CblasNoTrans, m_d.vlen, m_d.vlen, - 1, m_u.matrix, m_d.vlen, difference, 1, 0, temp_holder, 1); + SGVector temp_holder(m_d.vlen); + temp_holder.zero(); +#ifdef HAVE_LAPACK + cblas_dgemv( + CblasRowMajor, CblasNoTrans, m_d.vlen, m_d.vlen, 1, m_u.matrix, + m_d.vlen, difference, 1, 0, temp_holder, 1); +#else + linalg::dgemv(1, m_u, false, difference, 0, temp_holder); +#endif for (int32_t i=0; i point) else { for (int32_t i=0; i CGaussian::get_mean() @@ -287,41 +302,45 @@ void CGaussian::set_d(const SGVector d) SGMatrix CGaussian::get_cov() { - float64_t* cov=SG_MALLOC(float64_t, m_mean.vlen*m_mean.vlen); - memset(cov, 0, sizeof(float64_t)*m_mean.vlen*m_mean.vlen); + SGMatrix cov(m_mean.vlen, m_mean.vlen); + cov.zero(); if (m_cov_type==FULL) { if (!m_u.matrix) SG_ERROR("Unitary matrix not set\n") - float64_t* temp_holder=SG_MALLOC(float64_t, m_d.vlen*m_d.vlen); - float64_t* diag_holder=SG_MALLOC(float64_t, m_d.vlen*m_d.vlen); - memset(diag_holder, 0, sizeof(float64_t)*m_d.vlen*m_d.vlen); - for(int32_t i=0; i temp_holder(m_mean.vlen, m_mean.vlen); + SGMatrix diag_holder(m_mean.vlen, m_mean.vlen); + diag_holder.zero(); + for (int32_t i = 0; i < m_d.vlen; i++) + diag_holder(i, i) = m_d.vector[i]; +#ifdef HAVE_LAPACK + cblas_dgemm( + CblasRowMajor, CblasTrans, CblasNoTrans, m_d.vlen, m_d.vlen, + m_d.vlen, 1, m_u.matrix, m_d.vlen, diag_holder.matrix, m_d.vlen, 0, + temp_holder.matrix, m_d.vlen); + cblas_dgemm( + CblasRowMajor, CblasNoTrans, CblasNoTrans, m_d.vlen, m_d.vlen, + m_d.vlen, 1, temp_holder.matrix, m_d.vlen, m_u.matrix, m_d.vlen, 0, + cov.matrix, m_d.vlen); +#else + linalg::dgemm( + 1, m_u, diag_holder, true, false, 0, temp_holder); + linalg::dgemm(1, temp_holder, m_u, false, false, 0, cov); +#endif } - else if (m_cov_type==DIAG) + else if (m_cov_type == DIAG) { - for (int32_t i=0; i(cov, m_mean.vlen, m_mean.vlen, false);//fix needed + return cov; } void CGaussian::register_params() @@ -337,77 +356,94 @@ void CGaussian::decompose_cov(SGMatrix cov) { switch (m_cov_type) { - case FULL: - m_u=SGMatrix(cov.num_rows,cov.num_rows); - sg_memcpy(m_u.matrix, cov.matrix, sizeof(float64_t)*cov.num_rows*cov.num_rows); - - m_d.vector=SGMatrix::compute_eigenvectors(m_u.matrix, cov.num_rows, cov.num_rows); - m_d.vlen=cov.num_rows; - m_u.num_rows=cov.num_rows; - m_u.num_cols=cov.num_rows; - break; - case DIAG: - m_d=SGVector(cov.num_rows); - for (int32_t i=0; i(1); - m_d.vector[0]=cov.matrix[0]; - break; + case FULL: + { + m_u = SGMatrix(cov.num_rows, cov.num_rows); + m_u = cov.clone(); + m_d = SGVector(cov.num_rows); +#ifdef HAVE_LAPACK + m_d.vector = SGMatrix::compute_eigenvectors( + m_u.matrix, cov.num_rows, cov.num_rows); +#else + // FIXME use eigenvectors computeation warpper by micmn + typename SGMatrix::EigenMatrixXtMap eig = m_u; + typename SGVector::EigenVectorXtMap eigenvalues_eig = m_d; + + Eigen::EigenSolver::EigenMatrixXt> solver( + eig); + eigenvalues_eig = solver.eigenvalues().real(); +#endif + break; + } + case DIAG: + m_d = SGVector(cov.num_rows); + for (int32_t i = 0; i < cov.num_rows; i++) + m_d[i] = cov.matrix[i * cov.num_rows + i]; + + break; + case SPHERICAL: + m_d = SGVector(1); + m_d.vector[0] = cov.matrix[0]; + break; } } SGVector CGaussian::sample() { SG_DEBUG("Entering\n"); - float64_t* r_matrix=SG_MALLOC(float64_t, m_mean.vlen*m_mean.vlen); - memset(r_matrix, 0, m_mean.vlen*m_mean.vlen*sizeof(float64_t)); + SGMatrix r_matrix(m_mean.vlen, m_mean.vlen); + r_matrix.zero(); switch (m_cov_type) { - case FULL: - case DIAG: - for (int32_t i=0; i random_vec(m_mean.vlen); - for (int32_t i=0; i temp_matrix(m_d.vlen, m_d.vlen); + temp_matrix.zero(); +#ifdef HAVE_LAPACK + cblas_dgemm( + CblasRowMajor, CblasNoTrans, CblasNoTrans, m_d.vlen, m_d.vlen, + m_d.vlen, 1, m_u.matrix, m_d.vlen, r_matrix.matrix, m_d.vlen, 0, + temp_matrix.matrix, m_d.vlen); +#else + linalg::dgemm( + 1, m_u, r_matrix, false, false, 0, temp_matrix); +#endif + r_matrix = temp_matrix; } - float64_t* samp=SG_MALLOC(float64_t, m_mean.vlen); - - cblas_dgemv(CblasRowMajor, CblasNoTrans, m_mean.vlen, m_mean.vlen, - 1, r_matrix, m_mean.vlen, random_vec, 1, 0, samp, 1); - - for (int32_t i=0; i samp(m_mean.vlen); - SG_FREE(random_vec); - SG_FREE(r_matrix); +#ifdef HAVE_LAPACK + cblas_dgemv( + CblasRowMajor, CblasNoTrans, m_mean.vlen, m_mean.vlen, 1, + r_matrix.matrix, m_mean.vlen, random_vec.vector, 1, 0, samp.vector, 1); +#else + linalg::dgemv(1.0, r_matrix, false, random_vec, 0.0, samp); +#endif + for (int32_t i = 0; i < m_mean.vlen; i++) + samp.vector[i] += m_mean.vector[i]; SG_DEBUG("Leaving\n"); - return SGVector(samp, m_mean.vlen, false);//fix needed + return samp; } CGaussian* CGaussian::obtain_from_generic(CDistribution* distribution) @@ -423,5 +459,3 @@ CGaussian* CGaussian::obtain_from_generic(CDistribution* distribution) SG_REF(casted); return casted; } - -#endif // HAVE_LAPACK diff --git a/src/shogun/distributions/Gaussian.h b/src/shogun/distributions/Gaussian.h index e5e21839324..202c7f31cff 100644 --- a/src/shogun/distributions/Gaussian.h +++ b/src/shogun/distributions/Gaussian.h @@ -14,8 +14,6 @@ #include -#ifdef HAVE_LAPACK - #include #include #include @@ -245,5 +243,4 @@ class CGaussian : public CDistribution ECovType m_cov_type; }; } -#endif //HAVE_LAPACK #endif //_GAUSSIAN_H__ diff --git a/src/shogun/distributions/HMM.cpp b/src/shogun/distributions/HMM.cpp index 3bc2dd5c556..e5cbec82702 100644 --- a/src/shogun/distributions/HMM.cpp +++ b/src/shogun/distributions/HMM.cpp @@ -5586,7 +5586,9 @@ bool CHMM::baum_welch_viterbi_train(BaumWelchViterbiType type) float64_t prob_train=CMath::ALMOST_NEG_INFTY; iteration_count=iterations; - while (!converged(prob, prob_train) && (!CSignal::cancel_computations())) + // TODO: replace with the new signal + // while (!converged(prob, prob_train) && (!CSignal::cancel_computations())) + while (!converged(prob, prob_train)) { CMath::swap(working, estimate); prob=prob_train; diff --git a/src/shogun/evaluation/CrossValidation.cpp b/src/shogun/evaluation/CrossValidation.cpp index cc77dbadf7c..8c0336cc092 100644 --- a/src/shogun/evaluation/CrossValidation.cpp +++ b/src/shogun/evaluation/CrossValidation.cpp @@ -8,14 +8,15 @@ * Copyright (C) 2011 Berlin Institute of Technology and Max-Planck-Society */ +#include +#include #include -#include +#include #include #include -#include -#include -#include #include +#include +#include using namespace shogun; @@ -24,61 +25,44 @@ CCrossValidation::CCrossValidation() : CMachineEvaluation() init(); } -CCrossValidation::CCrossValidation(CMachine* machine, CFeatures* features, - CLabels* labels, CSplittingStrategy* splitting_strategy, - CEvaluation* evaluation_criterion, bool autolock) : - CMachineEvaluation(machine, features, labels, splitting_strategy, - evaluation_criterion, autolock) +CCrossValidation::CCrossValidation( + CMachine* machine, CFeatures* features, CLabels* labels, + CSplittingStrategy* splitting_strategy, CEvaluation* evaluation_criterion, + bool autolock) + : CMachineEvaluation( + machine, features, labels, splitting_strategy, evaluation_criterion, + autolock) { init(); } -CCrossValidation::CCrossValidation(CMachine* machine, CLabels* labels, - CSplittingStrategy* splitting_strategy, - CEvaluation* evaluation_criterion, bool autolock) : - CMachineEvaluation(machine, labels, splitting_strategy, evaluation_criterion, - autolock) +CCrossValidation::CCrossValidation( + CMachine* machine, CLabels* labels, CSplittingStrategy* splitting_strategy, + CEvaluation* evaluation_criterion, bool autolock) + : CMachineEvaluation( + machine, labels, splitting_strategy, evaluation_criterion, autolock) { init(); } CCrossValidation::~CCrossValidation() { - SG_UNREF(m_xval_outputs); } void CCrossValidation::init() { - m_num_runs=1; + m_num_runs = 1; - /* do reference counting for output objects */ - m_xval_outputs=new CList(true); - - SG_ADD(&m_num_runs, "num_runs", "Number of repetitions", - MS_NOT_AVAILABLE); - SG_ADD((CSGObject**)&m_xval_outputs, "m_xval_outputs", "List of output " - "classes for intermediade cross-validation results", - MS_NOT_AVAILABLE); + SG_ADD(&m_num_runs, "num_runs", "Number of repetitions", MS_NOT_AVAILABLE); } -CEvaluationResult* CCrossValidation::evaluate() +CEvaluationResult* CCrossValidation::evaluate_impl() { - SG_DEBUG("entering %s::evaluate()\n", get_name()) - - REQUIRE(m_machine, "%s::evaluate() is only possible if a machine is " - "attached\n", get_name()); - - REQUIRE(m_features, "%s::evaluate() is only possible if features are " - "attached\n", get_name()); - - REQUIRE(m_labels, "%s::evaluate() is only possible if labels are " - "attached\n", get_name()); - /* if for some reason the do_unlock_frag is set, unlock */ if (m_do_unlock) { m_machine->data_unlock(); - m_do_unlock=false; + m_do_unlock = false; } /* set labels in any case (no locking needs this) */ @@ -93,86 +77,78 @@ CEvaluationResult* CCrossValidation::evaluate() if (!m_machine->is_data_locked()) { m_machine->data_lock(m_labels, m_features); - m_do_unlock=true; + m_do_unlock = true; } } else { - SG_WARNING("%s does not support locking. Autolocking is skipped. " - "Set autolock flag to false to get rid of warning.\n", - m_machine->get_name()); + SG_WARNING( + "%s does not support locking. Autolocking is skipped. " + "Set autolock flag to false to get rid of warning.\n", + m_machine->get_name()); } } SGVector results(m_num_runs); - /* evtl. update xvalidation output class */ - CCrossValidationOutput* current=(CCrossValidationOutput*) - m_xval_outputs->get_first_element(); - while (current) - { - current->init_num_runs(m_num_runs); - current->init_num_folds(m_splitting_strategy->get_num_subsets()); - current->init_expose_labels(m_labels); - current->post_init(); - SG_UNREF(current); - current=(CCrossValidationOutput*) - m_xval_outputs->get_next_element(); - } - /* perform all the x-val runs */ SG_DEBUG("starting %d runs of cross-validation\n", m_num_runs) - for (index_t i=0; i get_first_element(); - while (current) - { - current->update_run_index(i); - SG_UNREF(current); - current=(CCrossValidationOutput*) - m_xval_outputs->get_next_element(); - } + SG_DEBUG("Creating CrossValidationStorage.\n") + CrossValidationStorage* storage = new CrossValidationStorage(); + SG_REF(storage) + storage->set_num_runs(m_num_runs); + storage->set_num_folds(m_splitting_strategy->get_num_subsets()); + storage->set_expose_labels(m_labels); + storage->post_init(); + SG_DEBUG("Ending CrossValidationStorage initilization.\n") SG_DEBUG("entering cross-validation run %d \n", i) - results[i]=evaluate_one_run(); + results[i] = evaluate_one_run(i, storage); SG_DEBUG("result of cross-validation run %d is %f\n", i, results[i]) + + /* Emit the value*/ + std::string obs_value_name{"cross_validation_run"}; + ObservedValue cv_data{i, obs_value_name, erase_type(storage), + CROSSVALIDATION}; + observe(cv_data); + SG_UNREF(storage) } /* construct evaluation result */ CCrossValidationResult* result = new CCrossValidationResult(); - result->mean=CStatistics::mean(results); - if (m_num_runs>1) - result->std_dev=CStatistics::std_deviation(results); + result->set_mean(CStatistics::mean(results)); + if (m_num_runs > 1) + result->set_std_dev(CStatistics::std_deviation(results)); else - result->std_dev=0; + result->set_std_dev(0); /* unlock machine if it was locked in this method */ if (m_machine->is_data_locked() && m_do_unlock) { m_machine->data_unlock(); - m_do_unlock=false; + m_do_unlock = false; } - SG_DEBUG("leaving %s::evaluate()\n", get_name()) - SG_REF(result); return result; } void CCrossValidation::set_num_runs(int32_t num_runs) { - if (num_runs <1) + if (num_runs < 1) SG_ERROR("%d is an illegal number of repetitions\n", num_runs) - m_num_runs=num_runs; + m_num_runs = num_runs; } -float64_t CCrossValidation::evaluate_one_run() +float64_t CCrossValidation::evaluate_one_run( + int64_t index, CrossValidationStorage* storage) { SG_DEBUG("entering %s::evaluate_one_run()\n", get_name()) - index_t num_subsets=m_splitting_strategy->get_num_subsets(); + index_t num_subsets = m_splitting_strategy->get_num_subsets(); SG_DEBUG("building index sets for %d-fold cross-validation\n", num_subsets) @@ -185,45 +161,38 @@ float64_t CCrossValidation::evaluate_one_run() /* different behavior whether data is locked or not */ if (m_machine->is_data_locked()) { + m_machine->set_store_model_features(true); SG_DEBUG("starting locked evaluation\n", get_name()) /* do actual cross-validation */ - for (index_t i=0; i get_first_element(); - while (current) - { - current->update_fold_index(i); - SG_UNREF(current); - current=(CCrossValidationOutput*) - m_xval_outputs->get_next_element(); - } + CrossValidationFoldStorage* fold = new CrossValidationFoldStorage(); + SG_REF(fold) + fold->set_run_index(index); + fold->set_fold_index(i); /* index subset for training, will be freed below */ SGVector inverse_subset_indices = - m_splitting_strategy->generate_subset_inverse(i); + m_splitting_strategy->generate_subset_inverse(i); /* train machine on training features */ m_machine->train_locked(inverse_subset_indices); /* feature subset for testing */ SGVector subset_indices = - m_splitting_strategy->generate_subset_indices(i); + m_splitting_strategy->generate_subset_indices(i); /* evtl. update xvalidation output class */ - current=(CCrossValidationOutput*)m_xval_outputs->get_first_element(); - while (current) - { - current->update_train_indices(inverse_subset_indices, "\t"); - current->update_trained_machine(m_machine, "\t"); - SG_UNREF(current); - current=(CCrossValidationOutput*) - m_xval_outputs->get_next_element(); - } + fold->set_train_indices(inverse_subset_indices); + auto fold_machine = (CMachine*)m_machine->clone(); + fold->set_trained_machine(fold_machine); + SG_UNREF(fold_machine) /* produce output for desired indices */ - CLabels* result_labels=m_machine->apply_locked(subset_indices); + CLabels* result_labels = m_machine->apply_locked(subset_indices); SG_REF(result_labels); /* set subset for testing labels */ @@ -231,27 +200,27 @@ float64_t CCrossValidation::evaluate_one_run() /* evaluate against own labels */ m_evaluation_criterion->set_indices(subset_indices); - results[i]=m_evaluation_criterion->evaluate(result_labels, m_labels); + results[i] = + m_evaluation_criterion->evaluate(result_labels, m_labels); /* evtl. update xvalidation output class */ - current=(CCrossValidationOutput*)m_xval_outputs->get_first_element(); - while (current) - { - current->update_test_indices(subset_indices, "\t"); - current->update_test_result(result_labels, "\t"); - current->update_test_true_result(m_labels, "\t"); - current->post_update_results(); - current->update_evaluation_result(results[i], "\t"); - SG_UNREF(current); - current=(CCrossValidationOutput*) - m_xval_outputs->get_next_element(); - } + fold->set_test_indices(subset_indices); + fold->set_test_result(result_labels); + CLabels* true_labels = (CLabels*)m_labels->clone(); + fold->set_test_true_result(true_labels); + SG_UNREF(true_labels) + fold->post_update_results(); + fold->set_evaluation_result(results[i]); /* remove subset to prevent side effects */ m_labels->remove_subset(); + /* Save fold into storage */ + storage->append_fold_result(fold); + /* clean up */ SG_UNREF(result_labels); + SG_UNREF(fold); SG_DEBUG("done locked evaluation\n", get_name()) } @@ -265,61 +234,57 @@ float64_t CCrossValidation::evaluate_one_run() /* do actual cross-validation */ - //TODO parallel xvalidation needs some serious fixing, see #3743 + // TODO parallel xvalidation needs some serious fixing, see #3743 //#pragma omp parallel for - for (index_t i=0; i get_num_threads()==1) + if (get_global_parallel()->get_num_threads() == 1) { - machine=m_machine; - features=m_features; - evaluation_criterion=m_evaluation_criterion; + machine = m_machine; + features = m_features; + evaluation_criterion = m_evaluation_criterion; } else { - machine=(CMachine*)m_machine->clone(); - features=(CFeatures*)m_features->clone(); - evaluation_criterion=(CEvaluation*)m_evaluation_criterion->clone(); + machine = (CMachine*)m_machine->clone(); + features = (CFeatures*)m_features->clone(); + evaluation_criterion = + (CEvaluation*)m_evaluation_criterion->clone(); } /* evtl. update xvalidation output class */ - CCrossValidationOutput* current; - #pragma omp critical - { - current=(CCrossValidationOutput*) - m_xval_outputs->get_first_element(); - while (current) - { - current->update_fold_index(i); - SG_UNREF(current); - current=(CCrossValidationOutput*) - m_xval_outputs->get_next_element(); - } - } + fold->set_run_index(index); + fold->set_fold_index(i); /* set feature subset for training */ - SGVector inverse_subset_indices= - m_splitting_strategy->generate_subset_inverse(i); + SGVector inverse_subset_indices = + m_splitting_strategy->generate_subset_inverse(i); features->add_subset(inverse_subset_indices); /* set label subset for training */ - if (get_global_parallel()->get_num_threads()==1) - labels=m_labels; + if (get_global_parallel()->get_num_threads() == 1) + labels = m_labels; else - labels=machine->get_labels(); + labels = machine->get_labels(); labels->add_subset(inverse_subset_indices); SG_DEBUG("training set %d:\n", i) - if (io->get_loglevel()==MSG_DEBUG) + if (io->get_loglevel() == MSG_DEBUG) { - SGVector::display_vector(inverse_subset_indices.vector, - inverse_subset_indices.vlen, "training indices"); + SGVector::display_vector( + inverse_subset_indices.vector, inverse_subset_indices.vlen, + "training indices"); } /* train machine on training features and remove subset */ @@ -328,69 +293,56 @@ float64_t CCrossValidation::evaluate_one_run() SG_DEBUG("finished training\n") /* evtl. update xvalidation output class */ - #pragma omp critical - { - current=(CCrossValidationOutput*)m_xval_outputs->get_first_element(); - while (current) - { - current->update_train_indices(inverse_subset_indices, "\t"); - current->update_trained_machine(machine, "\t"); - SG_UNREF(current); - current=(CCrossValidationOutput*) - m_xval_outputs->get_next_element(); - } - } + fold->set_train_indices(inverse_subset_indices); + auto fold_machine = (CMachine*)machine->clone(); + fold->set_trained_machine(fold_machine); + SG_UNREF(fold_machine) features->remove_subset(); labels->remove_subset(); - /* set feature subset for testing (subset method that stores pointer) */ + /* set feature subset for testing (subset method that stores + * pointer) */ SGVector subset_indices = - m_splitting_strategy->generate_subset_indices(i); + m_splitting_strategy->generate_subset_indices(i); features->add_subset(subset_indices); /* set label subset for testing */ labels->add_subset(subset_indices); SG_DEBUG("test set %d:\n", i) - if (io->get_loglevel()==MSG_DEBUG) + if (io->get_loglevel() == MSG_DEBUG) { - SGVector::display_vector(subset_indices.vector, - subset_indices.vlen, "test indices"); + SGVector::display_vector( + subset_indices.vector, subset_indices.vlen, "test indices"); } /* apply machine to test features and remove subset */ SG_DEBUG("starting evaluation\n") SG_DEBUG("%p\n", features) - CLabels* result_labels=machine->apply(features); + CLabels* result_labels = machine->apply(features); SG_DEBUG("finished evaluation\n") features->remove_subset(); SG_REF(result_labels); /* evaluate */ - results[i]=evaluation_criterion->evaluate(result_labels, labels); + results[i] = evaluation_criterion->evaluate(result_labels, labels); SG_DEBUG("result on fold %d is %f\n", i, results[i]) /* evtl. update xvalidation output class */ - #pragma omp critical - { - current=(CCrossValidationOutput*)m_xval_outputs->get_first_element(); - while (current) - { - current->update_test_indices(subset_indices, "\t"); - current->update_test_result(result_labels, "\t"); - current->update_test_true_result(labels, "\t"); - current->post_update_results(); - current->update_evaluation_result(results[i], "\t"); - SG_UNREF(current); - current=(CCrossValidationOutput*) - m_xval_outputs->get_next_element(); - } - } + fold->set_test_indices(subset_indices); + fold->set_test_result(result_labels); + CLabels* true_labels = (CLabels*)labels->clone(); + fold->set_test_true_result(true_labels); + SG_UNREF(true_labels) + fold->post_update_results(); + fold->set_evaluation_result(results[i]); + + storage->append_fold_result(fold); /* clean up, remove subsets */ labels->remove_subset(); - if (get_global_parallel()->get_num_threads()!=1) + if (get_global_parallel()->get_num_threads() != 1) { SG_UNREF(machine); SG_UNREF(features); @@ -398,20 +350,15 @@ float64_t CCrossValidation::evaluate_one_run() SG_UNREF(evaluation_criterion); } SG_UNREF(result_labels); + SG_UNREF(fold) } SG_DEBUG("done unlocked evaluation\n", get_name()) } /* build arithmetic mean of results */ - float64_t mean=CStatistics::mean(results); + float64_t mean = CStatistics::mean(results); SG_DEBUG("leaving %s::evaluate_one_run()\n", get_name()) return mean; } - -void CCrossValidation::add_cross_validation_output( - CCrossValidationOutput* cross_validation_output) -{ - m_xval_outputs->append_element(cross_validation_output); -} diff --git a/src/shogun/evaluation/CrossValidation.h b/src/shogun/evaluation/CrossValidation.h index 076c9590d5f..b0b07ef7323 100644 --- a/src/shogun/evaluation/CrossValidation.h +++ b/src/shogun/evaluation/CrossValidation.h @@ -19,20 +19,23 @@ namespace shogun { -class CMachineEvaluation; -class CCrossValidationOutput; -class CList; + class CMachineEvaluation; + class CCrossValidationOutput; + class CrossValidationStorage; + class CList; -/** @brief type to encapsulate the results of an evaluation run. - */ -class CCrossValidationResult : public CEvaluationResult -{ + /** @brief type to encapsulate the results of an evaluation run. + */ + class CCrossValidationResult : public CEvaluationResult + { public: CCrossValidationResult() { SG_ADD(&mean, "mean", "Mean of results", MS_NOT_AVAILABLE); - SG_ADD(&std_dev, "std_dev", - "Standard deviation of cross-validation folds", MS_NOT_AVAILABLE); + SG_ADD( + &std_dev, "std_dev", + "Standard deviation of cross-validation folds", + MS_NOT_AVAILABLE); mean = 0; std_dev = 0; @@ -52,24 +55,28 @@ class CCrossValidationResult : public CEvaluationResult * * @return name of the SGSerializable */ - virtual const char* get_name() const { return "CrossValidationResult"; } + virtual const char* get_name() const + { + return "CrossValidationResult"; + } /** helper method used to specialize a base class instance * * @param eval_result its dynamic type must be CCrossValidationResult */ - static CCrossValidationResult* obtain_from_generic( - CEvaluationResult* eval_result) + static CCrossValidationResult* + obtain_from_generic(CEvaluationResult* eval_result) { if (!eval_result) return NULL; - REQUIRE(eval_result->get_result_type()==CROSSVALIDATION_RESULT, - "CrossValidationResult::obtain_from_generic(): argument is" - "of wrong type!\n"); + REQUIRE( + eval_result->get_result_type() == CROSSVALIDATION_RESULT, + "CrossValidationResult::obtain_from_generic(): argument is" + "of wrong type!\n"); SG_REF(eval_result); - return (CCrossValidationResult*) eval_result; + return (CCrossValidationResult*)eval_result; } /** print result */ @@ -78,115 +85,155 @@ class CCrossValidationResult : public CEvaluationResult SG_SPRINT("%f+-%f\n", mean, std_dev); } - public: + /** + * Get the evaluations mean. + * @return mean + */ + float64_t get_mean() const + { + return mean; + } + + /** + * Get the standard deviation. + * @return standard deviation + */ + float64_t get_std_dev() const + { + return std_dev; + } + + /** + * Set the evaluations mean. + * @param mean the mean + */ + void set_mean(float64_t ev_mean) + { + this->mean = ev_mean; + } + + /** + * Set the standard deviation + * @param std_dev the standard deviation + */ + void set_std_dev(float64_t ev_std_dev) + { + this->std_dev = ev_std_dev; + } + + private: /** mean */ float64_t mean; /** Standard deviation of cross-validation folds */ float64_t std_dev; -}; - -/** @brief base class for cross-validation evaluation. - * Given a learning machine, a splitting strategy, an evaluation criterion, - * features and corresponding labels, this provides an interface for - * cross-validation. Results may be retrieved using the evaluate method. A - * number of repetitions may be specified for obtaining more accurate results. - * The arithmetic mean and standard deviation of different runs is returned. - * Default number of runs is one. - * - * This class calculates an evaluation criterion of every fold and then - * calculates the arithmetic mean of all folds. This is for example suitable - * for the AUC or for Accuracy. However, for example F1-measure may not be - * merged this way (result will be biased). To solve this, different sub-classes - * may average results of each cross validation fold differently by overwriting - * the evaluate_one_run method. - * - * See [Forman, G. and Scholz, M. (2009). Apples-to-apples in cross-validation - * studies: Pitfalls in classifier performance measurement. Technical report, - * HP Laboratories.] for details on this subject. - * - * Cross validation tries to lock underlying machines if that is possible to - * speed up computations. Can be turned off by the set_autolock() method. - * Locking in general may speed up things (eg for kernel machines the kernel - * matrix is precomputed), however, it is not always supported. - * - * Crossvalidation runs with current number of threads - * (Parallel::set_num_threads) for unlocked case, and currently duplicates all - * objects (might be changed later). - * - */ -class CCrossValidation: public CMachineEvaluation -{ -public: - /** constructor */ - CCrossValidation(); - - /** constructor - * @param machine learning machine to use - * @param features features to use for cross-validation - * @param labels labels that correspond to the features - * @param splitting_strategy splitting strategy to use - * @param evaluation_criterion evaluation criterion to use - * @param autolock whether machine should be auto-locked before evaluation - */ - CCrossValidation(CMachine* machine, CFeatures* features, CLabels* labels, - CSplittingStrategy* splitting_strategy, - CEvaluation* evaluation_criterion, bool autolock=true); - - /** constructor, for use with custom kernels (no features) - * @param machine learning machine to use - * @param labels labels that correspond to the features - * @param splitting_strategy splitting strategy to use - * @param evaluation_criterion evaluation criterion to use - * @param autolock autolock + }; + + /** @brief base class for cross-validation evaluation. + * Given a learning machine, a splitting strategy, an evaluation criterion, + * features and corresponding labels, this provides an interface for + * cross-validation. Results may be retrieved using the evaluate method. A + * number of repetitions may be specified for obtaining more accurate + * results. + * The arithmetic mean and standard deviation of different runs is returned. + * Default number of runs is one. + * + * This class calculates an evaluation criterion of every fold and then + * calculates the arithmetic mean of all folds. This is for example suitable + * for the AUC or for Accuracy. However, for example F1-measure may not be + * merged this way (result will be biased). To solve this, different + * sub-classes + * may average results of each cross validation fold differently by + * overwriting + * the evaluate_one_run method. + * + * See [Forman, G. and Scholz, M. (2009). Apples-to-apples in + * cross-validation + * studies: Pitfalls in classifier performance measurement. Technical + * report, + * HP Laboratories.] for details on this subject. + * + * Cross validation tries to lock underlying machines if that is possible to + * speed up computations. Can be turned off by the set_autolock() method. + * Locking in general may speed up things (eg for kernel machines the kernel + * matrix is precomputed), however, it is not always supported. + * + * Crossvalidation runs with current number of threads + * (Parallel::set_num_threads) for unlocked case, and currently duplicates + * all + * objects (might be changed later). + * */ - CCrossValidation(CMachine* machine, CLabels* labels, - CSplittingStrategy* splitting_strategy, - CEvaluation* evaluation_criterion, bool autolock=true); - - /** destructor */ - virtual ~CCrossValidation(); - - /** setter for the number of runs to use for evaluation */ - void set_num_runs(int32_t num_runs); + class CCrossValidation : public CMachineEvaluation + { + public: + /** constructor */ + CCrossValidation(); + + /** constructor + * @param machine learning machine to use + * @param features features to use for cross-validation + * @param labels labels that correspond to the features + * @param splitting_strategy splitting strategy to use + * @param evaluation_criterion evaluation criterion to use + * @param autolock whether machine should be auto-locked before + * evaluation + */ + CCrossValidation( + CMachine* machine, CFeatures* features, CLabels* labels, + CSplittingStrategy* splitting_strategy, + CEvaluation* evaluation_criterion, bool autolock = true); + + /** constructor, for use with custom kernels (no features) + * @param machine learning machine to use + * @param labels labels that correspond to the features + * @param splitting_strategy splitting strategy to use + * @param evaluation_criterion evaluation criterion to use + * @param autolock autolock + */ + CCrossValidation( + CMachine* machine, CLabels* labels, + CSplittingStrategy* splitting_strategy, + CEvaluation* evaluation_criterion, bool autolock = true); - /** evaluate */ - virtual CEvaluationResult* evaluate(); + /** destructor */ + virtual ~CCrossValidation(); - /** appends given cross validation output instance - * to the list of listeners - * - * @param cross_validation_output given cross validation output - */ - void add_cross_validation_output( - CCrossValidationOutput* cross_validation_output); + /** setter for the number of runs to use for evaluation */ + void set_num_runs(int32_t num_runs); - /** @return name of the SGSerializable */ - virtual const char* get_name() const - { - return "CrossValidation"; - } - -private: - void init(); - -protected: - /** Evaluates one single cross-validation run. - * Current implementation evaluates each fold separately and then calculates - * arithmetic mean. Suitable for accuracy and AUC for example. NOT for - * F1-measure. Has to be overridden by sub-classes if results have to be - * merged differently - * - * @return evaluation result of one cross-validation run - */ - virtual float64_t evaluate_one_run(); + /** @return name of the SGSerializable */ + virtual const char* get_name() const + { + return "CrossValidation"; + } - /** number of evaluation runs for one fold */ - int32_t m_num_runs; + private: + void init(); - /** xval output listeners */ - CList* m_xval_outputs; -}; + protected: + /** + * Does the actual evaluation. + * @return the cross-validation result + */ + virtual CEvaluationResult* evaluate_impl(); + + protected: + protected: + /** Evaluates one single cross-validation run. + * Current implementation evaluates each fold separately and then + * calculates + * arithmetic mean. Suitable for accuracy and AUC for example. NOT for + * F1-measure. Has to be overridden by sub-classes if results have to be + * merged differently + * + * @return evaluation result of one cross-validation run + */ + virtual float64_t + evaluate_one_run(int64_t index, CrossValidationStorage* storage); + /** number of evaluation runs for one fold */ + int32_t m_num_runs; + }; } #endif /* __CROSSVALIDATION_H_ */ diff --git a/src/shogun/evaluation/CrossValidationMKLStorage.cpp b/src/shogun/evaluation/CrossValidationMKLStorage.cpp deleted file mode 100644 index 17ac9a4400f..00000000000 --- a/src/shogun/evaluation/CrossValidationMKLStorage.cpp +++ /dev/null @@ -1,83 +0,0 @@ -/* - * This program is free software; you can redistribute it and/or modify - * it under the terms of the GNU General Public License as published by - * the Free Software Foundation; either version 3 of the License, or - * (at your option) any later version. - * - * Written (W) 2012 Sergey Lisitsyn - * Written (W) 2012 Heiko Strathmann - */ - -#include -#include -#include -#include - -using namespace shogun; - -void CCrossValidationMKLStorage::update_trained_machine( - CMachine* machine, const char* prefix) -{ - REQUIRE(machine, "%s::update_trained_machine(): Provided Machine is NULL!\n", - get_name()); - - CMKL* mkl=dynamic_cast(machine); - CMKLMulticlass* mkl_multiclass=dynamic_cast(machine); - REQUIRE(mkl || mkl_multiclass, "%s::update_trained_machine(): This method is only usable " - "with CMKL derived machines. This one is \"%s\"\n", get_name(), - machine->get_name()); - - CKernel* kernel = NULL; - if (mkl) - kernel = mkl->get_kernel(); - else - kernel = mkl_multiclass->get_kernel(); - - REQUIRE(kernel, "%s::update_trained_machine(): No kernel assigned to " - "machine of type \"%s\"\n", get_name(), machine->get_name()); - - CCombinedKernel* combined_kernel=dynamic_cast(kernel); - REQUIRE(combined_kernel, "%s::update_trained_machine(): This method is only" - " usable with CCombinedKernel on machines. This one is \"s\"\n", - get_name(), kernel->get_name()); - - SGVector w=combined_kernel->get_subkernel_weights(); - - /* evtl re-allocate memory (different number of runs from evaluation before) */ - if (m_mkl_weights.num_rows!=w.vlen || - m_mkl_weights.num_cols!=m_num_folds*m_num_runs) - { - if (m_mkl_weights.matrix) - { - SG_DEBUG("deleting memory for mkl weight matrix\n") - m_mkl_weights=SGMatrix(); - } - } - - /* evtl allocate memory (first call) */ - if (!m_mkl_weights.matrix) - { - SG_DEBUG("allocating memory for mkl weight matrix\n") - m_mkl_weights=SGMatrix(w.vlen,m_num_folds*m_num_runs); - } - - /* put current mkl weights into matrix, copy memory vector wise to make - * things fast. Compute index of address to where vector goes */ - - /* number of runs is w.vlen*m_num_folds shift */ - index_t run_shift=m_current_run_index*w.vlen*m_num_folds; - - /* fold shift is m_current_fold_index*w-vlen */ - index_t fold_shift=m_current_fold_index*w.vlen; - - /* add both index shifts */ - index_t first_idx=run_shift+fold_shift; - SG_DEBUG("run %d, fold %d, matrix index %d\n",m_current_run_index, - m_current_fold_index, first_idx); - - /* copy memory */ - sg_memcpy(&m_mkl_weights.matrix[first_idx], w.vector, - w.vlen*sizeof(float64_t)); - - SG_UNREF(kernel); -} diff --git a/src/shogun/evaluation/CrossValidationMKLStorage.h b/src/shogun/evaluation/CrossValidationMKLStorage.h deleted file mode 100644 index 8ad062650c9..00000000000 --- a/src/shogun/evaluation/CrossValidationMKLStorage.h +++ /dev/null @@ -1,61 +0,0 @@ -/* - * This program is free software; you can redistribute it and/or modify - * it under the terms of the GNU General Public License as published by - * the Free Software Foundation; either version 3 of the License, or - * (at your option) any later version. - * - * Written (W) 2012 Heiko Strathmann - * - */ - -#ifndef __CROSSVALIDATIONMKLSTORAGE_H_ -#define __CROSSVALIDATIONMKLSTORAGE_H_ - -#include - -#include -#include - -namespace shogun -{ - -class CMachine; -class CLabels; -class CEvaluation; - -/** @brief Class for storing MKL weights in every fold of cross-validation */ -class CCrossValidationMKLStorage: public CCrossValidationOutput -{ -public: - - /** constructor */ - CCrossValidationMKLStorage() : CCrossValidationOutput() {} - - /** destructor */ - virtual ~CCrossValidationMKLStorage() {}; - - /** @return name of SG_SERIALIZABLE */ - virtual const char* get_name() const { return "CrossValidationMKLStorage"; } - - /** update trained machine. Here, stores MKL weights in local matrix - * - * @param machine trained machine instance - * @param prefix prefix for output - */ - virtual void update_trained_machine(CMachine* machine, - const char* prefix=""); - - /** @return mkl weights matrix, one set of weights per column, - * num_folds*num_runs columns, one fold after another */ - virtual SGMatrix get_mkl_weights() { return m_mkl_weights; } - -protected: - /** storage for MKL weights, one set per column - * num_kernel times num_folds*num_runs matrix where all folds of a runs - * are added one after another */ - SGMatrix m_mkl_weights; -}; - -} - -#endif /* __CROSSVALIDATIONMKLSTORAGE_H_ */ diff --git a/src/shogun/evaluation/CrossValidationMulticlassStorage.cpp b/src/shogun/evaluation/CrossValidationMulticlassStorage.cpp deleted file mode 100644 index 38716bd74f3..00000000000 --- a/src/shogun/evaluation/CrossValidationMulticlassStorage.cpp +++ /dev/null @@ -1,153 +0,0 @@ -/* - * This program is free software; you can redistribute it and/or modify - * it under the terms of the GNU General Public License as published by - * the Free Software Foundation; either version 3 of the License, or - * (at your option) any later version. - * - * Copyright (C) 2012 Sergey Lisitsyn, Heiko Strathmann - */ - -#include -#include -#include -#include - -using namespace shogun; - -CCrossValidationMulticlassStorage::CCrossValidationMulticlassStorage(bool compute_ROC, bool compute_PRC, bool compute_conf_matrices) : - CCrossValidationOutput() -{ - m_initialized = false; - m_compute_ROC = compute_ROC; - m_compute_PRC = compute_PRC; - m_compute_conf_matrices = compute_conf_matrices; - m_pred_labels = NULL; - m_true_labels = NULL; - m_num_classes = 0; - m_binary_evaluations = new CDynamicObjectArray(); - - m_fold_ROC_graphs=NULL; - m_conf_matrices=NULL; -} - - -CCrossValidationMulticlassStorage::~CCrossValidationMulticlassStorage() -{ - if (m_compute_ROC && m_fold_ROC_graphs) - { - SG_FREE(m_fold_ROC_graphs); - } - - if (m_compute_PRC && m_fold_PRC_graphs) - { - SG_FREE(m_fold_PRC_graphs); - } - - if (m_compute_conf_matrices && m_conf_matrices) - { - SG_FREE(m_conf_matrices); - } - - if (m_binary_evaluations) - { - SG_UNREF(m_binary_evaluations); - } -}; - - -void CCrossValidationMulticlassStorage::post_init() -{ - if (m_initialized) - SG_ERROR("CrossValidationMulticlassStorage was already initialized once\n") - - if (m_compute_ROC) - { - SG_DEBUG("Allocating %d ROC graphs\n", m_num_folds*m_num_runs*m_num_classes) - m_fold_ROC_graphs = SG_MALLOC(SGMatrix, m_num_folds*m_num_runs*m_num_classes); - for (int32_t i=0; i(); - } - - if (m_compute_PRC) - { - SG_DEBUG("Allocating %d PRC graphs\n", m_num_folds*m_num_runs*m_num_classes) - m_fold_PRC_graphs = SG_MALLOC(SGMatrix, m_num_folds*m_num_runs*m_num_classes); - for (int32_t i=0; i(); - } - - if (m_binary_evaluations->get_num_elements()) - m_evaluations_results = SGVector(m_num_folds*m_num_runs*m_num_classes*m_binary_evaluations->get_num_elements()); - - m_accuracies = SGVector(m_num_folds*m_num_runs); - - if (m_compute_conf_matrices) - { - m_conf_matrices = SG_MALLOC(SGMatrix, m_num_folds*m_num_runs); - for (int32_t i=0; i(); - } - - m_initialized = true; -} - -void CCrossValidationMulticlassStorage::init_expose_labels(CLabels* labels) -{ - ASSERT((CMulticlassLabels*)labels) - m_num_classes = ((CMulticlassLabels*)labels)->get_num_classes(); -} - -void CCrossValidationMulticlassStorage::post_update_results() -{ - CROCEvaluation eval_ROC; - CPRCEvaluation eval_PRC; - int32_t n_evals = m_binary_evaluations->get_num_elements(); - for (int32_t c=0; cget_binary_for_class(c); - CBinaryLabels* true_labels_binary = m_true_labels->get_binary_for_class(c); - if (m_compute_ROC) - { - eval_ROC.evaluate(pred_labels_binary, true_labels_binary); - m_fold_ROC_graphs[m_current_run_index*m_num_folds*m_num_classes+m_current_fold_index*m_num_classes+c] = - eval_ROC.get_ROC(); - } - if (m_compute_PRC) - { - eval_PRC.evaluate(pred_labels_binary, true_labels_binary); - m_fold_PRC_graphs[m_current_run_index*m_num_folds*m_num_classes+m_current_fold_index*m_num_classes+c] = - eval_PRC.get_PRC(); - } - - for (int32_t i=0; iget_element_safe(i); - m_evaluations_results[m_current_run_index*m_num_folds*m_num_classes*n_evals+m_current_fold_index*m_num_classes*n_evals+c*n_evals+i] = - evaluator->evaluate(pred_labels_binary, true_labels_binary); - SG_UNREF(evaluator); - } - - SG_UNREF(pred_labels_binary); - SG_UNREF(true_labels_binary); - } - CMulticlassAccuracy accuracy; - - m_accuracies[m_current_run_index*m_num_folds+m_current_fold_index] = accuracy.evaluate(m_pred_labels, m_true_labels); - - if (m_compute_conf_matrices) - { - m_conf_matrices[m_current_run_index*m_num_folds+m_current_fold_index] = CMulticlassAccuracy::get_confusion_matrix(m_pred_labels, m_true_labels); - } -} - -void CCrossValidationMulticlassStorage::update_test_result(CLabels* results, const char* prefix) -{ - m_pred_labels = (CMulticlassLabels*)results; -} - -void CCrossValidationMulticlassStorage::update_test_true_result(CLabels* results, const char* prefix) -{ - m_true_labels = (CMulticlassLabels*)results; -} - diff --git a/src/shogun/evaluation/CrossValidationMulticlassStorage.h b/src/shogun/evaluation/CrossValidationMulticlassStorage.h deleted file mode 100644 index 04b09f403dc..00000000000 --- a/src/shogun/evaluation/CrossValidationMulticlassStorage.h +++ /dev/null @@ -1,226 +0,0 @@ -/* - * This program is free software; you can redistribute it and/or modify - * it under the terms of the GNU General Public License as published by - * the Free Software Foundation; either version 3 of the License, or - * (at your option) any later version. - * - * Written (W) 2012 Heiko Strathmann, Sergey Lisitsyn - * - */ - -#ifndef CROSSVALIDATIONMULTICLASSSTORAGE_H_ -#define CROSSVALIDATIONMULTICLASSSTORAGE_H_ - -#include - -#include -#include -#include -#include -#include - -namespace shogun -{ - -class CMachine; -class CLabels; -class CEvaluation; - -/** @brief Class for storing multiclass evaluation information in every fold of cross-validation. - * - * Be careful - can be very expensive memory-wise. - */ -class CCrossValidationMulticlassStorage: public CCrossValidationOutput -{ -public: - - /** constructor - * @param compute_ROC whether to compute ROCs - * @param compute_PRC whether to compute PRCs - * @param compute_conf_matrices whether to compute confusion matrices - */ - CCrossValidationMulticlassStorage(bool compute_ROC=true, bool compute_PRC=false, bool compute_conf_matrices=false); - - /** destructor */ - virtual ~CCrossValidationMulticlassStorage(); - - /** returns ROC of 1-v-R in given fold and run - * - * @param run run - * @param fold fold - * @param c class - * @return ROC of 'run' run, 'fold' fold and 'c' class - */ - SGMatrix get_fold_ROC(int32_t run, int32_t fold, int32_t c) - { - ASSERT(0<=run) - ASSERT(run get_fold_PRC(int32_t run, int32_t fold, int32_t c) - { - ASSERT(0<=run) - ASSERT(runpush_back(evaluation); - } - - /** returns binary evalution appended before - * - * @param idx - */ - CBinaryClassEvaluation* get_binary_evaluation(int32_t idx) - { - return (CBinaryClassEvaluation*)m_binary_evaluations->get_element_safe(idx); - } - - /** returns evaluation result of 1-v-R in given fold and run - * - * @param run run - * @param fold fold - * @param c class - * @param e evaluation number - */ - float64_t get_fold_evaluation_result(int32_t run, int32_t fold, int32_t c, int32_t e) - { - ASSERT(0<=run) - ASSERT(runget_num_elements(); - ASSERT(e get_fold_conf_matrix(int32_t run, int32_t fold) - { - ASSERT(0<=run) - ASSERT(run m_evaluations_results; - - /** accuracies */ - SGVector m_accuracies; - - /** whether compute ROCs */ - bool m_compute_ROC; - - /** fold ROC graphs */ - SGMatrix* m_fold_ROC_graphs; - - /** whether compute PRCs */ - bool m_compute_PRC; - - /** fold PRC graphs */ - SGMatrix* m_fold_PRC_graphs; - - /** whether compute confusion matrices */ - bool m_compute_conf_matrices; - - /** confusion matrices */ - SGMatrix* m_conf_matrices; - - /** predicted results */ - CMulticlassLabels* m_pred_labels; - - /** true labels */ - CMulticlassLabels* m_true_labels; - - /** number of classes */ - int32_t m_num_classes; - -}; - -} - -#endif /* CROSSVALIDATIONMULTICLASSSTORAGE_H_ */ diff --git a/src/shogun/evaluation/CrossValidationOutput.h b/src/shogun/evaluation/CrossValidationOutput.h deleted file mode 100644 index d14b107a8f5..00000000000 --- a/src/shogun/evaluation/CrossValidationOutput.h +++ /dev/null @@ -1,179 +0,0 @@ -/* - * This program is free software; you can redistribute it and/or modify - * it under the terms of the GNU General Public License as published by - * the Free Software Foundation; either version 3 of the License, or - * (at your option) any later version. - * - * Written (W) 2012 Sergey Lisitsyn - * Written (W) 2012 Heiko Strathmann - * - */ - -#ifndef __CROSSVALIDATIONOUTPUT_H_ -#define __CROSSVALIDATIONOUTPUT_H_ - -#include - -#include -#include - -namespace shogun -{ - -class CMachine; -class CLabels; -class CEvaluation; - -/** @brief Class for managing individual folds in cross-validation. - * - * It is often desired to print/save informations that occur during individual - * folds in cross-validation, such as indices, parameters of underlying - * machine etc. This abstract base class might be called from the - * CCrossValidation class after each fold in order to collect these things. - * Different implementations then could output the informations, or even store - * them to make them accessible later. Since is different for every underlying - * machine, individual sub-classes have to handle this separately. - * When writing new subclasses, try to make the design as inheritance based - * as possible, such that future sub-sub-classes can use yours. - * Note that subclasses only need to implement methods that they need since all - * methods in this class are already implemented empty, except for num/fold - * stuff. These number of runs/folds and their current indices are stored since - * these might be used from many subclasses. - */ -class CCrossValidationOutput: public CSGObject -{ -public: - - /** constructor */ - CCrossValidationOutput() : CSGObject() - { - m_current_run_index=0; - m_current_fold_index=0; - m_num_runs=0; - m_num_folds=0; - } - - /** destructor */ - virtual ~CCrossValidationOutput() {} - - /** @return name of SG_SERIALIZABLE */ - virtual const char* get_name() const=0; - - /** init number of runs (called once). saves them to local variable - * - * @param num_runs number of runs that will be performed - * @param prefix prefix for output - */ - virtual void init_num_runs(index_t num_runs, const char* prefix="") - { - m_num_runs=num_runs; - } - - /** init number of folds (called once). saves them to local variable - * @param num_folds number of folds that will be performed - * @param prefix prefix for output - */ - virtual void init_num_folds(index_t num_folds, const char* prefix="") - { - m_num_folds=num_folds; - } - - /** initially expose labels before usage - * @param labels labels to expose to CV output - */ - virtual void init_expose_labels(CLabels* labels) { } - - /** post init action (called once) */ - virtual void post_init() { } - - /** update run index (called every iteration). saves to local variable - * - * @param run_index index of current run - * @param prefix prefix for output - */ - virtual void update_run_index(index_t run_index, - const char* prefix="") - { - m_current_run_index=run_index; - } - - /** update fold index (called every iteration). saves to local variable - * - * @param fold_index index of current run - * @param prefix prefix for output - */ - virtual void update_fold_index(index_t fold_index, - const char* prefix="") - { - m_current_fold_index=fold_index; - } - - /** update train indices - * - * @param indices indices used for training - * @param prefix prefix for output - */ - virtual void update_train_indices(SGVector indices, - const char* prefix="") {} - - /** update test indices - * - * @param indices indices used for testing/validation - * @param prefix prefix for output - */ - virtual void update_test_indices(SGVector indices, - const char* prefix="") {} - - /** update trained machine - * - * @param machine trained machine instance - * @param prefix prefix for output - */ - virtual void update_trained_machine(CMachine* machine, - const char* prefix="") {} - - /** update test result - * - * @param results result labels for test/validation run - * @param prefix prefix for output - */ - virtual void update_test_result(CLabels* results, - const char* prefix="") {} - - /** update test true result - * - * @param results ground truth labels for test/validation run - * @param prefix prefix for output - */ - virtual void update_test_true_result(CLabels* results, - const char* prefix="") {} - - /** post update test and true results - */ - virtual void post_update_results() {} - - /** update evaluate result - * - * @param result evaluation result - * @param prefix prefix for output - */ - virtual void update_evaluation_result(float64_t result, - const char* prefix="") {} - -protected: - /** current run index is written here */ - index_t m_current_run_index; - - /** current fold index is written here */ - index_t m_current_fold_index; - - /** number of runs is initialised here */ - index_t m_num_runs; - - /** number of folds is initialised here */ - index_t m_num_folds; -}; - -} - -#endif /* __CROSSVALIDATIONOUTPUT_H_ */ diff --git a/src/shogun/evaluation/CrossValidationPrintOutput.cpp b/src/shogun/evaluation/CrossValidationPrintOutput.cpp deleted file mode 100644 index 1002ced9e3f..00000000000 --- a/src/shogun/evaluation/CrossValidationPrintOutput.cpp +++ /dev/null @@ -1,144 +0,0 @@ -/* - * This program is free software; you can redistribute it and/or modify - * it under the terms of the GNU General Public License as published by - * the Free Software Foundation; either version 3 of the License, or - * (at your option) any later version. - * - * Written (W) 2012 Sergey Lisitsyn - * Written (W) 2012 Heiko Strathmann - */ - -#include -#include -#include -#include -#include -#include -#include -#include - -using namespace shogun; - -void CCrossValidationPrintOutput::init_num_runs(index_t num_runs, - const char* prefix) -{ - SG_PRINT("%scross validation number of runs %d\n", prefix, num_runs) -} - -/** init number of folds */ -void CCrossValidationPrintOutput::init_num_folds(index_t num_folds, - const char* prefix) -{ - SG_PRINT("%scross validation number of folds %d\n", prefix, num_folds) -} - -void CCrossValidationPrintOutput::update_run_index(index_t run_index, - const char* prefix) -{ - SG_PRINT("%scross validation run %d\n", prefix, run_index) -} - -void CCrossValidationPrintOutput::update_fold_index(index_t fold_index, - const char* prefix) -{ - SG_PRINT("%sfold %d\n", prefix, fold_index) -} - -void CCrossValidationPrintOutput::update_train_indices( - SGVector indices, const char* prefix) -{ - indices.display_vector("train_indices", prefix); -} - -void CCrossValidationPrintOutput::update_test_indices( - SGVector indices, const char* prefix) -{ - indices.display_vector("test_indices", prefix); -} - -void CCrossValidationPrintOutput::update_trained_machine( - CMachine* machine, const char* prefix) -{ - if (dynamic_cast(machine)) - { - CLinearMachine* linear_machine=(CLinearMachine*)machine; - linear_machine->get_w().display_vector("learned_w", prefix); - SG_PRINT("%slearned_bias=%f\n", prefix, linear_machine->get_bias()) - } - - if (dynamic_cast(machine)) - { - CKernelMachine* kernel_machine=(CKernelMachine*)machine; - kernel_machine->get_alphas().display_vector("learned_alphas", prefix); - SG_PRINT("%slearned_bias=%f\n", prefix, kernel_machine->get_bias()) - } - - if (dynamic_cast(machine) - || dynamic_cast(machine)) - { - /* append one tab to prefix */ - char* new_prefix=append_tab_to_string(prefix); - - CMulticlassMachine* mc_machine=(CMulticlassMachine*)machine; - for (int i=0; iget_num_machines(); i++) - { - CMachine* sub_machine=mc_machine->get_machine(i); - //SG_PRINT("%smulti-class machine %d:\n", i, sub_machine) - this->update_trained_machine(sub_machine, new_prefix); - SG_UNREF(sub_machine); - } - - /* clean up */ - SG_FREE(new_prefix); - } - - if (dynamic_cast(machine)) - { - CMKL* mkl=(CMKL*)machine; - CCombinedKernel* kernel=dynamic_cast( - mkl->get_kernel()); - kernel->get_subkernel_weights().display_vector("MKL sub-kernel weights", - prefix); - SG_UNREF(kernel); - } - - if (dynamic_cast(machine)) - { - CMKLMulticlass* mkl=(CMKLMulticlass*)machine; - CCombinedKernel* kernel=dynamic_cast( - mkl->get_kernel()); - kernel->get_subkernel_weights().display_vector("MKL sub-kernel weights", - prefix); - SG_UNREF(kernel); - } -} - -void CCrossValidationPrintOutput::update_test_result(CLabels* results, - const char* prefix) -{ - results->get_values().display_vector("test_labels", prefix); -} - -void CCrossValidationPrintOutput::update_test_true_result(CLabels* results, - const char* prefix) -{ - results->get_values().display_vector("true_labels", prefix); -} - -void CCrossValidationPrintOutput::update_evaluation_result(float64_t result, - const char* prefix) -{ - SG_PRINT("%sevaluation result=%f\n", prefix, result) -} - -char* CCrossValidationPrintOutput::append_tab_to_string(const char* string) -{ - /* allocate memory, concatenate and add termination character */ - index_t len=strlen(string); - char* new_prefix=SG_MALLOC(char, len+2); - sg_memcpy(new_prefix, string, sizeof(char)*len); - new_prefix[len]='\t'; - new_prefix[len+1]='\0'; - - return new_prefix; -} diff --git a/src/shogun/evaluation/CrossValidationPrintOutput.h b/src/shogun/evaluation/CrossValidationPrintOutput.h deleted file mode 100644 index 58f798514dd..00000000000 --- a/src/shogun/evaluation/CrossValidationPrintOutput.h +++ /dev/null @@ -1,129 +0,0 @@ -/* - * This program is free software; you can redistribute it and/or modify - * it under the terms of the GNU General Public License as published by - * the Free Software Foundation; either version 3 of the License, or - * (at your option) any later version. - * - * Written (W) 2012 Sergey Lisitsyn - * Written (W) 2012 Heiko Strathmann - * - */ - -#ifndef __CROSSVALIDATIONPRINTOUTPUT_H_ -#define __CROSSVALIDATIONPRINTOUTPUT_H_ - -#include - -#include - -namespace shogun -{ - -class CMachine; -class CLabels; -class CEvaluation; - -/** @brief Class for outputting cross-validation intermediate results to the - * standard output. Simply prints all messages it gets */ -class CCrossValidationPrintOutput: public CCrossValidationOutput -{ -public: - - /** constructor */ - CCrossValidationPrintOutput() {}; - - /** destructor */ - virtual ~CCrossValidationPrintOutput() {}; - - /** @return name of SG_SERIALIZABLE */ - virtual const char* get_name() const { return "CrossValidationPrintOutput"; } - - /** init number of runs (called once) - * - * @param num_runs number of runs that will be performed - * @param prefix prefix for output - */ - virtual void init_num_runs(index_t num_runs, const char* prefix=""); - - /** init number of folds - * @param num_folds number of folds that will be performed - * @param prefix prefix for output - */ - virtual void init_num_folds(index_t num_folds, const char* prefix=""); - - /** update run index - * - * @param run_index index of current run - * @param prefix prefix for output - */ - virtual void update_run_index(index_t run_index, - const char* prefix=""); - - /** update fold index - * - * @param fold_index index of current run - * @param prefix prefix for output - */ - virtual void update_fold_index(index_t fold_index, - const char* prefix=""); - - /** update train indices - * - * @param indices indices used for training - * @param prefix prefix for output - */ - virtual void update_train_indices(SGVector indices, - const char* prefix=""); - - /** update test indices - * - * @param indices indices used for testing/validation - * @param prefix prefix for output - */ - virtual void update_test_indices(SGVector indices, - const char* prefix=""); - - /** update trained machine - * - * @param machine trained machine instance - * @param prefix prefix for output - */ - virtual void update_trained_machine(CMachine* machine, - const char* prefix=""); - - /** update test result - * - * @param results result labels for test/validation run - * @param prefix prefix for output - */ - virtual void update_test_result(CLabels* results, - const char* prefix=""); - - /** update test true result - * - * @param results ground truth labels for test/validation run - * @param prefix prefix for output - */ - virtual void update_test_true_result(CLabels* results, - const char* prefix=""); - - /** update evaluate result - * - * @param result evaluation result - * @param prefix prefix for output - */ - virtual void update_evaluation_result(float64_t result, - const char* prefix=""); - -protected: - /** returns a string which is the provided one plus a tab character - * - * @param string null-terminated string to append tab to - * @return null-terminated string with tab appended - */ - char* append_tab_to_string(const char* string); -}; - -} - -#endif /* __CROSSVALIDATIONPRINTOUTPUT_H_ */ diff --git a/src/shogun/evaluation/CrossValidationStorage.cpp b/src/shogun/evaluation/CrossValidationStorage.cpp new file mode 100644 index 00000000000..57cefc0fd9d --- /dev/null +++ b/src/shogun/evaluation/CrossValidationStorage.cpp @@ -0,0 +1,274 @@ +/* +* BSD 3-Clause License +* +* Copyright (c) 2017, Shogun-Toolbox e.V. +* All rights reserved. +* +* Redistribution and use in source and binary forms, with or without +* modification, are permitted provided that the following conditions are met: +* +* * Redistributions of source code must retain the above copyright notice, this +* list of conditions and the following disclaimer. +* +* * Redistributions in binary form must reproduce the above copyright notice, +* this list of conditions and the following disclaimer in the documentation +* and/or other materials provided with the distribution. +* +* * Neither the name of the copyright holder nor the names of its +* contributors may be used to endorse or promote products derived from +* this software without specific prior written permission. +* +* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +* AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE +* DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE +* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL +* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR +* SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER +* CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, +* OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +* OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +* +* Written (W) 2017 Giovanni De Toni +* +*/ + +#include "CrossValidationStorage.h" +#include +#include + +using namespace shogun; + +CrossValidationFoldStorage::CrossValidationFoldStorage() : CSGObject() +{ + m_current_run_index = 0; + m_current_fold_index = 0; + m_trained_machine = NULL; + m_test_result = NULL; + m_test_true_result = NULL; + + SG_ADD( + &m_current_run_index, "m_current_run_index", + "The current run index of this fold", MS_AVAILABLE) + SG_ADD( + &m_current_fold_index, "m_current_fold_index", "The current fold index", + MS_AVAILABLE) + SG_ADD( + (CSGObject**)&m_trained_machine, "m_trained_machine", + "The machine trained by this fold", MS_AVAILABLE) + SG_ADD( + (CSGObject**)&m_test_result, "m_test_result", + "The test result of this fold", MS_AVAILABLE) + SG_ADD( + (CSGObject**)&m_test_true_result, "m_test_true_result", + "The true test result for this fold", MS_AVAILABLE) +} + +CrossValidationFoldStorage::~CrossValidationFoldStorage() +{ + SG_UNREF(m_test_result); + SG_UNREF(m_test_true_result); + SG_UNREF(m_trained_machine); +} + +void CrossValidationFoldStorage::set_run_index(index_t run_index) +{ + m_current_run_index = run_index; +} + +void CrossValidationFoldStorage::set_fold_index(index_t fold_index) +{ + m_current_fold_index = fold_index; +} + +void CrossValidationFoldStorage::set_train_indices(SGVector indices) +{ + m_train_indices = indices; +} + +void CrossValidationFoldStorage::set_test_indices(SGVector indices) +{ + m_test_indices = indices; +} + +void CrossValidationFoldStorage::set_trained_machine(CMachine* machine) +{ + SG_REF(machine) + SG_UNREF(m_trained_machine) + m_trained_machine = machine; +} + +void CrossValidationFoldStorage::set_test_result(CLabels* results) +{ + SG_REF(results) + SG_UNREF(m_test_result) + m_test_result = results; +} + +void CrossValidationFoldStorage::set_test_true_result(CLabels* results) +{ + SG_REF(results) + SG_UNREF(m_test_true_result) + m_test_true_result = results; +} + +void CrossValidationFoldStorage::post_update_results() +{ +} + +void CrossValidationFoldStorage::set_evaluation_result(float64_t result) +{ + m_evaluation_result = result; +} + +index_t CrossValidationFoldStorage::get_current_run_index() const +{ + return m_current_run_index; +} + +index_t CrossValidationFoldStorage::get_current_fold_index() const +{ + return m_current_fold_index; +} + +const SGVector& CrossValidationFoldStorage::get_train_indices() const +{ + return m_train_indices; +} + +const SGVector& CrossValidationFoldStorage::get_test_indices() const +{ + return m_test_indices; +} + +CMachine* CrossValidationFoldStorage::get_trained_machine() const +{ + return m_trained_machine; +} + +CLabels* CrossValidationFoldStorage::get_test_result() const +{ + return m_test_result; +} + +CLabels* CrossValidationFoldStorage::get_test_true_result() const +{ + return m_test_true_result; +} + +float64_t CrossValidationFoldStorage::get_evaluation_result() const +{ + return m_evaluation_result; +} + +CrossValidationFoldStorage* CrossValidationStorage::get_fold(int fold) const +{ + REQUIRE( + fold < get_num_folds(), "The fold number must be less than %i", + get_num_folds()) + + CrossValidationFoldStorage* fld = m_folds_results[fold]; + SG_REF(fld); + return fld; +} + +bool CrossValidationFoldStorage:: +operator==(const CrossValidationFoldStorage& rhs) const +{ + return m_current_run_index == rhs.m_current_run_index && + m_current_fold_index == rhs.m_current_fold_index && + // m_train_indices.equals(rhs.m_train_indices) && + // m_test_indices.equals(rhs.m_test_indices) && + m_trained_machine->equals(rhs.m_trained_machine) && + m_test_result->equals(rhs.m_test_result) && + m_test_true_result->equals(rhs.m_test_true_result) && + m_evaluation_result == rhs.m_evaluation_result; +} + +/** CrossValidationStorage **/ + +CrossValidationStorage::CrossValidationStorage() : CSGObject() +{ + m_num_runs = 0; + m_num_folds = 0; + m_expose_labels = NULL; + + SG_ADD( + &m_num_runs, "m_num_runs", "The total number of cross-validation runs", + MS_AVAILABLE) + SG_ADD( + &m_num_folds, "m_num_folds", + "The total number of cross-validation folds", MS_AVAILABLE) + SG_ADD( + (CSGObject**)&m_expose_labels, "m_expose_labels", + "The labels used for this cross-validation", MS_AVAILABLE) +} + +CrossValidationStorage::~CrossValidationStorage() +{ + SG_UNREF(m_expose_labels); + for (auto i : m_folds_results) + SG_UNREF(i) +} + +void CrossValidationStorage::set_num_runs(index_t num_runs) +{ + m_num_runs = num_runs; +} + +void CrossValidationStorage::set_num_folds(index_t num_folds) +{ + m_num_folds = num_folds; +} + +void CrossValidationStorage::set_expose_labels(CLabels* labels) +{ + SG_REF(labels) + SG_UNREF(m_expose_labels) + m_expose_labels = labels; +} + +void CrossValidationStorage::post_init() +{ +} + +index_t CrossValidationStorage::get_num_runs() const +{ + return m_num_runs; +} + +index_t CrossValidationStorage::get_num_folds() const +{ + return m_num_folds; +} + +CLabels* CrossValidationStorage::get_expose_labels() const +{ + return m_expose_labels; +} + +void CrossValidationStorage::append_fold_result( + CrossValidationFoldStorage* result) +{ + SG_REF(result); + m_folds_results.push_back(result); +} + +bool CrossValidationStorage::operator==(const CrossValidationStorage& rhs) const +{ + auto member_vars = m_num_runs == rhs.m_num_runs && + m_num_folds == rhs.m_num_folds && + m_expose_labels->equals(rhs.m_expose_labels); + + if (!member_vars) + return member_vars; + + if (rhs.m_folds_results.size() != m_folds_results.size()) + return false; + for (index_t i = 0; i < m_folds_results.size(); i++) + { + if (!(m_folds_results[i] == rhs.m_folds_results[i])) + return false; + } + return member_vars; +} \ No newline at end of file diff --git a/src/shogun/evaluation/CrossValidationStorage.h b/src/shogun/evaluation/CrossValidationStorage.h new file mode 100644 index 00000000000..f06b21eba2d --- /dev/null +++ b/src/shogun/evaluation/CrossValidationStorage.h @@ -0,0 +1,293 @@ +/* +* BSD 3-Clause License +* +* Copyright (c) 2017, Shogun-Toolbox e.V. +* All rights reserved. +* +* Redistribution and use in source and binary forms, with or without +* modification, are permitted provided that the following conditions are met: +* +* * Redistributions of source code must retain the above copyright notice, this +* list of conditions and the following disclaimer. +* +* * Redistributions in binary form must reproduce the above copyright notice, +* this list of conditions and the following disclaimer in the documentation +* and/or other materials provided with the distribution. +* +* * Neither the name of the copyright holder nor the names of its +* contributors may be used to endorse or promote products derived from +* this software without specific prior written permission. +* +* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +* AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE +* DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE +* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL +* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR +* SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER +* CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, +* OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +* OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +* +* Written (W) 2017 Giovanni De Toni +* +*/ + +#ifndef SHOGUN_CROSSVALIDATIONSTORAGE_H +#define SHOGUN_CROSSVALIDATIONSTORAGE_H + +#include +#include +#include + +namespace shogun +{ + + class CMachine; + class CLabels; + class CEvaluation; + + /** + * Store information about a single fold run. + */ + class CrossValidationFoldStorage : public CSGObject + { + public: + CrossValidationFoldStorage(); + virtual ~CrossValidationFoldStorage(); + + /** Set run index. + * + * @param run_index index of current run + */ + virtual void set_run_index(index_t run_index); + + /** Set fold index. + * + * @param fold_index index of current run + */ + virtual void set_fold_index(index_t fold_index); + + /** Set train indices + * + * @param indices indices used for training + */ + virtual void set_train_indices(SGVector indices); + + /** Set test indices + * + * @param indices indices used for testing/validation + */ + virtual void set_test_indices(SGVector indices); + + /** Set trained machine + * + * @param machine trained machine instance + */ + virtual void set_trained_machine(CMachine* machine); + + /** Set test result + * + * @param results result labels for test/validation run + */ + virtual void set_test_result(CLabels* results); + + /** Set test true result + * + * @param results ground truth labels for test/validation run + */ + virtual void set_test_true_result(CLabels* results); + + /** post update test and true results + */ + virtual void post_update_results(); + + /** Set evaluate result + * + * @param result evaluation result + */ + virtual void set_evaluation_result(float64_t result); + + /** + * Get current run index + * @return index of the current run + */ + index_t get_current_run_index() const; + + /** + * Get current fold index + * @return index of the current fold + */ + index_t get_current_fold_index() const; + + /** + * Get train indices. + * @return train indices + */ + const SGVector& get_train_indices() const; + + /** + * Get test indices. + * @return test indices + */ + const SGVector& get_test_indices() const; + + /** + * Get trained machine on this fold + * @return trained machine + */ + CMachine* get_trained_machine() const; + + /** + * Get test result + * @return test result + */ + CLabels* get_test_result() const; + + /** + * Get ground truth (correct labels for this fold) + * @return ground truth + */ + CLabels* get_test_true_result() const; + + /** + * Get the evaluation result of this fold + * @return evaluation result + */ + float64_t get_evaluation_result() const; + + /** + * Operator == needed for Any comparison + * @param rhs other CrossValidationFoldStorage + * @return true if the objects are the same, false otherwise. + */ + bool operator==(const CrossValidationFoldStorage& rhs) const; + + /** + * Class name (used for serialization) + * @return class name + */ + virtual const char* get_name() const + { + return "CrossValidationFoldStorage"; + }; + + protected: + /** Current run index is written here */ + index_t m_current_run_index; + + /** Current fold index is written here */ + index_t m_current_fold_index; + + /** Train indices */ + SGVector m_train_indices; + + /** Test indices */ + SGVector m_test_indices; + + /** Trained machine */ + CMachine* m_trained_machine; + + /** Test results */ + CLabels* m_test_result; + + /** Ground truth */ + CLabels* m_test_true_result; + + /** Evaluation result for this fold */ + float64_t m_evaluation_result; + }; + + /** + * This class store some information about CrossValidation runs. + */ + class CrossValidationStorage : public CSGObject + { + public: + /** Constructor */ + CrossValidationStorage(); + + /** Destructor */ + virtual ~CrossValidationStorage(); + + /** + * Class name (used for serialization) + * @return class name + */ + virtual const char* get_name() const + { + return "CrossValidationStorage"; + }; + + /** Set number of runs. + * @param num_runs number of runs that will be performed + */ + virtual void set_num_runs(index_t num_runs); + + /** Set number of folds. + * @param num_folds number of folds that will be performed + */ + virtual void set_num_folds(index_t num_folds); + + /** Set labels before usage. + * @param labels labels to expose to CV output + */ + virtual void set_expose_labels(CLabels* labels); + + /** Post init action. */ + virtual void post_init(); + + /** + * Append a fold result to this storage + * @param result the result of a fold + */ + virtual void append_fold_result(CrossValidationFoldStorage* result); + + /** + * Get number of Cross Validation runs. + * @return Cross Validation's runs + */ + index_t get_num_runs() const; + + /** + * Get number of folds. + * @return + */ + index_t get_num_folds() const; + + /** + * Get original labels. + * @return labels + */ + CLabels* get_expose_labels() const; + + /** + * Get a specific fold result + * @param fold the fold index + * @return the CrossValidationFoldStorage object + */ + CrossValidationFoldStorage* get_fold(int fold) const; + + /** + * Operator == needed for Any comparison. + * @param rhs other CrossValidationStorage + * @return true if the objects are the same, false otherwise. + */ + bool operator==(const CrossValidationStorage& rhs) const; + + protected: + /** number of runs is initialised here */ + index_t m_num_runs; + + /** number of folds is initialised here */ + index_t m_num_folds; + + /** Original labels */ + CLabels* m_expose_labels; + + /** Vector with all the folds results */ + std::vector m_folds_results; + }; +} + +#endif // SHOGUN_CROSSVALIDATIONSTORAGE_H diff --git a/src/shogun/evaluation/GradientEvaluation.cpp b/src/shogun/evaluation/GradientEvaluation.cpp index ee87f556a69..09d18b5476c 100644 --- a/src/shogun/evaluation/GradientEvaluation.cpp +++ b/src/shogun/evaluation/GradientEvaluation.cpp @@ -49,7 +49,7 @@ void CGradientEvaluation::update_parameter_dictionary() SG_REF(m_parameter_dictionary); } -CEvaluationResult* CGradientEvaluation::evaluate() +CEvaluationResult* CGradientEvaluation::evaluate_impl() { if (parameter_hash_changed()) update_parameter_dictionary(); diff --git a/src/shogun/evaluation/GradientEvaluation.h b/src/shogun/evaluation/GradientEvaluation.h index 4a587c9141d..753555a6c9b 100644 --- a/src/shogun/evaluation/GradientEvaluation.h +++ b/src/shogun/evaluation/GradientEvaluation.h @@ -48,12 +48,6 @@ class CGradientEvaluation: public CMachineEvaluation */ virtual const char* get_name() const { return "GradientEvaluation"; } - /** evaluates differentiable function for value and derivative. - * - * @return GradientResult containing value and gradient - */ - virtual CEvaluationResult* evaluate(); - /** set differentiable function * * @param diff differentiable function @@ -79,6 +73,12 @@ class CGradientEvaluation: public CMachineEvaluation /** initialses and registers parameters */ void init(); + /** evaluates differentiable function for value and derivative. + * + * @return GradientResult containing value and gradient + */ + virtual CEvaluationResult* evaluate_impl(); + /** updates parameter dictionary of differentiable function */ void update_parameter_dictionary(); diff --git a/src/shogun/evaluation/MachineEvaluation.cpp b/src/shogun/evaluation/MachineEvaluation.cpp index 5711ea3d77a..0f73183f1e9 100644 --- a/src/shogun/evaluation/MachineEvaluation.cpp +++ b/src/shogun/evaluation/MachineEvaluation.cpp @@ -10,14 +10,18 @@ * Heiko Strathmann */ -#include "MachineEvaluation.h" +#include #include -#include #include +#include #include -#include +#include #include +#include +#include +#include + using namespace shogun; CMachineEvaluation::CMachineEvaluation() @@ -81,6 +85,8 @@ void CMachineEvaluation::init() m_evaluation_criterion = NULL; m_do_unlock = false; m_autolock = true; + m_cancel_computation = false; + m_pause_computation_flag = false; SG_ADD((CSGObject**)&m_machine, "machine", "Used learning machine", MS_NOT_AVAILABLE); @@ -101,6 +107,34 @@ void CMachineEvaluation::init() } +CEvaluationResult* CMachineEvaluation::evaluate() +{ + SG_DEBUG("entering %s::evaluate()\n", get_name()) + + REQUIRE( + m_machine, "%s::evaluate() is only possible if a machine is " + "attached\n", + get_name()); + + REQUIRE( + m_features, "%s::evaluate() is only possible if features are " + "attached\n", + get_name()); + + REQUIRE( + m_labels, "%s::evaluate() is only possible if labels are " + "attached\n", + get_name()); + + auto sub = connect_to_signal_handler(); + CEvaluationResult* result = evaluate_impl(); + sub.unsubscribe(); + reset_computation_variables(); + + SG_DEBUG("leaving %s::evaluate()\n", get_name()) + return result; +}; + CMachine* CMachineEvaluation::get_machine() const { SG_REF(m_machine); @@ -111,3 +145,17 @@ EEvaluationDirection CMachineEvaluation::get_evaluation_direction() { return m_evaluation_criterion->get_evaluation_direction(); } + +rxcpp::subscription CMachineEvaluation::connect_to_signal_handler() +{ + // Subscribe this algorithm to the signal handler + auto subscriber = rxcpp::make_subscriber( + [this](int i) { + if (i == SG_PAUSE_COMP) + this->on_pause(); + else + this->on_next(); + }, + [this]() { this->on_complete(); }); + return get_global_signal()->get_observable()->subscribe(subscriber); +} \ No newline at end of file diff --git a/src/shogun/evaluation/MachineEvaluation.h b/src/shogun/evaluation/MachineEvaluation.h index e1a38647da4..825645833fb 100644 --- a/src/shogun/evaluation/MachineEvaluation.h +++ b/src/shogun/evaluation/MachineEvaluation.h @@ -20,100 +20,193 @@ #include #include -namespace shogun -{ +#include +#include -class CMachine; -class CFeatures; -class CLabels; -class CSplittingStrategy; -class CEvaluation; - -/** @brief Machine Evaluation is an abstract class - * that evaluates a machine according to some criterion. - * - */ -class CMachineEvaluation: public CSGObject +namespace shogun { -public: +#define EVALUATION_CONTROLLERS \ + if (cancel_evaluation()) \ + continue; \ + pause_evaluation(); - CMachineEvaluation(); + class CMachine; + class CFeatures; + class CLabels; + class CSplittingStrategy; + class CEvaluation; - /** constructor - * @param machine learning machine to use - * @param features features to use for cross-validation - * @param labels labels that correspond to the features - * @param splitting_strategy splitting strategy to use - * @param evaluation_criterion evaluation criterion to use - * @param autolock whether machine should be auto-locked before evaluation - */ - CMachineEvaluation(CMachine* machine, CFeatures* features, CLabels* labels, - CSplittingStrategy* splitting_strategy, - CEvaluation* evaluation_criterion, bool autolock = true); - - /** constructor, for use with custom kernels (no features) - * @param machine learning machine to use - * @param labels labels that correspond to the features - * @param splitting_strategy splitting strategy to use - * @param evaluation_criterion evaluation criterion to use - * @param autolock autolock - */ - CMachineEvaluation(CMachine* machine, CLabels* labels, - CSplittingStrategy* splitting_strategy, - CEvaluation* evaluation_criterion, bool autolock = true); - - - virtual ~CMachineEvaluation(); - - /** @return in which direction is the best evaluation value? */ - EEvaluationDirection get_evaluation_direction(); - - /** method for evaluation. Performs cross-validation. - * Is repeated m_num_runs. If this number is larger than one, a confidence - * interval is calculated if m_conf_int_alpha is (0 lck(m_mutex); + while (m_pause_computation_flag.load()) + m_pause_computation.wait(lck); + } + } +#endif + +#ifndef SWIG + /** Resume current evaluation (sets the flag) */ + SG_FORCED_INLINE void resume_evaluation() + { + std::unique_lock lck(m_mutex); + m_pause_computation_flag = false; + m_pause_computation.notify_all(); + } +#endif + + protected: + /** Initialize Object */ + virtual void init(); + + /** + * Implementation of the evaluation procedure. Called + * by evaluate() method. This method has to SG_REF its result + * before returning it. + * @return the evaluation result + */ + virtual CEvaluationResult* evaluate_impl() = 0; + + /** connect the machine instance to the signal handler */ + rxcpp::subscription connect_to_signal_handler(); + + /** reset the computation variables */ + void reset_computation_variables() + { + m_cancel_computation = false; + m_pause_computation_flag = false; + } + + /** The action which will be done when the user decides to + * premature stop the CMachineEvaluation execution */ + virtual void on_next() + { + m_cancel_computation.store(true); + } + + /** The action which will be done when the user decides to + * pause the CMachineEvaluation execution */ + virtual void on_pause() + { + m_pause_computation_flag.store(true); + /* Here there should be the actual code*/ + resume_evaluation(); + } + + /** The action which will be done when the user decides to + * return to prompt and terminate the program execution */ + virtual void on_complete() + { + } + + protected: + /** Machine to be Evaluated */ + CMachine* m_machine; + + /** Features to be used*/ + CFeatures* m_features; + + /** Labels for the features */ + CLabels* m_labels; + + /** Splitting Strategy to be used */ + CSplittingStrategy* m_splitting_strategy; + + /** Criterion for evaluation */ + CEvaluation* m_evaluation_criterion; + + /** whether machine will automatically be locked before evaluation */ + bool m_autolock; + + /** whether machine should be unlocked after evaluation */ + bool m_do_unlock; + + /** Cancel evaluation */ + std::atomic m_cancel_computation; + + /** Pause evaluation flag */ + std::atomic m_pause_computation_flag; + + /** Conditional variable to make threads wait */ + std::condition_variable m_pause_computation; + + /** Mutex used to pause threads */ + std::mutex m_mutex; + }; } /* namespace shogun */ diff --git a/src/shogun/features/CombinedFeatures.cpp b/src/shogun/features/CombinedFeatures.cpp index 5fc4f508750..2e97688ca66 100644 --- a/src/shogun/features/CombinedFeatures.cpp +++ b/src/shogun/features/CombinedFeatures.cpp @@ -48,6 +48,9 @@ CCombinedFeatures::~CCombinedFeatures() CFeatures* CCombinedFeatures::get_feature_obj(int32_t idx) { + REQUIRE( + idx < get_num_feature_obj() && idx>=0, "Feature index (%d) must be within [%d, %d]", + idx, 0, get_num_feature_obj()-1); return (CFeatures*) feature_array->get_element(idx); } diff --git a/src/shogun/features/DataGenerator.cpp b/src/shogun/features/DataGenerator.cpp index 2341ade9882..f38983116bc 100644 --- a/src/shogun/features/DataGenerator.cpp +++ b/src/shogun/features/DataGenerator.cpp @@ -129,7 +129,7 @@ SGMatrix CDataGenerator::generate_sym_mix_gauss(index_t m, return result; } -#ifdef HAVE_LAPACK + SGMatrix CDataGenerator::generate_gaussians(index_t m, index_t n, index_t dim) { /* evtl. allocate space */ @@ -154,7 +154,6 @@ SGMatrix CDataGenerator::generate_gaussians(index_t m, index_t n, ind { SGVector v = g->sample(); sg_memcpy((result.matrix+j*result.num_rows+i*m*dim), v.vector, dim*sizeof(float64_t)); - SG_FREE(v.vector); } SG_UNREF(g); @@ -162,4 +161,3 @@ SGMatrix CDataGenerator::generate_gaussians(index_t m, index_t n, ind return result; } -#endif /* HAVE_LAPACK */ diff --git a/src/shogun/features/DataGenerator.h b/src/shogun/features/DataGenerator.h index b8d8fb24dbe..6e8bb6bcf09 100644 --- a/src/shogun/features/DataGenerator.h +++ b/src/shogun/features/DataGenerator.h @@ -85,7 +85,6 @@ class CDataGenerator: public CSGObject float64_t d, float64_t angle, SGMatrix target=SGMatrix()); -#ifdef HAVE_LAPACK /** Produces samples of gaussians * The functions produces m number of samples of each gaussians (n number) with * the given dimension. @@ -97,7 +96,6 @@ class CDataGenerator: public CSGObject * of the first gaussian, m number of second etc. */ static SGMatrix generate_gaussians(index_t m, index_t n, index_t dim); -#endif /* HAVE_LAPACK */ virtual const char* get_name() const { return "DataGenerator"; } diff --git a/src/shogun/features/DenseFeatures.cpp b/src/shogun/features/DenseFeatures.cpp index db716171ad3..1ef65dcc811 100644 --- a/src/shogun/features/DenseFeatures.cpp +++ b/src/shogun/features/DenseFeatures.cpp @@ -19,6 +19,7 @@ #include #include #include +#include #include #include @@ -76,7 +77,6 @@ template CDenseFeatures::~CDenseFeatures() template void CDenseFeatures::free_features() { - m_subset_stack->remove_all_subsets(); free_feature_matrix(); SG_UNREF(feature_cache); } @@ -504,8 +504,10 @@ template float64_t CDenseFeatures::dot(int32_t vec_idx1, CDotFeatu ST* vec1 = get_feature_vector(vec_idx1, len1, free1); ST* vec2 = sf->get_feature_vector(vec_idx2, len2, free2); + SGVector sg_vec1(vec1, len1, false); + SGVector sg_vec2(vec2, len2, false); - float64_t result = CMath::dot(vec1, vec2, len1); + float64_t result = linalg::dot(sg_vec1, sg_vec2); free_feature_vector(vec1, vec_idx1, free1); sf->free_feature_vector(vec2, vec_idx2, free2); @@ -943,9 +945,11 @@ template<> float64_t CDenseFeatures::dense_dot( int32_t vlen; bool vfree; float64_t* vec1 = get_feature_vector(vec_idx1, vlen, vfree); + SGVector sg_vec1(vec1, vlen, false); ASSERT(vlen == num_features) - float64_t result = CMath::dot(vec1, vec2, num_features); + SGVector tmp(const_cast(vec2), vec2_len, false); + float64_t result = linalg::dot(sg_vec1, tmp); free_feature_vector(vec1, vec_idx1, vfree); diff --git a/src/shogun/features/DotFeatures.cpp b/src/shogun/features/DotFeatures.cpp index 00eec29cf3d..c964e911497 100644 --- a/src/shogun/features/DotFeatures.cpp +++ b/src/shogun/features/DotFeatures.cpp @@ -8,13 +8,15 @@ * Copyright (C) 2009 Fraunhofer Institute FIRST and Max-Planck-Society */ +#include +#include +#include #include #include #include #include #include -#include -#include +#include #ifdef HAVE_OPENMP #include @@ -61,11 +63,10 @@ void CDotFeatures::dense_dot_range(float64_t* output, int32_t start, int32_t sto int32_t num_vectors=stop-start; ASSERT(num_vectors>0) - CSignal::clear_cancel(); - int32_t num_threads; int32_t step; - #pragma omp parallel shared(num_threads, step) + auto pb = progress(range(num_vectors), *this->io); +#pragma omp parallel shared(num_threads, step) { #ifdef HAVE_OPENMP #pragma omp single @@ -80,7 +81,6 @@ void CDotFeatures::dense_dot_range(float64_t* output, int32_t start, int32_t sto step=num_vectors; int32_t thread_num=0; #endif - bool progress=false; // (thread_num == 0); int32_t t_start=thread_num*step; int32_t t_stop=(thread_num==num_threads) ? stop : (thread_num+1)*step; @@ -88,23 +88,20 @@ void CDotFeatures::dense_dot_range(float64_t* output, int32_t start, int32_t sto #ifdef WIN32 for (int32_t i=t_start; idense_dot(i, vec, dim)+b; else output[i]=this->dense_dot(i, vec, dim)+b; - if (progress) - this->display_progress(t_start, t_stop, i); + pb.print_progress(); } } - -#ifndef WIN32 - if ( CSignal::cancel_computations() ) - SG_INFO("prematurely stopped. \n") -#endif + pb.complete(); } void CDotFeatures::dense_dot_range_subset(int32_t* sub_index, int32_t num, float64_t* output, float64_t* alphas, float64_t* vec, int32_t dim, float64_t b) @@ -112,8 +109,7 @@ void CDotFeatures::dense_dot_range_subset(int32_t* sub_index, int32_t num, float ASSERT(sub_index) ASSERT(output) - CSignal::clear_cancel(); - + auto pb = progress(range(num), *this->io); int32_t num_threads; int32_t step; #pragma omp parallel shared(num_threads, step) @@ -131,7 +127,6 @@ void CDotFeatures::dense_dot_range_subset(int32_t* sub_index, int32_t num, float step = num; int32_t thread_num=0; #endif - bool progress=false; // (thread_num == 0); int32_t t_start=thread_num*step; int32_t t_stop=(thread_num==num_threads) ? num : (thread_num+1)*step; @@ -139,23 +134,20 @@ void CDotFeatures::dense_dot_range_subset(int32_t* sub_index, int32_t num, float #ifdef WIN32 for (int32_t i=t_start; idense_dot(sub_index[i], vec, dim)+b; else output[i]=this->dense_dot(sub_index[i], vec, dim)+b; - if (progress) - this->display_progress(t_start, t_stop, i); + pb.print_progress(); } } - -#ifndef WIN32 - if ( CSignal::cancel_computations() ) - SG_INFO("prematurely stopped. \n") -#endif + pb.complete(); } SGMatrix CDotFeatures::get_computed_dot_feature_matrix() @@ -266,17 +258,18 @@ SGVector CDotFeatures::get_mean() ASSERT(dim>0) SGVector mean(dim); - memset(mean.vector, 0, sizeof(float64_t)*dim); + linalg::zero(mean); - for (int i = 0; i < num; i++) + for (int32_t i = 0; i < num; ++i) add_to_dense_vec(1, i, mean.vector, dim); - for (int j = 0; j < dim; j++) - mean.vector[j] /= num; + + linalg::scale(mean, mean, 1.0 / num); return mean; } -SGVector CDotFeatures::get_mean(CDotFeatures* lhs, CDotFeatures* rhs) +SGVector +CDotFeatures::compute_mean(CDotFeatures* lhs, CDotFeatures* rhs) { ASSERT(lhs && rhs) ASSERT(lhs->get_dim_feature_space() == rhs->get_dim_feature_space()) @@ -289,19 +282,20 @@ SGVector CDotFeatures::get_mean(CDotFeatures* lhs, CDotFeatures* rhs) ASSERT(dim>0) SGVector mean(dim); - memset(mean.vector, 0, sizeof(float64_t)*dim); + linalg::zero(mean); for (int i = 0; i < num_lhs; i++) lhs->add_to_dense_vec(1, i, mean.vector, dim); + for (int i = 0; i < num_rhs; i++) rhs->add_to_dense_vec(1, i, mean.vector, dim); - for (int j = 0; j < dim; j++) - mean.vector[j] /= (num_lhs+num_rhs); + + linalg::scale(mean, mean, 1.0 / (num_lhs + num_rhs)); return mean; } -SGMatrix CDotFeatures::get_cov() +SGMatrix CDotFeatures::get_cov(bool copy_data_for_speed) { int32_t num=get_num_vectors(); int32_t dim=get_dim_feature_space(); @@ -309,41 +303,44 @@ SGMatrix CDotFeatures::get_cov() ASSERT(dim>0) SGMatrix cov(dim, dim); - - memset(cov.matrix, 0, sizeof(float64_t)*dim*dim); - SGVector mean = get_mean(); - for (int i = 0; i < num; i++) + if (copy_data_for_speed) { - SGVector v = get_computed_dot_feature_vector(i); - SGVector::add(v.vector, 1, v.vector, -1, mean.vector, v.vlen); - for (int m = 0; m < v.vlen; m++) + SGMatrix centered_data(dim, num); + for (int i = 0; i < num; i++) { - for (int n = 0; n <= m ; n++) - { - (cov.matrix)[m*v.vlen+n] += v.vector[m]*v.vector[n]; - } + SGVector v = get_computed_dot_feature_vector(i); + centered_data.set_column(i, linalg::add(v, mean, 1.0, -1.0)); } + + cov = linalg::matrix_prod(centered_data, centered_data, false, true); } - for (int m = 0; m < dim; m++) + else { - for (int n = 0; n <= m ; n++) + linalg::zero(cov); + for (int i = 0; i < num; i++) { - (cov.matrix)[m*dim+n] /= num; + SGVector v = get_computed_dot_feature_vector(i); + linalg::add(v, mean, v, 1.0, -1.0); + for (int m = 0; m < v.vlen; m++) + linalg::add_col_vec(cov, m, v, cov, 1.0, v.vector[m]); } - } - for (int m = 0; m < dim-1; m++) - { - for (int n = m+1; n < dim; n++) + for (int m = 0; m < dim - 1; m++) { - (cov.matrix)[m*dim+n] = (cov.matrix)[n*dim+m]; + for (int n = m + 1; n < dim; n++) + { + (cov.matrix)[m * dim + n] = (cov.matrix)[n * dim + m]; + } } } + linalg::scale(cov, cov, 1.0 / num); + return cov; } -SGMatrix CDotFeatures::compute_cov(CDotFeatures* lhs, CDotFeatures* rhs) +SGMatrix CDotFeatures::compute_cov( + CDotFeatures* lhs, CDotFeatures* rhs, bool copy_data_for_speed) { CDotFeatures* feats[2]; feats[0]=lhs; @@ -364,53 +361,50 @@ SGMatrix CDotFeatures::compute_cov(CDotFeatures* lhs, CDotFeatures* r int32_t dim = dims[0]; SGMatrix cov(dim, dim); + SGVector mean = compute_mean(lhs, rhs); - memset(cov.matrix, 0, sizeof(float64_t)*dim*dim); - - SGVector mean=get_mean(lhs,rhs); - - for (int i = 0; i < 2; i++) + if (copy_data_for_speed) { - for (int j = 0; j < nums[i]; j++) + SGMatrix centered_data(dim, num); + for (int i = 0; i < num; i++) { - SGVector v = feats[i]->get_computed_dot_feature_vector(j); - SGVector::add(v.vector, 1, v.vector, -1, mean.vector, v.vlen); - for (int m = 0; m < v.vlen; m++) - { - for (int n = 0; n <= m; n++) - { - (cov.matrix)[m*v.vlen+n] += v.vector[m]*v.vector[n]; - } - } + SGVector v = + i < nums[0] ? lhs->get_computed_dot_feature_vector(i) + : rhs->get_computed_dot_feature_vector(i - nums[0]); + + centered_data.set_column(i, linalg::add(v, mean, 1.0, -1.0)); } + + cov = linalg::matrix_prod(centered_data, centered_data, false, true); } - for (int m = 0; m < dim; m++) + else { - for (int n = 0; n <= m; n++) + linalg::zero(cov); + for (int i = 0; i < 2; i++) { - (cov.matrix)[m*dim+n] /= num; + for (int j = 0; j < nums[i]; j++) + { + SGVector v = + feats[i]->get_computed_dot_feature_vector(j); + linalg::add(v, mean, v, 1.0, -1.0); + for (int m = 0; m < v.vlen; m++) + linalg::add_col_vec(cov, m, v, cov, 1.0, v.vector[m]); + } } - } - for (int m = 0; m < dim-1; m++) - { - for (int n = m+1; n < dim; n++) + + for (int m = 0; m < dim - 1; m++) { - (cov.matrix[m*dim+n]) = (cov.matrix)[n*dim+m]; + for (int n = m + 1; n < dim; n++) + { + (cov.matrix[m * dim + n]) = (cov.matrix)[n * dim + m]; + } } } + linalg::scale(cov, cov, 1.0 / num); return cov; } -void CDotFeatures::display_progress(int32_t start, int32_t stop, int32_t v) -{ - int32_t num_vectors=stop-start; - int32_t i=v-start; - - if ( (i% (num_vectors/100+1))== 0) - SG_PROGRESS(v, 0.0, num_vectors-1) -} - void CDotFeatures::init() { set_property(FP_DOT); diff --git a/src/shogun/features/DotFeatures.h b/src/shogun/features/DotFeatures.h index 96e361f2cc8..c2068cc5392 100644 --- a/src/shogun/features/DotFeatures.h +++ b/src/shogun/features/DotFeatures.h @@ -213,28 +213,31 @@ class CDotFeatures : public CFeatures * * @return mean returned */ - static SGVector get_mean(CDotFeatures* lhs, CDotFeatures* rhs); + static SGVector + compute_mean(CDotFeatures* lhs, CDotFeatures* rhs); /** get covariance * + * @param copy_data_for_speed if true, the method stores explicitly + * the centered data matrix and the covariance is calculated by matrix + * product of the centered data with its transpose, this make it + * possible to take advantage of multithreaded matrix product, + * this may not be possible if the data doesn't fit into memory, + * in such case set this parameter to false to compute iteratively + * the covariance matrix without storing the centered data. + * [default = true] * @return covariance */ - virtual SGMatrix get_cov(); + virtual SGMatrix get_cov(bool copy_data_for_speed = true); /** compute the covariance of two CDotFeatures together * + * @param copy_data_for_speed @see CDotFeatures::get_cov * @return covariance */ - static SGMatrix compute_cov(CDotFeatures* lhs, CDotFeatures* rhs); - - protected: - /** display progress output - * - * @param start minimum value - * @param stop maximum value - * @param v current value - */ - void display_progress(int32_t start, int32_t stop, int32_t v); + static SGMatrix compute_cov( + CDotFeatures* lhs, CDotFeatures* rhs, + bool copy_data_for_speed = true); private: void init(); diff --git a/src/shogun/features/ExplicitSpecFeatures.cpp b/src/shogun/features/ExplicitSpecFeatures.cpp index 770c3e37a54..12911537c86 100644 --- a/src/shogun/features/ExplicitSpecFeatures.cpp +++ b/src/shogun/features/ExplicitSpecFeatures.cpp @@ -10,6 +10,7 @@ #include #include +#include #include using namespace shogun; @@ -68,21 +69,21 @@ float64_t CExplicitSpecFeatures::dot(int32_t vec_idx1, CDotFeatures* df, int32_t ASSERT(vec_idx1 < num_strings) ASSERT(vec_idx2 < sf->num_strings) - float64_t* vec1=k_spectrum[vec_idx1]; - float64_t* vec2=sf->k_spectrum[vec_idx2]; + SGVector vec1(k_spectrum[vec_idx1], spec_size, false); + SGVector vec2(sf->k_spectrum[vec_idx2], spec_size, false); - return CMath::dot(vec1, vec2, spec_size); + return linalg::dot(vec1, vec2); } float64_t CExplicitSpecFeatures::dense_dot(int32_t vec_idx1, const float64_t* vec2, int32_t vec2_len) { ASSERT(vec2_len == spec_size) ASSERT(vec_idx1 < num_strings) - float64_t* vec1=k_spectrum[vec_idx1]; + SGVector vec1(k_spectrum[vec_idx1], spec_size, false); + SGVector vec2_wrapper(const_cast(vec2), vec2_len, false); float64_t result=0; - for (int32_t i=0; i #include #include +#include namespace shogun { @@ -109,6 +110,22 @@ class CFeatures : public CSGObject */ virtual EFeatureClass get_feature_class() const=0; +#ifndef SWIG + /** returns an iterator of indices + * from 0 to @ref CFeatures::get_num_vectors + * + * Should be used in algorithms in the following way: + * @code + * for (auto idx : features->index_iterator()) { ... } + * @endcode + * + */ + virtual Range index_iterator() const + { + return range(0, get_num_vectors()); + } +#endif + /** add preprocessor * * @param p preprocessor to set diff --git a/src/shogun/features/LBPPyrDotFeatures.cpp b/src/shogun/features/LBPPyrDotFeatures.cpp index 2f68a32cb9e..17047d43ab6 100644 --- a/src/shogun/features/LBPPyrDotFeatures.cpp +++ b/src/shogun/features/LBPPyrDotFeatures.cpp @@ -11,6 +11,7 @@ */ #include #include +#include using namespace shogun; @@ -105,7 +106,7 @@ float64_t CLBPPyrDotFeatures::dot(int32_t vec_idx1, CDotFeatures* df, int32_t ve SGVector vec1 = get_transformed_image(vec_idx1); SGVector vec2 = lbp_feat->get_transformed_image(vec_idx2); - return CMath::dot(vec1.vector, vec2.vector, vec_nDim); + return linalg::dot(vec1, vec2); } SGVector CLBPPyrDotFeatures::get_transformed_image(int32_t index) diff --git a/src/shogun/features/StringFeatures.cpp b/src/shogun/features/StringFeatures.cpp index f9b469373b5..1c409d742f9 100644 --- a/src/shogun/features/StringFeatures.cpp +++ b/src/shogun/features/StringFeatures.cpp @@ -1,11 +1,12 @@ +#include +#include #include -#include -#include #include #include -#include -#include #include +#include +#include +#include #include #include @@ -18,6 +19,7 @@ #include #else #include + #endif namespace shogun @@ -488,6 +490,7 @@ template void CStringFeatures::load_ascii_file(char* fname, bool r SG_DEBUG("block_size=%ld file_size=%ld\n", blocksize, fsize) + auto pb = progress(range(fsize), *this->io, "COUNTING: "); size_t sz=blocksize; while (sz == blocksize) { @@ -502,8 +505,9 @@ template void CStringFeatures::load_ascii_file(char* fname, bool r old_block_offs=block_offs; } } - SG_PROGRESS(block_offs, 0, fsize, 1, "COUNTING:\t") + pb.print_progress(); } + pb.complete(); SG_INFO("found %d strings\n", num_vectors) SG_FREE(dummy); @@ -512,6 +516,10 @@ template void CStringFeatures::load_ascii_file(char* fname, bool r overflow=SG_MALLOC(uint8_t, blocksize); features=SG_MALLOC(SGString, num_vectors); + auto pb2 = + PRange(range(num_vectors), *this->io, "LOADING: ", UTF8, []() { + return true; + }); rewind(f); sz=blocksize; int32_t lines=0; @@ -556,9 +564,11 @@ template void CStringFeatures::load_ascii_file(char* fname, bool r //CMath::display_vector(features[lines].string, len); old_sz=i+1; lines++; - SG_PROGRESS(lines, 0, num_vectors, 1, "LOADING:\t") + pb2.print_progress(); } } + pb2.complete(); + for (size_t i=old_sz; iget_num_elements(); ++i) { - m_active_subset=(CSubset*)other.m_active_subsets_stack->get_element(i); - m_active_subsets_stack->append_element(m_active_subset); + auto subset = other.m_active_subsets_stack->get_element(i); + m_active_subsets_stack->append_element(subset); + SG_UNREF(subset) } + m_active_subset = other.m_active_subset; + SG_REF(m_active_subset) } CSubsetStack::~CSubsetStack() @@ -63,6 +66,7 @@ void CSubsetStack::remove_all_subsets() m_active_subsets_stack->delete_element(i); SG_UNREF(m_active_subset); + m_active_subset = nullptr; } void CSubsetStack::init() diff --git a/src/shogun/features/hashed/HashedWDFeaturesTransposed.cpp b/src/shogun/features/hashed/HashedWDFeaturesTransposed.cpp index cda8ab0b491..e4061081243 100644 --- a/src/shogun/features/hashed/HashedWDFeaturesTransposed.cpp +++ b/src/shogun/features/hashed/HashedWDFeaturesTransposed.cpp @@ -8,13 +8,15 @@ * Copyright (C) 2010 Berlin Institute of Technology */ +#include +#include #include #include #include -#include #ifdef HAVE_PTHREAD #include + #endif using namespace shogun; @@ -31,6 +33,7 @@ struct HASHEDWD_THREAD_PARAM float64_t* vec; float64_t bias; bool progress; + PRange* progress_bar; uint32_t* index; }; #endif // DOXYGEN_SHOULD_SKIP_THIS @@ -219,14 +222,13 @@ void CHashedWDFeaturesTransposed::dense_dot_range(float64_t* output, int32_t sta #endif ASSERT(num_threads>0) - CSignal::clear_cancel(); - if (dim != w_dim) SG_ERROR("Dimensions don't match, vec_len=%d, w_dim=%d\n", dim, w_dim) if (num_threads < 2) { HASHEDWD_THREAD_PARAM params; + auto pb = progress(range(start, stop), *this->io); params.hf=this; params.sub_index=NULL; params.output=output; @@ -236,14 +238,17 @@ void CHashedWDFeaturesTransposed::dense_dot_range(float64_t* output, int32_t sta params.vec=vec; params.bias=b; params.progress=false; //true; - params.index=index; + params.progress_bar = &pb; + params.index = index; dense_dot_range_helper((void*) ¶ms); + pb.complete(); } #ifdef HAVE_PTHREAD else { pthread_t* threads = SG_MALLOC(pthread_t, num_threads-1); HASHEDWD_THREAD_PARAM* params = SG_MALLOC(HASHEDWD_THREAD_PARAM, num_threads); + auto pb = progress(range(start, stop), *this->io); int32_t step= num_vectors/num_threads; int32_t t; @@ -259,6 +264,7 @@ void CHashedWDFeaturesTransposed::dense_dot_range(float64_t* output, int32_t sta params[t].vec=vec; params[t].bias=b; params[t].progress = false; + params[t].progress_bar = &pb; params[t].index=index; pthread_create(&threads[t], NULL, CHashedWDFeaturesTransposed::dense_dot_range_helper, (void*)¶ms[t]); @@ -273,6 +279,7 @@ void CHashedWDFeaturesTransposed::dense_dot_range(float64_t* output, int32_t sta params[t].vec=vec; params[t].bias=b; params[t].progress = false; //true; + params[t].progress_bar = &pb; params[t].index=index; CHashedWDFeaturesTransposed::dense_dot_range_helper((void*) ¶ms[t]); @@ -284,11 +291,6 @@ void CHashedWDFeaturesTransposed::dense_dot_range(float64_t* output, int32_t sta } #endif SG_FREE(index); - -#ifndef WIN32 - if ( CSignal::cancel_computations() ) - SG_INFO("prematurely stopped. \n") -#endif } void CHashedWDFeaturesTransposed::dense_dot_range_subset(int32_t* sub_index, int num, float64_t* output, float64_t* alphas, float64_t* vec, int32_t dim, float64_t b) @@ -306,14 +308,13 @@ void CHashedWDFeaturesTransposed::dense_dot_range_subset(int32_t* sub_index, int #endif ASSERT(num_threads>0) - CSignal::clear_cancel(); - if (dim != w_dim) SG_ERROR("Dimensions don't match, vec_len=%d, w_dim=%d\n", dim, w_dim) if (num_threads < 2) { HASHEDWD_THREAD_PARAM params; + auto pb = progress(range(num), *this->io); params.hf=this; params.sub_index=sub_index; params.output=output; @@ -323,8 +324,10 @@ void CHashedWDFeaturesTransposed::dense_dot_range_subset(int32_t* sub_index, int params.vec=vec; params.bias=b; params.progress=false; //true; + params.progress_bar = &pb; params.index=index; dense_dot_range_helper((void*) ¶ms); + pb.complete(); } #ifdef HAVE_PTHREAD else @@ -332,7 +335,7 @@ void CHashedWDFeaturesTransposed::dense_dot_range_subset(int32_t* sub_index, int pthread_t* threads = SG_MALLOC(pthread_t, num_threads-1); HASHEDWD_THREAD_PARAM* params = SG_MALLOC(HASHEDWD_THREAD_PARAM, num_threads); int32_t step= num/num_threads; - + auto pb = progress(range(num), *this->io); int32_t t; for (t=0; tvec; float64_t bias=par->bias; bool progress=par->progress; + auto pb = par->progress_bar; uint32_t* index=par->index; int32_t string_length=hf->string_length; int32_t degree=hf->degree; @@ -440,7 +442,7 @@ void* CHashedWDFeaturesTransposed::dense_dot_range_helper(void* p) offs+=partial_w_dim*degree; if (progress) - hf->io->progress(i, 0,string_length, i); + pb->print_progress(); } for (int32_t j=start; jio->progress(i, 0,string_length, i); + pb->print_progress(); } for (int32_t j=start; j +#include #include #include @@ -264,7 +265,7 @@ float32_t CStreamingDenseFeatures::dot(CStreamingDotFeatures* df) SGVector other_vector=sf->get_vector(); - return CMath::dot(current_vector.vector, other_vector.vector, current_vector.vlen); + return linalg::dot(current_vector, other_vector); } template @@ -277,7 +278,7 @@ float32_t CStreamingDenseFeatures::dot(SGVector sgvec1) SG_ERROR( "Lengths %d and %d not equal while computing dot product!\n", len1, current_vector.vlen); - return CMath::dot(current_vector.vector, sgvec1.vector, len1); + return linalg::dot(current_vector, sgvec1); } template diff --git a/src/shogun/io/LibSVMFile.cpp b/src/shogun/io/LibSVMFile.cpp index 724a528fbea..bf6d93aa6ea 100644 --- a/src/shogun/io/LibSVMFile.cpp +++ b/src/shogun/io/LibSVMFile.cpp @@ -10,12 +10,13 @@ #include -#include -#include #include +#include #include #include #include +#include +#include using namespace shogun; @@ -119,7 +120,8 @@ void CLibSVMFile::get_sparse_matrix(SGSparseVector*& mat_feat, int32_t& for (int32_t i=0; i*& mat_feat, int32_t& num_feat, int32_t& num_vec, \ - SGVector*& multilabel, int32_t& num_classes, bool load_labels) \ -{ \ - num_feat=0; \ - \ - SG_INFO("counting line numbers in file %s\n", filename) \ - num_vec=get_num_lines(); \ - \ - int32_t current_line_ind=0; \ - SGVector line; \ - \ - int32_t num_feat_entries=0; \ - DynArray > entries_feat; \ - DynArray entries_label; \ - DynArray classes; \ - \ - mat_feat=SG_MALLOC(SGSparseVector, num_vec); \ - multilabel=SG_MALLOC(SGVector, num_vec); \ - \ - num_classes=0; \ - SG_SET_LOCALE_C; \ - \ - while (m_line_reader->has_next()) \ - { \ - num_feat_entries=0; \ - entries_feat.reset(SGVector(false)); \ - line=m_line_reader->read_line(); \ - \ - m_parser->set_tokenizer(m_whitespace_tokenizer); \ - m_parser->set_text(line); \ - \ - SGVector entry_label; \ - if (load_labels && m_parser->has_next()) \ - { \ - entry_label=m_parser->read_string(); \ - if (is_feat_entry(entry_label)) \ - { \ - entries_feat.push_back(entry_label); \ - num_feat_entries++; \ - entry_label=SGVector(0); \ - } \ - } \ - \ - while (m_parser->has_next()) \ - { \ - entries_feat.push_back(m_parser->read_string()); \ - num_feat_entries++; \ - } \ - \ - mat_feat[current_line_ind]=SGSparseVector(num_feat_entries); \ - for (int32_t i=0; iset_tokenizer(m_delimiter_feat_tokenizer); \ - m_parser->set_text(entries_feat[i]); \ - \ - int32_t feat_index=0; \ - \ - if (m_parser->has_next()) \ - feat_index=m_parser->read_int(); \ - \ - sg_type entry=0; \ - \ - if (m_parser->has_next()) \ - entry=m_parser->read_func(); \ - \ - if (feat_index>num_feat) \ - num_feat=feat_index; \ - \ - mat_feat[current_line_ind].features[i].feat_index=feat_index-1; \ - mat_feat[current_line_ind].features[i].entry=entry; \ - } \ - \ - if (load_labels) \ - { \ - m_parser->set_tokenizer(m_delimiter_label_tokenizer); \ - m_parser->set_text(entry_label); \ - \ - int32_t num_label_entries=0; \ - entries_label.reset(0); \ - \ - while (m_parser->has_next()) \ - { \ - num_label_entries++; \ - float64_t label_val=m_parser->read_real(); \ - \ - if (classes.find_element(label_val)==-1) \ - classes.push_back(label_val); \ - \ - entries_label.push_back(label_val); \ - } \ - multilabel[current_line_ind]=SGVector(num_label_entries); \ - \ - for (int32_t j=0; j < num_label_entries; j++) \ - multilabel[current_line_ind][j]=entries_label[j]; \ - \ - } \ - \ - current_line_ind++; \ - SG_PROGRESS(current_line_ind, 0, num_vec, 1, "LOADING:\t") \ - } \ - num_classes=classes.get_num_elements(); \ - \ - SG_RESET_LOCALE; \ - \ - SG_INFO("file successfully read\n") \ -} +#define GET_MULTI_LABELED_SPARSE_MATRIX(read_func, sg_type) \ + void CLibSVMFile::get_sparse_matrix( \ + SGSparseVector*& mat_feat, int32_t& num_feat, \ + int32_t& num_vec, SGVector*& multilabel, \ + int32_t& num_classes, bool load_labels) \ + { \ + num_feat = 0; \ + \ + SG_INFO("counting line numbers in file %s.\n", filename) \ + num_vec = get_num_lines(); \ + SG_INFO("File %s has %d lines.\n", filename, num_vec) \ + \ + int32_t current_line_ind = 0; \ + SGVector line; \ + \ + int32_t num_feat_entries = 0; \ + DynArray> entries_feat; \ + DynArray entries_label; \ + DynArray classes; \ + \ + mat_feat = SG_MALLOC(SGSparseVector, num_vec); \ + multilabel = SG_MALLOC(SGVector, num_vec); \ + \ + auto pb = progress(range(0, num_vec), *this->io, "LOADING: "); \ + num_classes = 0; \ + SG_SET_LOCALE_C; \ + \ + while (m_line_reader->has_next()) \ + { \ + num_feat_entries = 0; \ + entries_feat.reset(SGVector(false)); \ + line = m_line_reader->read_line(); \ + \ + m_parser->set_tokenizer(m_whitespace_tokenizer); \ + m_parser->set_text(line); \ + \ + SGVector entry_label; \ + if (load_labels && m_parser->has_next()) \ + { \ + entry_label = m_parser->read_string(); \ + if (is_feat_entry(entry_label)) \ + { \ + entries_feat.push_back(entry_label); \ + num_feat_entries++; \ + entry_label = SGVector(0); \ + } \ + } \ + \ + while (m_parser->has_next()) \ + { \ + entries_feat.push_back(m_parser->read_string()); \ + num_feat_entries++; \ + } \ + \ + mat_feat[current_line_ind] = \ + SGSparseVector(num_feat_entries); \ + for (int32_t i = 0; i < num_feat_entries; i++) \ + { \ + m_parser->set_tokenizer(m_delimiter_feat_tokenizer); \ + m_parser->set_text(entries_feat[i]); \ + \ + int32_t feat_index = 0; \ + \ + if (m_parser->has_next()) \ + feat_index = m_parser->read_int(); \ + \ + sg_type entry = 0; \ + \ + if (m_parser->has_next()) \ + entry = m_parser->read_func(); \ + \ + if (feat_index > num_feat) \ + num_feat = feat_index; \ + \ + mat_feat[current_line_ind].features[i].feat_index = \ + feat_index - 1; \ + mat_feat[current_line_ind].features[i].entry = entry; \ + } \ + \ + if (load_labels) \ + { \ + m_parser->set_tokenizer(m_delimiter_label_tokenizer); \ + m_parser->set_text(entry_label); \ + \ + int32_t num_label_entries = 0; \ + entries_label.reset(0); \ + \ + while (m_parser->has_next()) \ + { \ + num_label_entries++; \ + float64_t label_val = m_parser->read_real(); \ + \ + if (classes.find_element(label_val) == -1) \ + classes.push_back(label_val); \ + \ + entries_label.push_back(label_val); \ + } \ + multilabel[current_line_ind] = \ + SGVector(num_label_entries); \ + \ + for (int32_t j = 0; j < num_label_entries; j++) \ + multilabel[current_line_ind][j] = entries_label[j]; \ + } \ + \ + current_line_ind++; \ + pb.print_progress(); \ + } \ + pb.complete(); \ + num_classes = classes.get_num_elements(); \ + \ + SG_RESET_LOCALE; \ + \ + SG_INFO("file successfully read\n") \ + } GET_MULTI_LABELED_SPARSE_MATRIX(read_bool, bool) GET_MULTI_LABELED_SPARSE_MATRIX(read_char, int8_t) diff --git a/src/shogun/io/SGIO.cpp b/src/shogun/io/SGIO.cpp index f98b6ad38cb..4d6a44c4d24 100644 --- a/src/shogun/io/SGIO.cpp +++ b/src/shogun/io/SGIO.cpp @@ -53,20 +53,17 @@ char SGIO::file_buffer[FBUFSIZE]; char SGIO::directory_name[FBUFSIZE]; SGIO::SGIO() -: target(stdout), last_progress_time(0), progress_start_time(0), - last_progress(0), show_progress(false), location_info(MSG_NONE), - syntax_highlight(true), loglevel(MSG_WARN) + : target(stdout), show_progress(false), location_info(MSG_NONE), + syntax_highlight(true), loglevel(MSG_WARN) { m_refcount = new RefCount(); } SGIO::SGIO(const SGIO& orig) -: target(orig.get_target()), last_progress_time(0), - progress_start_time(0), last_progress(0), - show_progress(orig.get_show_progress()), - location_info(orig.get_location_info()), - syntax_highlight(orig.get_syntax_highlight()), - loglevel(orig.get_loglevel()) + : target(orig.get_target()), show_progress(orig.get_show_progress()), + location_info(orig.get_location_info()), + syntax_highlight(orig.get_syntax_highlight()), + loglevel(orig.get_loglevel()) { m_refcount = new RefCount(); } @@ -152,129 +149,6 @@ void SGIO::buffered_message(EMessageType prio, const char *fmt, ... ) const } } -void SGIO::progress( - float64_t current_val, float64_t min_val, float64_t max_val, - int32_t decimals, const char* prefix) -{ - if (!show_progress) - return; - - float64_t runtime = CTime::get_curtime(); - - char str[1000]; - float64_t v=-1, estimate=0, total_estimate=0 ; - - if (max_val-min_val>0.0) - v=100*(current_val-min_val+1)/(max_val-min_val+1); - else - return; - - v=CMath::clamp(v,1e-5,100.0); - - if (decimals < 1) - decimals = 1; - - if (last_progress==0) - { - last_progress_time = runtime; - progress_start_time = runtime; - last_progress = v; - } - else - { - last_progress = v-1e-6; - if ((v!=100.0) && (runtime - last_progress_time<0.5)) - { - - // This is made to display correctly the percentage - // if the algorithm execution is too fast - if (current_val >= max_val-1) - { - v = 100; - last_progress=v-1e-6; - snprintf(str, sizeof(str), "%%s %%%d.%df%%%% %%1.1f seconds remaining %%1.1f seconds total ",decimals+3, decimals); - message(MSG_MESSAGEONLY, "", "", -1, str, prefix, v, estimate, total_estimate); - message(MSG_MESSAGEONLY, "", "", -1, "\n"); - } - return; - } - - last_progress_time = runtime; - estimate = (1-v/100)*(last_progress_time-progress_start_time)/(v/100); - total_estimate = (last_progress_time-progress_start_time)/(v/100); - } - - if (estimate>120) - { - snprintf(str, sizeof(str), "%%s %%%d.%df%%%% %%1.1f minutes remaining %%1.1f minutes total \r",decimals+3, decimals); - message(MSG_MESSAGEONLY, "", "", -1, str, prefix, v, estimate/60, total_estimate/60); - } - else - { - snprintf(str, sizeof(str), "%%s %%%d.%df%%%% %%1.1f seconds remaining %%1.1f seconds total \r",decimals+3, decimals); - message(MSG_MESSAGEONLY, "", "", -1, str, prefix, v, estimate, total_estimate); - } - - // Print a new line if the execution is completed - // to prevent bad display - if (current_val >= max_val-1) - { - message(MSG_MESSAGEONLY, "", "", -1, "\n"); - } - - fflush(target); -} - -void SGIO::absolute_progress( - float64_t current_val, float64_t val, float64_t min_val, float64_t max_val, - int32_t decimals, const char* prefix) -{ - if (!show_progress) - return; - - float64_t runtime = CTime::get_curtime(); - - char str[1000]; - float64_t v=-1, estimate=0, total_estimate=0 ; - - if (max_val-min_val>0) - v=100*(val-min_val+1)/(max_val-min_val+1); - - if (decimals < 1) - decimals = 1; - - if (last_progress>v) - { - last_progress_time = runtime; - progress_start_time = runtime; - last_progress = v; - } - else - { - v=CMath::clamp(v,1e-5,100.0); - last_progress = v-1e-6; - - if ((v!=100.0) && (runtime - last_progress_time<100)) - return; - - last_progress_time = runtime; - estimate = (1-v/100)*(last_progress_time-progress_start_time)/(v/100); - total_estimate = (last_progress_time-progress_start_time)/(v/100); - } - - if (estimate>120) - { - snprintf(str, sizeof(str), "%%s %%%d.%df %%1.1f minutes remaining %%1.1f minutes total \r",decimals+3, decimals); - message(MSG_MESSAGEONLY, "", "", -1, str, prefix, current_val, estimate/60, total_estimate/60); - } - else - { - snprintf(str, sizeof(str), "%%s %%%d.%df %%1.1f seconds remaining %%1.1f seconds total \r",decimals+3, decimals); - message(MSG_MESSAGEONLY, "", "", -1, str, prefix, current_val, estimate, total_estimate); - } - - fflush(target); -} void SGIO::done() { diff --git a/src/shogun/io/SGIO.h b/src/shogun/io/SGIO.h index 27086b9d971..e041bc1dac3 100644 --- a/src/shogun/io/SGIO.h +++ b/src/shogun/io/SGIO.h @@ -136,23 +136,9 @@ __FILE__ ":" func ": Unstable method! Please report if it seems to " \ #define SG_PRINT(...) { io->message(MSG_MESSAGEONLY, __PRETTY_FUNCTION__, __FILE__, __LINE__, __VA_ARGS__); } #define SG_OBJ_PRINT(o, ...) { o->io->message(MSG_MESSAGEONLY, __PRETTY_FUNCTION__, __FILE__, __LINE__, __VA_ARGS__); } #define SG_NOTIMPLEMENTED { io->not_implemented(__PRETTY_FUNCTION__, __FILE__, __LINE__); } +#define SG_GPL_ONLY { io->gpl_only(__PRETTY_FUNCTION__, __FILE__, __LINE__); } #define SG_DEPRECATED { io->deprecated(__PRETTY_FUNCTION__, __FILE__, __LINE__); } -#define SG_PROGRESS(...) { \ - if (SG_UNLIKELY(io->get_show_progress())) \ - io->progress(__VA_ARGS__); \ -} - -#define SG_OBJ_PROGRESS(o, ...) { \ - if (SG_UNLIKELY(o->io->get_show_progress()))\ - o->io->progress(__VA_ARGS__); \ -} - -#define SG_ABS_PROGRESS(...) { \ - if (SG_UNLIKELY(io->get_show_progress())) \ - io->absolute_progress(__VA_ARGS__); \ -} - #define SG_DONE() { \ if (SG_UNLIKELY(io->get_show_progress())) \ io->done(); \ @@ -178,23 +164,13 @@ __FILE__ ":" func ": Unstable method! Please report if it seems to " \ #define SG_SERROR(...) { sg_io->message(MSG_ERROR,__PRETTY_FUNCTION__, __FILE__, __LINE__, __VA_ARGS__); } #define SG_SPRINT(...) { sg_io->message(MSG_MESSAGEONLY,__PRETTY_FUNCTION__, __FILE__, __LINE__, __VA_ARGS__); } - -#define SG_SPROGRESS(...) { \ - if (SG_UNLIKELY(sg_io->get_show_progress())) \ - sg_io->progress(__VA_ARGS__); \ -} - -#define SG_SABS_PROGRESS(...) { \ - if (SG_UNLIKELY(sg_io->get_show_progress())) \ - sg_io->absolute_progress(__VA_ARGS__); \ -} - #define SG_SDONE() { \ if (SG_UNLIKELY(sg_io->get_show_progress())) \ sg_io->done(); \ } #define SG_SNOTIMPLEMENTED { sg_io->not_implemented(__PRETTY_FUNCTION__, __FILE__, __LINE__); } +#define SG_SGPL_ONLY { sg_io->gpl_only(__PRETTY_FUNCTION__, __FILE__, __LINE__); } #define SG_SDEPRECATED { sg_io->deprecated(__PRETTY_FUNCTION__, __FILE__, __LINE__); } #define ASSERT(x) { \ @@ -288,12 +264,6 @@ class SGIO return location_info; } - /** @return last progress as a percentage */ - inline float64_t get_last_progress() const - { - return last_progress; - } - /** get syntax highlight * * @return if syntax highlighting is enabled @@ -317,33 +287,6 @@ class SGIO void message(EMessageType prio, const char* function, const char* file, int32_t line, const char *fmt, ... ) const; - /** print progress bar - * - * @param current_val current value - * @param min_val minimum value - * @param max_val maximum value - * @param decimals decimals - * @param prefix message prefix - */ - void progress( - float64_t current_val, - float64_t min_val=0.0, float64_t max_val=1.0, int32_t decimals=1, - const char* prefix="PROGRESS:\t"); - - /** print absolute progress bar - * - * @param current_val current value - * @param val value - * @param min_val minimum value - * @param max_val maximum value - * @param decimals decimals - * @param prefix message prefix - */ - void absolute_progress( - float64_t current_val, float64_t val, - float64_t min_val=0.0, float64_t max_val=1.0, int32_t decimals=1, - const char* prefix="PROGRESS:\t"); - /** print 'done' with priority INFO, * but only if progress bar is enabled * @@ -356,6 +299,12 @@ class SGIO message(MSG_ERROR, function, file, line, "Sorry, not yet implemented .\n"); } + /** print error message 'Only available with GPL parts.' */ + inline void gpl_only(const char* function, const char* file, int32_t line) const + { + message(MSG_ERROR, function, file, line, "This feature is only available if Shogun is built with GPL codes.\n"); + } + /** print warning message 'function deprecated' */ inline void deprecated(const char* function, const char* file, int32_t line) const { @@ -567,12 +516,6 @@ class SGIO protected: /** target file */ FILE* target; - /** last progress time */ - float64_t last_progress_time; - /** progress start time */ - float64_t progress_start_time; - /** last progress */ - float64_t last_progress; /** if progress bar shall be shown */ bool show_progress; /** if each print function should append filename and linenumber of diff --git a/src/shogun/io/TBOutputFormat.cpp b/src/shogun/io/TBOutputFormat.cpp new file mode 100644 index 00000000000..9478dde26d6 --- /dev/null +++ b/src/shogun/io/TBOutputFormat.cpp @@ -0,0 +1,156 @@ +/* +* BSD 3-Clause License +* +* Copyright (c) 2017, Shogun-Toolbox e.V. +* All rights reserved. +* +* Redistribution and use in source and binary forms, with or without +* modification, are permitted provided that the following conditions are met: +* +* * Redistributions of source code must retain the above copyright notice, this +* list of conditions and the following disclaimer. +* +* * Redistributions in binary form must reproduce the above copyright notice, +* this list of conditions and the following disclaimer in the documentation +* and/or other materials provided with the distribution. +* +* * Neither the name of the copyright holder nor the names of its +* contributors may be used to endorse or promote products derived from +* this software without specific prior written permission. +* +* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +* AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE +* DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE +* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL +* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR +* SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER +* CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, +* OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +* OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +* +* Written (W) 2017 Giovanni De Toni +* +*/ +#include +#ifdef HAVE_TFLOGGER + +#include +#include +#include +#include +#include + +using namespace shogun; + +#define CHECK_TYPE(type) \ + else if ( \ + value.first.get_value().type_info().hash_code() == \ + typeid(type).hash_code()) \ + { \ + summaryValue->set_simple_value( \ + recall_type(value.first.get_value())); \ + } + +#define CHECK_TYPE_HISTO(type) \ + else if ( \ + value.first.get_value().type_info().hash_code() == \ + typeid(type).hash_code()) \ + { \ + tensorflow::histogram::Histogram h; \ + tensorflow::HistogramProto* hp = new tensorflow::HistogramProto(); \ + auto v = recall_type(value.first.get_value()); \ + for (auto value_v : v) \ + h.Add(value_v); \ + h.EncodeToProto(hp, true); \ + summaryValue->set_allocated_histo(hp); \ + } + +TBOutputFormat::TBOutputFormat(){}; + +TBOutputFormat::~TBOutputFormat(){}; + +tensorflow::Event TBOutputFormat::convert_scalar( + const TimedObservedValue& value, std::string& node_name) +{ + tensorflow::Event e; + std::time_t now_t = convert_to_millis(value.second); + e.set_wall_time(now_t); + e.set_step(value.first.get_step()); + + tensorflow::Summary* summary = e.mutable_summary(); + auto summaryValue = summary->add_value(); + summaryValue->set_tag(value.first.get_name()); + summaryValue->set_node_name(node_name); + + if (value.first.get_value().type_info().hash_code() == + typeid(int8_t).hash_code()) + { + summaryValue->set_simple_value( + recall_type(value.first.get_value())); + } + CHECK_TYPE(uint8_t) + CHECK_TYPE(int16_t) + CHECK_TYPE(uint16_t) + CHECK_TYPE(int32_t) + CHECK_TYPE(uint32_t) + CHECK_TYPE(int64_t) + CHECK_TYPE(uint64_t) + CHECK_TYPE(float32_t) + CHECK_TYPE(float64_t) + CHECK_TYPE(floatmax_t) + CHECK_TYPE(char) + else + { + SG_ERROR( + "Unsupported type %s", value.first.get_value().type_info().name()); + } + + return e; +} + +tensorflow::Event TBOutputFormat::convert_vector( + const TimedObservedValue& value, std::string& node_name) +{ + tensorflow::Event e; + std::time_t now_t = convert_to_millis(value.second); + e.set_wall_time(now_t); + e.set_step(value.first.get_step()); + + tensorflow::Summary* summary = e.mutable_summary(); + auto summaryValue = summary->add_value(); + summaryValue->set_tag(value.first.get_name()); + summaryValue->set_node_name(node_name); + + if (value.first.get_value().type_info().hash_code() == + typeid(std::vector).hash_code()) + { + tensorflow::histogram::Histogram h; + tensorflow::HistogramProto* hp = new tensorflow::HistogramProto(); + auto v = recall_type>(value.first.get_value()); + for (auto value_v : v) + h.Add(value_v); + h.EncodeToProto(hp, true); + summaryValue->set_allocated_histo(hp); + } + CHECK_TYPE_HISTO(std::vector) + CHECK_TYPE_HISTO(std::vector) + CHECK_TYPE_HISTO(std::vector) + CHECK_TYPE_HISTO(std::vector) + CHECK_TYPE_HISTO(std::vector) + CHECK_TYPE_HISTO(std::vector) + CHECK_TYPE_HISTO(std::vector) + CHECK_TYPE_HISTO(std::vector) + CHECK_TYPE_HISTO(std::vector) + CHECK_TYPE_HISTO(std::vector) + CHECK_TYPE_HISTO(std::vector) + else + { + SG_ERROR( + "Unsupported type %s", value.first.get_value().type_info().name()); + } + + return e; +} + +#endif // HAVE_TFLOGGER diff --git a/src/shogun/io/TBOutputFormat.h b/src/shogun/io/TBOutputFormat.h new file mode 100644 index 00000000000..bffcab33ae0 --- /dev/null +++ b/src/shogun/io/TBOutputFormat.h @@ -0,0 +1,82 @@ +/* +* BSD 3-Clause License +* +* Copyright (c) 2017, Shogun-Toolbox e.V. +* All rights reserved. +* +* Redistribution and use in source and binary forms, with or without +* modification, are permitted provided that the following conditions are met: +* +* * Redistributions of source code must retain the above copyright notice, this +* list of conditions and the following disclaimer. +* +* * Redistributions in binary form must reproduce the above copyright notice, +* this list of conditions and the following disclaimer in the documentation +* and/or other materials provided with the distribution. +* +* * Neither the name of the copyright holder nor the names of its +* contributors may be used to endorse or promote products derived from +* this software without specific prior written permission. +* +* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +* AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE +* DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE +* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL +* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR +* SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER +* CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, +* OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +* OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +* +* Written (W) 2017 Giovanni De Toni +* +*/ +#include +#ifdef HAVE_TFLOGGER + +#ifndef SHOGUN_OUTPUTFORMAT_H +#define SHOGUN_OUTPUTFORMAT_H + +#include +#include +#include +#include + +#include + +namespace shogun +{ + /** + * Convert an std::pair to a tensorflow::Event, + * which can be written to file and used with tools like Tensorboard. + */ + class TBOutputFormat : public CSGObject + { + + public: + TBOutputFormat(); + ~TBOutputFormat(); + + /** + * Generate a tensorflow::Event object give some informations + * @param event_step the current event step + * @param value the value which will be converted to tensorflow::Event + * @param node_name the node name (default: node) + * @return the newly created tensorflow::Event + */ + tensorflow::Event + convert_scalar(const TimedObservedValue& value, std::string& node_name); + + tensorflow::Event + convert_vector(const TimedObservedValue& value, std::string& node_name); + + virtual const char* get_name() const + { + return "TFLogger"; + } + }; +} + +#endif // SHOGUN_OUTPUTFORMAT_H +#endif // HAVE_TFLOGGER diff --git a/src/shogun/io/UAIFile.cpp b/src/shogun/io/UAIFile.cpp index f93e82f05cb..e12ac98ef00 100644 --- a/src/shogun/io/UAIFile.cpp +++ b/src/shogun/io/UAIFile.cpp @@ -104,30 +104,34 @@ void CUAIFile::init_with_defaults() SG_REF(m_line_reader); } -#define GET_VECTOR(read_func, sg_type) \ -void CUAIFile::get_vector(sg_type*& vector, int32_t& len) \ -{ \ - if (!m_line_reader->has_next()) \ - return; \ - \ - SGVector line; \ - int32_t num_elements = 0; \ - \ - line = m_line_reader->read_line(); \ - m_tokenizer->set_text(line); \ - while (m_tokenizer->has_next()) \ - { \ - int32_t temp_start; \ - m_tokenizer->next_token_idx(temp_start); \ - num_elements++; \ - } \ - \ - vector = SG_MALLOC(sg_type, num_elements); \ - m_parser->set_text(line); \ - for (int32_t i=0; iread_func(); \ - len = num_elements; \ -} +#define GET_VECTOR(read_func, sg_type) \ + void CUAIFile::get_vector(sg_type*& vector, int32_t& len) \ + { \ + SG_SET_LOCALE_C; \ + \ + if (!m_line_reader->has_next()) \ + return; \ + \ + SGVector line; \ + int32_t num_elements = 0; \ + \ + line = m_line_reader->read_line(); \ + m_tokenizer->set_text(line); \ + while (m_tokenizer->has_next()) \ + { \ + int32_t temp_start; \ + m_tokenizer->next_token_idx(temp_start); \ + num_elements++; \ + } \ + \ + vector = SG_MALLOC(sg_type, num_elements); \ + m_parser->set_text(line); \ + for (int32_t i = 0; i < num_elements; i++) \ + vector[i] = m_parser->read_func(); \ + len = num_elements; \ + \ + SG_RESET_LOCALE; \ + } GET_VECTOR(read_char, int8_t) GET_VECTOR(read_byte, uint8_t) diff --git a/src/shogun/kernel/CombinedKernel.cpp b/src/shogun/kernel/CombinedKernel.cpp index 9c39a2c8fb7..cfab3c28aad 100644 --- a/src/shogun/kernel/CombinedKernel.cpp +++ b/src/shogun/kernel/CombinedKernel.cpp @@ -121,8 +121,13 @@ bool CCombinedKernel::init(CFeatures* l, CFeatures* r) // skip over features - the custom kernel does not need any if (k->get_kernel_type() != K_CUSTOM) { - lf = ((CCombinedFeatures*) l)->get_feature_obj(f_idx); - rf = ((CCombinedFeatures*) r)->get_feature_obj(f_idx); + if (((CCombinedFeatures*)l)->get_num_feature_obj() > f_idx && + ((CCombinedFeatures*)r)->get_num_feature_obj() > f_idx) + { + lf = ((CCombinedFeatures*)l)->get_feature_obj(f_idx); + rf = ((CCombinedFeatures*)r)->get_feature_obj(f_idx); + } + f_idx++; if (!lf || !rf) { diff --git a/src/shogun/kernel/CombinedKernel.h b/src/shogun/kernel/CombinedKernel.h index b139a0ac029..185c4fb3e93 100644 --- a/src/shogun/kernel/CombinedKernel.h +++ b/src/shogun/kernel/CombinedKernel.h @@ -124,7 +124,14 @@ class CCombinedKernel : public CKernel */ inline CKernel* get_kernel(int32_t idx) { - return (CKernel*) kernel_array->get_element(idx); + if (idx < get_num_kernels()) + { + return (CKernel*)kernel_array->get_element(idx); + } + else + { + return 0; + } } /** get last kernel diff --git a/src/shogun/kernel/ExponentialARDKernel.cpp b/src/shogun/kernel/ExponentialARDKernel.cpp index 349ef96c800..564f1c65c8e 100644 --- a/src/shogun/kernel/ExponentialARDKernel.cpp +++ b/src/shogun/kernel/ExponentialARDKernel.cpp @@ -14,7 +14,6 @@ #include #include #include -#include using namespace shogun; @@ -89,11 +88,7 @@ void CExponentialARDKernel::lazy_update_weights() if (m_ARD_type==KT_SCALAR || m_ARD_type==KT_DIAG) { SGMatrix log_weights(m_log_weights.vector,1,m_log_weights.vlen,false); - m_weights_raw=linalg::elementwise_compute(m_log_weights, - [ ](float64_t& value) - { - return CMath::exp(value); - }); + m_weights_raw = linalg::exponent(m_log_weights); } else if (m_ARD_type==KT_FULL) { @@ -235,12 +230,8 @@ SGMatrix CExponentialARDKernel::get_weighted_vector(SGVector rtmp(vec.vector,vec.vlen,1,false); - SGMatrix weights=linalg::elementwise_compute(m_log_weights, - [ ](float64_t& value) - { - return CMath::exp(value); - }); - res=linalg::element_prod(weights, rtmp); + SGMatrix weights(linalg::exponent(m_log_weights)); + res = linalg::element_prod(weights, rtmp); } return res; } diff --git a/src/shogun/kernel/Kernel.cpp b/src/shogun/kernel/Kernel.cpp index 2f8303d799d..41b4b0baaa0 100644 --- a/src/shogun/kernel/Kernel.cpp +++ b/src/shogun/kernel/Kernel.cpp @@ -12,12 +12,13 @@ * Copyright (C) 1999-2009 Fraunhofer Institute FIRST and Max-Planck-Society */ -#include -#include -#include +#include #include -#include +#include #include +#include +#include +#include #include @@ -984,8 +985,6 @@ template struct K_THREAD_PARAM int32_t end; /** start (unit number of elements) */ int64_t total_start; - /** end (unit number of elements) */ - int64_t total_end; /** m */ int32_t m; /** n */ @@ -996,6 +995,8 @@ template struct K_THREAD_PARAM bool symmetric; /** output progress */ bool verbose; + /* Progress bar*/ + PRange* pb; }; } @@ -1274,8 +1275,8 @@ template void* CKernel::get_kernel_matrix_helper(void* p) int32_t m=params->m; bool verbose=params->verbose; int64_t total_start=params->total_start; - int64_t total_end=params->total_end; int64_t total=total_start; + PRange* pb = params->pb; for (int32_t i=i_start; i void* CKernel::get_kernel_matrix_helper(void* p) if (symmetric && i!=j) total++; - if (total%100 == 0) - SG_OBJ_PROGRESS(k, total, total_start, total_end) + pb->print_progress(); - if (CSignal::cancel_computations()) - break; + // TODO: replace with the new signal + // if (CSignal::cancel_computations()) + // break; } } @@ -1335,7 +1336,8 @@ SGMatrix CKernel::get_kernel_matrix() K_THREAD_PARAM params; int64_t step = total_num/num_threads; index_t t = 0; - #pragma omp parallel for lastprivate(t) private(params) + auto pb = progress(range(total_num), *this->io); +#pragma omp parallel for lastprivate(t) private(params) for (t = 0; t < num_threads; ++t) { params.kernel = this; @@ -1343,11 +1345,11 @@ SGMatrix CKernel::get_kernel_matrix() params.start = compute_row_start(t*step, n, symmetric); params.end = compute_row_start((t+1)*step, n, symmetric); params.total_start=t*step; - params.total_end=(t+1)*step; params.n=n; params.m=m; params.symmetric=symmetric; params.verbose=false; + params.pb = &pb; CKernel::get_kernel_matrix_helper((void*)¶ms); } @@ -1358,15 +1360,15 @@ SGMatrix CKernel::get_kernel_matrix() params.start = compute_row_start(t*step, n, symmetric); params.end = m; params.total_start=t*step; - params.total_end=total_num; params.n=n; params.m=m; params.symmetric=symmetric; params.verbose=false; + params.pb = &pb; CKernel::get_kernel_matrix_helper((void*)¶ms); } - SG_DONE() + pb.complete(); return SGMatrix(result,m,n,true); } @@ -1377,4 +1379,3 @@ template SGMatrix CKernel::get_kernel_matrix(); template void* CKernel::get_kernel_matrix_helper(void* p); template void* CKernel::get_kernel_matrix_helper(void* p); - diff --git a/src/shogun/kernel/Kernel.h b/src/shogun/kernel/Kernel.h index 88e66fb413b..4768cd9f3eb 100644 --- a/src/shogun/kernel/Kernel.h +++ b/src/shogun/kernel/Kernel.h @@ -18,7 +18,6 @@ #include #include -#include #include #include #include diff --git a/src/shogun/kernel/string/CommUlongStringKernel.cpp b/src/shogun/kernel/string/CommUlongStringKernel.cpp index 6936dab8967..bfba00ee012 100644 --- a/src/shogun/kernel/string/CommUlongStringKernel.cpp +++ b/src/shogun/kernel/string/CommUlongStringKernel.cpp @@ -8,11 +8,13 @@ * Copyright (C) 1999-2009 Fraunhofer Institute FIRST and Max-Planck-Society */ -#include -#include -#include +#include #include #include +#include +#include + +#include using namespace shogun; @@ -231,11 +233,8 @@ bool CCommUlongStringKernel::init_optimization( SG_DEBUG("initializing CCommUlongStringKernel optimization\n") - for (int32_t i=0; iio)) { - if ( (i % (count/10+1)) == 0) - SG_PROGRESS(i, 0, count) - add_to_normal(IDX[i], weights[i]); } diff --git a/src/shogun/kernel/string/CommWordStringKernel.cpp b/src/shogun/kernel/string/CommWordStringKernel.cpp index 35450c378ba..f2dfb0f53de 100644 --- a/src/shogun/kernel/string/CommWordStringKernel.cpp +++ b/src/shogun/kernel/string/CommWordStringKernel.cpp @@ -8,14 +8,15 @@ * Copyright (C) 1999-2009 Fraunhofer Institute FIRST and Max-Planck-Society */ -#include -#include - #include +#include +#include +#include +#include #include + #include -#include using namespace shogun; @@ -45,20 +46,17 @@ CCommWordStringKernel::CCommWordStringKernel( bool CCommWordStringKernel::init_dictionary(int32_t size) { - dictionary_size= size; - SG_FREE(dictionary_weights); - dictionary_weights=SG_MALLOC(float64_t, size); + dictionary_weights=SGVector(size); SG_DEBUG("using dictionary of %d words\n", size) clear_normal(); - return dictionary_weights!=NULL; + return dictionary_weights.vector!=NULL; } CCommWordStringKernel::~CCommWordStringKernel() { cleanup(); - SG_FREE(dictionary_weights); SG_FREE(dict_diagonal_optimization); } @@ -97,7 +95,7 @@ float64_t CCommWordStringKernel::compute_diag(int32_t idx_a) ASSERT((1<<(sizeof(uint16_t)*8)) > alen) int32_t num_symbols=(int32_t) l->get_num_symbols(); - ASSERT(num_symbols<=dictionary_size) + ASSERT(num_symbols<=dictionary_weights.vlen) int32_t* dic = dict_diagonal_optimization; memset(dic, 0, num_symbols*sizeof(int32_t)); @@ -285,7 +283,7 @@ void CCommWordStringKernel::add_to_normal(int32_t vec_idx, float64_t weight) void CCommWordStringKernel::clear_normal() { - memset(dictionary_weights, 0, dictionary_size*sizeof(float64_t)); + dictionary_weights.zero(); set_is_initialized(false); } @@ -303,11 +301,8 @@ bool CCommWordStringKernel::init_optimization( SG_DEBUG("initializing CCommWordStringKernel optimization\n") - for (int32_t i=0; iio)) { - if ( (i % (count/10+1)) == 0) - SG_PROGRESS(i, 0, count) - add_to_normal(IDX[i], weights[i]); } @@ -602,9 +597,6 @@ char* CCommWordStringKernel::compute_consensus( void CCommWordStringKernel::init() { - dictionary_size=0; - dictionary_weights=NULL; - use_sign=false; use_dict_diagonal_optimization=false; dict_diagonal_optimization=NULL; @@ -613,8 +605,8 @@ void CCommWordStringKernel::init() init_dictionary(1<<(sizeof(uint16_t)*8)); set_normalizer(new CSqrtDiagKernelNormalizer(use_dict_diagonal_optimization)); - m_parameters->add_vector(&dictionary_weights, &dictionary_size, "dictionary_weights", - "Dictionary for applying kernel."); + SG_ADD(&dictionary_weights, "dictionary_weights", + "Dictionary for applying kernel.", MS_NOT_AVAILABLE); SG_ADD(&use_sign, "use_sign", "If signum(counts) is used instead of counts.", MS_AVAILABLE); SG_ADD(&use_dict_diagonal_optimization, diff --git a/src/shogun/kernel/string/CommWordStringKernel.h b/src/shogun/kernel/string/CommWordStringKernel.h index 841bb31ccee..3011c5f30fa 100644 --- a/src/shogun/kernel/string/CommWordStringKernel.h +++ b/src/shogun/kernel/string/CommWordStringKernel.h @@ -149,13 +149,11 @@ class CCommWordStringKernel : public CStringKernel /** get dictionary * - * @param dsize dictionary size will be stored in here - * @param dweights dictionary weights will be stored in here + * @return dictionary weights */ - void get_dictionary(int32_t& dsize, float64_t*& dweights) + SGVector get_dictionary() const { - dsize=dictionary_size; - dweights = dictionary_weights; + return dictionary_weights; } /** compute scoring @@ -240,11 +238,9 @@ class CCommWordStringKernel : public CStringKernel void init(); protected: - /** size of dictionary (number of possible strings) */ - int32_t dictionary_size; /** dictionary weights - array to hold counters for all possible * strings */ - float64_t* dictionary_weights; + SGVector dictionary_weights; /** if sign shall be used */ bool use_sign; diff --git a/src/shogun/kernel/string/DistantSegmentsKernel.cpp b/src/shogun/kernel/string/DistantSegmentsKernel.cpp deleted file mode 100644 index a9e9407a7d4..00000000000 --- a/src/shogun/kernel/string/DistantSegmentsKernel.cpp +++ /dev/null @@ -1,128 +0,0 @@ -/* - * This program is free software; you can redistribute it and/or modify - * it under the terms of the GNU General Public License as published by - * the Free Software Foundation; either version 3 of the License, or - * (at your option) any later version. - * - * Written (W) 2011 Heiko Strathmann - * DS-Kernel implementation Written (W) 2008 Sébastien Boisvert under GPLv3 - * Copyright (C) 2011 Berlin Institute of Technology and Max-Planck-Society - */ - -#include -#include - -using namespace shogun; - -CDistantSegmentsKernel::CDistantSegmentsKernel() : CStringKernel(), - m_delta(0), m_theta(0) -{ - init(); -} - -CDistantSegmentsKernel::CDistantSegmentsKernel(int32_t size, int32_t delta, - int32_t theta) : CStringKernel(size), m_delta(delta), - m_theta(theta) -{ - init(); -} - -CDistantSegmentsKernel::CDistantSegmentsKernel(CStringFeatures* l, - CStringFeatures* r, int32_t size, int32_t delta, int32_t theta) : - CStringKernel(size), m_delta(delta), m_theta(theta) -{ - init(); - CStringKernel::init(l, r); -} - -bool CDistantSegmentsKernel::init(CFeatures* l, CFeatures* r) -{ - CKernel::init(l, r); - return init_normalizer(); -} - -void CDistantSegmentsKernel::init() -{ - SG_ADD(&m_delta, "delta", "Delta parameter of the DS-Kernel", MS_AVAILABLE); - SG_ADD(&m_theta, "theta", "Theta parameter of the DS-Kernel", MS_AVAILABLE); -} - -float64_t CDistantSegmentsKernel::compute(int32_t idx_a, int32_t idx_b) -{ - bool free_a, free_b; - int32_t aLength=0, bLength=0; - char* a=((CStringFeatures*) lhs)->get_feature_vector(idx_a, aLength, - free_a); - char* b=((CStringFeatures*) rhs)->get_feature_vector(idx_b, bLength, - free_b); - ASSERT(a && b) - - if ((aLength<1)||(bLength<1)) - SG_ERROR("Empty sequences") - - float64_t result=compute(a, aLength, b, bLength, m_delta, m_theta); - - ((CStringFeatures*) lhs)->free_feature_vector(a, idx_a, free_a); - ((CStringFeatures*) rhs)->free_feature_vector(b, idx_b, free_b); - - return result; -} - -int32_t CDistantSegmentsKernel::bin(int32_t j, int32_t i) -{ - if (i>j) - return 0; - if (i==3 && j>=3) - { - return j*(j-1)*(j-2)/6; - } - else if (i==2 && j>=2) - { - return j*(j-1)/2; - } - return 0; -} - -int32_t CDistantSegmentsKernel::compute(char* s, int32_t sLength, char* t, - int32_t tLength, int32_t delta_m, int32_t theta_m) -{ - int32_t c=0; - int32_t* i_=SG_MALLOC(int32_t, delta_m+1); - int32_t* l_=SG_MALLOC(int32_t, delta_m+1); - for (int32_t j_s=0; j_s<=(int32_t) sLength-1; j_s++) - { - for (int32_t j_t=0; j_t<=(int32_t) tLength-1; j_t++) - { - if (s[j_s-1+1]==t[j_t-1+1]) - { - int32_t n=CMath::min(CMath::min(sLength-j_s, tLength-j_t), delta_m); - int32_t k=-1; - int32_t i=1; - while (i<=n) - { - k++; - i_[2*k]=i; - i++; - while (i<=n&&s[j_s-1+i]==t[j_t-1+i]) - i++; - i_[2*k+1]=i; - l_[k]=i_[2*k+1]-i_[2*k]+1; - i++; - while (i<=n&&s[j_s-1+i]!=t[j_t-1+i]) - i++; - } - c+=bin(l_[0], 3)-2*bin(l_[0]-theta_m, 3) - +bin(l_[0]-2*theta_m, 3); - int32_t c1=0; - for (int32_t r=1; r<=k; r++) - { - c1+=bin(l_[r], 2)-bin(l_[r]-theta_m, 2); - } - c+=CMath::min(theta_m, i_[1]-i_[0])*c1; - } - } - } - SG_FREE(l_); - SG_FREE(i_); - return c; -} diff --git a/src/shogun/kernel/string/DistantSegmentsKernel.h b/src/shogun/kernel/string/DistantSegmentsKernel.h deleted file mode 100644 index aedd5b5595d..00000000000 --- a/src/shogun/kernel/string/DistantSegmentsKernel.h +++ /dev/null @@ -1,135 +0,0 @@ -/* - * This program is free software; you can redistribute it and/or modify - * it under the terms of the GNU General Public License as published by - * the Free Software Foundation; either version 3 of the License, or - * (at your option) any later version. - * - * Written (W) 2011 Heiko Strathmann - * DS-Kernel implementation Written (W) 2008 Sébastien Boisvert under GPLv3 - * Copyright (C) 2011 Berlin Institute of Technology and Max-Planck-Society - */ - -#ifndef DISTANTSEGMENTSKERNEL_H_ -#define DISTANTSEGMENTSKERNEL_H_ - -#include - -#include - -namespace shogun -{ - -/** - * @brief The distant segments kernel is a string kernel, - * which counts the number of substrings, so-called segments, - * at a certain distance from each other. - * - * The implementation is taken from - * http://www.retrovirology.com/content/5/1/110/ and - * only adjusted to work with shogun. See that page for any details. - * - * Reference: Sebastien Boisvert, Mario Marchand, Francois Laviolette, - * and Jacques Corbeil. Hiv-1 coreceptor usage prediction without - * multiple alignments: an application of string kernels. - * Retrovirology, 5(1):110, Dec 2008. - */ -class CDistantSegmentsKernel: public CStringKernel -{ -public: - /** default constructor */ - CDistantSegmentsKernel(); - - /** constructor - * - * @param size cache size - * @param delta \f[\delta\f]-parameter of the DS-kernel - * @param theta \f[\theta\f]-parameter of the DS-kernel - */ - CDistantSegmentsKernel(int32_t size, int32_t delta, int32_t theta); - - /** constructor - * - * @param l features of left-side - * @param r features of right-side - * @param size cache size - * @param delta \f[\delta\f]-parameter of the DS-kernel - * @param theta \f[\theta\f]-parameter of the DS-kernel - */ - CDistantSegmentsKernel(CStringFeatures* l, CStringFeatures* r, - int32_t size, int32_t delta, int32_t theta); - - /** initialize kernel with features - * - * @param l features of left-side - * @param r features of right-side - * @return true if successful - */ - virtual bool init(CFeatures* l, CFeatures* r); - - /** - * @return kernel type - */ - virtual EKernelType get_kernel_type() - { - return K_DISTANTSEGMENTS; - } - - /** - * @return name of kernel - */ - virtual const char* get_name() const - { - return "DistantSegmentsKernel"; - } - -protected: - /** - * compute kernel function for features a and b - * idx_{a,b} denote the index of the feature vectors - * in the corresponding feature object - * - * @param idx_a index a - * @param idx_b index b - * @return computed kernel function at indices a,b - */ - virtual float64_t compute(int32_t idx_a, int32_t idx_b); - -private: - /** initializes kernel parameters and registers them */ - void init(); - - /** - * helper function taken from - * http://www.retrovirology.com/content/5/1/110/ - * */ - int32_t bin(int32_t j, int32_t i); - - /** - * Computes the DS-kernel for the given strings and parameters. - * Taken from http://www.retrovirology.com/content/5/1/110/ - * with little adjustments. - * - * @param s first string for kernel computation - * @param sLength length of that string - * @param b second string for kernel computation - * @param bLength length of that string - * @param delta_m delta parameter - * @param theta_m theta parameter - * @return computed kernel function of the given strings and parameters - */ - int32_t compute(char* s, int32_t sLength, char* b, int32_t bLength, - int32_t delta_m, int32_t theta_m); - -protected: - /** delta parameter of DS-kernel */ - int32_t m_delta; - - /** theta parameter of DS-kernel */ - int32_t m_theta; - - -}; - -} - -#endif /* DISTANTSEGMENTSKERNEL_H_ */ diff --git a/src/shogun/kernel/string/LinearStringKernel.cpp b/src/shogun/kernel/string/LinearStringKernel.cpp index 1d0fddf3fd3..2b43cecf2a6 100644 --- a/src/shogun/kernel/string/LinearStringKernel.cpp +++ b/src/shogun/kernel/string/LinearStringKernel.cpp @@ -10,20 +10,20 @@ #include #include -#include +#include #include #include using namespace shogun; CLinearStringKernel::CLinearStringKernel() -: CStringKernel(0), normal(NULL) +: CStringKernel(0) { } CLinearStringKernel::CLinearStringKernel( CStringFeatures* l, CStringFeatures* r) -: CStringKernel(0), normal(NULL) +: CStringKernel(0) { init(l, r); } @@ -48,7 +48,7 @@ void CLinearStringKernel::cleanup() void CLinearStringKernel::clear_normal() { - memset(normal, 0, lhs->get_num_vectors()*sizeof(float64_t)); + memset(m_normal.vector, 0, lhs->get_num_vectors()*sizeof(float64_t)); } void CLinearStringKernel::add_to_normal(int32_t idx, float64_t weight) @@ -58,7 +58,7 @@ void CLinearStringKernel::add_to_normal(int32_t idx, float64_t weight) char* vec = ((CStringFeatures*) lhs)->get_feature_vector(idx, vlen, vfree); for (int32_t i=0; inormalize_lhs(vec[i], idx); + m_normal.vector[i] += weight*normalizer->normalize_lhs(vec[i], idx); ((CStringFeatures*) lhs)->free_feature_vector(vec, idx, vfree); } @@ -71,7 +71,9 @@ float64_t CLinearStringKernel::compute(int32_t idx_a, int32_t idx_b) char* avec = ((CStringFeatures*) lhs)->get_feature_vector(idx_a, alen, free_avec); char* bvec = ((CStringFeatures*) rhs)->get_feature_vector(idx_b, blen, free_bvec); ASSERT(alen==blen) - float64_t result=CMath::dot(avec, bvec, alen); + SGVector a_wrap(avec, alen, false); + SGVector b_wrap(bvec, blen, false); + float64_t result = linalg::dot(a_wrap, b_wrap); ((CStringFeatures*) lhs)->free_feature_vector(avec, idx_a, free_avec); ((CStringFeatures*) rhs)->free_feature_vector(bvec, idx_b, free_bvec); return result; @@ -83,8 +85,8 @@ bool CLinearStringKernel::init_optimization( int32_t num_feat = ((CStringFeatures*) lhs)->get_max_vector_length(); ASSERT(num_feat) - normal = SG_MALLOC(float64_t, num_feat); - ASSERT(normal) + m_normal = SGVector(num_feat); + ASSERT(m_normal.vector) clear_normal(); for (int32_t i = 0; inormalize_lhs(((float64_t) avec[j]), sv_idx[i]); } ((CStringFeatures*) lhs)->free_feature_vector(avec, sv_idx[i], free_avec); @@ -107,8 +109,7 @@ bool CLinearStringKernel::init_optimization( bool CLinearStringKernel::delete_optimization() { - SG_FREE(normal); - normal = NULL; + m_normal = SGVector(); set_is_initialized(false); return true; } @@ -118,7 +119,10 @@ float64_t CLinearStringKernel::compute_optimized(int32_t idx_b) int32_t blen; bool free_bvec; char* bvec = ((CStringFeatures*) rhs)->get_feature_vector(idx_b, blen, free_bvec); - float64_t result=normalizer->normalize_rhs(CMath::dot(normal, bvec, blen), idx_b); + float64_t dot = 0.0; + for (auto i = 0; m_normal.vlen; ++i) + dot += m_normal[i]*(float64_t)bvec[i]; + float64_t result=normalizer->normalize_rhs(dot, idx_b); ((CStringFeatures*) rhs)->free_feature_vector(bvec, idx_b, free_bvec); return result; } diff --git a/src/shogun/kernel/string/LinearStringKernel.h b/src/shogun/kernel/string/LinearStringKernel.h index 5b59c13ea78..3c03db01d5f 100644 --- a/src/shogun/kernel/string/LinearStringKernel.h +++ b/src/shogun/kernel/string/LinearStringKernel.h @@ -117,7 +117,7 @@ class CLinearStringKernel: public CStringKernel protected: /** normal vector (used in case of optimized kernel) */ - float64_t* normal; + SGVector m_normal; }; } #endif /* _LINEARSTRINGKERNEL_H___ */ diff --git a/src/shogun/kernel/string/WeightedDegreePositionStringKernel.cpp b/src/shogun/kernel/string/WeightedDegreePositionStringKernel.cpp index 4841b13e0d9..f5253b81401 100644 --- a/src/shogun/kernel/string/WeightedDegreePositionStringKernel.cpp +++ b/src/shogun/kernel/string/WeightedDegreePositionStringKernel.cpp @@ -9,11 +9,12 @@ * Copyright (C) 1999-2009 Fraunhofer Institute FIRST and Max-Planck-Society */ -#include +#include +#include #include #include #include -#include +#include #include #include @@ -24,6 +25,7 @@ #ifdef HAVE_PTHREAD #include + #endif using namespace shogun; @@ -268,12 +270,10 @@ bool CWeightedDegreePositionStringKernel::init_optimization( if (tree_num<0) SG_DEBUG("initializing CWeightedDegreePositionStringKernel optimization\n") - for (int32_t i=0; iio)) { if (tree_num<0) { - if ( (i % (p_count/10+1)) == 0) - SG_PROGRESS(i,0,p_count) add_example_to_tree(IDX[i], alphas[i]); } else @@ -283,9 +283,6 @@ bool CWeightedDegreePositionStringKernel::init_optimization( } } - if (tree_num<0) - SG_DONE() - set_is_initialized(true) ; return true ; } @@ -1255,35 +1252,41 @@ void CWeightedDegreePositionStringKernel::compute_batch( if (num_threads < 2) { - CSignal::clear_cancel(); - for (int32_t j=0; j params; - params.vec=vec; - params.result=result; - params.weights=weights; - params.kernel=this; - params.tries=&tries; - params.factor=factor; - params.j=j; - params.start=0; - params.end=num_vec; - params.length=length; - params.max_shift=max_shift; - params.shift=shift; - params.vec_idx=vec_idx; - compute_batch_helper((void*) ¶ms); - - SG_PROGRESS(j,0,num_feat) - } + + auto pb = progress(range(num_feat), *this->io); + // TODO: replace with the new signal + // for (int32_t j=0; j params; + params.vec = vec; + params.result = result; + params.weights = weights; + params.kernel = this; + params.tries = &tries; + params.factor = factor; + params.j = j; + params.start = 0; + params.end = num_vec; + params.length = length; + params.max_shift = max_shift; + params.shift = shift; + params.vec_idx = vec_idx; + compute_batch_helper((void*)¶ms); + + pb.print_progress(); + } + pb.complete(); } #ifdef HAVE_PTHREAD else { - CSignal::clear_cancel(); - for (int32_t j=0; jio); + // TODO: replace with the new signal + // for (int32_t j=0; jio); for( k = 0; k < max_degree; ++k ) { const int32_t nofKmers = nofsKmers[ k ]; @@ -1439,7 +1443,7 @@ float64_t* CWeightedDegreePositionStringKernel::compute_scoring( x[j] = -1; } tries.traverse( tree, p, info, 0, x, k ); - SG_PROGRESS(i++,0,num_feat*max_degree) + pb.print_progress(); } // --- add partial overlap scores @@ -1478,6 +1482,7 @@ float64_t* CWeightedDegreePositionStringKernel::compute_scoring( // end; // end; } + pb.complete(); // === return a vector num_feat=1; @@ -1521,7 +1526,7 @@ char* CWeightedDegreePositionStringKernel::compute_consensus( table[i]=new DynArray(num_suppvec/10); //compute consensus via dynamic programming - for (int32_t i=0; iio)) { bool cumulative=false; @@ -1537,8 +1542,6 @@ char* CWeightedDegreePositionStringKernel::compute_consensus( tries.fill_backtracking_table(i, NULL, table[i], cumulative, weights); else tries.fill_backtracking_table(i, table[i-1], table[i], cumulative, weights); - - SG_PROGRESS(i,0,num_feat) } diff --git a/src/shogun/kernel/string/WeightedDegreeStringKernel.cpp b/src/shogun/kernel/string/WeightedDegreeStringKernel.cpp index 50632dac655..b325867d290 100644 --- a/src/shogun/kernel/string/WeightedDegreeStringKernel.cpp +++ b/src/shogun/kernel/string/WeightedDegreeStringKernel.cpp @@ -9,12 +9,13 @@ * Copyright (C) 1999-2009 Fraunhofer Institute FIRST and Max-Planck-Society */ -#include +#include +#include +#include #include #include #include -#include -#include +#include #include #include @@ -23,6 +24,7 @@ #ifdef HAVE_PTHREAD #include + #endif using namespace shogun; @@ -213,12 +215,10 @@ bool CWeightedDegreeStringKernel::init_optimization(int32_t count, int32_t* IDX, if (tree_num<0) SG_DEBUG("initializing CWeightedDegreeStringKernel optimization\n") - for (int32_t i=0; iio)) { if (tree_num<0) { - if ( (i % (count/10+1)) == 0) - SG_PROGRESS(i, 0, count) if (max_mismatch==0) add_example_to_tree(IDX[i], alphas[i]) ; @@ -236,9 +236,6 @@ bool CWeightedDegreeStringKernel::init_optimization(int32_t count, int32_t* IDX, } } - if (tree_num<0) - SG_DONE() - //tries.compact_nodes(NO_CHILD, 0, weights) ; set_is_initialized(true) ; @@ -883,11 +880,13 @@ void CWeightedDegreeStringKernel::compute_batch( #endif ASSERT(num_threads>0) int32_t* vec=SG_MALLOC(int32_t, num_threads*num_feat); + auto pb = progress(range(num_feat), *this->io); if (num_threads < 2) { - CSignal::clear_cancel(); - for (int32_t j=0; j src, float64_t threshold) : CDe SGVector labels(src.vlen); for (int32_t i = 0; i < labels.vlen; i++) { - labels[i] = src[i] + threshold >= 0 ? +1.0 : -1.0; + labels[i] = src[i] >= threshold ? +1.0 : -1.0; } set_labels(labels); set_values(src); diff --git a/src/shogun/labels/DenseLabels.cpp b/src/shogun/labels/DenseLabels.cpp index 525f43b3d5b..01bd9ef8c4d 100644 --- a/src/shogun/labels/DenseLabels.cpp +++ b/src/shogun/labels/DenseLabels.cpp @@ -10,13 +10,14 @@ * Copyright (C) 1999-2009 Fraunhofer Institute FIRST and Max-Planck-Society */ -#include -#include -#include +#include #include #include +#include +#include +#include #include -#include +#include using namespace shogun; @@ -32,6 +33,8 @@ CDenseLabels::CDenseLabels(int32_t num_lab) init(); m_labels = SGVector(num_lab); m_current_values=SGVector(num_lab); + linalg::zero(m_labels); + linalg::zero(m_current_values); } CDenseLabels::CDenseLabels(CFile* loader) diff --git a/src/shogun/labels/Labels.cpp b/src/shogun/labels/Labels.cpp index 78cc90c046d..1c6c3b24fb5 100644 --- a/src/shogun/labels/Labels.cpp +++ b/src/shogun/labels/Labels.cpp @@ -31,7 +31,9 @@ void CLabels::init() { SG_ADD((CSGObject **)&m_subset_stack, "subset_stack", "Current subset stack", MS_NOT_AVAILABLE); - + SG_ADD( + &m_current_values, "current_values", "current active value vector", + MS_NOT_AVAILABLE) m_subset_stack = new CSubsetStack(); SG_REF(m_subset_stack); } diff --git a/src/shogun/labels/MulticlassLabels.cpp b/src/shogun/labels/MulticlassLabels.cpp index 8219d152c19..29d3ff7c906 100644 --- a/src/shogun/labels/MulticlassLabels.cpp +++ b/src/shogun/labels/MulticlassLabels.cpp @@ -25,6 +25,15 @@ CMulticlassLabels::CMulticlassLabels(CFile* loader) : CDenseLabels(loader) init(); } +CMulticlassLabels::CMulticlassLabels(CBinaryLabels* labels) + : CDenseLabels(labels->get_num_labels()) +{ + init(); + + for (index_t i = 0; i < labels->get_num_labels(); ++i) + m_labels[i] = (labels->get_label(i) == 1 ? 1 : 0); +} + CMulticlassLabels::~CMulticlassLabels() { } @@ -41,8 +50,7 @@ void CMulticlassLabels::set_multiclass_confidences(int32_t i, "%s::set_multiclass_confidences(): Length of confidences should " "match size of the matrix", get_name()); - for (index_t j=0; j CMulticlassLabels::get_multiclass_confidences(int32_t i) @@ -147,3 +155,19 @@ CLabels* CMulticlassLabels::shallow_subset_copy() return shallow_copy_labels; } + +CMulticlassLabels* CMulticlassLabels::obtain_from_generic(CLabels* labels) +{ + if (labels == NULL) + return NULL; + + if (labels->get_label_type() != LT_MULTICLASS) + { + SG_SERROR("The Labels passed cannot be casted to CMulticlassLabels!") + return NULL; + } + + CMulticlassLabels* casted = dynamic_cast(labels); + SG_REF(casted) + return casted; +} \ No newline at end of file diff --git a/src/shogun/labels/MulticlassLabels.h b/src/shogun/labels/MulticlassLabels.h index 5ea4b0fb20f..45673fd51c1 100644 --- a/src/shogun/labels/MulticlassLabels.h +++ b/src/shogun/labels/MulticlassLabels.h @@ -51,12 +51,21 @@ class CMulticlassLabels : public CDenseLabels */ CMulticlassLabels(SGVector src); + /** constructor * * @param loader File object via which to load data */ CMulticlassLabels(CFile* loader); + /** + * Convert binary labels to multiclass labels, + * namely -1 is mapped to 0 and 1 to 1. + * + * @param labels Binary labels + */ + CMulticlassLabels(CBinaryLabels* labels); + /** destructor */ ~CMulticlassLabels(); @@ -126,6 +135,12 @@ class CMulticlassLabels : public CDenseLabels #ifndef SWIG // SWIG should skip this part virtual CLabels* shallow_subset_copy(); #endif + /** + * Cast a generic label object to a multiclass one + * @param labels generic CLabels instance + * @return the casted pointer (already SG_REF'ed) + */ + static CMulticlassLabels* obtain_from_generic(CLabels* labels); private: /** initialises and register parameters */ diff --git a/src/shogun/latent/LatentSOSVM.cpp b/src/shogun/latent/LatentSOSVM.cpp deleted file mode 100644 index 01e0b303e1e..00000000000 --- a/src/shogun/latent/LatentSOSVM.cpp +++ /dev/null @@ -1,71 +0,0 @@ -/* - * This program is free software; you can redistribute it and/or modify - * it under the terms of the GNU General Public License as published by - * the Free Software Foundation; either version 3 of the License, or - * (at your option) any later version. - * - * Written (W) 2012 Viktor Gal - * Copyright (C) 2012 Viktor Gal - */ - -#include -#ifdef USE_GPL_SHOGUN -#include - -using namespace shogun; - -CLatentSOSVM::CLatentSOSVM() - : CLinearLatentMachine() -{ - register_parameters(); - m_so_solver=NULL; -} - -CLatentSOSVM::CLatentSOSVM(CLatentModel* model, CLinearStructuredOutputMachine* so_solver, float64_t C) - : CLinearLatentMachine(model, C) -{ - register_parameters(); - set_so_solver(so_solver); -} - -CLatentSOSVM::~CLatentSOSVM() -{ - SG_UNREF(m_so_solver); -} - -CLatentLabels* CLatentSOSVM::apply_latent() -{ - return NULL; -} - -void CLatentSOSVM::set_so_solver(CLinearStructuredOutputMachine* so) -{ - SG_REF(so); - SG_UNREF(m_so_solver); - m_so_solver = so; -} - -float64_t CLatentSOSVM::do_inner_loop(float64_t cooling_eps) -{ - float64_t lambda = 1/m_C; - CDualLibQPBMSOSVM* so = new CDualLibQPBMSOSVM(); - so->set_lambda(lambda); - so->train(); - - /* copy the resulting w */ - set_w(so->get_w().clone()); - - /* get the primal objective value */ - float64_t po = so->get_result().Fp; - - SG_UNREF(so); - - return po; -} - -void CLatentSOSVM::register_parameters() -{ - m_parameters->add((CSGObject**)&m_so_solver, "so_solver", "Structured Output Solver."); -} - -#endif //USE_GPL_SHOGUN diff --git a/src/shogun/latent/LatentSOSVM.h b/src/shogun/latent/LatentSOSVM.h deleted file mode 100644 index f5c0c6be3a4..00000000000 --- a/src/shogun/latent/LatentSOSVM.h +++ /dev/null @@ -1,77 +0,0 @@ -/* - * This program is free software; you can redistribute it and/or modify - * it under the terms of the GNU General Public License as published by - * the Free Software Foundation; either version 3 of the License, or - * (at your option) any later version. - * - * Written (W) 2012 Viktor Gal - * Copyright (C) 2012 Viktor Gal - */ - -#ifndef __LATENTSOSVM_H__ -#define __LATENTSOSVM_H__ - -#include -#ifdef USE_GPL_SHOGUN - -#include -#include - -namespace shogun -{ - /** @brief class Latent Structured Output SVM, - * an structured output based machine for classification - * problems with latent variables. - */ - class CLatentSOSVM: public CLinearLatentMachine - { - public: - /** default ctor*/ - CLatentSOSVM(); - - /** - * - * @param model - * @param so_solver - * @param C - */ - CLatentSOSVM(CLatentModel* model, CLinearStructuredOutputMachine* so_solver, float64_t C); - - virtual ~CLatentSOSVM(); - - /** apply linear machine to data - * - * @return classified labels - */ - virtual CLatentLabels* apply_latent(); - - /** set SO solver that is going to be used - * - * @param so SO machine - */ - void set_so_solver(CLinearStructuredOutputMachine* so); - - /** Returns the name of the SGSerializable instance. - * - * @return name of the SGSerializable - */ - virtual const char* get_name() const { return "LatentSOSVM"; } - - protected: - /** do inner loop with given cooling epsilon - * - * @param cooling_eps cooling epsilon - */ - virtual float64_t do_inner_loop(float64_t cooling_eps); - - private: - void register_parameters(); - - private: - /** Linear Structured Solver */ - CLinearStructuredOutputMachine* m_so_solver; - }; -} -#endif //USE_GPL_SHOGUN -#endif /* __LATENTSOSVM_H__ */ - diff --git a/src/shogun/latent/LatentSVM.cpp b/src/shogun/latent/LatentSVM.cpp deleted file mode 100644 index c0b99d10294..00000000000 --- a/src/shogun/latent/LatentSVM.cpp +++ /dev/null @@ -1,83 +0,0 @@ -/* - * This program is free software; you can redistribute it and/or modify - * it under the terms of the GNU General Public License as published by - * the Free Software Foundation; either version 3 of the License, or - * (at your option) any later version. - * - * Written (W) 2012 Viktor Gal - * Copyright (C) 2012 Viktor Gal - */ - -#include -#ifdef USE_GPL_SHOGUN - -#include - -#include -#include - -using namespace shogun; - -CLatentSVM::CLatentSVM() - : CLinearLatentMachine() -{ -} - -CLatentSVM::CLatentSVM(CLatentModel* model, float64_t C) - : CLinearLatentMachine(model, C) -{ -} - -CLatentSVM::~CLatentSVM() -{ -} - -CLatentLabels* CLatentSVM::apply_latent() -{ - if (!m_model) - SG_ERROR("LatentModel is not set!\n") - - if (m_model->get_num_vectors() < 1) - return NULL; - - SGVector w = get_w(); - index_t num_examples = m_model->get_num_vectors(); - CLatentLabels* hs = new CLatentLabels(num_examples); - CBinaryLabels* ys = new CBinaryLabels(num_examples); - hs->set_labels(ys); - m_model->set_labels(hs); - - for (index_t i = 0; i < num_examples; ++i) - { - /* find h for the example */ - CData* h = m_model->infer_latent_variable(w, i); - hs->add_latent_label(h); - } - - /* compute the y labels */ - CDotFeatures* x = m_model->get_psi_feature_vectors(); - x->dense_dot_range(ys->get_labels().vector, 0, num_examples, NULL, w.vector, w.vlen, 0.0); - - return hs; -} - -float64_t CLatentSVM::do_inner_loop(float64_t cooling_eps) -{ - CLabels* ys = m_model->get_labels()->get_labels(); - CDotFeatures* feats = (m_model->get_caching() ? - m_model->get_cached_psi_features() : - m_model->get_psi_feature_vectors()); - CSVMOcas svm(m_C, feats, ys); - svm.set_epsilon(cooling_eps); - svm.train(); - SG_UNREF(ys); - SG_UNREF(feats); - - /* copy the resulting w */ - set_w(svm.get_w().clone()); - - return svm.compute_primal_objective(); -} - -#endif //USE_GPL_SHOGUN - diff --git a/src/shogun/latent/LatentSVM.h b/src/shogun/latent/LatentSVM.h deleted file mode 100644 index 35e43e6d0b4..00000000000 --- a/src/shogun/latent/LatentSVM.h +++ /dev/null @@ -1,83 +0,0 @@ -/* - * This program is free software; you can redistribute it and/or modify - * it under the terms of the GNU General Public License as published by - * the Free Software Foundation; either version 3 of the License, or - * (at your option) any later version. - * - * Written (W) 2012 Viktor Gal - * Copyright (C) 2012 Viktor Gal - */ - -#ifndef __LATENTSVM_H__ -#define __LATENTSVM_H__ - - -#include -#ifdef USE_GPL_SHOGUN -#include -#include - -namespace shogun -{ - class LatentModel; - - /** @brief LatentSVM class - * Latent SVM implementation based on [1]. - * For optimization this implementation uses SVMOcas. - * - * User must provide a her own CLatentModel which implements the PSI(x_i,h_i) - * function for the given problem. - * - * [1] P. F. Felzenszwalb, R. B. Girshick, D. McAllester, and D. Ramanan, - * "Object detection with discriminatively trained part-based models," - * Pattern Analysis and Machine Intelligence, - * IEEE Transactions on, vol. 32, no. 9, pp. 1627-1645, 2010. - * - */ - class CLatentSVM: public CLinearLatentMachine - { - public: - /** default contstructor */ - CLatentSVM(); - - /** constructor - * - * @param model the user defined CLatentModel object. - * @param C regularization constant - */ - CLatentSVM(CLatentModel* model, float64_t C); - - virtual ~CLatentSVM(); - - /** apply linear machine to all examples - * - * @return resulting labels - */ - virtual CLatentLabels* apply_latent(); - - using CLinearLatentMachine::apply_latent; - - /** Returns the name of the SGSerializable instance. - * - * @return name of the SGSerializable - */ - virtual const char* get_name() const { return "LatentSVM"; } - - protected: - /** inner loop of the latent machine - * - * The optimization part after finding the argmax_h for the - * positive examples in the outter loop. It uses SVMOcas for - * finding the cutting plane. - * - * @param cooling_eps epsilon - * @return primal objective value - */ - virtual float64_t do_inner_loop(float64_t cooling_eps); - }; -} - -#endif /* __LATENTSVM_H__ */ - -#endif //USE_GPL_SHOGUN - diff --git a/src/shogun/lib/CircularBuffer.cpp b/src/shogun/lib/CircularBuffer.cpp index ec886221938..50703228bc0 100644 --- a/src/shogun/lib/CircularBuffer.cpp +++ b/src/shogun/lib/CircularBuffer.cpp @@ -66,7 +66,7 @@ int32_t CCircularBuffer::push(SGVector source) // determine which part of the memory block is free to read if (m_end_pos>=m_begin_pos) { - int32_t bytes_to_memory_end=m_buffer.vlen-(m_end_pos-m_buffer.vector); + auto bytes_to_memory_end=m_buffer.vlen-std::distance(m_buffer.vector, m_end_pos); if (bytes_to_memory_end=m_begin_pos) { - int32_t bytes_to_memory_end=m_buffer.vlen-(m_end_pos-m_buffer.vector); + int32_t bytes_to_memory_end=m_buffer.vlen-std::distance(m_buffer.vector, m_end_pos); if (bytes_to_memory_end 0) + ? temp + : has_next_locally(m_buffer.vector+m_last_idx-head_length, m_end_pos); } } else @@ -225,8 +223,8 @@ index_t CCircularBuffer::next_token_idx(index_t &start) if (m_bytes_count==0) return m_bytes_count; - int32_t tail_length=m_end_pos-m_buffer.vector; - int32_t head_length=m_buffer_end-m_begin_pos; + auto tail_length=std::distance(m_buffer.vector, m_end_pos); + auto head_length=std::distance(m_begin_pos, m_buffer_end); // determine position of finder pointer in memory block if (m_last_idx= num_chars) + move_pointer(&m_begin_pos, m_begin_pos+num_chars); + else + move_pointer(&m_begin_pos, m_buffer.vector+num_chars-head_length); m_last_idx-=num_chars; if (m_last_idx<0) @@ -377,7 +379,7 @@ void CCircularBuffer::detach_chunk(char** dest, int32_t* dest_size, int32_t dest bool CCircularBuffer::has_next_locally(char* part_begin, char* part_end) { - int32_t num_bytes_to_search=part_end-part_begin; + auto num_bytes_to_search=std::distance(part_begin, part_end); SGVector buffer_part(part_begin, num_bytes_to_search, false); m_tokenizer->set_text(buffer_part); @@ -388,7 +390,7 @@ bool CCircularBuffer::has_next_locally(char* part_begin, char* part_end) index_t CCircularBuffer::next_token_idx_locally(index_t &start, char* part_begin, char* part_end) { index_t end=0; - int32_t num_bytes_to_search=part_end-part_begin; + auto num_bytes_to_search=std::distance(part_begin, part_end); if (num_bytes_to_search<=0) { start=0; @@ -411,7 +413,7 @@ index_t CCircularBuffer::next_token_idx_locally(index_t &start, char* part_begin void CCircularBuffer::move_pointer(char** pointer, char* new_position) { - *pointer=new_position; - if (*pointer>=m_buffer.vector+m_buffer.vlen) - *pointer=m_buffer.vector; + *pointer = (new_position >= m_buffer_end) + ? m_buffer.vector + : new_position; } diff --git a/src/shogun/lib/DynInt.h b/src/shogun/lib/DynInt.h deleted file mode 100644 index 594475bc7cd..00000000000 --- a/src/shogun/lib/DynInt.h +++ /dev/null @@ -1,579 +0,0 @@ -/* - * This program is free software; you can redistribute it and/or modify - * it under the terms of the GNU General Public License as published by - * the Free Software Foundation; either version 3 of the License, or - * (at your option) any later version. - * - * Written (W) 2009 Soeren Sonnenburg - * Copyright (C) 2009 Fraunhofer Institute FIRST and Max Planck Society - */ - -#ifndef __DYNINT_H__ -#define __DYNINT_H__ - -#include - -#include -#include -#include - -namespace shogun -{ -/** @brief integer type of dynamic size - * - * This object can be used to create huge integers. These integers can be used - * directly instead of the usual int32_t etc types since operators are properly - * overloaded. - * - * An exampe use would be 512 wide unsigned ints consisting of four uint64's: - * - * CDynInt int512; - * - * This data type is mostly used as a (efficient) storage container for - * bit-mapped strings. Therefore, currently only comparison, assignment and - * bit operations are implemented. - * - * TODO: implement add,mul,div - */ -template class CDynInt -{ -public: - /** default constructor - * - * creates a DynInt that is all zero. - */ - CDynInt() - { - for (int i=0; i &x) - { - for (int i=0; i& operator=(const CDynInt& x) - { - for (int i=0; i operator|(const CDynInt& x) const - { - CDynInt r; - - for (int i=sz-1; i>=0; i--) - r.integer[i]=integer[i] | x.integer[i]; - - return r; - } - - /** overload & operator and return x & y - * - * @param x x - */ - const CDynInt operator&(const CDynInt& x) const - { - CDynInt r; - - for (int i=sz-1; i>=0; i--) - r.integer[i]=integer[i] & x.integer[i]; - - return r; - } - - /** overload << operator - * - * perform bit shift to the left - * - * @param shift shift by this amount - */ - CDynInt operator<<(int shift) - { - CDynInt r=*this; - - while (shift>0) - { - int s=CMath::min(shift, 8*((int) sizeof(T))-1); - - for (int i=0; i> (sizeof(T)*8 - s); - r.integer[i]= (r.integer[i] << s) | overflow; - } - - shift-=s; - } - - return r; - } - - /** overload >> operator - * - * perform bit shift to the right - * - * @param shift shift by this amount - */ - CDynInt operator>>(int shift) - { - CDynInt r=*this; - - while (shift>0) - { - int s=CMath::min(shift, 8*((int) sizeof(T))-1); - - for (int i=sz-1; i>=0; i--) - { - T overflow=0; - if (i>0) - overflow = (r.integer[i-1] << (sizeof(T)*8 - s)); - r.integer[i]= (r.integer[i] >> s) | overflow; - } - - shift-=s; - } - - return r; - } - - /** overload ^ operator and return x ^ y - * - * @param x x - */ - const CDynInt operator^(const CDynInt& x) const - { - CDynInt r; - - for (int i=sz-1; i>=0; i--) - r.integer[i]=integer[i] ^ x.integer[i]; - - return r; - } - - /** overload + operator and return x + y - * - * @param x x - */ - const CDynInt operator+(const CDynInt &x) const - { - CDynInt r; - - T overflow=0; - for (int i=sz-1; i>=0; i--) - { - r.integer[i]=integer[i]+x.integer[i]+overflow; - if (r.integer[i] < CMath::max(integer[i], x.integer[i])) - overflow=1; - else - overflow=0; - } - - return x; - } - - /** overload - operator and return x - y - * - * @param x x - */ - const CDynInt operator-(const CDynInt &x) const - { - return NULL; - } - - /** overload / operator and return x / y - * - * @param x x - */ - const CDynInt operator/(const CDynInt &x) const - { - return NULL; - } - - /** overload * operator and return x * y - * - * @param x x - */ - const CDynInt operator*(const CDynInt &x) const - { - return NULL; - } - - /** overload += operator; add x to current DynInt - * - * @param x x - */ - CDynInt& operator+=(const CDynInt &x) - { - return NULL; - } - - /** overload -= operator; substract x from current DynInt - * - * @param x x - */ - CDynInt& operator-=(const CDynInt &x) - { - return NULL; - } - - /** overload *= operator; multiple x to with current DynInt - * - * @param x x - */ - CDynInt& operator*=(const CDynInt &x) - { - return NULL; - } - - /** overload /= operator; divide current object by x - * - * @param x x - */ - CDynInt& operator/=(const CDynInt &x) - { - return NULL; - } - - /** overload == operator; test if current object equals x - * - * @param x x - */ - bool operator==(const CDynInt &x) const - { - for (int i=sz-1; i>=0; i--) - { - if (integer[i]!=x.integer[i]) - return false; - } - - return true; - } - - /** overload >= operator; test if current object greater equal x - * - * @param x x - */ - bool operator>=(const CDynInt &x) const - { - for (int i=0; ix.integer[i]) - return true; - if (integer[i] &x) const - { - for (int i=0; ix.integer[i]) - return false; - } - return true; - } - - /** overload > operator; test if current object is bigger than x - * - * @param x x - */ - bool operator>(const CDynInt &x) const - { - for (int i=0; ix.integer[i]) - return true; - if (integer[i] &x) const - { - for (int i=0; ix.integer[i]) - return false; - } - return false; - } - - /** overload ! operator; test if current object is not equal to x - * - * @param x x - */ - bool operator!=(const CDynInt &x) const - { - for (int i=sz-1; i>=0; i--) - { - if (integer[i]!=x.integer[i]) - return true; - } - return false; - } - - /** overload |= operator - * - * perform bitwise or with current DynInt and x - * - * @param x x - */ - CDynInt& operator|=(const CDynInt& x) - { - for (int i=sz-1; i>=0; i--) - integer[i]|=x.integer[i]; - - return *this; - } - - /** overload &= operator - * - * perform bitwise and with current DynInt and x - * - * @param x x - */ - CDynInt& operator&=(const CDynInt& x) - { - for (int i=sz-1; i>=0; i--) - integer[i]&=x.integer[i]; - - return *this; - } - - /** overload ^= operator - * - * perform bitwise xor with current DynInt and x - * - * @param x x - */ - CDynInt& operator^=(const CDynInt& x) - { - for (int i=sz-1; i>=0; i--) - integer[i]^=x.integer[i]; - - return *this; - } - - /** overload <<= operator - * - * perform bit shift to the left - * - * @param shift shift by this amount - */ - CDynInt& operator<<=(int shift) - { - *this=*this<>= operator - * - * perform bit shift to the right - * - * @param shift shift by this amount - */ - CDynInt& operator>>=(int shift) - { - *this=*this>>shift; - return *this; - } - - /** negate DynInt */ - CDynInt& operator~() - { - for (int i=sz-1; i>=0; i--) - integer[i]= ~integer[i]; - return *this; - } - - /** cast to least significant word *dangerous* */ - operator T() { return integer[sz-1]; } - - /** decrement DynInt by one */ - CDynInt& operator--() - { - T overflow=0; - for (int i=sz-1; i>=0; i--) - { - T x = integer[i]-1-overflow; - overflow=0; - if (integer[i]>x) - overflow=1; - integer[i]=x; - } - return *this; - } - - /** increment DynInt by one */ - CDynInt& operator++() - { - T overflow=0; - for (int i=sz-1; i>=0; i--) - { - T x = integer[i]+1+overflow; - overflow=0; - if (integer[i]>x) - overflow=1; - integer[i]=x; - } - return *this; - } - - /** print the current long integer in hex (without carriage return */ - void print_hex() const - { - for (int i=0; i uint192_t; - -/// 256 bit integer constructed out of 4 64bit uint64_t's -typedef CDynInt uint256_t; - -/// 512 bit integer constructed out of 8 64bit uint64_t's -typedef CDynInt uint512_t; - -/// 1024 bit integer constructed out of 16 64bit uint64_t's -typedef CDynInt uint1024_t; -//@} -} -#endif // __DYNINT_H__ diff --git a/src/shogun/lib/RxCppHeader.h b/src/shogun/lib/RxCppHeader.h new file mode 100644 index 00000000000..33e069c59a7 --- /dev/null +++ b/src/shogun/lib/RxCppHeader.h @@ -0,0 +1,27 @@ +#ifndef SHOGUN_RXCPPHEADER_H +#define SHOGUN_RXCPPHEADER_H + +/** +* Rx namespace +*/ +namespace rxcpp +{ + template + class observer; + + namespace subjects + { + template + class subject; + } + + template + class dynamic_observable; + template + class observable; + template + class subscriber; + class subscription; +} + +#endif // SHOGUN_RXCPPHEADER_H diff --git a/src/shogun/lib/SGMatrix.cpp b/src/shogun/lib/SGMatrix.cpp index acfafa150bc..12a2df5bd84 100644 --- a/src/shogun/lib/SGMatrix.cpp +++ b/src/shogun/lib/SGMatrix.cpp @@ -18,11 +18,11 @@ #include #include #include +#include #include #include #include #include -#include namespace shogun { @@ -388,8 +388,31 @@ void SGMatrix::create_diagonal_matrix(T* matrix, T* v,int32_t size) } template -float64_t SGMatrix::trace( - float64_t* mat, int32_t cols, int32_t rows) +SGMatrix SGMatrix::submatrix(index_t col_start, index_t col_end) const +{ + assert_on_cpu(); + return SGMatrix( + get_column_vector(col_start), num_rows, col_end - col_start, false); +} + +template +SGVector SGMatrix::get_column(index_t col) const +{ + assert_on_cpu(); + return SGVector(get_column_vector(col), num_rows, false); +} + +template +void SGMatrix::set_column(index_t col, const SGVector vec) +{ + assert_on_cpu(); + ASSERT(!vec.on_gpu()) + ASSERT(vec.vlen == num_rows) + sg_memcpy(&matrix[num_rows * col], vec.vector, sizeof(T) * num_rows); +} + +template +float64_t SGMatrix::trace(float64_t* mat, int32_t cols, int32_t rows) { float64_t trace=0; for (int64_t i=0; i SGMatrix::create_identity_matrix(index_t size, char scale) template <> SGMatrix SGMatrix::create_identity_matrix(index_t size, int8_t scale) { - SGMatrix I(size, size); + SGMatrix identity_matrix(size, size); for (index_t i=0; i SGMatrix SGMatrix::create_identity_matrix(index_t size, uint8_t scale) { - SGMatrix I(size, size); + SGMatrix identity_matrix(size, size); for (index_t i=0; i SGMatrix SGMatrix::create_identity_matrix(index_t size, bool scale) { - SGMatrix I(size, size); + SGMatrix identity_matrix(size, size); for (index_t i=0; i SGMatrix SGMatrix::create_identity_matrix(index_t size, int16_t scale) { - SGMatrix I(size, size); + SGMatrix identity_matrix(size, size); for (index_t i=0; i SGMatrix SGMatrix::create_identity_matrix(index_t size, uint16_t scale) { - SGMatrix I(size, size); + SGMatrix identity_matrix(size, size); for (index_t i=0; i SGMatrix SGMatrix::create_identity_matrix(index_t size, int32_t scale) { - SGMatrix I(size, size); + SGMatrix identity_matrix(size, size); for (index_t i=0; i SGMatrix SGMatrix::create_identity_matrix(index_t size, uint32_t scale) { - SGMatrix I(size, size); + SGMatrix identity_matrix(size, size); for (index_t i=0; i SGMatrix SGMatrix::create_identity_matrix(index_t size, int64_t scale) { - SGMatrix I(size, size); + SGMatrix identity_matrix(size, size); for (index_t i=0; i SGMatrix SGMatrix::create_identity_matrix(index_t size, uint64_t scale) { - SGMatrix I(size, size); + SGMatrix identity_matrix(size, size); for (index_t i=0; i SGMatrix SGMatrix::create_identity_matrix(index_t size, float32_t scale) { - SGMatrix I(size, size); + SGMatrix identity_matrix(size, size); for (index_t i=0; i SGMatrix SGMatrix::create_identity_matrix(index_t size, float64_t scale) { - SGMatrix I(size, size); + SGMatrix identity_matrix(size, size); for (index_t i=0; i SGMatrix SGMatrix::create_identity_matrix(index_t size, floatmax_t scale) { - SGMatrix I(size, size); + SGMatrix identity_matrix(size, size); for (index_t i=0; i SGMatrix SGMatrix::create_identity_matrix(index_t size, complex128_t scale) { - SGMatrix I(size, size); + SGMatrix identity_matrix(size, size); for (index_t i=0; i #include #include +#include #include #include @@ -39,6 +40,9 @@ template class SGMatrix : public SGReferencedData { friend class LinalgBackendEigen; + public: + typedef RandomIterator iterator; + public: typedef Eigen::Matrix EigenMatrixXt; typedef Eigen::Map > EigenMatrixXtMap; @@ -132,7 +136,7 @@ template class SGMatrix : public SGReferencedData /** Empty destructor */ virtual ~SGMatrix(); -#ifndef SWIG // SWIG should skip this part +#ifndef SWIG // SWIG should skip this parts /** Get a column vector * @param col column index * @return the column vector for index col @@ -144,6 +148,28 @@ template class SGMatrix : public SGReferencedData return &matrix[c*num_rows]; } + /** Given a range of columns (start, end), return a view + * of the matrix from column start to end excluded. + * \warning The returned SGMatrix is non-owning! + * @param col_start column index (inclusive) + * @param col_end column index (excluded) + * @return the submatrix + */ + SGMatrix submatrix(index_t col_start, index_t col_end) const; + + /** Map a column to a SGVector + * \warning The returned SGVector is non-owning! + * @param col column index + * @return the column vector for index col + */ + SGVector get_column(index_t col) const; + + /** Copy the content of a vector into a column + * @param col column index + * @param vec vector + */ + void set_column(index_t col, const SGVector vec); + /** Get a row vector * * @param row row index @@ -197,6 +223,12 @@ template class SGMatrix : public SGReferencedData return matrix[index]; } + /** Returns an iterator to the first element of the container. */ + iterator begin() noexcept { return iterator(matrix); } + + /** Returns an iterator to the element following the last element of the container. */ + iterator end() noexcept { return iterator(matrix + (num_rows * num_cols)); } + #endif // SWIG should skip this part /** Get element at index @@ -240,9 +272,9 @@ template class SGMatrix : public SGReferencedData } /** The size */ - inline uint64_t size() const + inline int64_t size() const { - const uint64_t c=num_cols; + const int64_t c=num_cols; return num_rows*c; } diff --git a/src/shogun/lib/SGVector.cpp b/src/shogun/lib/SGVector.cpp index 89b4b0730bd..dc8388d75a4 100644 --- a/src/shogun/lib/SGVector.cpp +++ b/src/shogun/lib/SGVector.cpp @@ -21,11 +21,10 @@ #include #include +#include #include #include -#include - #define COMPLEX128_ERROR_NOARG(function) \ template <> \ void SGVector::function() \ @@ -97,14 +96,30 @@ SGVector::SGVector(index_t len, bool ref_counting) m_on_gpu.store(false, std::memory_order_release); } -template +template +SGVector::SGVector(SGMatrix matrix) + : SGReferencedData(matrix), vlen(matrix.num_cols * matrix.num_rows), + gpu_ptr(NULL) +{ + ASSERT(!matrix.on_gpu()) + vector = matrix.data(); + m_on_gpu.store(false, std::memory_order_release); +} + +template SGVector::SGVector(GPUMemoryBase* gpu_vector, index_t len) - : SGReferencedData(true), vector(NULL), vlen(len), - gpu_ptr(std::shared_ptr>(gpu_vector)) + : SGReferencedData(true), vector(NULL), vlen(len), + gpu_ptr(std::shared_ptr>(gpu_vector)) { m_on_gpu.store(true, std::memory_order_release); } +template +SGVector::SGVector(std::initializer_list il): + SGVector(il.begin(), il.end()) +{ +} + template SGVector::SGVector(const SGVector &orig) : SGReferencedData(orig) { @@ -722,13 +737,12 @@ float32_t SGVector::twonorm(const float32_t* x, int32_t len) template <> float64_t SGVector::twonorm(const float64_t* v, int32_t n) { - float64_t norm = 0.0; #ifdef HAVE_LAPACK - norm = cblas_dnrm2(n, v, 1); + return cblas_dnrm2(n, v, 1); #else - norm = CMath::sqrt(CMath::dot(v, v, n)); + SGVector wrapper(const_cast(v), n, false); + return CMath::sqrt(linalg::dot(wrapper, wrapper)); #endif - return norm; } template <> diff --git a/src/shogun/lib/SGVector.h b/src/shogun/lib/SGVector.h index 8b7fedfa673..3b2a7344e9d 100644 --- a/src/shogun/lib/SGVector.h +++ b/src/shogun/lib/SGVector.h @@ -19,10 +19,12 @@ #include #include #include +#include #include #include #include +#include namespace Eigen { @@ -43,6 +45,9 @@ template class SGVector : public SGReferencedData { friend class LinalgBackendEigen; + public: + typedef RandomIterator iterator; + public: typedef Eigen::Matrix EigenVectorXt; typedef Eigen::Matrix EigenRowVectorXt; @@ -65,6 +70,9 @@ template class SGVector : public SGReferencedData /** Constructor to create new vector in memory */ SGVector(index_t len, bool ref_counting=true); + /** Constructor to create new vector from a SGMatrix */ + SGVector(SGMatrix matrix); + /** Construct SGVector from GPU memory. * * @param vector GPUMemoryBase pointer @@ -86,12 +94,24 @@ template class SGVector : public SGReferencedData } #ifndef SWIG // SWIG should skip this part -#if defined(HAVE_CXX0X) || defined(HAVE_CXX11) /** The container type for a given template argument */ template using container_type = SGVector; -#endif // define (HAVE_CXX0X) || defined(HAVE_CXX11) + /** Construct SGVector from InputIterator list */ + template + SGVector(InputIt begin, InputIt end): + SGReferencedData(true), + vlen(std::distance(begin, end)), + gpu_ptr(nullptr) + { + vector = SG_MALLOC(T, vlen); + std::copy(begin, end, vector); + m_on_gpu.store(false, std::memory_order_release); + } + + /** Construct SGVector from initializer list */ + SGVector(std::initializer_list il); /** Wraps a matrix around the data of an Eigen3 column vector */ SGVector(EigenVectorXt& vec); @@ -142,6 +162,12 @@ template class SGVector : public SGReferencedData return vector; } + /** Returns an iterator to the first element of the container. */ + iterator begin() noexcept { return iterator(vector); } + + /** Returns an iterator to the element following the last element of the container. */ + iterator end() noexcept { return iterator(vector + vlen); } + SGVector& operator=(const SGVector&); /** Cast to pointer */ diff --git a/src/shogun/lib/ShogunException.cpp b/src/shogun/lib/ShogunException.cpp index 5da565c39e8..3a21aa4dc7f 100644 --- a/src/shogun/lib/ShogunException.cpp +++ b/src/shogun/lib/ShogunException.cpp @@ -35,7 +35,6 @@ ShogunException::init(const char* str) ShogunException::ShogunException(const char* str) { #ifndef WIN32 - CSignal::unset_handler(); #endif init(str); diff --git a/src/shogun/lib/Signal.cpp b/src/shogun/lib/Signal.cpp index 1afc6b96cbb..fb5102f9357 100644 --- a/src/shogun/lib/Signal.cpp +++ b/src/shogun/lib/Signal.cpp @@ -5,174 +5,87 @@ * (at your option) any later version. * * Written (W) 1999-2009 Soeren Sonnenburg + * Written (W) 2017 Giovanni De Toni * Copyright (C) 1999-2009 Fraunhofer Institute FIRST and Max-Planck-Society */ -#include - +#include #include -#include +#include #include #include -#include using namespace shogun; +using namespace rxcpp; + +bool CSignal::m_active = false; +CSignal::SGSubjectS* CSignal::m_subject = new rxcpp::subjects::subject(); -int CSignal::signals[NUMTRAPPEDSIGS]={SIGINT, SIGURG}; -struct sigaction CSignal::oldsigaction[NUMTRAPPEDSIGS]; -bool CSignal::active=false; -bool CSignal::cancel_computation=false; -bool CSignal::cancel_immediately=false; +CSignal::SGObservableS* CSignal::m_observable = + new CSignal::SGObservableS(CSignal::m_subject->get_observable()); +CSignal::SGSubscriberS* CSignal::m_subscriber = + new CSignal::SGSubscriberS(CSignal::m_subject->get_subscriber()); CSignal::CSignal() -: CSGObject() { } CSignal::~CSignal() { - if (!unset_handler()) - SG_PRINT("error uninitalizing signal handler\n") } void CSignal::handler(int signal) { - if (signal == SIGINT) - { - SG_SPRINT("\nImmediately return to prompt / Prematurely finish computations / Do nothing (I/P/D)? ") - char answer=fgetc(stdin); + /* If the handler is not enabled, then return */ + if (!m_active) + return; - if (answer == 'I') - { - unset_handler(); - set_cancel(true); - if (sg_print_error) - sg_print_error(stdout, "sg stopped by SIGINT\n"); - } - else if (answer == 'P') - set_cancel(); - else - SG_SPRINT("Continuing...\n") - } - else if (signal == SIGURG) - set_cancel(); - else - SG_SPRINT("unknown signal %d received\n", signal) -} - -bool CSignal::set_handler() -{ - if (!active) + if (signal == SIGINT) { - struct sigaction act; - sigset_t st; - - sigemptyset(&st); - for (int32_t i=0; i=0; j--) - sigaction(signals[i], &oldsigaction[i], NULL); - - clear(); - return false; - } + case 'I': + SG_SPRINT("[ShogunSignalHandler] Killing the application...\n"); + m_subscriber->on_completed(); + exit(0); + break; + case 'C': + SG_SPRINT( + "[ShogunSignalHandler] Terminating" + " prematurely current algorithm...\n"); + m_subscriber->on_next(SG_BLOCK_COMP); + break; + case 'P': + SG_SPRINT("[ShogunSignalHandler] Pausing current computation...") + m_subscriber->on_next(SG_PAUSE_COMP); + break; + default: + SG_SPRINT("[ShogunSignalHandler] Continuing...\n") + break; } - - active=true; - return true; } else - return false; -} - -bool CSignal::unset_handler() -{ - if (active) { - bool result=true; - - for (int32_t i=0; i=NSIG) -Sigfunc *handlers[NSIG]={0}; - -int sigaddset(sigset_t *set, int signo) -{ - if (SIGBAD(signo)) { - errno = EINVAL; - return -1; + SG_SPRINT("[ShogunSignalHandler] Unknown signal %d received\n", signal) } - *set |= 1 << (signo-1); - return 0; } -int sigaction(int signo, const struct sigaction *act, struct sigaction *oact) +void CSignal::reset_handler() { - if (SIGBAD(signo)) { - errno = EINVAL; - return -1; - } - - if(oact){ - oact->sa_handler = handlers[signo]; - oact->sa_mask = 0; - oact->sa_flags =0; - } - if (act) - handlers[signo]=act->sa_handler; + delete m_subject; + delete m_observable; + delete m_subscriber; - return 0; + m_subject = new rxcpp::subjects::subject(); + m_observable = new CSignal::SGObservableS(m_subject->get_observable()); + m_subscriber = new CSignal::SGSubscriberS(m_subject->get_subscriber()); } -#endif diff --git a/src/shogun/lib/Signal.h b/src/shogun/lib/Signal.h index 764ac345a6c..f63b5d297b3 100644 --- a/src/shogun/lib/Signal.h +++ b/src/shogun/lib/Signal.h @@ -5,139 +5,99 @@ * (at your option) any later version. * * Written (W) 1999-2009 Soeren Sonnenburg + * Written (W) 2017 Giovanni De Toni * Copyright (C) 1999-2009 Fraunhofer Institute FIRST and Max-Planck-Society */ #ifndef __SIGNAL__H_ #define __SIGNAL__H_ -#include - -#if defined(__MINGW64__) || defined(_MSC_VER) -typedef unsigned long sigset_t; -#endif -#if defined(__MINGW32__) && !defined(__MINGW64__) -typedef int sigset_t; -#endif - -#ifndef SIGURG -#define SIGURG -16 -#endif - -#if defined(__MINGW64__) || defined(_MSC_VER) || defined(__MINGW32__) -typedef void Sigfunc (int); - -struct sigaction { - Sigfunc *sa_handler; - sigset_t sa_mask; - int sa_flags; -}; - -#define sigemptyset(ptr) (*(ptr) = 0) -#define sigfillset(ptr) ( *(ptr) = ~(sigset_t)0,0) - -int sigaddset(sigset_t*, int); -int sigaction(int signo, const struct sigaction *act, struct sigaction *oact); -#endif - -#ifndef DISABLE_CANCEL_CALLBACK -namespace shogun -{ -extern void (*sg_cancel_computations)(bool &delayed, bool &immediately); -} -#endif - -#include #include -#include -#define NUMTRAPPEDSIGS 2 - namespace shogun { -/** @brief Class Signal implements signal handling to e.g. allow ctrl+c to cancel a - * long running process. - * - * This is done in two ways: - * - * -# A signal handler is attached to trap the SIGINT and SIGURG signal. - * Pressing ctrl+c or sending the SIGINT (kill ...) signal to the shogun - * process will make shogun print a message asking to immediately exit the - * running method and to fall back to the command line. - * -# When an URG signal is received or ctrl+c P is pressed shogun will - * prematurely stop a method and continue execution. For example when an SVM - * solver takes a long time without progressing much, one might still be - * interested in the result and should thus send SIGURG or interactively - * prematurely stop the method - */ -class CSignal : public CSGObject -{ + /** + * Possible Shogun signal types. + */ + enum sg_signals_types + { + SG_BLOCK_COMP, + SG_PAUSE_COMP + }; + + /** @brief Class Signal implements signal handling to e.g. allow CTRL+C to + * cancel a long running process. + * + * -# A signal handler is attached to trap the SIGINT signal. + * Pressing CTRL+C or sending the SIGINT (kill ...) signal to the shogun + * process will make shogun print a message asking the user to choose an + * option bewteen: immediately exit the running method and fall back to + * the command line, prematurely stop the current algoritmh and do nothing. + */ + class CSignal : CSGObject + { public: - /** default constructor */ + typedef rxcpp::subjects::subject SGSubjectS; + typedef rxcpp::observable> + SGObservableS; + typedef rxcpp::subscriber> + SGSubscriberS; + CSignal(); virtual ~CSignal(); - /** handler + /** Signal handler. Need to be registered with std::signal. * * @param signal signal number */ static void handler(int signal); - /** set handler - * - * @return if setting was successful - */ - static bool set_handler(); - - /** unset handler - * - * @return if unsetting was successful - */ - static bool unset_handler(); - - /** clear signals */ - static void clear(); - - /** clear cancel flag signals */ - static void clear_cancel(); - - /** set cancel flag signals */ - static void set_cancel(bool immediately=false); +#ifndef SWIG // SWIG should skip this part + /** + * Get observable + * @return RxCpp observable + */ + SGObservableS* get_observable() + { + return m_observable; + }; +#endif - /** cancel computations - * - * @return if computations should be cancelled - */ - static inline bool cancel_computations() +#ifndef SWIG // SWIG should skip this part + /** + * Get subscriber + * @return RxCpp subscriber + */ + SGSubscriberS* get_subscriber() { -#ifndef DISABLE_CANCEL_CALLBACK - if (sg_cancel_computations) - sg_cancel_computations(cancel_computation, cancel_immediately); + return m_subscriber; + }; #endif - if (cancel_immediately) - throw ShogunException("Computations have been cancelled immediately"); - return cancel_computation; + /** Enable signal handler + */ + void enable_handler() + { + m_active = true; } + /** + * Reset handler in case of multiple instantiation + */ + static void reset_handler(); /** @return object name */ virtual const char* get_name() const { return "Signal"; } - protected: - /** signals; handling external lib */ - static int signals[NUMTRAPPEDSIGS]; - - /** signal actions */ - static struct sigaction oldsigaction[NUMTRAPPEDSIGS]; - - /** active signal */ - static bool active; + private: + /** Active signal */ + static bool m_active; - /** if computation should be cancelled */ - static bool cancel_computation; - - /** if shogun should return ASAP */ - static bool cancel_immediately; + public: + /** Observable */ + static SGSubjectS* m_subject; + static SGObservableS* m_observable; + static SGSubscriberS* m_subscriber; }; } #endif // __SIGNAL__H_ diff --git a/src/shogun/lib/any.h b/src/shogun/lib/any.h index 5bb8e6caccd..9d61551725d 100644 --- a/src/shogun/lib/any.h +++ b/src/shogun/lib/any.h @@ -35,8 +35,9 @@ #ifndef _ANY_H_ #define _ANY_H_ -#include #include +#include +#include #include #ifdef HAVE_CXA_DEMANGLE #include @@ -44,393 +45,429 @@ namespace shogun { - /** Converts compiler-dependent name of class to - * something human readable. - * @return human readable name of class - */ - template - std::string demangledType() - { + /** Converts compiler-dependent name of class to + * something human readable. + * @return human readable name of class + */ + template + std::string demangledType() + { #ifdef HAVE_CXA_DEMANGLE size_t length; int status; - char* demangled = abi::__cxa_demangle(typeid(T).name(), nullptr, &length, &status); + char* demangled = + abi::__cxa_demangle(typeid(T).name(), nullptr, &length, &status); std::string demangled_string(demangled); free(demangled); #else std::string demangled_string(typeid(T).name()); #endif - return demangled_string; - } - - enum class PolicyType { - OWNING, - NON_OWNING - }; - - /** @brief An interface for a policy to store a value. - * Value can be any data like primitive data-types, shogun objects, etc. - * Policy defines how to handle this data. It works with a - * provided memory region and is able to set value, clear it - * and return the type-name as string. - */ - class BaseAnyPolicy - { - public: - /** Puts provided value pointed by v (untyped to be generic) to storage. - * @param storage pointer to a pointer to storage - * @param v pointer to value - */ - virtual void set(void** storage, const void* v) const = 0; - - /** Clears storage. - * @param storage pointer to a pointer to storage - */ - virtual void clear(void** storage) const = 0; - - /** Returns type-name as string. - * @return name of type class - */ - virtual std::string type() const = 0; - - /** Compares type. - * @param ti type information - * @return true if type matches - */ - virtual bool matches(const std::type_info& ti) const = 0; - - /** Compares two storages. - * @param storage pointer to a pointer to storage - * @param other_storage pointer to a pointer to another storage - * @return true if both storages have same value - */ - virtual bool equals(void** storage, void** other_storage) const = 0; - - /** Returns the name of policy. - * @return name of policy - */ - virtual std::string policy_name() const = 0; - - /** Returns the type of policy. - * @return type of policy - */ - virtual PolicyType policy_type() const = 0; - }; - - /** @brief This is one concrete implementation of policy that - * uses void pointers to store values. - */ - template - class PointerValueAnyPolicy : public BaseAnyPolicy - { - public: - /** Puts provided value pointed by v (untyped to be generic) to storage. - * @param storage pointer to a pointer to storage - * @param v pointer to value - */ - virtual void set(void** storage, const void* v) const - { - *(storage) = new T(*reinterpret_cast(v)); - } - - /** Clears storage. - * @param storage pointer to a pointer to storage - */ - virtual void clear(void** storage) const - { - delete reinterpret_cast(*storage); - } - - /** Returns type-name as string. - * @return name of type class - */ - virtual std::string type() const - { - return demangledType(); - } - - /** Compares type. - * @param ti type information - * @return true if type matches - */ - virtual bool matches(const std::type_info& ti) const - { - return typeid(T) == ti; - } - - /** Compares two storages. - * @param storage pointer to a pointer to storage - * @param other_storage pointer to a pointer to another storage - * @return true if both storages have same value - */ - bool equals(void** storage, void** other_storage) const - { - T typed_storage = *(reinterpret_cast(*storage)); - T typed_other_storage = *(reinterpret_cast(*other_storage)); - return typed_storage == typed_other_storage; - } - - virtual std::string policy_name() const { - return "owning"; - } - - virtual PolicyType policy_type() const { - return PolicyType::OWNING; - } - }; - - template - class NonOwningAnyPolicy : public BaseAnyPolicy - { - public: - /** Puts provided value pointed by v (untyped to be generic) to storage. - * @param storage pointer to a pointer to storage - * @param v pointer to value - */ - virtual void set(void** storage, const void* v) const - { - *(storage) = const_cast(v); - } - - /** Clears storage. - * @param storage pointer to a pointer to storage - */ - virtual void clear(void** storage) const - { - } - - /** Returns type-name as string. - * @return name of type class - */ - virtual std::string type() const - { - return demangledType(); - } - - /** Compares type. - * @param ti type information - * @return true if type matches - */ - virtual bool matches(const std::type_info& ti) const - { - return typeid(T) == ti; - } - - /** Compares two storages. - * @param storage pointer to a pointer to storage - * @param other_storage pointer to a pointer to another storage - * @return true if both storages have same value - */ - bool equals(void** storage, void** other_storage) const - { - T typed_storage = *(reinterpret_cast(*storage)); - T typed_other_storage = *(reinterpret_cast(*other_storage)); - return typed_storage == typed_other_storage; - } - - virtual std::string policy_name() const { - return "non owning"; - } - - virtual PolicyType policy_type() const { - return PolicyType::NON_OWNING; - } - }; - - template - static BaseAnyPolicy* owning_policy() - { - typedef PointerValueAnyPolicy Policy; - static Policy policy; - return &policy; - } - - template - static BaseAnyPolicy* non_owning_policy() - { - typedef NonOwningAnyPolicy Policy; - static Policy policy; - return &policy; - } - - /** @brief Allows to store objects of arbitrary types - * by using a BaseAnyPolicy and provides a type agnostic API. - * See its usage in CSGObject::Self, CSGObject::set(), CSGObject::get() - * and CSGObject::has(). - * . - */ - class Any - { - public: - /** Used to denote an empty Any object */ - struct Empty; - - /** Empty value constructor */ - Any() : Any(owning_policy(), nullptr) - { - } - - /** Constructor to copy value */ - template - explicit Any(const T& v) : Any(owning_policy(), nullptr) - { - policy->set(&storage, &v); - } - - /** Base constructor */ - Any(BaseAnyPolicy* the_policy, void* the_storage) : policy(the_policy), storage(the_storage) - { - } - - /** Copy constructor */ - Any(const Any& other) : Any(other.policy, nullptr) - { - assert_same_policy_type(other.policy); - policy->set(&storage, other.storage); - } - - /** Assignment operator - * @param other another Any object - * @return Any object - */ - Any& operator=(const Any& other) - { - assert_same_policy_type(other.policy); - policy->clear(&storage); - policy = other.policy; - policy->set(&storage, other.storage); - return *(this); - } - - /** Equality operator - * @param lhs Any object on left hand side - * @param rhs Any object on right hand side - * @return true if both are equal - */ - friend inline bool operator==(const Any& lhs, const Any& rhs); - - /** Inequality operator - * @param lhs Any object on left hand side - * @param rhs Any object on right hand side - * @return false if both are equal - */ - friend inline bool operator!=(const Any& lhs, const Any& rhs); - - /** Destructor */ - ~Any() - { - policy->clear(&storage); - } - - /** Casts hidden value to provided type, fails otherwise. - * @return type-casted value - */ - template - T& as() const - { - if (same_type()) - { - return *(reinterpret_cast(storage)); - } - else - { - throw std::logic_error("Bad cast to " + demangledType() + - " but the type is " + policy->type()); - } - } - - /** @return true if type is same. */ - template - inline bool same_type() const - { - return (policy == owning_policy()) || (policy == non_owning_policy()) || same_type_fallback(); - } - - /** @return true if type-id is same. */ - template - bool same_type_fallback() const - { - return policy->matches(typeid(T)); - } - - /** @return true if Any object is empty. */ - bool empty() const - { - return same_type(); - } - - private: - - void assert_same_policy_type(BaseAnyPolicy* other_policy) { - if (policy->policy_type() != other_policy->policy_type()) { - throw std::logic_error("The policies are different: " + - policy->policy_name() + " and " + - other_policy->policy_name()); - } - } - - private: - BaseAnyPolicy* policy; - void* storage; - }; - - inline bool operator==(const Any& lhs, const Any& rhs) - { - void* lhs_storage = lhs.storage; - void* rhs_storage = rhs.storage; - return lhs.policy == rhs.policy && - lhs.policy->equals(&lhs_storage, &rhs_storage); - } - - inline bool operator!=(const Any& lhs, const Any& rhs) - { - return !(lhs == rhs); - } - - /** Used to denote an empty Any object */ - struct Any::Empty - { - /** Equality operator */ - bool operator==(const Empty& other) const - { - return true; - } - }; - - /** Erases value type i.e. converts it to Any - * For input object of any type, it returns an Any object - * which stores the input object's raw value. It saves the type - * information internally to be recalled later by using recall_type(). - * - * @param v value - * @return Any object with the input value - */ - template - inline Any erase_type(const T& v) - { - return Any(v); - } - - template - inline Any erase_type_non_owning(T* v) - { - return Any(non_owning_policy(), v); - } - - /** Tries to recall Any type, fails when type is wrong. - * Any stores type information of an object internally in a BaseAnyPolicy. - * This function returns type-casted value if the internal type information - * matches with the provided typename, otherwise throws std::logic_error. - * - * @param any object of Any - * @return type-casted value - */ - template - inline T recall_type(const Any& any) - { - return any.as(); - } - + return demangled_string; + } + + enum class PolicyType + { + OWNING, + NON_OWNING + }; + + /** @brief An interface for a policy to store a value. + * Value can be any data like primitive data-types, shogun objects, etc. + * Policy defines how to handle this data. It works with a + * provided memory region and is able to set value, clear it + * and return the type-name as string. + */ + class BaseAnyPolicy + { + public: + /** Puts provided value pointed by v (untyped to be generic) to storage. + * @param storage pointer to a pointer to storage + * @param v pointer to value + */ + virtual void set(void** storage, const void* v) const = 0; + + /** Clears storage. + * @param storage pointer to a pointer to storage + */ + virtual void clear(void** storage) const = 0; + + /** Returns type-name as string. + * @return name of type class + */ + virtual std::string type() const = 0; + + /** Returns type info + * @return type info of value's type + */ + virtual const std::type_info& type_info() const = 0; + + /** Compares type. + * @param ti type information + * @return true if type matches + */ + virtual bool matches(const std::type_info& ti) const = 0; + + /** Compares two storages. + * @param storage pointer to a pointer to storage + * @param other_storage pointer to a pointer to another storage + * @return true if both storages have same value + */ + virtual bool equals(void** storage, void** other_storage) const = 0; + + /** Returns the name of policy. + * @return name of policy + */ + virtual std::string policy_name() const = 0; + + /** Returns the type of policy. + * @return type of policy + */ + virtual PolicyType policy_type() const = 0; + }; + + /** @brief This is one concrete implementation of policy that + * uses void pointers to store values. + */ + template + class PointerValueAnyPolicy : public BaseAnyPolicy + { + public: + /** Puts provided value pointed by v (untyped to be generic) to storage. + * @param storage pointer to a pointer to storage + * @param v pointer to value + */ + virtual void set(void** storage, const void* v) const + { + *(storage) = new T(*reinterpret_cast(v)); + } + + /** Clears storage. + * @param storage pointer to a pointer to storage + */ + virtual void clear(void** storage) const + { + delete reinterpret_cast(*storage); + } + + /** Returns type-name as string. + * @return name of type class + */ + virtual std::string type() const + { + return demangledType(); + } + + /** Returns type info + * @return type info of value's type + */ + virtual const std::type_info& type_info() const + { + return typeid(T); + } + + /** Compares type. + * @param ti type information + * @return true if type matches + */ + virtual bool matches(const std::type_info& ti) const + { + return typeid(T) == ti; + } + + /** Compares two storages. + * @param storage pointer to a pointer to storage + * @param other_storage pointer to a pointer to another storage + * @return true if both storages have same value + */ + bool equals(void** storage, void** other_storage) const + { + T typed_storage = *(reinterpret_cast(*storage)); + T typed_other_storage = *(reinterpret_cast(*other_storage)); + return typed_storage == typed_other_storage; + } + + virtual std::string policy_name() const + { + return "owning"; + } + + virtual PolicyType policy_type() const + { + return PolicyType::OWNING; + } + }; + + template + class NonOwningAnyPolicy : public BaseAnyPolicy + { + public: + /** Puts provided value pointed by v (untyped to be generic) to storage. + * @param storage pointer to a pointer to storage + * @param v pointer to value + */ + virtual void set(void** storage, const void* v) const + { + *(storage) = const_cast(v); + } + + /** Clears storage. + * @param storage pointer to a pointer to storage + */ + virtual void clear(void** storage) const + { + } + + /** Returns type-name as string. + * @return name of type class + */ + virtual std::string type() const + { + return demangledType(); + } + + /** Returns type info + * @return type info of value's type + */ + virtual const std::type_info& type_info() const + { + return typeid(T); + } + + /** Compares type. + * @param ti type information + * @return true if type matches + */ + virtual bool matches(const std::type_info& ti) const + { + return typeid(T) == ti; + } + + /** Compares two storages. + * @param storage pointer to a pointer to storage + * @param other_storage pointer to a pointer to another storage + * @return true if both storages have same value + */ + bool equals(void** storage, void** other_storage) const + { + T typed_storage = *(reinterpret_cast(*storage)); + T typed_other_storage = *(reinterpret_cast(*other_storage)); + return typed_storage == typed_other_storage; + } + + virtual std::string policy_name() const + { + return "non owning"; + } + + virtual PolicyType policy_type() const + { + return PolicyType::NON_OWNING; + } + }; + + template + static BaseAnyPolicy* owning_policy() + { + typedef PointerValueAnyPolicy Policy; + static Policy policy; + return &policy; + } + + template + static BaseAnyPolicy* non_owning_policy() + { + typedef NonOwningAnyPolicy Policy; + static Policy policy; + return &policy; + } + + /** @brief Allows to store objects of arbitrary types + * by using a BaseAnyPolicy and provides a type agnostic API. + * See its usage in CSGObject::Self, CSGObject::set(), CSGObject::get() + * and CSGObject::has(). + * . + */ + class Any + { + public: + /** Used to denote an empty Any object */ + struct Empty; + + /** Empty value constructor */ + Any() : Any(owning_policy(), nullptr) + { + } + + /** Constructor to copy value */ + template + explicit Any(const T& v) : Any(owning_policy(), nullptr) + { + policy->set(&storage, &v); + } + + /** Base constructor */ + Any(BaseAnyPolicy* the_policy, void* the_storage) + : policy(the_policy), storage(the_storage) + { + } + + /** Copy constructor */ + Any(const Any& other) : Any(other.policy, nullptr) + { + assert_same_policy_type(other.policy); + policy->set(&storage, other.storage); + } + + /** Assignment operator + * @param other another Any object + * @return Any object + */ + Any& operator=(const Any& other) + { + assert_same_policy_type(other.policy); + policy->clear(&storage); + policy = other.policy; + policy->set(&storage, other.storage); + return *(this); + } + + /** Equality operator + * @param lhs Any object on left hand side + * @param rhs Any object on right hand side + * @return true if both are equal + */ + friend inline bool operator==(const Any& lhs, const Any& rhs); + + /** Inequality operator + * @param lhs Any object on left hand side + * @param rhs Any object on right hand side + * @return false if both are equal + */ + friend inline bool operator!=(const Any& lhs, const Any& rhs); + + /** Destructor */ + ~Any() + { + policy->clear(&storage); + } + + /** Casts hidden value to provided type, fails otherwise. + * @return type-casted value + */ + template + T& as() const + { + if (same_type()) + { + return *(reinterpret_cast(storage)); + } + else + { + throw std::logic_error( + "Bad cast to " + demangledType() + " but the type is " + + policy->type()); + } + } + + /** @return true if type is same. */ + template + inline bool same_type() const + { + return (policy == owning_policy()) || + (policy == non_owning_policy()) || + same_type_fallback(); + } + + /** @return true if type-id is same. */ + template + bool same_type_fallback() const + { + return policy->matches(typeid(T)); + } + + /** @return true if Any object is empty. */ + bool empty() const + { + return same_type(); + } + + const std::type_info& type_info() const + { + return policy->type_info(); + } + + private: + void assert_same_policy_type(BaseAnyPolicy* other_policy) + { + if (policy->policy_type() != other_policy->policy_type()) + { + throw std::logic_error( + "The policies are different: " + policy->policy_name() + + " and " + other_policy->policy_name()); + } + } + + private: + BaseAnyPolicy* policy; + void* storage; + }; + + inline bool operator==(const Any& lhs, const Any& rhs) + { + void* lhs_storage = lhs.storage; + void* rhs_storage = rhs.storage; + return lhs.policy == rhs.policy && + lhs.policy->equals(&lhs_storage, &rhs_storage); + } + + inline bool operator!=(const Any& lhs, const Any& rhs) + { + return !(lhs == rhs); + } + + /** Used to denote an empty Any object */ + struct Any::Empty + { + /** Equality operator */ + bool operator==(const Empty& other) const + { + return true; + } + }; + + /** Erases value type i.e. converts it to Any + * For input object of any type, it returns an Any object + * which stores the input object's raw value. It saves the type + * information internally to be recalled later by using recall_type(). + * + * @param v value + * @return Any object with the input value + */ + template + inline Any erase_type(const T& v) + { + return Any(v); + } + + template + inline Any erase_type_non_owning(T* v) + { + return Any(non_owning_policy(), v); + } + + /** Tries to recall Any type, fails when type is wrong. + * Any stores type information of an object internally in a BaseAnyPolicy. + * This function returns type-casted value if the internal type information + * matches with the provided typename, otherwise throws std::logic_error. + * + * @param any object of Any + * @return type-casted value + */ + template + inline T recall_type(const Any& any) + { + return any.as(); + } } -#endif //_ANY_H_ +#endif //_ANY_H_ diff --git a/src/shogun/lib/config.h.in b/src/shogun/lib/config.h.in index e3699f1d55c..2cf9d91a036 100644 --- a/src/shogun/lib/config.h.in +++ b/src/shogun/lib/config.h.in @@ -10,11 +10,11 @@ #cmakedefine HAVE_CURL 1 #cmakedefine HAVE_JSON 1 #cmakedefine HAVE_XML 1 -#cmakedefine HAVE_LARGEFILE 1 #cmakedefine HAVE_DOXYGEN 1 #cmakedefine HAVE_LAPACK 1 #cmakedefine HAVE_MVEC 1 #cmakedefine HAVE_PROTOBUF 1 +#cmakedefine HAVE_TFLOGGER 1 #cmakedefine HAVE_ARPACK 1 #cmakedefine HAVE_VIENNACL 1 @@ -42,6 +42,13 @@ /* Tells ViennaCL to use OpenCL as computation backend */ #cmakedefine VIENNACL_WITH_OPENCL 1 +/* Eigen Lapack optimization flags */ +#cmakedefine EIGEN_USE_BLAS 1 +#cmakedefine EIGEN_USE_LAPACKE 1 +#cmakedefine EIGEN_USE_LAPACKE_STRICT 1 +#cmakedefine EIGEN_USE_MKL_VML 1 +#cmakedefine EIGEN_USE_MKL_ALL 1 + /* for linear algebra global backend setups */ #cmakedefine USE_EIGEN3_GLOBAL 1 #cmakedefine USE_VIENNACL_GLOBAL 1 @@ -90,7 +97,6 @@ #cmakedefine HAVE_FDOPEN 1 -#cmakedefine USE_SPINLOCKS 1 #cmakedefine USE_SHORTREAL_KERNELCACHE 1 #cmakedefine USE_BIGSTATES 1 @@ -110,7 +116,6 @@ #cmakedefine USE_GZIP 1 #cmakedefine USE_BZIP2 1 #cmakedefine USE_LZMA 1 -#cmakedefine USE_REFERENCE_COUNTING 1 #cmakedefine USE_SNAPPY 1 #cmakedefine HAVE_SSE2 1 @@ -126,8 +131,6 @@ #cmakedefine HAVE_CXX0X 1 #cmakedefine HAVE_CXX11 1 -#cmakedefine HAVE_CXX11_ATOMIC 1 -#cmakedefine HAVE_STD_UNORDERED_MAP 1 /* does the compiler support abi::__cxa_demangle */ #cmakedefine HAVE_CXA_DEMANGLE 1 diff --git a/src/shogun/lib/cpu.h b/src/shogun/lib/cpu.h index 0916756d799..fa434779062 100644 --- a/src/shogun/lib/cpu.h +++ b/src/shogun/lib/cpu.h @@ -36,9 +36,23 @@ SG_FORCED_INLINE static void CpuRelax() { #ifdef _MSC_VER - _mm_pause(); + _mm_pause(); +#elif defined(__i386__) || defined(__x86_64__) + asm volatile("pause"); +#elif defined(__arm__) || defined(__aarch64__) + asm volatile("yield"); +#elif defined(__powerpc__) || defined(__ppc__) + asm volatile("or 27,27,27"); +#elif defined(__s390__) || defined(__s390x__) + asm volatile("" : : : "memory"); #else - asm("pause"); +#warning "Unknown architecture, defaulting to delaying loop." + static uint32_t bar = 13; + static uint32_t* foo = &bar; + for (unsigned int i = 0; i < 100000; i++) + { + *foo = (*foo * 33) + 17; + } #endif } diff --git a/src/shogun/lib/external/brent.cpp b/src/shogun/lib/external/brent.cpp deleted file mode 100644 index 72b7ec6c1a7..00000000000 --- a/src/shogun/lib/external/brent.cpp +++ /dev/null @@ -1,1525 +0,0 @@ -# include -# include -# include -# include - -using namespace std; - -# include "brent.h" - -namespace shogun -{ - -//****************************************************************************80 - -double glomin ( double a, double b, double c, double m, double e, double t, - func_base& f, double &x ) - -//****************************************************************************80 -// -// Purpose: -// -// GLOMIN seeks a global minimum of a function F(X) in an interval [A,B]. -// -// Discussion: -// -// This function assumes that F(X) is twice continuously differentiable -// over [A,B] and that F''(X) <= M for all X in [A,B]. -// -// Licensing: -// -// This code is distributed under the GNU LGPL license. -// -// Modified: -// -// 06 May 2012 -// -// Author: -// -// Original FORTRAN77 version by Richard Brent. -// C++ version by John Burkardt. -// Modifications by John Denker. -// -// Reference: -// -// Richard Brent, -// Algorithms for Minimization Without Derivatives, -// Dover, 2002, -// ISBN: 0-486-41998-3, -// LC: QA402.5.B74. -// -// Parameters: -// -// Input, double A, B, the endpoints of the interval. -// It must be the case that A < B. -// -// Input, double C, an initial guess for the global -// minimizer. If no good guess is known, C = A or B is acceptable. -// -// Input, double M, the bound on the second derivative. -// -// Input, double E, a positive tolerance, a bound for the -// absolute error in the evaluation of F(X) for any X in [A,B]. -// -// Input, double T, a positive error tolerance. -// -// Input, func_base& F, a user-supplied c++ functor whose -// global minimum is being sought. The input and output -// of F() are of type double. -// -// Output, double &X, the estimated value of the abscissa -// for which F attains its global minimum value in [A,B]. -// -// Output, double GLOMIN, the value F(X). -// -{ - double a0; - double a2; - double a3; - double d0; - double d1; - double d2; - double h; - int k; - double m2; - double macheps; - double p; - double q; - double qs; - double r; - double s; - double sc; - double y; - double y0; - double y1; - double y2; - double y3; - double yb; - double z0; - double z1; - double z2; - - a0 = b; - x = a0; - a2 = a; - y0 = f ( b ); - yb = y0; - y2 = f ( a ); - y = y2; - - if ( y0 < y ) - { - y = y0; - } - else - { - x = a; - } - - if ( m <= 0.0 || b <= a ) - { - return y; - } - - macheps = r8_epsilon ( ); - - m2 = 0.5 * ( 1.0 + 16.0 * macheps ) * m; - - if ( c <= a || b <= c ) - { - sc = 0.5 * ( a + b ); - } - else - { - sc = c; - } - - y1 = f ( sc ); - k = 3; - d0 = a2 - sc; - h = 9.0 / 11.0; - - if ( y1 < y ) - { - x = sc; - y = y1; - } -// -// Loop. -// - for ( ; ; ) - { - d1 = a2 - a0; - d2 = sc - a0; - z2 = b - a2; - z0 = y2 - y1; - z1 = y2 - y0; - r = d1 * d1 * z0 - d0 * d0 * z1; - p = r; - qs = 2.0 * ( d0 * z1 - d1 * z0 ); - q = qs; - - if ( k < 1000000 || y2 <= y ) - { - for ( ; ; ) - { - if ( q * ( r * ( yb - y2 ) + z2 * q * ( ( y2 - y ) + t ) ) < - z2 * m2 * r * ( z2 * q - r ) ) - { - a3 = a2 + r / q; - y3 = f ( a3 ); - - if ( y3 < y ) - { - x = a3; - y = y3; - } - } - k = ( ( 1611 * k ) % 1048576 ); - q = 1.0; - r = ( b - a ) * 0.00001 * ( double ) ( k ); - - if ( z2 <= r ) - { - break; - } - } - } - else - { - k = ( ( 1611 * k ) % 1048576 ); - q = 1.0; - r = ( b - a ) * 0.00001 * ( double ) ( k ); - - while ( r < z2 ) - { - if ( q * ( r * ( yb - y2 ) + z2 * q * ( ( y2 - y ) + t ) ) < - z2 * m2 * r * ( z2 * q - r ) ) - { - a3 = a2 + r / q; - y3 = f ( a3 ); - - if ( y3 < y ) - { - x = a3; - y = y3; - } - } - k = ( ( 1611 * k ) % 1048576 ); - q = 1.0; - r = ( b - a ) * 0.00001 * ( double ) ( k ); - } - } - - r = m2 * d0 * d1 * d2; - s = sqrt ( ( ( y2 - y ) + t ) / m2 ); - h = 0.5 * ( 1.0 + h ); - p = h * ( p + 2.0 * r * s ); - q = q + 0.5 * qs; - r = - 0.5 * ( d0 + ( z0 + 2.01 * e ) / ( d0 * m2 ) ); - - if ( r < s || d0 < 0.0 ) - { - r = a2 + s; - } - else - { - r = a2 + r; - } - - if ( 0.0 < p * q ) - { - a3 = a2 + p / q; - } - else - { - a3 = r; - } - - for ( ; ; ) - { - a3 = r8_max ( a3, r ); - - if ( b <= a3 ) - { - a3 = b; - y3 = yb; - } - else - { - y3 = f ( a3 ); - } - - if ( y3 < y ) - { - x = a3; - y = y3; - } - - d0 = a3 - a2; - - if ( a3 <= r ) - { - break; - } - - p = 2.0 * ( y2 - y3 ) / ( m * d0 ); - - if ( ( 1.0 + 9.0 * macheps ) * d0 <= r8_abs ( p ) ) - { - break; - } - - if ( 0.5 * m2 * ( d0 * d0 + p * p ) <= ( y2 - y ) + ( y3 - y ) + 2.0 * t ) - { - break; - } - a3 = 0.5 * ( a2 + a3 ); - h = 0.9 * h; - } - - if ( b <= a3 ) - { - break; - } - - a0 = sc; - sc = a2; - a2 = a3; - y0 = y1; - y1 = y2; - y2 = y3; - } - - return y; -} -//****************************************************************************80 - -double local_min ( double a, double b, double t, func_base& f, - double &x ) - -//****************************************************************************80 -// -// Purpose: -// -// LOCAL_MIN seeks a local minimum of a function F(X) in an interval [A,B]. -// -// Discussion: -// -// The method used is a combination of golden section search and -// successive parabolic interpolation. Convergence is never much slower -// than that for a Fibonacci search. If F has a continuous second -// derivative which is positive at the minimum (which is not at A or -// B), then convergence is superlinear, and usually of the order of -// about 1.324.... -// -// The values EPS and T define a tolerance TOL = EPS * abs ( X ) + T. -// F is never evaluated at two points closer than TOL. -// -// If F is a unimodal function and the computed values of F are always -// unimodal when separated by at least SQEPS * abs ( X ) + (T/3), then -// LOCAL_MIN approximates the abscissa of the global minimum of F on the -// interval [A,B] with an error less than 3*SQEPS*abs(LOCAL_MIN)+T. -// -// If F is not unimodal, then LOCAL_MIN may approximate a local, but -// perhaps non-global, minimum to the same accuracy. -// -// Licensing: -// -// This code is distributed under the GNU LGPL license. -// -// Modified: -// -// 17 July 2011 -// -// Author: -// -// Original FORTRAN77 version by Richard Brent. -// C++ version by John Burkardt. -// Modifications by John Denker. -// -// Reference: -// -// Richard Brent, -// Algorithms for Minimization Without Derivatives, -// Dover, 2002, -// ISBN: 0-486-41998-3, -// LC: QA402.5.B74. -// -// Parameters: -// -// Input, double A, B, the endpoints of the interval. -// -// Input, double T, a positive absolute error tolerance. -// -// Input, func_base& F, a user-supplied c++ functor whose -// local minimum is being sought. The input and output -// of F() are of type double. -// -// Output, double &X, the estimated value of an abscissa -// for which F attains a local minimum value in [A,B]. -// -// Output, double LOCAL_MIN, the value F(X). -// -{ - double c; - double d = 0.0; - double e; - double eps; - double fu; - double fv; - double fw; - double fx; - double m; - double p; - double q; - double r; - double sa; - double sb; - double t2; - double tol; - double u; - double v; - double w; -// -// C is the square of the inverse of the golden ratio. -// - c = 0.5 * ( 3.0 - sqrt ( 5.0 ) ); - - eps = sqrt ( r8_epsilon ( ) ); - - sa = a; - sb = b; - x = sa + c * ( b - a ); - w = x; - v = w; - e = 0.0; - fx = f ( x ); - fw = fx; - fv = fw; - - for ( ; ; ) - { - m = 0.5 * ( sa + sb ) ; - tol = eps * r8_abs ( x ) + t; - t2 = 2.0 * tol; -// -// Check the stopping criterion. -// - if ( r8_abs ( x - m ) <= t2 - 0.5 * ( sb - sa ) ) - { - break; - } -// -// Fit a parabola. -// - r = 0.0; - q = r; - p = q; - - if ( tol < r8_abs ( e ) ) - { - r = ( x - w ) * ( fx - fv ); - q = ( x - v ) * ( fx - fw ); - p = ( x - v ) * q - ( x - w ) * r; - q = 2.0 * ( q - r ); - if ( 0.0 < q ) - { - p = - p; - } - q = r8_abs ( q ); - r = e; - e = d; - } - - if ( r8_abs ( p ) < r8_abs ( 0.5 * q * r ) && - q * ( sa - x ) < p && - p < q * ( sb - x ) ) - { -// -// Take the parabolic interpolation step. -// - d = p / q; - u = x + d; -// -// F must not be evaluated too close to A or B. -// - if ( ( u - sa ) < t2 || ( sb - u ) < t2 ) - { - if ( x < m ) - { - d = tol; - } - else - { - d = - tol; - } - } - } -// -// A golden-section step. -// - else - { - if ( x < m ) - { - e = sb - x; - } - else - { - e = sa - x; - } - d = c * e; - } -// -// F must not be evaluated too close to X. -// - if ( tol <= r8_abs ( d ) ) - { - u = x + d; - } - else if ( 0.0 < d ) - { - u = x + tol; - } - else - { - u = x - tol; - } - - fu = f ( u ); -// -// Update A, B, V, W, and X. -// - if ( fu <= fx ) - { - if ( u < x ) - { - sb = x; - } - else - { - sa = x; - } - v = w; - fv = fw; - w = x; - fw = fx; - x = u; - fx = fu; - } - else - { - if ( u < x ) - { - sa = u; - } - else - { - sb = u; - } - - if ( fu <= fw || w == x ) - { - v = w; - fv = fw; - w = u; - fw = fu; - } - else if ( fu <= fv || v == x || v== w ) - { - v = u; - fv = fu; - } - } - } - return fx; -} -//****************************************************************************80 - -double local_min_rc ( double &a, double &b, int &status, double value ) - -//****************************************************************************80 -// -// Purpose: -// -// LOCAL_MIN_RC seeks a minimizer of a scalar function of a scalar variable. -// -// Discussion: -// -// This routine seeks an approximation to the point where a function -// F attains a minimum on the interval (A,B). -// -// The method used is a combination of golden section search and -// successive parabolic interpolation. Convergence is never much -// slower than that for a Fibonacci search. If F has a continuous -// second derivative which is positive at the minimum (which is not -// at A or B), then convergence is superlinear, and usually of the -// order of about 1.324... -// -// The routine is a revised version of the Brent local minimization -// algorithm, using reverse communication. -// -// It is worth stating explicitly that this routine will NOT be -// able to detect a minimizer that occurs at either initial endpoint -// A or B. If this is a concern to the user, then the user must -// either ensure that the initial interval is larger, or to check -// the function value at the returned minimizer against the values -// at either endpoint. -// -// Licensing: -// -// This code is distributed under the GNU LGPL license. -// -// Modified: -// -// 17 July 2011 -// -// Author: -// -// John Burkardt -// -// Reference: -// -// Richard Brent, -// Algorithms for Minimization Without Derivatives, -// Dover, 2002, -// ISBN: 0-486-41998-3, -// LC: QA402.5.B74. -// -// David Kahaner, Cleve Moler, Steven Nash, -// Numerical Methods and Software, -// Prentice Hall, 1989, -// ISBN: 0-13-627258-4, -// LC: TA345.K34. -// -// Parameters -// -// Input/output, double &A, &B. On input, the left and right -// endpoints of the initial interval. On output, the lower and upper -// bounds for an interval containing the minimizer. It is required -// that A < B. -// -// Input/output, int &STATUS, used to communicate between -// the user and the routine. The user only sets STATUS to zero on the first -// call, to indicate that this is a startup call. The routine returns STATUS -// positive to request that the function be evaluated at ARG, or returns -// STATUS as 0, to indicate that the iteration is complete and that -// ARG is the estimated minimizer. -// -// Input, double VALUE, the function value at ARG, as requested -// by the routine on the previous call. -// -// Output, double LOCAL_MIN_RC, the currently considered point. -// On return with STATUS positive, the user is requested to evaluate the -// function at this point, and return the value in VALUE. On return with -// STATUS zero, this is the routine's estimate for the function minimizer. -// -// Local parameters: -// -// C is the squared inverse of the golden ratio. -// -// EPS is the square root of the relative machine precision. -// -{ - static double arg; - static double c; - static double d; - static double e; - static double eps; - static double fu; - static double fv; - static double fw; - static double fx; - static double midpoint; - static double p; - static double q; - static double r; - static double tol; - static double tol1; - static double tol2; - static double u; - static double v; - static double w; - static double x; -// -// STATUS (INPUT) = 0, startup. -// - if ( status == 0 ) - { - if ( b <= a ) - { - cout << "\n"; - cout << "LOCAL_MIN_RC - Fatal error!\n"; - cout << " A < B is required, but\n"; - cout << " A = " << a << "\n"; - cout << " B = " << b << "\n"; - status = -1; - exit ( 1 ); - } - c = 0.5 * ( 3.0 - sqrt ( 5.0 ) ); - - eps = sqrt ( r8_epsilon ( ) ); - tol = r8_epsilon ( ); - - v = a + c * ( b - a ); - w = v; - x = v; - e = 0.0; - - status = 1; - arg = x; - - return arg; - } -// -// STATUS (INPUT) = 1, return with initial function value of FX. -// - else if ( status == 1 ) - { - fx = value; - fv = fx; - fw = fx; - } -// -// STATUS (INPUT) = 2 or more, update the data. -// - else if ( 2 <= status ) - { - fu = value; - - if ( fu <= fx ) - { - if ( x <= u ) - { - a = x; - } - else - { - b = x; - } - v = w; - fv = fw; - w = x; - fw = fx; - x = u; - fx = fu; - } - else - { - if ( u < x ) - { - a = u; - } - else - { - b = u; - } - - if ( fu <= fw || w == x ) - { - v = w; - fv = fw; - w = u; - fw = fu; - } - else if ( fu <= fv || v == x || v == w ) - { - v = u; - fv = fu; - } - } - } -// -// Take the next step. -// - midpoint = 0.5 * ( a + b ); - tol1 = eps * r8_abs ( x ) + tol / 3.0; - tol2 = 2.0 * tol1; -// -// If the stopping criterion is satisfied, we can exit. -// - if ( r8_abs ( x - midpoint ) <= ( tol2 - 0.5 * ( b - a ) ) ) - { - status = 0; - return arg; - } -// -// Is golden-section necessary? -// - if ( r8_abs ( e ) <= tol1 ) - { - if ( midpoint <= x ) - { - e = a - x; - } - else - { - e = b - x; - } - d = c * e; - } -// -// Consider fitting a parabola. -// - else - { - r = ( x - w ) * ( fx - fv ); - q = ( x - v ) * ( fx - fw ); - p = ( x - v ) * q - ( x - w ) * r; - q = 2.0 * ( q - r ); - if ( 0.0 < q ) - { - p = - p; - } - q = r8_abs ( q ); - r = e; - e = d; -// -// Choose a golden-section step if the parabola is not advised. -// - if ( - ( r8_abs ( 0.5 * q * r ) <= r8_abs ( p ) ) || - ( p <= q * ( a - x ) ) || - ( q * ( b - x ) <= p ) ) - { - if ( midpoint <= x ) - { - e = a - x; - } - else - { - e = b - x; - } - d = c * e; - } -// -// Choose a parabolic interpolation step. -// - else - { - d = p / q; - u = x + d; - - if ( ( u - a ) < tol2 ) - { - d = tol1 * r8_sign ( midpoint - x ); - } - - if ( ( b - u ) < tol2 ) - { - d = tol1 * r8_sign ( midpoint - x ); - } - } - } -// -// F must not be evaluated too close to X. -// - if ( tol1 <= r8_abs ( d ) ) - { - u = x + d; - } - if ( r8_abs ( d ) < tol1 ) - { - u = x + tol1 * r8_sign ( d ); - } -// -// Request value of F(U). -// - arg = u; - status = status + 1; - - return arg; -} -//****************************************************************************80 - -double r8_abs ( double x ) - -//****************************************************************************80 -// -// Purpose: -// -// R8_ABS returns the absolute value of an R8. -// -// Licensing: -// -// This code is distributed under the GNU LGPL license. -// -// Modified: -// -// 07 May 2006 -// -// Author: -// -// John Burkardt -// -// Parameters: -// -// Input, double X, the quantity whose absolute value is desired. -// -// Output, double R8_ABS, the absolute value of X. -// -{ - double value; - - if ( 0.0 <= x ) - { - value = x; - } - else - { - value = - x; - } - return value; -} -//****************************************************************************80 - -double r8_epsilon ( ) - -//****************************************************************************80 -// -// Purpose: -// -// R8_EPSILON returns the R8 round off unit. -// -// Discussion: -// -// R8_EPSILON is a number R which is a power of 2 with the property that, -// to the precision of the computer's arithmetic, -// 1 < 1 + R -// but -// 1 = ( 1 + R / 2 ) -// -// Licensing: -// -// This code is distributed under the GNU LGPL license. -// -// Modified: -// -// 08 May 2006 -// -// Author: -// -// John Burkardt -// -// Parameters: -// -// Output, double R8_EPSILON, the double precision round-off unit. -// -{ - double r; - - r = 1.0; - - while ( 1.0 < ( double ) ( 1.0 + r ) ) - { - r = r / 2.0; - } - - return ( 2.0 * r ); -} -//****************************************************************************80 - -double r8_max ( double x, double y ) - -//****************************************************************************80 -// -// Purpose: -// -// R8_MAX returns the maximum of two R8's. -// -// Licensing: -// -// This code is distributed under the GNU LGPL license. -// -// Modified: -// -// 18 August 2004 -// -// Author: -// -// John Burkardt -// -// Parameters: -// -// Input, double X, Y, the quantities to compare. -// -// Output, double R8_MAX, the maximum of X and Y. -// -{ - double value; - - if ( y < x ) - { - value = x; - } - else - { - value = y; - } - return value; -} -//****************************************************************************80 - -double r8_sign ( double x ) - -//****************************************************************************80 -// -// Purpose: -// -// R8_SIGN returns the sign of an R8. -// -// Licensing: -// -// This code is distributed under the GNU LGPL license. -// -// Modified: -// -// 18 October 2004 -// -// Author: -// -// John Burkardt -// -// Parameters: -// -// Input, double X, the number whose sign is desired. -// -// Output, double R8_SIGN, the sign of X. -// -{ - double value; - - if ( x < 0.0 ) - { - value = -1.0; - } - else - { - value = 1.0; - } - return value; -} -//****************************************************************************80 - -void timestamp ( ) - -//****************************************************************************80 -// -// Purpose: -// -// TIMESTAMP prints the current YMDHMS date as a time stamp. -// -// Example: -// -// 31 May 2001 09:45:54 AM -// -// Licensing: -// -// This code is distributed under the GNU LGPL license. -// -// Modified: -// -// 24 September 2003 -// -// Author: -// -// John Burkardt -// -// Parameters: -// -// None -// -{ -# define TIME_SIZE 40 - - static char time_buffer[TIME_SIZE]; - const struct tm *tm; - time_t now; - - now = time ( NULL ); - tm = localtime ( &now ); - - strftime ( time_buffer, TIME_SIZE, "%d %B %Y %I:%M:%S %p", tm ); - - cout << time_buffer << "\n"; - - return; -# undef TIME_SIZE -} -//****************************************************************************80 - -double zero ( double a, double b, double t, func_base& f ) - -//****************************************************************************80 -// -// Purpose: -// -// ZERO seeks the root of a function F(X) in an interval [A,B]. -// -// Discussion: -// -// The interval [A,B] must be a change of sign interval for F. -// That is, F(A) and F(B) must be of opposite signs. Then -// assuming that F is continuous implies the existence of at least -// one value C between A and B for which F(C) = 0. -// -// The location of the zero is determined to within an accuracy -// of 6 * MACHEPS * r8_abs ( C ) + 2 * T. -// -// Licensing: -// -// This code is distributed under the GNU LGPL license. -// -// Modified: -// -// 06 May 2012 -// -// Author: -// -// Original FORTRAN77 version by Richard Brent. -// C++ version by John Burkardt. -// Modifications by John Denker. -// -// Reference: -// -// Richard Brent, -// Algorithms for Minimization Without Derivatives, -// Dover, 2002, -// ISBN: 0-486-41998-3, -// LC: QA402.5.B74. -// -// Parameters: -// -// Input, double A, B, the endpoints of the change of sign interval. -// -// Input, double T, a positive error tolerance. -// -// Input, func_base& F, the name of a user-supplied c++ functor -// whose zero is being sought. The input and output -// of F() are of type double. -// -// Output, double ZERO, the estimated value of a zero of -// the function F. -// -{ - double c; - double d; - double e; - double fa; - double fb; - double fc; - double m; - double macheps; - double p; - double q; - double r; - double s; - double sa; - double sb; - double tol; -// -// Make local copies of A and B. -// - sa = a; - sb = b; - fa = f ( sa ); - fb = f ( sb ); - - c = sa; - fc = fa; - e = sb - sa; - d = e; - - macheps = r8_epsilon ( ); - - for ( ; ; ) - { - if ( r8_abs ( fc ) < r8_abs ( fb ) ) - { - sa = sb; - sb = c; - c = sa; - fa = fb; - fb = fc; - fc = fa; - } - - tol = 2.0 * macheps * r8_abs ( sb ) + t; - m = 0.5 * ( c - sb ); - - if ( r8_abs ( m ) <= tol || fb == 0.0 ) - { - break; - } - - if ( r8_abs ( e ) < tol || r8_abs ( fa ) <= r8_abs ( fb ) ) - { - e = m; - d = e; - } - else - { - s = fb / fa; - - if ( sa == c ) - { - p = 2.0 * m * s; - q = 1.0 - s; - } - else - { - q = fa / fc; - r = fb / fc; - p = s * ( 2.0 * m * a * ( q - r ) - ( sb - sa ) * ( r - 1.0 ) ); - q = ( q - 1.0 ) * ( r - 1.0 ) * ( s - 1.0 ); - } - - if ( 0.0 < p ) - { - q = - q; - } - else - { - p = - p; - } - - s = e; - e = d; - - if ( 2.0 * p < 3.0 * m * q - r8_abs ( tol * q ) && - p < r8_abs ( 0.5 * s * q ) ) - { - d = p / q; - } - else - { - e = m; - d = e; - } - } - sa = sb; - fa = fb; - - if ( tol < r8_abs ( d ) ) - { - sb = sb + d; - } - else if ( 0.0 < m ) - { - sb = sb + tol; - } - else - { - sb = sb - tol; - } - - fb = f ( sb ); - - if ( ( 0.0 < fb && 0.0 < fc ) || ( fb <= 0.0 && fc <= 0.0 ) ) - { - c = sa; - fc = fa; - e = sb - sa; - d = e; - } - } - return sb; -} -//****************************************************************************80 - -void zero_rc ( double a, double b, double t, double &arg, int &status, - double value ) - -//****************************************************************************80 -// -// Purpose: -// -// ZERO_RC seeks the root of a function F(X) using reverse communication. -// -// Discussion: -// -// The interval [A,B] must be a change of sign interval for F. -// That is, F(A) and F(B) must be of opposite signs. Then -// assuming that F is continuous implies the existence of at least -// one value C between A and B for which F(C) = 0. -// -// The location of the zero is determined to within an accuracy -// of 6 * MACHEPS * r8_abs ( C ) + 2 * T. -// -// The routine is a revised version of the Brent zero finder -// algorithm, using reverse communication. -// -// Licensing: -// -// This code is distributed under the GNU LGPL license. -// -// Modified: -// -// 17 July 2011 -// -// Author: -// -// John Burkardt -// -// Reference: -// -// Richard Brent, -// Algorithms for Minimization Without Derivatives, -// Dover, 2002, -// ISBN: 0-486-41998-3, -// LC: QA402.5.B74. -// -// Parameters: -// -// Input, double A, B, the endpoints of the change of sign interval. -// -// Input, double T, a positive error tolerance. -// -// Output, double &ARG, the currently considered point. The user -// does not need to initialize this value. On return with STATUS positive, -// the user is requested to evaluate the function at ARG, and return -// the value in VALUE. On return with STATUS zero, ARG is the routine's -// estimate for the function's zero. -// -// Input/output, int &STATUS, used to communicate between -// the user and the routine. The user only sets STATUS to zero on the first -// call, to indicate that this is a startup call. The routine returns STATUS -// positive to request that the function be evaluated at ARG, or returns -// STATUS as 0, to indicate that the iteration is complete and that -// ARG is the estimated zero -// -// Input, double VALUE, the function value at ARG, as requested -// by the routine on the previous call. -// -{ - static double c; - static double d; - static double e; - static double fa; - static double fb; - static double fc; - double m; - static double macheps; - double p; - double q; - double r; - double s; - static double sa; - static double sb; - double tol; -// -// Input STATUS = 0. -// Initialize, request F(A). -// - if ( status == 0 ) - { - macheps = r8_epsilon ( ); - - sa = a; - sb = b; - e = sb - sa; - d = e; - - status = 1; - arg = a; - return; - } -// -// Input STATUS = 1. -// Receive F(A), request F(B). -// - else if ( status == 1 ) - { - fa = value; - status = 2; - arg = sb; - return; - } -// -// Input STATUS = 2 -// Receive F(B). -// - else if ( status == 2 ) - { - fb = value; - - if ( 0.0 < fa * fb ) - { - status = -1; - return; - } - c = sa; - fc = fa; - } - else - { - fb = value; - - if ( ( 0.0 < fb && 0.0 < fc ) || ( fb <= 0.0 && fc <= 0.0 ) ) - { - c = sa; - fc = fa; - e = sb - sa; - d = e; - } - } -// -// Compute the next point at which a function value is requested. -// - if ( r8_abs ( fc ) < r8_abs ( fb ) ) - { - sa = sb; - sb = c; - c = sa; - fa = fb; - fb = fc; - fc = fa; - } - - tol = 2.0 * macheps * r8_abs ( sb ) + t; - m = 0.5 * ( c - sb ); - - if ( r8_abs ( m ) <= tol || fb == 0.0 ) - { - status = 0; - arg = sb; - return; - } - - if ( r8_abs ( e ) < tol || r8_abs ( fa ) <= r8_abs ( fb ) ) - { - e = m; - d = e; - } - else - { - s = fb / fa; - - if ( sa == c ) - { - p = 2.0 * m * s; - q = 1.0 - s; - } - else - { - q = fa / fc; - r = fb / fc; - p = s * ( 2.0 * m * a * ( q - r ) - ( sb - sa ) * ( r - 1.0 ) ); - q = ( q - 1.0 ) * ( r - 1.0 ) * ( s - 1.0 ); - } - - if ( 0.0 < p ) - { - q = - q; - } - else - { - p = - p; - } - s = e; - e = d; - - if ( 2.0 * p < 3.0 * m * q - r8_abs ( tol * q ) && - p < r8_abs ( 0.5 * s * q ) ) - { - d = p / q; - } - else - { - e = m; - d = e; - } - } - - sa = sb; - fa = fb; - - if ( tol < r8_abs ( d ) ) - { - sb = sb + d; - } - else if ( 0.0 < m ) - { - sb = sb + tol; - } - else - { - sb = sb - tol; - } - - arg = sb; - status = status + 1; - - return; -} - -// ====================================================================== -// === Simple wrapper functions -// === for convenience and/or compatibility. -// -// === The three functions are the same as above, -// === except that they take a plain function F -// === instead of a c++ functor. In all cases, the -// === input and output of F() are of type double. - -typedef double DoubleOfDouble (double); - -class func_wrapper : public func_base { - DoubleOfDouble* func; -public: - func_wrapper(DoubleOfDouble* f) { - func = f; - } - virtual double operator() (double x){ - return func(x); - } -}; - -//****************************************************************************80 - -double glomin ( double a, double b, double c, double m, double e, - double t, double f ( double x ), double &x ){ - func_wrapper foo(f); - return glomin(a, b, c, m, e, t, foo, x); -} - -//****************************************************************************80 - -double local_min ( double a, double b, double t, double f ( double x ), - double &x ){ - func_wrapper foo(f); - return local_min(a, b, t, foo, x); -} - -//****************************************************************************80 - -double zero ( double a, double b, double t, double f ( double x ) ){ - func_wrapper foo(f); - return zero(a, b, t, foo); -} - -// ====================================================================== -// Generally useful functor to evaluate a monic polynomial. -// For details, see class definition in brent.hpp - -double monicPoly::operator()(double x){ - double rslt(1); - for (int ii = coeff.size()-1; ii >= 0; ii--){ - rslt *= x; - rslt += coeff[ii]; - } - return rslt; -} - -// Similarly, evaluate a general polynomial (not necessarily monic): -double Poly::operator()(double x){ - double rslt(0); - for (int ii = coeff.size()-1; ii >= 0; ii--){ - rslt *= x; - rslt += coeff[ii]; - } - return rslt; -} - -} diff --git a/src/shogun/lib/external/brent.h b/src/shogun/lib/external/brent.h deleted file mode 100644 index 4568561c848..00000000000 --- a/src/shogun/lib/external/brent.h +++ /dev/null @@ -1,58 +0,0 @@ -#include -namespace shogun -{ -#ifndef DOXYGEN_SHOULD_SKIP_THIS -class func_base{ -public: - virtual double operator() (double) = 0; -}; - -class monicPoly : public func_base { -public: - std::vector coeff; - virtual double operator() (double x); -// constructors: - monicPoly(const size_t degree) - : coeff(degree) {} - monicPoly(const std::vector& v) - : coeff(v) {} - monicPoly(const double* c, size_t degree) - : coeff(std::vector(c, c+degree)) {} -}; - -class Poly : public func_base { -public: - std::vector coeff; // a vector of size nterms i.e. 1+degree - virtual double operator() (double x); -// constructors: - Poly(const size_t degree) - : coeff(1+degree) {} - Poly(const std::vector& v) - : coeff(v) {} - Poly(const double* c, size_t degree) - : coeff(std::vector(c, 1+c+degree)) {} -}; - -double glomin ( double a, double b, double c, double m, double e, double t, - func_base& f, double &x ); -double local_min ( double a, double b, double t, func_base& f, - double &x ); -double local_min_rc ( double &a, double &b, int &status, double value ); -double r8_abs ( double x ); -double r8_epsilon ( ); -double r8_max ( double x, double y ); -double r8_sign ( double x ); -void timestamp ( ); -double zero ( double a, double b, double t, func_base& f ); -void zero_rc ( double a, double b, double t, double &arg, int &status, - double value ); - -// === simple wrapper functions -// === for convenience and/or compatibility -double glomin ( double a, double b, double c, double m, double e, double t, - double f ( double x ), double &x ); -double local_min ( double a, double b, double t, double f ( double x ), - double &x ); -double zero ( double a, double b, double t, double f ( double x ) ); -#endif -} diff --git a/src/shogun/lib/external/cdflib.cpp b/src/shogun/lib/external/cdflib.cpp index a89515ffc99..3983ddace6c 100644 --- a/src/shogun/lib/external/cdflib.cpp +++ b/src/shogun/lib/external/cdflib.cpp @@ -10049,7 +10049,7 @@ void negative_binomial_cdf_values ( int *n_data, int *f, int *s, double *p, 1, 2, 3, 0, 1, 2 }; - if ( n_data < 0 ) + if ( *n_data < 0 ) { *n_data = 0; } diff --git a/src/shogun/lib/external/gpdt.cpp b/src/shogun/lib/external/gpdt.cpp deleted file mode 100644 index d140b3e9a2f..00000000000 --- a/src/shogun/lib/external/gpdt.cpp +++ /dev/null @@ -1,322 +0,0 @@ -/****************************************************************************** - *** GPDT - Gradient Projection Decomposition Technique *** - ****************************************************************************** - *** *** - *** GPDT is a C++ software designed to train large-scale Support Vector *** - *** Machines for binary classification in both scalar and distributed *** - *** memory parallel environments. It uses the Joachims' problem *** - *** decomposition technique to split the whole quadratic programming (QP) *** - *** problem into a sequence of smaller QP subproblems, each one being *** - *** solved by a suitable gradient projection method (GPM). The presently *** - *** implemented GPMs are the Generalized Variable Projection Method *** - *** GVPM (T. Serafini, G. Zanghirati, L. Zanni, "Gradient Projection *** - *** Methods for Quadratic Programs and Applications in Training Support *** - *** Vector Machines"; Optim. Meth. Soft. 20, 2005, 353-378) and the *** - *** Dai-Fletcher Method DFGPM (Y. Dai and R. Fletcher,"New Algorithms for *** - *** Singly Linear Constrained Quadratic Programs Subject to Lower and *** - *** Upper Bounds"; Math. Prog. to appear). *** - *** *** - *** Authors: *** - *** Thomas Serafini, Luca Zanni *** - *** Dept. of Mathematics, University of Modena and Reggio Emilia - ITALY *** - *** serafini.thomas@unimo.it, zanni.luca@unimo.it *** - *** Gaetano Zanghirati *** - *** Dept. of Mathematics, University of Ferrara - ITALY *** - *** g.zanghirati@unife.it *** - *** *** - *** Software homepage: http://dm.unife.it/gpdt *** - *** *** - *** This work is supported by the Italian FIRB Projects *** - *** 'Statistical Learning: Theory, Algorithms and Applications' *** - *** (grant RBAU01877P), http://slipguru.disi.unige.it/ASTA *** - *** and *** - *** 'Parallel Algorithms and Numerical Nonlinear Optimization' *** - *** (grant RBAU01JYPN), http://dm.unife.it/pn2o *** - *** *** - *** Copyright (C) 2004-2008 by T. Serafini, G. Zanghirati, L. Zanni. *** - *** *** - *** COPYRIGHT NOTIFICATION *** - *** *** - *** Permission to copy and modify this software and its documentation *** - *** for internal research use is granted, provided that this notice is *** - *** retained thereon and on all copies or modifications. The authors and *** - *** their respective Universities makes no representations as to the *** - *** suitability and operability of this software for any purpose. It is *** - *** provided "as is" without express or implied warranty. *** - *** Use of this software for commercial purposes is expressly prohibited *** - *** without contacting the authors. *** - *** *** - *** This program is free software; you can redistribute it and/or modify *** - *** it under the terms of the GNU General Public License as published by *** - *** the Free Software Foundation; either version 3 of the License, or *** - *** (at your option) any later version. *** - *** *** - *** This program is distributed in the hope that it will be useful, *** - *** but WITHOUT ANY WARRANTY; without even the implied warranty of *** - *** MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the *** - *** GNU General Public License for more details. *** - *** *** - *** You should have received a copy of the GNU General Public License *** - *** along with this program; if not, write to the Free Software *** - *** Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. *** - *** *** - *** File: gpdt.cpp *** - *** Type: scalar *** - *** Version: 1.0 *** - *** Date: October, 2005 *** - *** Revision: 1 *** - *** *** - *** SHOGUN adaptions Written (W) 2006-2008 Soeren Sonnenburg *** - ******************************************************************************/ -#include -#ifdef USE_GPL_SHOGUN -#include -#include -#include -#include -#include -#include -#include - -using namespace shogun; - -#ifndef DOXYGEN_SHOULD_SKIP_THIS -void fatalError(const char *msg1, const char *msg2); - -/******************************************************************************/ -/*** Class constructor ***/ -/******************************************************************************/ -QPproblem::QPproblem() -{ - /*** set problem defaults ***/ - maxmw = 40; - c_const = 10.0; - projection_solver = SOLVER_FLETCHER; - projection_projector = 1; - PreprocessMode = 0; - delta = 1.0e-3; - DELTAsv = EPS_SV; - ker_type = 2; - chunk_size = 400; - q = -1; - y = NULL; - tau_proximal = 0.0; - dim = 1; -} - -/******************************************************************************/ -/*** Class destructor ***/ -/******************************************************************************/ -QPproblem::~QPproblem() -{ - //if (y != NULL) free(y); -} - -/******************************************************************************/ -/*** Setter method for the subproblem features ***/ -/******************************************************************************/ -void QPproblem::copy_subproblem(QPproblem* dst, QPproblem* p, int32_t len, int32_t *perm) -{ - int32_t k; - - *dst=*p; - dst->ell = len; - - dst->KER->SetSubproblem(p->KER, len, perm); - dst->y = SG_MALLOC(int32_t, len); - for (k = 0; k < len; k++) - dst->y[k] = p->y[perm[k]]; -} - -namespace shogun -{ -/******************************************************************************/ -/*** Extract the samples information from an SVMlight-compliant data file ***/ -/******************************************************************************/ -int32_t prescan_document(char *file, int32_t *lines, int32_t *vlen, int32_t *ll) -{ - FILE *fl; - int32_t ic; - char c; - int64_t current_length, current_vlen; - - if ((fl = fopen (file, "r")) == NULL) - return(-1); - current_length = 0; - current_vlen = 0; - - *ll = 0; /* length of the longest input line (the read buffer should - be allocated with this size) */ - *lines = 1; /* number of lines in the file */ - *vlen = 0; /* max number of nonzero components in a vector */ - - while ((ic = getc(fl)) != EOF) - { - c = (char)ic; - current_length++; - - if (c == ' ') - current_vlen++; - - if (c == '\n') - { - (*lines)++; - if (current_length > (*ll)) - *ll = current_length; - if (current_vlen > (*vlen)) - *vlen = current_vlen; - current_length = 0; - current_vlen = 0; - } - } - fclose(fl); - return(0); -} -} -/******************************************************************************/ -/*** return 1 if problem is single class, 0 if two-class ***/ -/******************************************************************************/ -int32_t QPproblem::Check2Class() -{ - int32_t i; - - for (i = 1; i < ell; i++) - if (y[i] != y[0]) - return 0; - return 1; -} - -namespace shogun -{ -/******************************************************************************/ -/*** Compute the size of data splitting for preprocessing ***/ -/******************************************************************************/ -void SplitParts( - int32_t n, int32_t part, int32_t parts, int32_t *dim, int32_t *off) -{ - int32_t r; - - r = n % parts; - *dim = n / parts; - - if (part < r) - { - (*dim)++; - *off = *dim * part; - } - else - *off = *dim * part + r; -} -} -/******************************************************************************/ -/*** Kernel class constructor ***/ -/******************************************************************************/ -sKernel::sKernel (CKernel* k, int32_t l) -{ - kernel=k; - ell=l; - nor = NULL; - vaux = NULL; - lx = NULL; - ix = NULL; - x = NULL; - IsSubproblem = 0; - KernelEvaluations = 0.0; -} - -/******************************************************************************/ -/*** Set the problem data for kernel evaluation ***/ -/******************************************************************************/ -void sKernel::SetData( - float32_t **x_, int32_t **ix_, int32_t *lx_, int32_t _ell, int32_t _dim) -{ - int32_t i, j, k; - - dim = _dim; - ell = _ell; - nor = SG_MALLOC(float64_t, ell); - vaux = SG_CALLOC(float32_t, dim); - - IsSubproblem = 0; - x = x_; - ix = ix_; - lx = lx_; - - // unroll one (sparse) vector - vauxRow = 0; - i = vauxRow; - for (k = 0; k < lx[i]; k++) - vaux[ix[i][k]] = x[i][k]; - - // compute the squared Euclidean norm of each vector - for (i = 0; i < ell; i++) - { - nor[i] = 0.0; - for (j = 0; j < lx[i]; j++) - nor[i] += (float64_t)(x[i][j]*x[i][j]); - } -} - -/******************************************************************************/ -/*** Set the subproblem data ***/ -/******************************************************************************/ -void sKernel::SetSubproblem(sKernel* ker, int32_t len, int32_t *perm) -{ - int32_t k; - - /* arrays allocations */ - nor = SG_MALLOC(float64_t, len); - vaux = SG_CALLOC(float32_t, ker->dim); - - lx = SG_MALLOC(int32_t, len); - ix = SG_MALLOC(int32_t*, len); - x = SG_MALLOC(float32_t*, len); - IsSubproblem = 1; - - for (k = 0; k < len; k++) - { - x[k] = ker->x[perm[k]]; - ix[k] = ker->ix[perm[k]]; - lx[k] = ker->lx[perm[k]]; - nor[k] = ker->nor[perm[k]]; - } - - // unroll one (sparse) vector - vauxRow = 0; - for (k = 0; k < lx[vauxRow]; k++) - vaux[ix[vauxRow][k]] = x[vauxRow][k]; -} - -/******************************************************************************/ -/*** Kernel class destructor ***/ -/******************************************************************************/ -sKernel::~sKernel() -{ - int32_t i; - - SG_FREE(nor); - SG_FREE(vaux); - - SG_FREE(lx); - if (ix != NULL) - { - if (!IsSubproblem) - for (i = 0; i < ell; i++) - SG_FREE(ix[i]); - SG_FREE(ix); - } - if (x != NULL) - { - if (!IsSubproblem) - for (i = 0; i < ell; i++) - SG_FREE(x[i]); - SG_FREE(x); - } -} - -#endif // DOXYGEN_SHOULD_SKIP_THIS - -/******************************************************************************/ -/*** End of gpdt.cpp file ***/ -/******************************************************************************/ -#endif //USE_GPL_SHOGUN diff --git a/src/shogun/lib/external/gpdt.h b/src/shogun/lib/external/gpdt.h deleted file mode 100644 index 992eaefc246..00000000000 --- a/src/shogun/lib/external/gpdt.h +++ /dev/null @@ -1,203 +0,0 @@ -/****************************************************************************** - *** GPDT - Gradient Projection Decomposition Technique *** - ****************************************************************************** - *** *** - *** GPDT is a C++ software designed to train large-scale Support Vector *** - *** Machines for binary classification in both scalar and distributed *** - *** memory parallel environments. It uses the Joachims' problem *** - *** decomposition technique to split the whole quadratic programming (QP) *** - *** problem into a sequence of smaller QP subproblems, each one being *** - *** solved by a suitable gradient projection method (GPM). The presently *** - *** implemented GPMs are the Generalized Variable Projection Method *** - *** GVPM (T. Serafini, G. Zanghirati, L. Zanni, "Gradient Projection *** - *** Methods for Quadratic Programs and Applications in Training Support *** - *** Vector Machines"; Optim. Meth. Soft. 20, 2005, 353-378) and the *** - *** Dai-Fletcher Method DFGPM (Y. Dai and R. Fletcher,"New Algorithms for *** - *** Singly Linear Constrained Quadratic Programs Subject to Lower and *** - *** Upper Bounds"; Math. Prog. to appear). *** - *** *** - *** Authors: *** - *** Thomas Serafini, Luca Zanni *** - *** Dept. of Mathematics, University of Modena and Reggio Emilia - ITALY *** - *** serafini.thomas@unimo.it, zanni.luca@unimo.it *** - *** Gaetano Zanghirati *** - *** Dept. of Mathematics, University of Ferrara - ITALY *** - *** g.zanghirati@unife.it *** - *** *** - *** Software homepage: http://dm.unife.it/gpdt *** - *** *** - *** This work is supported by the Italian FIRB Projects *** - *** 'Statistical Learning: Theory, Algorithms and Applications' *** - *** (grant RBAU01877P), http://slipguru.disi.unige.it/ASTA *** - *** and *** - *** 'Parallel Algorithms and Numerical Nonlinear Optimization' *** - *** (grant RBAU01JYPN), http://dm.unife.it/pn2o *** - *** *** - *** Copyright (C) 2004 by T. Serafini, G. Zanghirati, L. Zanni. *** - *** *** - *** COPYRIGHT NOTIFICATION *** - *** *** - *** Permission to copy and modify this software and its documentation *** - *** for internal research use is granted, provided that this notice is *** - *** retained thereon and on all copies or modifications. The authors and *** - *** their respective Universities makes no representations as to the *** - *** suitability and operability of this software for any purpose. It is *** - *** provided "as is" without express or implied warranty. *** - *** Use of this software for commercial purposes is expressly prohibited *** - *** without contacting the authors. *** - *** *** - *** This program is free software; you can redistribute it and/or modify *** - *** it under the terms of the GNU General Public License as published by *** - *** the Free Software Foundation; either version 3 of the License, or *** - *** (at your option) any later version. *** - *** *** - *** This program is distributed in the hope that it will be useful, *** - *** but WITHOUT ANY WARRANTY; without even the implied warranty of *** - *** MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the *** - *** GNU General Public License for more details. *** - *** *** - *** You should have received a copy of the GNU General Public License *** - *** along with this program; if not, write to the Free Software *** - *** Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. *** - *** *** - *** File: gpdt.h *** - *** Type: scalar *** - *** Version: 1.0 *** - *** Date: October, 2005 *** - *** Revision: 1 *** - *** *** - ******************************************************************************/ -#include -#ifdef USE_GPL_SHOGUN -#include - -#ifndef DOXYGEN_SHOULD_SKIP_THIS - -namespace shogun -{ -#define MAXLENGTH 256 -#define cachetype KERNELCACHE_ELEM -#define EPS_SV 1.0e-9 /* precision for multipliers */ - -enum { - SOLVER_VPM = 0, - SOLVER_FLETCHER = 1 -}; - -/** s kernel */ -class sKernel -{ -public: - /** kernel type */ - int32_t ker_type; - /** lx */ - int32_t *lx; - /** ix */ - int32_t **ix; - /** x */ - float32_t **x; - /** nor */ - float64_t *nor; - /** sigma */ - float64_t sigma; - /** degree */ - float64_t degree; - /** normalization factor */ - float64_t norm; - /** c poly */ - float64_t c_poly; - /** kernel evaluations */ - float64_t KernelEvaluations; - - /** call kernel fun - * - * @param i - * @param j - * @return something floaty - */ - float64_t (sKernel::*kernel_fun)(int32_t i, int32_t j); - - /** constructor - * - * @param k kernel - * @param ell ell - */ - sKernel (shogun::CKernel* k, int32_t ell); - ~sKernel(); - - /** set data - * - * @param x_ new x - * @param ix_ new ix - * @param lx_ new lx - * @param ell new ell - * @param dim dim - */ - void SetData( - float32_t **x_, int32_t **ix_, int32_t *lx_, int32_t ell, int32_t dim); - - /** set subproblem - * - * @param ker kernel - * @param len len - * @param perm perm - */ - void SetSubproblem (sKernel* ker, int32_t len, int32_t *perm); - - /** get an item from the kernel - * - * @param i index i - * @param j index j - * @return item from kernel at index i, j - */ - float64_t Get(int32_t i, int32_t j) - { - KernelEvaluations += 1.0F; - return kernel->kernel(i, j); - } - - /** add something - * - * @param v v - * @param j j - * @param mul mul - */ - void Add (float64_t *v, int32_t j, float64_t mul); - - /** prod something - * - * @param v v - * @param j j - * @return something floaty - */ - float64_t Prod (float64_t *v, int32_t j); - - /** get kernel - * - * @return kernel - */ - inline CKernel* get_kernel() - { - return kernel; - } - -private: - CKernel* kernel; - int32_t vauxRow; - int32_t IsSubproblem; - int32_t ell, dim; - float32_t *vaux; - - float64_t dot (int32_t i, int32_t j); -}; - -void SplitParts ( - int32_t n, int32_t part, int32_t parts, int32_t *dim, int32_t *off); -void SplitNum (int32_t n, int32_t *nloc, int32_t *noff); -} -#endif // DOXYGEN_SHOULD_SKIP_THIS - -/******************************************************************************/ -/*** End of gpdt.h file ***/ -/******************************************************************************/ -#endif //USE_GPL_SHOGUN diff --git a/src/shogun/lib/external/gpdtsolve.cpp b/src/shogun/lib/external/gpdtsolve.cpp deleted file mode 100644 index 031670a78e8..00000000000 --- a/src/shogun/lib/external/gpdtsolve.cpp +++ /dev/null @@ -1,1580 +0,0 @@ -/****************************************************************************** - *** GPDT - Gradient Projection Decomposition Technique *** - ****************************************************************************** - *** *** - *** GPDT is a C++ software designed to train large-scale Support Vector *** - *** Machines for binary classification in both scalar and distributed *** - *** memory parallel environments. It uses the Joachims' problem *** - *** decomposition technique to split the whole quadratic programming (QP) *** - *** problem into a sequence of smaller QP subproblems, each one being *** - *** solved by a suitable gradient projection method (GPM). The presently *** - *** implemented GPMs are the Generalized Variable Projection Method *** - *** GVPM (T. Serafini, G. Zanghirati, L. Zanni, "Gradient Projection *** - *** Methods for Quadratic Programs and Applications in Training Support *** - *** Vector Machines"; Optim. Meth. Soft. 20, 2005, 353-378) and the *** - *** Dai-Fletcher Method DFGPM (Y. Dai and R. Fletcher,"New Algorithms for *** - *** Singly Linear Constrained Quadratic Programs Subject to Lower and *** - *** Upper Bounds"; Math. Prog. to appear). *** - *** *** - *** Authors: *** - *** Thomas Serafini, Luca Zanni *** - *** Dept. of Mathematics, University of Modena and Reggio Emilia - ITALY *** - *** serafini.thomas@unimo.it, zanni.luca@unimo.it *** - *** Gaetano Zanghirati *** - *** Dept. of Mathematics, University of Ferrara - ITALY *** - *** g.zanghirati@unife.it *** - *** *** - *** Software homepage: http://dm.unife.it/gpdt *** - *** *** - *** This work is supported by the Italian FIRB Projects *** - *** 'Statistical Learning: Theory, Algorithms and Applications' *** - *** (grant RBAU01877P), http://slipguru.disi.unige.it/ASTA *** - *** and *** - *** 'Parallel Algorithms and Numerical Nonlinear Optimization' *** - *** (grant RBAU01JYPN), http://dm.unife.it/pn2o *** - *** *** - *** Copyright (C) 2004-2008 by T. Serafini, G. Zanghirati, L. Zanni. *** - *** *** - *** COPYRIGHT NOTIFICATION *** - *** *** - *** Permission to copy and modify this software and its documentation *** - *** for internal research use is granted, provided that this notice is *** - *** retained thereon and on all copies or modifications. The authors and *** - *** their respective Universities makes no representations as to the *** - *** suitability and operability of this software for any purpose. It is *** - *** provided "as is" without express or implied warranty. *** - *** Use of this software for commercial purposes is expressly prohibited *** - *** without contacting the authors. *** - *** *** - *** This program is free software; you can redistribute it and/or modify *** - *** it under the terms of the GNU General Public License as published by *** - *** the Free Software Foundation; either version 3 of the License, or *** - *** (at your option) any later version. *** - *** *** - *** This program is distributed in the hope that it will be useful, *** - *** but WITHOUT ANY WARRANTY; without even the implied warranty of *** - *** MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the *** - *** GNU General Public License for more details. *** - *** *** - *** You should have received a copy of the GNU General Public License *** - *** along with this program; if not, write to the Free Software *** - *** Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. *** - *** *** - *** File: gpdtsolve.cpp *** - *** Type: scalar *** - *** Version: 1.0 *** - *** Date: November, 2006 *** - *** Revision: 2 *** - *** *** - *** SHOGUN adaptions Written (W) 2006-2009 Soeren Sonnenburg *** - ******************************************************************************/ - -#include -#ifdef USE_GPL_SHOGUN -#include -#include -#include -#include -#include - -#include -#include -#include -#include -#include - -using namespace shogun; - -#ifndef DOXYGEN_SHOULD_SKIP_THIS -namespace shogun -{ -#define y_in(i) y[index_in[(i)]] -#define y_out(i) y[index_out[(i)]] -#define alpha_in(i) alpha[index_in[(i)]] -#define alpha_out(i) alpha[index_out[(i)]] -#define minfty (-1.0e+30) // minus infinity - -uint32_t Randnext = 1; - -#define ThRand (Randnext = Randnext * 1103515245L + 12345L) -#define ThRandPos ((Randnext = Randnext * 1103515245L + 12345L) & 0x7fffffff) - -FILE *fp; - -/* utility routines prototyping */ -void quick_si (int32_t a[], int32_t k); -void quick_s3 (int32_t a[], int32_t k, int32_t ia[]); -void quick_s2 (float64_t a[], int32_t k, int32_t ia[]); - -/******************************************************************************/ -/*** Class for caching strategy implementation ***/ -/******************************************************************************/ -class sCache -{ - -public: - sCache (sKernel* sk, int32_t Mbyte, int32_t ell); - ~sCache (); - - cachetype *FillRow (int32_t row, int32_t IsC = 0); - cachetype *GetRow (int32_t row); - - int32_t DivideMP (int32_t *out, int32_t *in, int32_t n); - - /*** Itarations counter ***/ - void Iteration() { nit++; } - - /*** Cache size control ***/ - int32_t CheckCycle() - { - int32_t us; - cache_entry *pt = first_free->next; - - for (us = 0; pt != first_free; us++, pt = pt->next); - if (us != maxmw-1) - return 1; - else - return 0; - } - -private: - - struct cache_entry - { - int32_t row; // unused row - int32_t last_access_it; - cache_entry *prev, *next; - cachetype *data; - }; - - sKernel* KER; - int32_t maxmw, ell; - int32_t nit; - - cache_entry *mw; - cache_entry *first_free; - cache_entry **pindmw; // 0 if unused row - cachetype *onerow; - - cachetype *FindFree(int32_t row, int32_t IsC); -}; - - -/******************************************************************************/ -/*** Cache class constructor ***/ -/******************************************************************************/ -sCache::sCache(sKernel* sk, int32_t Mbyte, int32_t _ell) : KER(sk), ell(_ell) -{ - int32_t i; - - // size in dwords of one cache row - maxmw = (sizeof(cache_entry) + sizeof(cache_entry *) - + ell*sizeof(cachetype)) / 4; - // number of cache rows - maxmw = Mbyte*1024*(1024/4) / maxmw; - - /* arrays allocation */ - mw = SG_MALLOC(cache_entry, maxmw); - pindmw = SG_MALLOC(cache_entry*, ell); - onerow = SG_MALLOC(cachetype, ell); - - /* arrays initialization */ - for (i = 0; i < maxmw; i++) - { - mw[i].prev = (i == 0 ? &mw[maxmw-1] : &mw[i-1]); - mw[i].next = (i == maxmw-1 ? &mw[0] : &mw[i+1]); - mw[i].data = SG_MALLOC(cachetype, ell); - mw[i].row = -1; // unused row - mw[i].last_access_it = -1; - } - for (i = 0; i < ell; i++) - pindmw[i] = 0; - - first_free = &mw[0]; - nit = 0; -} - -/******************************************************************************/ -/*** Cache class destructor ***/ -/******************************************************************************/ -sCache::~sCache() -{ - int32_t i; - - for (i = maxmw-1; i >= 0; i--) - SG_FREE(mw[i].data); - - SG_FREE(onerow); - SG_FREE(pindmw); - SG_FREE(mw); -} - - -/******************************************************************************/ -/*** Retrieve a cached row ***/ -/******************************************************************************/ -cachetype *sCache::GetRow(int32_t row) -{ - cache_entry *c; - - c = pindmw[row]; - if (c == NULL) - return NULL; - - c->last_access_it = nit; - if (c == first_free) - { - first_free = first_free->next; - } - else - { - // move "row" as the most recently used. - c->prev->next = c->next; - c->next->prev = c->prev; - c->next = first_free; - c->prev = first_free->prev; - first_free->prev = c; - c->prev->next = c; - } - - return c->data; -} - -/****************************************************************************** - *** Look for a free cache row *** - *** IMPORTANT: call this method only if you are sure that "row" *** - *** is not already in the cache ( i.e. after calling GetRow() ) *** - ******************************************************************************/ -cachetype *sCache::FindFree(int32_t row, int32_t IsC) -{ - cachetype *pt; - - if (first_free->row != -1) // cache row already contains data - { - if (first_free->last_access_it == nit || IsC) - return 0; - else - pindmw[first_free->row] = 0; - } - first_free->row = row; - first_free->last_access_it = nit; - pindmw[row] = first_free; - - pt = first_free->data; - first_free = first_free->next; - - return pt; -} - - -/******************************************************************************/ -/*** Enter data in a cache row ***/ -/******************************************************************************/ -cachetype *sCache::FillRow(int32_t row, int32_t IsC) -{ - int32_t j; - cachetype *pt; - - pt = GetRow(row); - if (pt != NULL) - return pt; - - pt = FindFree(row, IsC); - if (pt == 0) - pt = onerow; - - // Compute all the row elements - for (j = 0; j < ell; j++) - pt[j] = (cachetype)KER->Get(row, j); - return pt; -} - - -/******************************************************************************/ -/*** Expand a sparse row in a full cache row ***/ -/******************************************************************************/ -int32_t sCache::DivideMP(int32_t *out, int32_t *in, int32_t n) -{ - /******************************************************************** - * Input meaning: * - * in = vector containing row to be extracted in the cache * - * n = size of in * - * out = the indexes of "in" of the components to be computed * - * by this processor (first those in the cache, then the * - * ones not yet computed) * - * Returns: the number of components of this processor * - ********************************************************************/ - - int32_t *remained, nremained, k; - int32_t i; - - remained = SG_MALLOC(int32_t, n); - - nremained = 0; - k = 0; - for (i = 0; i < n; i++) - { - if (pindmw[in[i]] != NULL) - out[k++] = i; - else - remained[nremained++] = i; - } - for (i = 0; i < nremained; i++) - out[k++] = remained[i]; - - SG_FREE(remained); - return n; -} - -/******************************************************************************/ -/*** Check solution optimality ***/ -/******************************************************************************/ -int32_t QPproblem::optimal() -{ - /*********************************************************************** - * Returns 1 if the computed solution is optimal, otherwise returns 0. * - * To verify the optimality it checks the KKT optimality conditions. * - ***********************************************************************/ - register int32_t i, j, margin_sv_number, z, k, s, kin, z1, znew=0, nnew; - - float64_t gx_i, aux, s1, s2; - - /* sort -y*grad and store in ing the swaps done */ - for (j = 0; j < ell; j++) - { - grad[j] = y[j] - st[j]; - ing[j] = j; - } - - quick_s2(grad,ell,ing); - - /* compute bee */ - margin_sv_number = 0; - - for (i = chunk_size - 1; i >= 0; i--) - if (is_free(index_in[i])) - { - margin_sv_number++; - bee = y_in(i) - st[index_in[i]]; - break; - } - - if (margin_sv_number > 0) - { - aux=-1.0; - for (i = nb-1; i >= 0; i--) - { - gx_i = bee + st[index_out[i]]; - if ((is_zero(index_out[i]) && (gx_i*y_out(i) < (1.0-delta))) || - (is_bound(index_out[i]) && (gx_i*y_out(i) > (1.0+delta))) || - (is_free(index_out[i]) && - ((gx_i*y_out(i) < 1.0-delta) || (gx_i*y_out(i) > 1.0+delta)))) - { - if (fabs(gx_i*y_out(i) - 1.0) > aux) - aux = fabs(gx_i*y_out(i) - 1.0); - } - } - } - else - { - for (i = ell - 1; i >= 0; i--) - if (is_free(i)) - { - margin_sv_number++; - bee = y[i] - st[i]; - break; - } - if (margin_sv_number == 0) - { - s1 = -minfty; - s2 = -minfty; - for (j = 0; j < ell; j++) - if ( (alpha[ing[j]] > DELTAsv) && (y[ing[j]] >= 0) ) - { - s1 = grad[j]; - break; - } - for (j = 0; j < ell; j++) - if ( (alpha[ing[j]] < c_const-DELTAsv) && (y[ing[j]] <= 0) ) - { - s2 = grad[j]; - break; - } - if (s1 < s2) - aux = s1; - else - aux = s2; - - s1 = minfty; - s2 = minfty; - for (j = ell-1; j >=0; j--) - if ( (alpha[ing[j]] > DELTAsv) && (y[ing[j]] <= 0) ) - { - s1 = grad[j]; - break; - } - for (j = ell-1; j >=0; j--) - if ( (alpha[ing[j]] < c_const-DELTAsv) && (y[ing[j]] >= 0) ) - { - s2 = grad[j]; - break; - } - if (s2 > s1) s1 = s2; - - bee = 0.5 * (s1+aux); - } - - aux = -1.0; - for (i = ell-1; i >= 0; i--) - { - gx_i = bee + st[i]; - if ((is_zero(i) && (gx_i*y[i] < (1.0-delta))) || - (is_bound(i) && (gx_i*y[i] > (1.0+delta))) || - (is_free(i) && - ((gx_i*y[i] < 1.0-delta) || (gx_i*y[i] > 1.0+delta)))) - { - if (fabs(gx_i*y[i] - 1.0) > aux) - aux = fabs(gx_i*y[i] - 1.0); - } - } - } - - if (aux < 0.0) - return 1; - else - { - if (verbosity > 1) - SG_SINFO(" Max KKT violation: %lf\n", aux) - else if (verbosity > 0) - SG_SINFO(" %lf\n", aux) - - if (fabs(kktold-aux) < delta*0.01 && aux < delta*2.0) - { - if (DELTAvpm > InitialDELTAvpm*0.1) - { - DELTAvpm = (DELTAvpm*0.5 > InitialDELTAvpm*0.1 ? - DELTAvpm*0.5 : InitialDELTAvpm*0.1); - SG_SINFO("Inner tolerance changed to: %lf\n", DELTAvpm) - } - } - - kktold = aux; - - /***************************************************************************** - *** Update the working set (T. Serafini, L. Zanni, "On the Working Set *** - *** Selection in Gradient Projection-based Decomposition Techniques for *** - *** Support Vector Machines"; Optim. Meth. Soft. 20, 2005). *** - *****************************************************************************/ - for (j = 0; j < chunk_size; j++) - inold[j] = index_in[j]; - - z = -1; /* index of the last entered component from the top */ - z1 = ell; /* index of the last entered component from the bottom */ - k = 0; - j = 0; - while (k < q) - { - i = z + 1; /* index of the candidate components from the top */ - while (i < z1) - { - if ( is_free(ing[i]) || - (-y[ing[i]]>=0 && is_zero(ing[i])) || - (-y[ing[i]]<=0 && is_bound(ing[i])) - ) - { - znew = i; /* index of the component to select from the top */ - break; - } - i++; - } - if (i == z1) break; - - s = z1 - 1; - while (znew < s) - { - if ( is_free(ing[s]) || - (y[ing[s]]>=0 && is_zero(ing[s])) || - (y[ing[s]]<=0 && is_bound(ing[s])) - ) - { - z1 = s; - z = znew; - break; - } - s--; - } - if (znew == s) break; - - index_in[k++] = ing[z]; - index_in[k++] = ing[z1]; - } - - if (k < q) - { - if (verbosity > 1) - SG_SINFO(" New q: %i\n", k) - q = k; - } - - quick_si(index_in, q); - - s = 0; - j = 0; - for (k = 0; k < chunk_size; k++) - { - z = inold[k]; - for (i = j; i < q; i++) - if (z <= index_in[i]) - break; - - if (i == q) - { - for (i = k; i < chunk_size; i++) - { - ing[s] = inold[i]; /* older components not in the new basis */ - s = s+1; - } - break; - } - - if (z == index_in[i]) - j = i + 1; /* the component is still in the basis */ - else - { - ing[s] = z; /* older component not in the new basis */ - s = s + 1; - j = i; - } - } - - for (i = 0; i < s; i++) - { - bmemrid[i] = bmem[ing[i]]; - pbmr[i] = i; - } - - quick_s3(bmemrid, s, pbmr); - - /* check if support vectors not at bound enter the basis */ - j = q; - i = 0; - while (j < chunk_size && i < s) - { - if (is_free(ing[pbmr[i]])) - { - index_in[j] = ing[pbmr[i]]; - j++; - } - i++; - } - - /* choose which bound variables keep in basis (alpha = 0 or alpha = C) */ - if (j < chunk_size) - { - i = 0; - while (j < chunk_size && i < s) - { - if (is_zero(ing[pbmr[i]])) - { - index_in[j] = ing[pbmr[i]]; - j++; - } - i++; - } - if (j < chunk_size) - { - i = 0; - while (j < chunk_size && i < s) - { - if (is_bound(ing[pbmr[i]])) - { - index_in[j] = ing[pbmr[i]]; - j++; - } - i++; - } - } - } - - quick_si(index_in, chunk_size); - - for (i = 0; i < chunk_size; i++) - bmem[index_in[i]]++; - - j = 0; - k = 0; - for (i = 0; i < chunk_size; i++) - { - for (z = j; z < index_in[i]; z++) - { - index_out[k] = z; - k++; - } - j = index_in[i]+1; - } - for (z = j; z < ell; z++) - { - index_out[k] = z; - k++; - } - - for (i = 0; i < nb; i++) - bmem[index_out[i]] = 0; - - kin = 0; - j = 0; - for (k = 0; k < chunk_size; k++) - { - z = index_in[k]; - for (i = j; i < chunk_size; i++) - if (z <= inold[i]) - break; - if (i == chunk_size) - { - for (s = k; s < chunk_size; s++) - { - incom[s] = -1; - cec[index_in[s]]++; - } - kin = kin + chunk_size - k ; - break; - } - - if (z == inold[i]) - { - incom[k] = i; - j = i+1; - } - else - { - incom[k] = -1; - j = i; - kin = kin + 1; - cec[index_in[k]]++; - } - } - - nnew = kin & (~1); - if (nnew < 10) - nnew = 10; - if (nnew < chunk_size/10) - nnew = chunk_size/10; - if (nnew < q) - { - q = nnew; - q = q & (~1); - } - - if (kin == 0) - { - DELTAkin *= 0.1; - if (DELTAkin < 1.0e-6) - { - SG_SINFO("\n***ERROR***: GPDT stops with tolerance") - SG_SINFO( - " %lf because it is unable to change the working set.\n", kktold); - return 1; - } - else - { - SG_SINFO("Inner tolerance temporary changed to:") - SG_SINFO(" %e\n", DELTAvpm*DELTAkin) - } - } - else - DELTAkin = 1.0; - - if (verbosity > 1) - { - SG_SINFO(" Working set: new components: %i", kin) - SG_SINFO(", new parameter n: %i\n", q) - } - - return 0; - } -} - -/******************************************************************************/ -/*** Optional preprocessing: random distribution ***/ -/******************************************************************************/ -int32_t QPproblem::Preprocess0(int32_t *aux, int32_t *sv) -{ - int32_t i, j; - - Randnext = 1; - memset(sv, 0, ell*sizeof(int32_t)); - for (i = 0; i < chunk_size; i++) - { - do - { - j = ThRandPos % ell; - } while (sv[j] != 0); - sv[j] = 1; - } - return(chunk_size); -} - -/******************************************************************************/ -/*** Optional preprocessing: block parallel distribution ***/ -/******************************************************************************/ -int32_t QPproblem::Preprocess1(sKernel* kernel, int32_t *aux, int32_t *sv) -{ - int32_t s; // elements owned by the processor - int32_t sl; // elements of the n-1 subproblems - int32_t n, i, off, j, k, ll; - int32_t nsv, nbsv; - int32_t *sv_loc, *bsv_loc, *sp_y; - float32_t *sp_D=NULL; - float64_t *sp_alpha, *sp_h; - - s = ell; - /* divide the s elements into n blocks smaller than preprocess_size */ - n = (s + preprocess_size - 1) / preprocess_size; - sl = 1 + s / n; - - if (verbosity > 0) - { - SG_SINFO(" Preprocessing: examples = %d", s) - SG_SINFO(", subp. = %d", n) - SG_SINFO(", size = %d\n",sl) - } - - sv_loc = SG_MALLOC(int32_t, s); - bsv_loc = SG_MALLOC(int32_t, s); - sp_alpha = SG_MALLOC(float64_t, sl); - sp_h = SG_MALLOC(float64_t, sl); - sp_y = SG_MALLOC(int32_t, sl); - - if (sl < 500) - sp_D = SG_MALLOC(float32_t, sl*sl); - - for (i = 0; i < sl; i++) - sp_h[i] = -1.0; - memset(alpha, 0, ell*sizeof(float64_t)); - - /* randomly reorder the component to select */ - for (i = 0; i < ell; i++) - aux[i] = i; - Randnext = 1; - for (i = 0; i < ell; i++) - { - j = ThRandPos % ell; - k = ThRandPos % ell; - ll = aux[j]; aux[j] = aux[k]; aux[k] = ll; - } - - nbsv = nsv = 0; - for (i = 0; i < n; i++) - { - if (verbosity > 0) - SG_SINFO("%d...", i) - SplitParts(s, i, n, &ll, &off); - - if (sl < 500) - { - for (j = 0; j < ll; j++) - { - sp_y[j] = y[aux[j+off]]; - for (k = j; k < ll; k++) - sp_D[k*sl + j] = sp_D[j*sl + k] - = y[aux[j+off]] * y[aux[k+off]] - * (float32_t)kernel->Get(aux[j+off], aux[k+off]); - } - - memset(sp_alpha, 0, sl*sizeof(float64_t)); - - /* call the gradient projection QP solver */ - gpm_solver(projection_solver, projection_projector, ll, sp_D, sp_h, - c_const, 0.0, sp_y, sp_alpha, delta*10, NULL); - } - else - { - QPproblem p2; - QPproblem::copy_subproblem(&p2, this, ll, aux + off); - p2.chunk_size = (int32_t) ((float64_t)chunk_size / sqrt((float64_t)n)); - p2.q = (int32_t) ((float64_t)q / sqrt((float64_t)n)); - p2.maxmw = ll*ll*4 / (1024 * 1024); - if (p2.maxmw > maxmw / 2) - p2.maxmw = maxmw / 2; - p2.verbosity = 0; - p2.delta = delta * 10.0; - p2.PreprocessMode = 0; - kernel->KernelEvaluations += p2.gpdtsolve(sp_alpha); - } - - for (j = 0; j < ll; j++) - { - /* modify bound support vector approximation */ - if (sp_alpha[j] < (c_const-DELTAsv)) - sp_alpha[j] = 0.0; - else - sp_alpha[j] = c_const; - if (sp_alpha[j] > DELTAsv) - { - if (sp_alpha[j] < (c_const-DELTAsv)) - sv_loc[nsv++] = aux[j+off]; - else - bsv_loc[nbsv++] = aux[j+off]; - alpha[aux[j+off]] = sp_alpha[j]; - } - } - } - - Randnext = 1; - - /* add the known support vectors to the working set */ - memset(sv, 0, ell*sizeof(int32_t)); - ll = (nsv < chunk_size ? nsv : chunk_size); - for (i = 0; i < ll; i++) - { - do { - j = sv_loc[ThRandPos % nsv]; - } while (sv[j] != 0); - sv[j] = 1; - } - - /* add the known bound support vectors to the working set */ - ll = ((nsv+nbsv) < chunk_size ? (nsv+nbsv) : chunk_size); - for (; i < ll; i++) - { - do { - j = bsv_loc[ThRandPos % nbsv]; - } while (sv[j] != 0); - sv[j] = 1; - } - - /* eventually fill up the working set with other components - randomly chosen */ - for (; i < chunk_size; i++) - { - do { - j = ThRandPos % ell; - } while (sv[j] != 0); - sv[j] = 1; - } - - - /* dealloc temporary arrays */ - if (sl < 500) SG_FREE(sp_D); - SG_FREE(sp_y ); - SG_FREE(sp_h ); - SG_FREE(sv_loc ); - SG_FREE(bsv_loc ); - SG_FREE(sp_alpha); - - if (verbosity > 0) - { - SG_SINFO("\n Preprocessing: SV = %d", nsv) - SG_SINFO(", BSV = %d\n", nbsv) - } - - return(nsv); -} - -/******************************************************************************/ -/*** Compute the QP problem solution ***/ -/******************************************************************************/ -float64_t QPproblem::gpdtsolve(float64_t *solution) -{ - int32_t i, j, k, z, iin, jin, nit, tot_vpm_iter, lsCount; - int32_t tot_vpm_secant, projCount, proximal_count; - int32_t vpmWarningThreshold; - int32_t nzin, nzout; - int32_t *sp_y; /* labels vector */ - int32_t *indnzin, *indnzout; /* nonzero components indices vectors */ - float32_t *sp_D; /* quadratic part of the objective function */ - float64_t *sp_h, *sp_hloc, /* linear part of the objective function */ - *sp_alpha,*stloc; /* variables and gradient updating vectors */ - float64_t sp_e, aux, fval, tau_proximal_this, dfval; - float64_t *vau; - float64_t *weight; - float64_t tot_prep_time, tot_vpm_time, tot_st_time, total_time; - sCache *Cache; - cachetype *ptmw; - clock_t t, ti; - - Cache = new sCache(KER, maxmw, ell); - if (chunk_size > ell) chunk_size = ell; - - if (chunk_size <= 20) - vpmWarningThreshold = 30*chunk_size; - else if (chunk_size <= 200) - vpmWarningThreshold = 20*chunk_size + 200; - else - vpmWarningThreshold = 10*chunk_size + 2200; - - kktold = 10000.0; - if (delta <= 5e-3) - { - if ( (chunk_size <= 20) | ((float64_t)chunk_size/ell <= 0.001) ) - DELTAvpm = delta * 0.1; - else if ( (chunk_size <= 200) | ((float64_t)chunk_size/ell <= 0.005) ) - DELTAvpm = delta * 0.5; - else - DELTAvpm = delta; - } - else - { - if ( (chunk_size <= 200) | ((float64_t)chunk_size/ell <= 0.005) ) - DELTAvpm = (1e-3 < delta*0.1) ? 1e-3 : delta*0.1; - else - DELTAvpm = 5e-3; - } - - InitialDELTAvpm = DELTAvpm; - DELTAsv = EPS_SV * c_const; - DELTAkin = 1.0; - - q = q & (~1); - nb = ell - chunk_size; - tot_vpm_iter = 0; - tot_vpm_secant = 0; - - tot_prep_time = tot_vpm_time = tot_st_time = total_time = 0.0; - - ti = clock(); - - /* arrays allocation */ - SG_SDEBUG("ell:%d, chunk_size:%d, nb:%d dim:%d\n", ell, chunk_size,nb, dim) - ing = SG_MALLOC(int32_t, ell); - inaux = SG_MALLOC(int32_t, ell); - index_in = SG_MALLOC(int32_t, chunk_size); - index_out = SG_MALLOC(int32_t, ell); - indnzout = SG_MALLOC(int32_t, nb); - alpha = SG_MALLOC(float64_t, ell); - - memset(alpha, 0, ell*sizeof(float64_t)); - memset(ing, 0, ell*sizeof(int32_t)); - - if (verbosity > 0 && PreprocessMode != 0) - SG_SINFO("\n*********** Begin setup step...\n") - t = clock(); - - switch(PreprocessMode) - { - case 1: Preprocess1(KER, inaux, ing); break; - case 0: - default: - Preprocess0(inaux, ing); break; - } - - for (j = k = i = 0; i < ell; i++) - if (ing[i] == 0) - index_out[j++] = i; - else - index_in[k++] = i; - - t = clock() - t; - if (verbosity > 0 && PreprocessMode != 0) - { - SG_SINFO(" Time for setup: %.2lf\n", (float64_t)t/CLOCKS_PER_SEC) - SG_SINFO( - "\n\n*********** Begin decomposition technique...\n"); - } - - /* arrays allocation */ - bmem = SG_MALLOC(int32_t, ell); - bmemrid = SG_MALLOC(int32_t, chunk_size); - pbmr = SG_MALLOC(int32_t, chunk_size); - cec = SG_MALLOC(int32_t, ell); - indnzin = SG_MALLOC(int32_t, chunk_size); - inold = SG_MALLOC(int32_t, chunk_size); - incom = SG_MALLOC(int32_t, chunk_size); - vau = SG_MALLOC(float64_t, ell); - grad = SG_MALLOC(float64_t, ell); - weight = SG_MALLOC(float64_t, dim); - st = SG_MALLOC(float64_t, ell); - stloc = SG_MALLOC(float64_t, ell); - - for (i = 0; i < ell; i++) - { - bmem[i] = 0; - cec[i] = 0; - st[i] = 0; - } - - sp_y = SG_MALLOC(int32_t, chunk_size); - sp_D = SG_MALLOC(float32_t, chunk_size*chunk_size); - sp_alpha = SG_MALLOC(float64_t, chunk_size); - sp_h = SG_MALLOC(float64_t, chunk_size); - sp_hloc = SG_MALLOC(float64_t, chunk_size); - - for (i = 0; i < chunk_size; i++) - cec[index_in[i]] = cec[index_in[i]]+1; - - for (i = chunk_size-1; i >= 0; i--) - { - incom[i] = -1; - sp_alpha[i] = 0.0; - bmem[index_in[i]] = 1; - } - - if (verbosity == 1) - { - SG_SINFO(" IT | Prep Time | Solver IT | Solver Time |") - SG_SINFO(" Grad Time | KKT violation\n") - SG_SINFO("------+-----------+-----------+-------------+") - SG_SINFO("-----------+--------------\n") - } - - /***************************************************************************/ - /* Begin the problem resolution loop */ - nit = 0; - do - { - t = clock(); - if ((nit % 10) == 9) - { - if ((t-ti) > 0) - total_time += (float64_t)(t-ti) / CLOCKS_PER_SEC; - else - total_time += (float64_t)(ti-t) / CLOCKS_PER_SEC; - ti = t; - } - - if (verbosity > 1) - SG_SINFO("\n*********** ITERATION: %d\n", nit + 1) - else if (verbosity > 0) - SG_SINFO("%5d |", nit + 1) - else - SG_SINFO(".") - fflush(stdout); - - nzout = 0; - for (k = 0; k < nb; k++) - if (alpha_out(k)>DELTAsv) - { - indnzout[nzout] = index_out[k]; - nzout++; - } - - sp_e = 0.; - for (j = 0; j < nzout; j++) - { - vau[j] = y[indnzout[j]]*alpha[indnzout[j]]; - sp_e += vau[j]; - } - - if (verbosity > 1) - SG_SINFO(" spe: %e ", sp_e) - - for (i = 0; i < chunk_size; i++) - sp_y[i] = y_in(i); - - /* Construct the objective function Hessian */ - for (i = 0; i < chunk_size; i++) - { - iin = index_in[i]; - ptmw = Cache->GetRow(iin); - if (ptmw != 0) - { - for (j = 0; j <= i; j++) - sp_D[i*chunk_size + j] = sp_y[i]*sp_y[j] * ptmw[index_in[j]]; - } - else if (incom[i] == -1) - for (j = 0; j <= i; j++) - sp_D[i*chunk_size + j] = sp_y[i]*sp_y[j] - * (float32_t)KER->Get(iin, index_in[j]); - else - { - for (j = 0; j < i; j++) - if (incom[j] == -1) - sp_D[i*chunk_size + j] - = sp_y[i]*sp_y[j] * (float32_t)KER->Get(iin, index_in[j]); - else - sp_D[i*chunk_size + j] - = sp_D[incom[j]*chunk_size + incom[i]]; - sp_D[i*chunk_size + i] - = sp_y[i]*sp_y[i] * (float32_t)KER->Get(iin, index_in[i]); - } - } - for (i = 0; i < chunk_size; i++) - { - for (j = 0; j < i; j++) - sp_D[j*chunk_size + i] = sp_D[i*chunk_size + j]; - } - - if (nit == 0 && PreprocessMode > 0) - { - for (i = 0; i < chunk_size; i++) - { - iin = index_in[i]; - aux = 0.; - ptmw = Cache->GetRow(iin); - if (ptmw == NULL) - for (j = 0; j < nzout; j++) - aux += vau[j] * KER->Get(iin, indnzout[j]); - else - for (j = 0; j < nzout; j++) - aux += vau[j] * ptmw[indnzout[j]]; - sp_h[i] = y_in(i) * aux - 1.0; - } - } - else - { - for (i = 0; i < chunk_size; i++) - vau[i] = alpha_in(i); - for (i = 0; i < chunk_size; i++) - { - aux = 0.0; - for (j = 0; j < chunk_size; j++) - aux += sp_D[i*chunk_size + j] * vau[j]; - sp_h[i] = st[index_in[i]] * y_in(i) - 1.0 - aux; - } - } - - t = clock() - t; - if (verbosity > 1) - SG_SINFO(" Preparation Time: %.2lf\n", (float64_t)t/CLOCKS_PER_SEC) - else if (verbosity > 0) - SG_SINFO(" %8.2lf |", (float64_t)t/CLOCKS_PER_SEC) - tot_prep_time += (float64_t)t/CLOCKS_PER_SEC; - - /*** Proximal point modification: first type ***/ - - if (tau_proximal < 0.0) - tau_proximal_this = 0.0; - else - tau_proximal_this = tau_proximal; - proximal_count = 0; - do { - t = clock(); - for (i = 0; i < chunk_size; i++) - { - vau[i] = sp_D[i*chunk_size + i]; - sp_h[i] -= tau_proximal_this * alpha_in(i); - sp_D[i*chunk_size + i] += (float32_t)tau_proximal_this; - } - - if (kktold < delta*100) - for (i = 0; i < chunk_size; i++) - sp_alpha[i] = alpha_in(i); - else - for (i = 0; i < chunk_size; i++) - sp_alpha[i] = 0.0; - - /*** call the chosen inner gradient projection QP solver ***/ - i = gpm_solver(projection_solver, projection_projector, chunk_size, - sp_D, sp_h, c_const, sp_e, sp_y, sp_alpha, - DELTAvpm*DELTAkin, &lsCount, &projCount); - - if (i > vpmWarningThreshold) - { - if (ker_type == 2) - { - SG_SINFO("\n WARNING: inner subproblem hard to solve;") - SG_SINFO(" setting a smaller -q or") - SG_SINFO(" tuning -c and -g options might help.\n") - } - else - { - SG_SINFO("\n WARNING: inner subproblem hard to solve;") - SG_SINFO(" set a smaller -q or") - SG_SINFO(" try a better data scaling.\n") - } - } - - t = clock() - t; - tot_vpm_iter += i; - tot_vpm_secant += projCount; - tot_vpm_time += (float64_t)t/CLOCKS_PER_SEC; - if (verbosity > 1) - { - SG_SINFO(" Solver it: %d", i) - SG_SINFO(", ls: %d", lsCount) - SG_SINFO(", time: %.2lf\n", (float64_t)t/CLOCKS_PER_SEC) - } - else if (verbosity > 0) - { - SG_SINFO(" %6d", i) - SG_SINFO(" | %8.2lf |", (float64_t)t/CLOCKS_PER_SEC) - } - - /*** Proximal point modification: second type ***/ - - for (i = 0; i < chunk_size; i++) - sp_D[i*chunk_size + i] = (float32_t)vau[i]; - tau_proximal_this = 0.0; - if (tau_proximal < 0.0) - { - dfval = 0.0; - for (i = 0; i < chunk_size; i++) - { - aux = 0.0; - for (j = 0; j < chunk_size; j++) - aux += sp_D[i*chunk_size + j]*(alpha_in(j) - sp_alpha[j]); - dfval += (0.5*aux - st[index_in[i]]*y_in(i) + 1.0) * (alpha_in(i) - sp_alpha[i]); - } - - aux=0.0; - for (i = 0; i < chunk_size; i++) - aux += (alpha_in(i) - sp_alpha[i])*(alpha_in(i) - sp_alpha[i]); - - if ((-dfval/aux) < -0.5*tau_proximal) - { - tau_proximal_this = -tau_proximal; - if (verbosity > 0) - SG_SDEBUG("tau threshold: %lf ", -dfval/aux) - } - } - proximal_count++; - } while (tau_proximal_this != 0.0 && proximal_count < 2); // Proximal point loop - - t = clock(); - - nzin = 0; - for (j = 0; j < chunk_size; j++) - { - if (nit == 0) - aux = sp_alpha[j]; - else - aux = sp_alpha[j] - alpha_in(j); - if (fabs(aux) > DELTAsv) - { - indnzin[nzin] = index_in[j]; - grad[nzin] = aux * y_in(j); - nzin++; - } - } - - // in case of LINADD enabled use faster linadd variant - if (KER->get_kernel()->has_property(KP_LINADD) && get_linadd_enabled()) - { - KER->get_kernel()->clear_normal() ; - - for (j = 0; j < nzin; j++) - KER->get_kernel()->add_to_normal(indnzin[j], grad[j]); - - if (nit == 0 && PreprocessMode > 0) - { - for (j = 0; j < nzout; j++) - { - jin = indnzout[j]; - KER->get_kernel()->add_to_normal(jin, alpha[jin] * y[jin]); - } - } - - for (i = 0; i < ell; i++) - st[i] += KER->get_kernel()->compute_optimized(i); - } - else // nonlinear kernel - { - k = Cache->DivideMP(ing, indnzin, nzin); - for (j = 0; j < k; j++) - { - ptmw = Cache->FillRow(indnzin[ing[j]]); - for (i = 0; i < ell; i++) - st[i] += grad[ing[j]] * ptmw[i]; - } - - if (PreprocessMode > 0 && nit == 0) - { - clock_t ti2; - - ti2 = clock(); - for (j = 0; j < nzout; j++) - { - jin = indnzout[j]; - ptmw = Cache->FillRow(jin); - for (i = 0; i < ell; i++) - st[i] += alpha[jin] * y[jin] * ptmw[i]; - } - if (verbosity > 1) - SG_SINFO( - " G*x0 time: %.2lf\n", (float64_t)(clock()-ti2)/CLOCKS_PER_SEC); - } - } - - /*** sort the vectors for cache managing ***/ - - t = clock() - t; - if (verbosity > 1) - SG_SINFO(" Gradient updating time: %.2lf\n", (float64_t)t/CLOCKS_PER_SEC) - else if (verbosity > 0) - SG_SINFO(" %8.2lf |", (float64_t)t/CLOCKS_PER_SEC) - tot_st_time += (float64_t)t/CLOCKS_PER_SEC; - - /*** global updating of the solution vector ***/ - for (i = 0; i < chunk_size; i++) - alpha_in(i) = sp_alpha[i]; - - if (verbosity > 1) - { - j = k = 0; - for (i = 0; i < ell; i++) - { - if (is_free(i)) j++; - if (is_bound(i)) k++; - } - SG_SINFO(" SV: %d", j+k) - SG_SINFO(", BSV: %d\n", k) - } - Cache->Iteration(); - nit = nit+1; - } while (!optimal() && !(CSignal::cancel_computations())); - /* End of the problem resolution loop */ - /***************************************************************************/ - - t = clock(); - if ((t-ti) > 0) - total_time += (float64_t)(t-ti) / CLOCKS_PER_SEC; - else - total_time += (float64_t)(ti-t) / CLOCKS_PER_SEC; - ti = t; - - sg_memcpy(solution, alpha, ell * sizeof(float64_t)); - - /* objective function evaluation */ - fval = 0.0; - for (i = 0; i < ell; i++) - fval += alpha[i]*(y[i]*st[i]*0.5 - 1.0); - - SG_SINFO("\n------+-----------+-----------+-------------+") - SG_SINFO("-----------+--------------\n") - SG_SINFO( - "\n- TOTAL ITERATIONS: %i\n", nit); - - if (verbosity > 1) - { - j = 0; - k = 0; - z = 0; - for (i = ell - 1; i >= 0; i--) - { - if (cec[i] > 0) j++; - if (cec[i] > 1) k++; - if (cec[i] > 2) z++; - } - SG_SINFO( - "- Variables entering the working set at least one time: %i\n", j); - SG_SINFO( - "- Variables entering the working set at least two times: %i\n", k); - SG_SINFO( - "- Variables entering the working set at least three times: %i\n", z); - } - - - SG_FREE(bmem); - SG_FREE(bmemrid); - SG_FREE(pbmr); - SG_FREE(cec); - SG_FREE(ing); - SG_FREE(inaux); - SG_FREE(indnzin); - SG_FREE(index_in); - SG_FREE(inold); - SG_FREE(incom); - SG_FREE(indnzout); - SG_FREE(index_out); - SG_FREE(vau); - SG_FREE(alpha); - SG_FREE(weight); - SG_FREE(grad); - SG_FREE(stloc); - SG_FREE(st); - SG_FREE(sp_h); - SG_FREE(sp_hloc); - SG_FREE(sp_y); - SG_FREE(sp_D); - SG_FREE(sp_alpha); - delete Cache; - - aux = KER->KernelEvaluations; - SG_SINFO("- Total CPU time: %lf\n", total_time) - if (verbosity > 0) - { - SG_SINFO( - "- Total kernel evaluations: %.0lf\n", aux); - SG_SINFO( - "- Total inner solver iterations: %i\n", tot_vpm_iter); - if (projection_projector == 1) - SG_SINFO( - "- Total projector iterations: %i\n", tot_vpm_secant); - SG_SINFO( - "- Total preparation time: %lf\n", tot_prep_time); - SG_SINFO( - "- Total inner solver time: %lf\n", tot_vpm_time); - SG_SINFO( - "- Total gradient updating time: %lf\n", tot_st_time); - } - SG_SINFO("- Objective function value: %lf\n", fval) - objective_value=fval; - return aux; -} - -/******************************************************************************/ -/*** Quick sort for integer vectors ***/ -/******************************************************************************/ -void quick_si(int32_t a[], int32_t n) -{ - int32_t i, j, s, d, l, x, w, ps[20], pd[20]; - - l = 0; - ps[0] = 0; - pd[0] = n-1; - do - { - s = ps[l]; - d = pd[l]; - l--; - do - { - i = s; - j = d; - x = a[(s+d)/2]; - do - { - while (a[i] < x) i++; - while (a[j] > x) j--; - if (i <= j) - { - w = a[i]; - a[i] = a[j]; - i++; - a[j] = w; - j--; - } - } while(i<=j); - if (j-s > d-i) - { - l++; - ps[l] = s; - pd[l] = j; - s = i; - } - else - { - if (i < d) - { - l++; - ps[l] = i; - pd[l] = d; - } - d = j; - } - } while (s < d); - } while (l >= 0); -} - -/******************************************************************************/ -/*** Quick sort for real vectors returning also the exchanges ***/ -/******************************************************************************/ -void quick_s2(float64_t a[], int32_t n, int32_t ia[]) -{ - int32_t i, j, s, d, l, iw, ps[20], pd[20]; - float64_t x, w; - - l = 0; - ps[0] = 0; - pd[0] = n-1; - do - { - s = ps[l]; - d = pd[l]; - l--; - do - { - i = s; - j = d; - x = a[(s+d)/2]; - do - { - while (a[i] < x) i++; - while (a[j] > x) j--; - if (i <= j) - { - iw = ia[i]; - w = a[i]; - ia[i] = ia[j]; - a[i] = a[j]; - i++; - ia[j] = iw; - a[j] = w; - j--; - } - } while (i <= j); - if (j-s > d-i) - { - l++; - ps[l] = s; - pd[l] = j; - s = i; - } - else - { - if (i < d) - { - l++; - ps[l] = i; - pd[l] = d; - } - d = j; - } - } while (s < d); - } while(l>=0); -} - -/******************************************************************************/ -/*** Quick sort for integer vectors returning also the exchanges ***/ -/******************************************************************************/ -void quick_s3(int32_t a[], int32_t n, int32_t ia[]) -{ - int32_t i, j, s, d, l, iw, w, x, ps[20], pd[20]; - - l = 0; - ps[0] = 0; - pd[0] = n-1; - do - { - s = ps[l]; - d = pd[l]; - l--; - do - { - i = s; - j = d; - x = a[(s+d)/2]; - do - { - while (a[i] < x) i++; - while (a[j] > x) j--; - if (i <= j) - { - iw = ia[i]; - w = a[i]; - ia[i] = ia[j]; - a[i] = a[j]; - i++; - ia[j] = iw; - a[j] = w; - j--; - } - } while (i <= j); - if (j-s > d-i) - { - l++; - ps[l] = s; - pd[l] = j; - s = i; - } - else - { - if (i < d) - { - l++; - ps[l] = i; - pd[l] = d; - } - d = j; - } - } while (s < d); - } while (l >= 0); -} -} - -#endif // DOXYGEN_SHOULD_SKIP_THIS - -/******************************************************************************/ -/*** End of gpdtsolve.cpp file ***/ -/******************************************************************************/ -#endif //USE_GPL_SHOGUN diff --git a/src/shogun/lib/external/gpdtsolve.h b/src/shogun/lib/external/gpdtsolve.h deleted file mode 100644 index fceba025b2a..00000000000 --- a/src/shogun/lib/external/gpdtsolve.h +++ /dev/null @@ -1,175 +0,0 @@ -/* - * This program is free software; you can redistribute it and/or modify - * it under the terms of the GNU General Public License as published by - * the Free Software Foundation; either version 3 of the License, or - * (at your option) any later version. - * - *** Authors: *** - *** Thomas Serafini, Luca Zanni *** - *** Dept. of Mathematics, University of Modena and Reggio Emilia - ITALY *** - *** serafini.thomas@unimo.it, zanni.luca@unimo.it *** - *** Gaetano Zanghirati *** - *** Dept. of Mathematics, University of Ferrara - ITALY *** - *** g.zanghirati@unife.it *** - *** *** - *** Software homepage: http://dm.unife.it/gpdt *** - *** *** - *** This work is supported by the Italian FIRB Projects *** - *** 'Statistical Learning: Theory, Algorithms and Applications' *** - *** (grant RBAU01877P), http://slipguru.disi.unige.it/ASTA *** - *** and *** - *** 'Parallel Algorithms and Numerical Nonlinear Optimization' *** - *** (grant RBAU01JYPN), http://dm.unife.it/pn2o *** - *** *** - *** Copyright (C) 2004 by T. Serafini, G. Zanghirati, L. Zanni. *** - *** *** - *** SHOGUN adaptions Written (W) 2006-2008 Soeren Sonnenburg *** - */ - -#include -#ifdef USE_GPL_SHOGUN - -#include - -#ifndef DOXYGEN_SHOULD_SKIP_THIS -namespace shogun -{ -/** class QProblem */ -class QPproblem -{ -// ----------------- Public Data --------------- -public: - /** chunk size */ - int32_t chunk_size; - /** ell */ - int32_t ell; - /** y */ - int32_t *y; - /** delta sv */ - float64_t DELTAsv; - /** q */ - int32_t q; - /** max mw */ - int32_t maxmw; - /** c const */ - float64_t c_const; - /** bee */ - float64_t bee; - /** delta */ - float64_t delta; - /** linadd */ - bool linadd; - - /** kernel */ - sKernel* KER; - /** kernel type */ - int32_t ker_type; - /** projection solver */ - int32_t projection_solver; - /** projection projector */ - int32_t projection_projector; - /** preprocess mode */ - int32_t PreprocessMode; - /** preprocess size */ - int32_t preprocess_size; - /** verbosity */ - int32_t verbosity; - /** tau proximal */ - float64_t tau_proximal; - /** objective value */ - float64_t objective_value; - -// ----------------- Public Methods --------------- - /** constructor */ - QPproblem (); - ~QPproblem(); - - /** read SVM file - * - * @param fInput input filename - * @return an int - */ - int32_t ReadSVMFile (char *fInput); - - /** read GPDT binary - * - * @param fName input filename - * @return an int - */ - int32_t ReadGPDTBinary(char *fName); - - /** check if 2-class - * - * @return an int - */ - int32_t Check2Class (); - - /** subproblem - * - * @param dst where to copy subproblem to - * @param ker problem kernel - * @param len length - * @param perm perm - */ - static void copy_subproblem(QPproblem* dst, QPproblem* ker, int32_t len, int32_t *perm); - - /** PrepMP */ - void PrepMP (); - - /** solve gpdt - * - * @param solution - * @return something floaty - */ - float64_t gpdtsolve (float64_t *solution); - - /** solve pgpdt - * - * @param solution - * @return something floaty - */ - float64_t pgpdtsolve (float64_t *solution); - - /** check if lineadd is enabled - * - * @return if lineadd is enabled - */ - inline bool get_linadd_enabled() - { - return linadd; - } - - /** @return object name */ - virtual const char* get_name() const { return "QPproblem"; } - -// ----------------- Private Data --------------- -private: - int32_t dim; - int32_t *index_in, *index_out; - int32_t *ing, *inaux, *inold, *incom; - int32_t *cec; - int32_t nb; - int32_t *bmem, *bmemrid, *pbmr; - int32_t my_chunk_size; // chunk_size for the current processor - int32_t my_spD_offset; // offset of the current processor into sp_D matrix - int32_t recvl[32], displ[32]; - float64_t kktold; - float64_t DELTAvpm, InitialDELTAvpm, DELTAkin; - float64_t *alpha; - float64_t *grad, *st; - -// ----------------- Private Methods --------------- -private: - int32_t Preprocess0 (int32_t *aux, int32_t *sv); - int32_t Preprocess1 (sKernel* KER, int32_t *aux, int32_t *sv); - int32_t optimal (); - - bool is_zero(int32_t i) { return (alpha[i] < DELTAsv); } - bool is_free(int32_t i) - { return (alpha[i] > DELTAsv && alpha[i] < (c_const - DELTAsv)); } - bool is_bound(int32_t i) { return (alpha[i] > (c_const - DELTAsv)); } - -}; -} -#endif // DOXYGEN_SHOULD_SKIP_THIS -#endif //USE_GPL_SHOGUN diff --git a/src/shogun/lib/external/gpm.cpp b/src/shogun/lib/external/gpm.cpp deleted file mode 100644 index f3108a16338..00000000000 --- a/src/shogun/lib/external/gpm.cpp +++ /dev/null @@ -1,1246 +0,0 @@ -/****************************************************************************** - *** GPDT - Gradient Projection Decomposition Technique *** - ****************************************************************************** - *** *** - *** GPDT is a C++ software designed to train large-scale Support Vector *** - *** Machines for binary classification in both scalar and distributed *** - *** memory parallel environments. It uses the Joachims' problem *** - *** decomposition technique to split the whole quadratic programming (QP) *** - *** problem into a sequence of smaller QP subproblems, each one being *** - *** solved by a suitable gradient projection method (GPM). The presently *** - *** implemented GPMs are the Generalized Variable Projection Method *** - *** GVPM (T. Serafini, G. Zanghirati, L. Zanni, "Gradient Projection *** - *** Methods for Quadratic Programs and Applications in Training Support *** - *** Vector Machines"; Optim. Meth. Soft. 20, 2005, 353-378) and the *** - *** Dai-Fletcher Method DFGPM (Y. Dai and R. Fletcher,"New Algorithms for *** - *** Singly Linear Constrained Quadratic Programs Subject to Lower and *** - *** Upper Bounds"; Math. Prog. to appear). *** - *** *** - *** Authors: *** - *** Thomas Serafini, Luca Zanni *** - *** Dept. of Mathematics, University of Modena and Reggio Emilia - ITALY *** - *** serafini.thomas@unimo.it, zanni.luca@unimo.it *** - *** Gaetano Zanghirati *** - *** Dept. of Mathematics, University of Ferrara - ITALY *** - *** g.zanghirati@unife.it *** - *** *** - *** Software homepage: http://dm.unife.it/gpdt *** - *** *** - *** This work is supported by the Italian FIRB Projects *** - *** 'Statistical Learning: Theory, Algorithms and Applications' *** - *** (grant RBAU01877P), http://slipguru.disi.unige.it/ASTA *** - *** and *** - *** 'Parallel Algorithms and Numerical Nonlinear Optimization' *** - *** (grant RBAU01JYPN), http://dm.unife.it/pn2o *** - *** *** - *** Copyright (C) 2004-2008 by T. Serafini, G. Zanghirati, L. Zanni. *** - *** *** - *** COPYRIGHT NOTIFICATION *** - *** *** - *** Permission to copy and modify this software and its documentation *** - *** for internal research use is granted, provided that this notice is *** - *** retained thereon and on all copies or modifications. The authors and *** - *** their respective Universities makes no representations as to the *** - *** suitability and operability of this software for any purpose. It is *** - *** provided "as is" without express or implied warranty. *** - *** Use of this software for commercial purposes is expressly prohibited *** - *** without contacting the authors. *** - *** *** - *** This program is free software; you can redistribute it and/or modify *** - *** it under the terms of the GNU General Public License as published by *** - *** the Free Software Foundation; either version 3 of the License, or *** - *** (at your option) any later version. *** - *** *** - *** This program is distributed in the hope that it will be useful, *** - *** but WITHOUT ANY WARRANTY; without even the implied warranty of *** - *** MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the *** - *** GNU General Public License for more details. *** - *** *** - *** You should have received a copy of the GNU General Public License *** - *** along with this program; if not, write to the Free Software *** - *** Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. *** - *** *** - *** File: gpm.cpp *** - *** Type: scalar *** - *** Version: 1.0 *** - *** Date: October, 2005 *** - *** Revision: 1 *** - *** *** - *** SHOGUN adaptions Written (W) 2006-2008 Soeren Sonnenburg *** - ******************************************************************************/ -#include -#ifdef USE_GPL_SHOGUN -#include -#include -#include -#include -#include -#include - -namespace shogun -{ -#define maxvpm 30000 /* max number of method iterations allowed */ -#define maxprojections 200 -#define in 8000 /* max size of the QP problem to solve */ -#define alpha_max 1e10 -#define alpha_min 1e-10 - -extern uint32_t Randnext; -#define ThRand (Randnext = Randnext * 1103515245L + 12345L) -#define ThRandPos ((Randnext = Randnext * 1103515245L + 12345L) & 0x7fffffff) - -int32_t InnerProjector( - int32_t method, int32_t n, int32_t *iy, float64_t e, float64_t *qk, float64_t l, - float64_t u, float64_t *x, float64_t &lambda); - -/* Uncomment if you want to allocate vectors on the stack * - * instead of the heap. On some architectures this helps * - * improving speed, but may generate a stack overflow */ -// #define VARIABLES_ON_STACK - -/* Uncomment if you want to use the adaptive steplength * - in the GVPM solver */ -#define VPM_ADA - - -/****************************************************************************** - *** Generalized Variable Projection Method (T. Serafini, G. Zanghirati, *** - *** L. Zanni, "Gradient Projection Methods for Quadratic Programs and *** - *** Applications in Training Support Vector Machines"; *** - *** Optim. Meth. Soft. 20, 2005, 353-378) *** - ******************************************************************************/ -int32_t gvpm( - int32_t Projector, int32_t n, float32_t *vecA, float64_t *b, float64_t c, - float64_t e, int32_t *iy, float64_t *x, float64_t tol, int32_t *ls, - int32_t *proj) -{ - int32_t i, j, iter, it, it2, luv, info; - float64_t gd, max, normd, dAd, lam, lamnew, alpha, kktlam, ak, bk; - - int32_t lscount = 0, projcount = 0; - float64_t eps = 1.0e-16; - float64_t DELTAsv, ProdDELTAsv; - float64_t lam_ext; - - /* solver-specific settings */ -#ifdef VPM_ADA - int32_t nc = 1, ia1 = -1; - float64_t alpha1, alpha2; -#endif - - /* allocation-dependant settings */ -#ifdef VARIABLES_ON_STACK - - int32_t ipt[in], ipt2[in], uv[in]; - float64_t g[in], y[in], tempv[in], d[in], Ad[in], t[in]; - -#else - - int32_t *ipt, *ipt2, *uv; - float64_t *g, *y, *tempv, *d, *Ad, *t; - - /*** array allocations ***/ - ipt = SG_MALLOC(int32_t, n); - ipt2 = SG_MALLOC(int32_t, n); - uv = SG_MALLOC(int32_t, n); - g = SG_MALLOC(float64_t, n); - y = SG_MALLOC(float64_t, n); - d = SG_MALLOC(float64_t, n); - Ad = SG_MALLOC(float64_t, n); - t = SG_MALLOC(float64_t, n); - tempv = SG_MALLOC(float64_t, n); - -#endif - - DELTAsv = EPS_SV * c; - if (tol <= 1.0e-5 || n <= 20) - ProdDELTAsv = 0.0F; - else - ProdDELTAsv = EPS_SV * c; - - for (i = 0; i < n; i++) - tempv[i] = -x[i]; - lam_ext = 0.0; - projcount += InnerProjector(Projector, n, iy, e, tempv, 0, c, x, lam_ext); - - /* compute g = A*x + b in sparse form * - * (inline computation for better perfomrance) */ - { - float32_t *tempA; - - it = 0; - for (i = 0; i < n; i++) - if (fabs(x[i]) > ProdDELTAsv*1e-2) - ipt[it++] = i; - - memset(t, 0, n*sizeof(float64_t)); - for (i = 0; i < it; i++) - { - tempA = vecA + ipt[i]*n; - for (j = 0; j < n; j++) - t[j] += (tempA[j] * x[ipt[i]]); - } - } - - for (i = 0; i < n; i++) - { - g[i] = t[i] + b[i], - y[i] = g[i] - x[i]; - } - - projcount += InnerProjector(Projector, n, iy, e, y, 0, c, tempv, lam_ext); - - max = alpha_min; - for (i = 0; i < n; i++) - { - y[i] = tempv[i] - x[i]; - if (fabs(y[i]) > max) - max = fabs(y[i]); - } - - if (max < c*tol*1e-3) - { - lscount = 0; - iter = 0; - goto Clean; - } - - alpha = 1.0 / max; - - for (iter = 1; iter <= maxvpm; iter++) - { - for (i = 0; i < n; i++) - tempv[i] = alpha*g[i] - x[i]; - - projcount += InnerProjector(Projector, n, iy, e, tempv, 0, c, y, lam_ext); - - gd = 0.0; - for (i = 0; i < n; i++) - { - d[i] = y[i] - x[i]; - gd += d[i] * g[i]; - } - - /* compute Ad = A*d or Ad = Ay-t depending on their sparsity * - * (inline computation for better perfomrance) */ - { - float32_t *tempA; - - it = it2 = 0; - for (i = 0; i < n; i++) - if (fabs(d[i]) > (ProdDELTAsv*1.0e-2)) - ipt[it++] = i; - for (i = 0; i < n; i++) - if (fabs(y[i]) > ProdDELTAsv) - ipt2[it2++] = i; - - memset(Ad, 0, n*sizeof(float64_t)); - if (it < it2) // Ad = A*d - { - for (i = 0; i < it; i++) - { - tempA = vecA + ipt[i]*n; - for (j = 0; j < n; j++) - Ad[j] += (tempA[j] * d[ipt[i]]); - } - } - else // Ad = A*y - t - { - for (i = 0; i < it2; i++) - { - tempA = vecA + ipt2[i]*n; - for (j = 0; j < n; j++) - Ad[j] += (tempA[j] * y[ipt2[i]]); - } - for (j = 0; j < n; j++) - Ad[j] -= t[j]; - } - } - - normd = 0.0; - for (i = 0; i < n; i++) - normd += d[i] * d[i]; - - dAd = 0.0; - for (i = 0; i < n; i++) - dAd += d[i]*Ad[i]; - - if (dAd > eps*normd && gd < 0.0) - { - lam = lamnew = -gd/dAd; - if (lam > 1.0 || lam < 0.0) - lam = 1.0; - else - lscount++; - -#ifdef VPM_ADA - - /*** use the adaptive switching rule for steplength selection ***/ - - // compute alpha1 = (d'* (d.*diaga)) / (d'*Ad); - alpha1 = normd / dAd; - - // alpha2 = d'*Ad / (Ad' * (Ad./diaga)); - alpha2 = 0.0; - for (i = 0; i < n; i++) - alpha2 += Ad[i] * Ad[i]; - alpha2 = dAd / alpha2; - - if ( (nc > 2 - && ( - (ia1 == 1 - && ( - lamnew < 0.1 || (alpha1 > alpha && alpha2 < alpha) - ) - ) - || - (ia1 == -1 - && ( - lamnew > 5.0 || (alpha1 > alpha && alpha2 < alpha) - ) - ) - ) - ) - || nc > 9 ) - { - ia1 = -ia1; - nc = 0; - } - - if (ia1 == 1) - alpha = alpha1; - else - alpha = alpha2; - - if (alpha < alpha_min) - alpha = alpha_min; - else if (alpha > alpha_max) - alpha = alpha_max; - - nc++; - -#else - - /*** use the fixed switching rule for steplength selection ***/ - - if ((iter % 6) < 3) // alpha = d'*Ad / (Ad' * (Ad./diaga)); - { - alpha = 0.0; - for (i = 0; i < n; i++) - alpha += Ad[i] * Ad[i]; - alpha = dAd / alpha; - } - else // alpha = (d'* (d.*diaga)) / (d'*Ad); - { - alpha = 0.0; - for (i = 0; i < n; i++) - alpha += d[i] * d[i]; - alpha = alpha / dAd; - } - -#endif - - } - else - { - lam = 1.0; - alpha = alpha_max; - } - - for (i = 0; i < n; i++) - { - x[i] = x[i] + lam*d[i]; - t[i] = t[i] + lam*Ad[i]; - g[i] = b[i] + t[i]; - } - - /*** stopping criterion based on KKT conditions ***/ - bk = 0.0; - ak = 0.0; - for (i = 0; i < n; i++) - { - bk += x[i] * x[i]; - ak += d[i] * d[i]; - } - - if (lam*sqrt(ak) < tol*10 * sqrt(bk)) - { - it = 0; - luv = 0; - kktlam = 0.0; - for (i = 0; i < n; i++) - { - if (x[i] > DELTAsv && x[i] < c-DELTAsv) - { - ipt[it++] = i; - kktlam = kktlam - iy[i]*g[i]; - } - else - uv[luv++] = i; - } - - if (it == 0) - { - if (lam*sqrt(ak) < tol*0.5 * sqrt(bk)) - goto Clean; - } - else - { - kktlam = kktlam/it; - info = 1; - for (i = 0; i < it; i++) - { - if (fabs(iy[ipt[i]]*g[ipt[i]]+kktlam) > tol) - { - info = 0; - break; - } - } - - if (info == 1) - { - for (i = 0; i < luv; i++) - { - if (x[uv[i]] <= DELTAsv) - { - if (g[uv[i]] + kktlam*iy[uv[i]] < -tol) - { - info = 0; - break; - } - } - else - { - if (g[uv[i]] + kktlam*iy[uv[i]] > tol) - { - info = 0; - break; - } - } - } - } - - if (info == 1) - goto Clean; - } - } // stopping rule based on the norm of d_k - } - - SG_SWARNING("GVPM exits after maxvpm = %d iterations.\n", maxvpm) - -Clean: - - /*** allocation-dependant freeing ***/ -#ifndef VARIABLES_ON_STACK - SG_FREE(t); - SG_FREE(uv); - SG_FREE(ipt2); - SG_FREE(ipt); - SG_FREE(g); - SG_FREE(y); - SG_FREE(tempv); - SG_FREE(d); - SG_FREE(Ad); -#endif - - if (ls != NULL) *ls = lscount; - if (proj != NULL) *proj = projcount; - return(iter); -} - -/****************************************************************************** - *** Dai-Fletcher QP solver (Y. Dai and R. Fletcher,"New Algorithms for *** - *** Singly Linear Constrained Quadratic Programs Subject to Lower and *** - *** Upper Bounds"; Math. Prog. to appear) *** - ******************************************************************************/ -int32_t FletcherAlg2A( - int32_t Projector, int32_t n, float32_t *vecA, float64_t *b, float64_t c, - float64_t e, int32_t *iy, float64_t *x, float64_t tol, int32_t *ls, - int32_t *proj) -{ - int32_t i, j, iter, it, it2, luv, info, lscount = 0, projcount = 0; - float64_t gd, max, ak, bk, akold, bkold, lamnew, alpha, kktlam, lam_ext; - float64_t eps = 1.0e-16; - float64_t DELTAsv, ProdDELTAsv; - - /*** variables for the adaptive nonmonotone linesearch ***/ - int32_t L, llast; - float64_t fr, fbest, fv, fc, fv0; - - /*** allocation-dependant settings ***/ -#ifdef VARIABLES_ON_STACK - - int32_t ipt[in], ipt2[in], uv[in]; - float64_t g[in], y[in], tempv[in], d[in], Ad[in], t[in], - xplus[in], tplus[in], sk[in], yk[in]; -#else - - int32_t *ipt, *ipt2, *uv; - float64_t *g, *y, *tempv, *d, *Ad, *t, *xplus, *tplus, *sk, *yk; - - /*** arrays allocation ***/ - ipt = SG_MALLOC(int32_t, n); - ipt2 = SG_MALLOC(int32_t, n); - uv = SG_MALLOC(int32_t, n); - g = SG_MALLOC(float64_t, n); - y = SG_MALLOC(float64_t, n); - tempv = SG_MALLOC(float64_t, n); - d = SG_MALLOC(float64_t, n); - Ad = SG_MALLOC(float64_t, n); - t = SG_MALLOC(float64_t, n); - xplus = SG_MALLOC(float64_t, n); - tplus = SG_MALLOC(float64_t, n); - sk = SG_MALLOC(float64_t, n); - yk = SG_MALLOC(float64_t, n); - -#endif - - DELTAsv = EPS_SV * c; - if (tol <= 1.0e-5 || n <= 20) - ProdDELTAsv = 0.0F; - else - ProdDELTAsv = EPS_SV * c; - - for (i = 0; i < n; i++) - tempv[i] = -x[i]; - - lam_ext = 0.0; - - projcount += InnerProjector(Projector, n, iy, e, tempv, 0, c, x, lam_ext); - - // g = A*x + b; - // SparseProd(n, t, A, x, ipt); - { - float32_t *tempA; - - it = 0; - for (i = 0; i < n; i++) - if (fabs(x[i]) > ProdDELTAsv) - ipt[it++] = i; - - memset(t, 0, n*sizeof(float64_t)); - for (i = 0; i < it; i++) - { - tempA = vecA + ipt[i] * n; - for (j = 0; j < n; j++) - t[j] += (tempA[j]*x[ipt[i]]); - } - } - - for (i = 0; i < n; i++) - { - g[i] = t[i] + b[i], - y[i] = g[i] - x[i]; - } - - projcount += InnerProjector(Projector, n, iy, e, y, 0, c, tempv, lam_ext); - - max = alpha_min; - for (i = 0; i < n; i++) - { - y[i] = tempv[i] - x[i]; - if (fabs(y[i]) > max) - max = fabs(y[i]); - } - - if (max < c*tol*1e-3) - { - lscount = 0; - iter = 0; - goto Clean; - } - - alpha = 1.0 / max; - - fv0 = 0.0; - for (i = 0; i < n; i++) - fv0 += x[i] * (0.5*t[i] + b[i]); - - /*** adaptive nonmonotone linesearch ***/ - L = 2; - fr = alpha_max; - fbest = fv0; - fc = fv0; - llast = 0; - akold = bkold = 0.0; - - for (iter = 1; iter <= maxvpm; iter++) - { - for (i = 0; i < n; i++) - tempv[i] = alpha*g[i] - x[i]; - - projcount += InnerProjector(Projector, n, iy, e, tempv, 0, c, y, lam_ext); - - gd = 0.0; - for (i = 0; i < n; i++) - { - d[i] = y[i] - x[i]; - gd += d[i] * g[i]; - } - - /* compute Ad = A*d or Ad = A*y - t depending on their sparsity */ - { - float32_t *tempA; - - it = it2 = 0; - for (i = 0; i < n; i++) - if (fabs(d[i]) > (ProdDELTAsv*1.0e-2)) - ipt[it++] = i; - for (i = 0; i < n; i++) - if (fabs(y[i]) > ProdDELTAsv) - ipt2[it2++] = i; - - memset(Ad, 0, n*sizeof(float64_t)); - if (it < it2) // compute Ad = A*d - { - for (i = 0; i < it; i++) - { - tempA = vecA + ipt[i]*n; - for (j = 0; j < n; j++) - Ad[j] += (tempA[j] * d[ipt[i]]); - } - } - else // compute Ad = A*y-t - { - for (i = 0; i < it2; i++) - { - tempA = vecA + ipt2[i]*n; - for (j = 0; j < n; j++) - Ad[j] += (tempA[j] * y[ipt2[i]]); - } - for (j = 0; j < n; j++) - Ad[j] -= t[j]; - } - } - - ak = 0.0; - for (i = 0; i < n; i++) - ak += d[i] * d[i]; - - bk = 0.0; - for (i = 0; i < n; i++) - bk += d[i]*Ad[i]; - - if (bk > eps*ak && gd < 0.0) // ak is normd - lamnew = -gd/bk; - else - lamnew = 1.0; - - fv = 0.0; - for (i = 0; i < n; i++) - { - xplus[i] = x[i] + d[i]; - tplus[i] = t[i] + Ad[i]; - fv += xplus[i] * (0.5*tplus[i] + b[i]); - } - - if ((iter == 1 && fv >= fv0) || (iter > 1 && fv >= fr)) - { - lscount++; - fv = 0.0; - for (i = 0; i < n; i++) - { - xplus[i] = x[i] + lamnew*d[i]; - tplus[i] = t[i] + lamnew*Ad[i]; - fv += xplus[i] * (0.5*tplus[i] + b[i]); - } - } - - for (i = 0; i < n; i++) - { - sk[i] = xplus[i] - x[i]; - yk[i] = tplus[i] - t[i]; - x[i] = xplus[i]; - t[i] = tplus[i]; - g[i] = t[i] + b[i]; - } - - // update the line search control parameters - - if (fv < fbest) - { - fbest = fv; - fc = fv; - llast = 0; - } - else - { - fc = (fc > fv ? fc : fv); - llast++; - if (llast == L) - { - fr = fc; - fc = fv; - llast = 0; - } - } - - ak = bk = 0.0; - for (i = 0; i < n; i++) - { - ak += sk[i] * sk[i]; - bk += sk[i] * yk[i]; - } - - if (bk < eps*ak) - alpha = alpha_max; - else - { - if (bkold < eps*akold) - alpha = ak/bk; - else - alpha = (akold+ak)/(bkold+bk); - - if (alpha > alpha_max) - alpha = alpha_max; - else if (alpha < alpha_min) - alpha = alpha_min; - } - - akold = ak; - bkold = bk; - - /*** stopping criterion based on KKT conditions ***/ - - bk = 0.0; - for (i = 0; i < n; i++) - bk += x[i] * x[i]; - - if (sqrt(ak) < tol*10 * sqrt(bk)) - { - it = 0; - luv = 0; - kktlam = 0.0; - for (i = 0; i < n; i++) - { - if ((x[i] > DELTAsv) && (x[i] < c-DELTAsv)) - { - ipt[it++] = i; - kktlam = kktlam - iy[i]*g[i]; - } - else - uv[luv++] = i; - } - - if (it == 0) - { - if (sqrt(ak) < tol*0.5 * sqrt(bk)) - goto Clean; - } - else - { - - kktlam = kktlam/it; - info = 1; - for (i = 0; i < it; i++) - if ( fabs(iy[ipt[i]] * g[ipt[i]] + kktlam) > tol ) - { - info = 0; - break; - } - - if (info == 1) - { - for (i = 0; i < luv; i++) - { - if (x[uv[i]] <= DELTAsv) - { - if (g[uv[i]] + kktlam*iy[uv[i]] < -tol) - { - info = 0; - break; - } - } - else - { - if (g[uv[i]] + kktlam*iy[uv[i]] > tol) - { - info = 0; - break; - } - } - } - } - - if (info == 1) - goto Clean; - } - } - } - - SG_SWARNING("Dai-Fletcher method exits after maxvpm = %d iterations.\n", maxvpm) - -Clean: - -#ifndef VARIABLES_ON_STACK - SG_FREE(sk); - SG_FREE(yk); - SG_FREE(tplus); - SG_FREE(xplus); - SG_FREE(t); - SG_FREE(uv); - SG_FREE(ipt2); - SG_FREE(ipt); - SG_FREE(g); - SG_FREE(y); - SG_FREE(tempv); - SG_FREE(d); - SG_FREE(Ad); -#endif - - if (ls != NULL) *ls = lscount; - if (proj != NULL) *proj = projcount; - return(iter); - -} - -/******************************************************************************/ -/*** Encapsulating method to call the chosen Gradient Projection Method ***/ -/******************************************************************************/ -int32_t gpm_solver( - int32_t Solver, int32_t Projector, int32_t n, float32_t *A, float64_t *b, - float64_t c, float64_t e, int32_t *iy, float64_t *x, float64_t tol, - int32_t *ls, int32_t *proj) -{ - /*** Uncomment the following if you need to scale the QP Hessian matrix - *** before calling the chosen solver - int32_t i, j; - float32_t *ptrA; - float64_t max, s; - - max = fabs(A[0][0]); - for (i = 1; i < n; i++) - if (fabs(A[i][i]) > max) - max = fabs(A[i][i]); - - s = 1.0 / max; - ptrA = vecA; - for (i = 0; i < n; i++) - for (j = 0;j < n; j++) - *ptrA++ = (float32_t)(A[i][j]*s); - - if (Solver == SOLVER_FLETCHER) - j = FletcherAlg2A(n, vecA, b, c/s, e/s, iy, x, tol, ls); - else - j = gvpm(n, vecA, b, c/s, e/s, iy, x, tol, ls); - - for (i = 0; i < n; i++) - x[i] *= s; - ***/ - - /*** calling the chosen solver with unscaled data ***/ - if (Solver == SOLVER_FLETCHER) - return FletcherAlg2A(Projector, n, A, b, c, e, iy, x, tol, ls, proj); - else - return gvpm(Projector, n, A, b, c, e, iy, x, tol, ls, proj); -} - -/****************************************************************************** - *** Piecewise linear monotone target function for the Dai-Fletcher *** - *** projector (Y. Dai and R. Fletcher, "New Algorithms for Singly Linear *** - *** Constrained Quadratic Programs Subject to Lower and Upper Bounds"; *** - *** Math. Prog. to appear) *** - ******************************************************************************/ -float64_t ProjectR( - float64_t *x, int32_t n, float64_t lambda, int32_t *a, float64_t b, - float64_t *c, float64_t l, float64_t u) -{ - int32_t i; - float64_t r = 0.0; - - for (i = 0; i < n; i++) - { - x[i] = -c[i] + lambda*(float64_t)a[i]; - if (x[i] >= u) x[i] = u; - else if (x[i] < l) x[i] = l; - r += (float64_t)a[i]*x[i]; - } - - return (r - b); -} - -/****************************************************************************** - *** Dai-Fletcher QP projector (Y. Dai and R. Fletcher, "New Algorithms for *** - *** Singly Linear Constrained Quadratic Programs Subject to Lower and *** - *** Upper Bounds"; Math. Prog. to appear) *** - ******************************************************************************/ -/*** *** - *** Solves the problem min x'*x/2 + c'*x *** - *** subj to a'*x - b = 0 *** - *** l <= x <= u *** - ******************************************************************************/ -int32_t ProjectDai( - int32_t n, int32_t *a, float64_t b, float64_t *c, float64_t l, float64_t u, - float64_t *x, float64_t &lam_ext) -{ - float64_t lambda, lambdal, lambdau, dlambda, lambda_new, tol_lam; - float64_t r, rl, ru, s, tol_r; - int32_t iter; - - tol_lam = 1.0e-11; - tol_r = 1.0e-10 * sqrt((u-l)*(float64_t)n); - lambda = lam_ext; - dlambda = 0.5; - iter = 1; - b = -b; - - // Bracketing Phase - r = ProjectR(x, n, lambda, a, b, c, l, u); - if (fabs(r) < tol_r) - return 0; - - if (r < 0.0) - { - lambdal = lambda; - rl = r; - lambda = lambda + dlambda; - r = ProjectR(x, n, lambda, a, b, c, l, u); - while (r < 0.0) - { - lambdal = lambda; - s = rl/r - 1.0; - if (s < 0.1) s = 0.1; - dlambda = dlambda + dlambda/s; - lambda = lambda + dlambda; - rl = r; - r = ProjectR(x, n, lambda, a, b, c, l, u); - } - lambdau = lambda; - ru = r; - } - else - { - lambdau = lambda; - ru = r; - lambda = lambda - dlambda; - r = ProjectR(x, n, lambda, a, b, c, l, u); - while (r > 0.0) - { - lambdau = lambda; - s = ru/r - 1.0; - if (s < 0.1) s = 0.1; - dlambda = dlambda + dlambda/s; - lambda = lambda - dlambda; - ru = r; - r = ProjectR(x, n, lambda, a, b, c, l, u); - } - lambdal = lambda; - rl = r; - } - - - // Secant Phase - s = 1.0 - rl/ru; - dlambda = dlambda/s; - lambda = lambdau - dlambda; - r = ProjectR(x, n, lambda, a, b, c, l, u); - - while ( fabs(r) > tol_r - && dlambda > tol_lam * (1.0 + fabs(lambda)) - && iter < maxprojections ) - { - iter++; - if (r > 0.0) - { - if (s <= 2.0) - { - lambdau = lambda; - ru = r; - s = 1.0 - rl/ru; - dlambda = (lambdau - lambdal) / s; - lambda = lambdau - dlambda; - } - else - { - s = ru/r-1.0; - if (s < 0.1) s = 0.1; - dlambda = (lambdau - lambda) / s; - lambda_new = 0.75*lambdal + 0.25*lambda; - if (lambda_new < (lambda - dlambda)) - lambda_new = lambda - dlambda; - lambdau = lambda; - ru = r; - lambda = lambda_new; - s = (lambdau - lambdal) / (lambdau - lambda); - } - } - else - { - if (s >= 2.0) - { - lambdal = lambda; - rl = r; - s = 1.0 - rl/ru; - dlambda = (lambdau - lambdal) / s; - lambda = lambdau - dlambda; - } - else - { - s = rl/r - 1.0; - if (s < 0.1) s = 0.1; - dlambda = (lambda-lambdal) / s; - lambda_new = 0.75*lambdau + 0.25*lambda; - if (lambda_new > (lambda + dlambda)) - lambda_new = lambda + dlambda; - lambdal = lambda; - rl = r; - lambda = lambda_new; - s = (lambdau - lambdal) / (lambdau-lambda); - } - } - r = ProjectR(x, n, lambda, a, b, c, l, u); - } - - lam_ext = lambda; - if (iter >= maxprojections) - SG_SERROR("Projector exits after max iterations: %d\n", iter) - - return (iter); -} - -#define SWAP(a,b) { register float64_t t=(a);(a)=(b);(b)=t; } - -/*** Median computation using Quick Select ***/ -float64_t quick_select(float64_t *arr, int32_t n) -{ - int32_t low, high; - int32_t median; - int32_t middle, l, h; - - low = 0; - high = n-1; - median = (low + high) / 2; - - for (;;) - { - if (high <= low) - return arr[median]; - - if (high == low + 1) - { - if (arr[low] > arr[high]) - SWAP(arr[low], arr[high]); - return arr[median]; - } - - middle = (low + high) / 2; - if (arr[middle] > arr[high]) SWAP(arr[middle], arr[high]); - if (arr[low] > arr[high]) SWAP(arr[low], arr[high]); - if (arr[middle] > arr[low]) SWAP(arr[middle], arr[low]); - - SWAP(arr[middle], arr[low+1]); - - l = low + 1; - h = high; - for (;;) - { - do l++; while (arr[low] > arr[l]); - do h--; while (arr[h] > arr[low]); - if (h < l) - break; - SWAP(arr[l], arr[h]); - } - - SWAP(arr[low], arr[h]); - if (h <= median) - low = l; - if (h >= median) - high = h - 1; - } -} - -/****************************************************************************** - *** Pardalos-Kovoor projector (P.M. Pardalos and N. Kovoor, "An Algorithm *** - *** for a Singly Constrained Class of Quadratic Programs Subject to Upper *** - *** and Lower Bounds"; Math. Prog. 46, 1990, 321-328). *** - ****************************************************************************** - *** Solves the problem *** - *** min x'*x/2 + qk'*x *** - *** subj to iy'*x + e = 0 *** - *** l <= x <= u *** - *** iy(i) ~= 0 *** - ******************************************************************************/ - -int32_t Pardalos( - int32_t n, int32_t *iy, float64_t e, float64_t *qk, float64_t low, - float64_t up, float64_t *x) -{ - int32_t i, l, iter; /* conters */ - int32_t luv, lxint; /* dimensions */ - float64_t d, xmin, xmax, xmold, xmid, xx, ts, sw, s, s1, testsum; - - /*** allocation-dependant settings ***/ -#ifdef VARIABLES_ON_STACK - int32_t uv[in], uvt[in]; - float64_t xint[2*in+2], xint2[2*in+2], a[in], b[in], at[in], bt[in]; - float64_t newdia[in], newdt[in]; -#else - - int32_t *uv, *uvt; - float64_t *xint, *xint2, *a, *b, *at, *bt, *newdia, *newdt; - - /*** arrays allocation ***/ - uv = SG_MALLOC(int32_t, n); - uvt = SG_MALLOC(int32_t, n); - a = SG_MALLOC(float64_t, n); - b = SG_MALLOC(float64_t, n); - at = SG_MALLOC(float64_t, n); - bt = SG_MALLOC(float64_t, n); - newdia = SG_MALLOC(float64_t, n); - newdt = SG_MALLOC(float64_t, n); - xint = SG_MALLOC(float64_t, (2*n + 2)); - xint2 = SG_MALLOC(float64_t, (2*n + 2)); - -#endif - - d = 0.0; - for (i = 0; i < n; i++) - d += iy[i] * qk[i]; - d = 0.5 * (d-e); - - for (i = 0; i < n; i++) - { - /* The following computations should divide by iy[i] instead * - * of multiply by iy[i], but this is correct for binary classification * - * with labels -1 and 1. */ - if (iy[i] > 0) - { - a[i] = ((qk[i] + low) * iy[i]) * 0.5; - b[i] = ((up + qk[i]) * iy[i]) * 0.5; - } - else - { - b[i] = ((qk[i] + low) * iy[i]) * 0.5; - a[i] = ((up + qk[i]) * iy[i]) * 0.5; - } - newdia[i] = (iy[i]*iy[i]); - } - - xmin = -1e33; - xmax = 1e33; - - /* arrays initialization */ - for (i = 0; i < n; i++) - { - uv[i] = i; /* contains the unset variables */ - xint[i] = a[i]; - xint[n+i] = b[i]; - } - - xmid = xmin; - xint[2*n] = xmin; - xint[2*n+1] = xmax; - ts = 0.0; - sw = 0.0; - luv = n; - lxint = 2*n+2; - - iter = 0; - do { - for (i = 0; i < luv; i++) - { - at[i] = a[uv[i]]; - bt[i] = b[uv[i]]; - newdt[i] = newdia[uv[i]]; - } - - xmold = xmid; - xmid = quick_select(xint, lxint); - if (xmold == xmid) - xmid = xint[(int32_t)(ThRandPos % lxint)]; - - s = ts; - s1 = sw; - for (i = 0; i < luv; i++) - { - if (xmid > bt[i]) - s += newdt[i]*bt[i]; - else if (xmid < at[i]) - s += newdt[i]*at[i]; - else - s1 += newdt[i]; - } - - testsum = s + s1*xmid; - if (testsum <= (d+(1e-15))) - xmin = xmid; - if (testsum >= (d-(1e-15))) - xmax = xmid; - - l = 0; - for (i = 0; i < lxint; i++) - if((xint[i] >= xmin) && (xint[i] <= xmax)) - xint2[l++] = xint[i]; - lxint = l; - sg_memcpy(xint, xint2, lxint*sizeof(float64_t)); - - l = 0; - for (i = 0; i < luv; i++) - { - if (xmin >= bt[i]) - ts += newdt[i]*bt[i]; - else if (xmax <= at[i]) - ts += newdt[i]*at[i]; - else if ((at[i] <= xmin) && (bt[i] >= xmax)) - sw += newdt[i]; - else - uvt[l++] = uv[i]; - } - luv = l; - sg_memcpy(uv, uvt, luv*sizeof(int32_t)); - iter++; - } while(luv != 0 && iter < maxprojections); - - if (sw == 0) - xx = xmin; - else - xx = (d-ts) / sw; - - for (i = 0; i < n; i++) - { - if (b[i] <= xmin) - x[i] = b[i]; - else if (a[i] >= xmax) - x[i] = a[i]; - else if ((a[i]<=xmin) && (xmax<=b[i])) - x[i] = xx; - else - SG_SWARNING("Inner solver troubles...\n") - } - - for (i = 0; i < n; i++) - x[i] = (2.0*x[i]*iy[i]-qk[i]); - -#ifndef VARIABLES_ON_STACK - SG_FREE(newdt); - SG_FREE(newdia); - SG_FREE(a); - SG_FREE(b); - SG_FREE(uvt); - SG_FREE(uv); - SG_FREE(bt); - SG_FREE(at); - SG_FREE(xint2); - SG_FREE(xint); -#endif - - return(iter); -} - -/******************************************************************************/ -/*** Wrapper method to call the selected inner projector ***/ -/******************************************************************************/ -int32_t InnerProjector( - int32_t method, int32_t n, int32_t *iy, float64_t e, float64_t *qk, - float64_t l, float64_t u, float64_t *x, float64_t &lambda) -{ - if (method == 0) - return Pardalos(n, iy, e, qk, l, u, x); - else - return ProjectDai(n, iy, e, qk, l, u, x, lambda); -} -} -/******************************************************************************/ -/*** End of gpm.cpp file ***/ -/******************************************************************************/ -#endif //USE_GPL_SHOGUN diff --git a/src/shogun/lib/external/gpm.h b/src/shogun/lib/external/gpm.h deleted file mode 100644 index eba58bcc2b4..00000000000 --- a/src/shogun/lib/external/gpm.h +++ /dev/null @@ -1,54 +0,0 @@ -/* - * This program is free software; you can redistribute it and/or modify - * it under the terms of the GNU General Public License as published by - * the Free Software Foundation; either version 3 of the License, or - * (at your option) any later version. - * - *** Authors: *** - *** Thomas Serafini, Luca Zanni *** - *** Dept. of Mathematics, University of Modena and Reggio Emilia - ITALY *** - *** serafini.thomas@unimo.it, zanni.luca@unimo.it *** - *** Gaetano Zanghirati *** - *** Dept. of Mathematics, University of Ferrara - ITALY *** - *** g.zanghirati@unife.it *** - *** *** - *** Software homepage: http://dm.unife.it/gpdt *** - *** *** - *** This work is supported by the Italian FIRB Projects *** - *** 'Statistical Learning: Theory, Algorithms and Applications' *** - *** (grant RBAU01877P), http://slipguru.disi.unige.it/ASTA *** - *** and *** - *** 'Parallel Algorithms and Numerical Nonlinear Optimization' *** - *** (grant RBAU01JYPN), http://dm.unife.it/pn2o *** - *** *** - *** Copyright (C) 2004-2008 by T. Serafini, G. Zanghirati, L. Zanni. *** - *** *** - *** SHOGUN adaptions Written (W) 2006-2008 Soeren Sonnenburg *** - */ - -#include -#ifdef USE_GPL_SHOGUN -#include - -namespace shogun -{ -/** gpm solver - * @param Solver - * @param Projector - * @param n - * @param A - * @param b - * @param c - * @param e - * @param iy - * @param x - * @param tol - * @param ls - * @param proj - */ -int32_t gpm_solver( - int32_t Solver, int32_t Projector, int32_t n, float32_t *A, float64_t *b, - float64_t c, float64_t e, int32_t *iy, float64_t *x, float64_t tol, - int32_t *ls = 0, int32_t *proj = 0); -} -#endif //USE_GPL_SHOGUN diff --git a/src/shogun/lib/external/libocas.cpp b/src/shogun/lib/external/libocas.cpp deleted file mode 100644 index e577fa157ce..00000000000 --- a/src/shogun/lib/external/libocas.cpp +++ /dev/null @@ -1,1977 +0,0 @@ -/* - * This program is free software; you can redistribute it and/or modify - * it under the terms of the GNU General Public License as published by - * the Free Software Foundation; either version 3 of the License, or - * (at your option) any later version. - * - * libocas.c: Implementation of the OCAS solver for training - * linear SVM classifiers. - * - * Copyright (C) 2008 Vojtech Franc, xfrancv@cmp.felk.cvut.cz - * Soeren Sonnenburg, soeren.sonnenburg@first.fraunhofer.de - *-------------------------------------------------------------------- */ - -#include -#ifdef USE_GPL_SHOGUN - -#include -#include -#include -#ifndef _WIN32 -#include -#endif -#include -#include -#include - -#include -#include -#include -#include - -namespace shogun -{ -#define MU 0.1 /* must be from (0,1> 1..means that OCAS becomes equivalent to CPA */ - /* see paper Franc&Sonneburg JMLR 2009 */ - -static const uint32_t QPSolverMaxIter = 10000000; - -static float64_t *H; -static uint32_t BufSize; - -/*---------------------------------------------------------------------- - Returns pointer at i-th column of Hessian matrix. - ----------------------------------------------------------------------*/ -static const float64_t *get_col( uint32_t i) -{ - return( &H[ BufSize*i ] ); -} - -/*---------------------------------------------------------------------- - Returns time of the day in seconds. - ----------------------------------------------------------------------*/ -static float64_t get_time() -{ - struct timeval tv; - if (gettimeofday(&tv, NULL)==0) - return tv.tv_sec+((float64_t)(tv.tv_usec))/1e6; - else - return 0.0; -} - - -/*---------------------------------------------------------------------- - Linear binary Ocas-SVM solver with additinal contraint enforceing - a subset of weights (indices of the weights given by num_nnw/nnw_idx) - to be non-negative. - - ----------------------------------------------------------------------*/ -ocas_return_value_T svm_ocas_solver_nnw( - float64_t C, - uint32_t nData, - uint32_t num_nnw, - uint32_t* nnw_idx, - float64_t TolRel, - float64_t TolAbs, - float64_t QPBound, - float64_t MaxTime, - uint32_t _BufSize, - uint8_t Method, - int (*add_pw_constr)(uint32_t, uint32_t, void*), - void (*clip_neg_W)(uint32_t, uint32_t*, void*), - void (*compute_W)(float64_t*, float64_t*, float64_t*, uint32_t, void*), - float64_t (*update_W)(float64_t, void*), - int (*add_new_cut)(float64_t*, uint32_t*, uint32_t, uint32_t, void*), - int (*compute_output)(float64_t*, void* ), - int (*sort)(float64_t*, float64_t*, uint32_t), - void (*ocas_print)(ocas_return_value_T), - void* user_data) -{ - ocas_return_value_T ocas={0,0,0,0,0,0,0,0,0,0,0,0,0,0,0}; - float64_t *b, *alpha, *diag_H; - float64_t *output, *old_output; - float64_t xi, sq_norm_W, QPSolverTolRel, dot_prod_WoldW, sq_norm_oldW; - float64_t A0, B0, GradVal, t, t1, t2, *Ci, *Bi, *hpf, *hpb; - float64_t start_time, ocas_start_time; - uint32_t cut_length; - uint32_t i, *new_cut; - uint32_t *I; -/* uint8_t S = 1; */ - libqp_state_T qp_exitflag; - - float64_t max_cp_norm; - float64_t max_b; - float64_t Cs[2]; - uint8_t S[2]; - - ocas_start_time = get_time(); - ocas.qp_solver_time = 0; - ocas.output_time = 0; - ocas.sort_time = 0; - ocas.add_time = 0; - ocas.w_time = 0; - ocas.print_time = 0; - - BufSize = _BufSize; - - QPSolverTolRel = TolRel*0.5; - - H=NULL; - b=NULL; - alpha=NULL; - new_cut=NULL; - I=NULL; - diag_H=NULL; - output=NULL; - old_output=NULL; - hpf=NULL; - hpb = NULL; - Ci=NULL; - Bi=NULL; - - /* Hessian matrix contains dot product of normal vectors of selected cutting planes */ - H = (float64_t*)LIBOCAS_CALLOC(BufSize*BufSize, float64_t); - if(H == NULL) - { - ocas.exitflag=-2; - goto cleanup; - } - - /* bias of cutting planes */ - b = (float64_t*)LIBOCAS_CALLOC(BufSize, float64_t); - if(b == NULL) - { - ocas.exitflag=-2; - goto cleanup; - } - - alpha = (float64_t*)LIBOCAS_CALLOC(BufSize, float64_t); - if(alpha == NULL) - { - ocas.exitflag=-2; - goto cleanup; - } - - /* indices of examples which define a new cut */ - new_cut = (uint32_t*)LIBOCAS_CALLOC(nData, uint32_t); - if(new_cut == NULL) - { - ocas.exitflag=-2; - goto cleanup; - } - - I = (uint32_t*)LIBOCAS_CALLOC(BufSize, uint32_t); - if(I == NULL) - { - ocas.exitflag=-2; - goto cleanup; - } - - for(i=0; i< BufSize; i++) I[i] = 2; - - diag_H = (float64_t*)LIBOCAS_CALLOC(BufSize, float64_t); - if(diag_H == NULL) - { - ocas.exitflag=-2; - goto cleanup; - } - - output = (float64_t*)LIBOCAS_CALLOC(nData, float64_t); - if(output == NULL) - { - ocas.exitflag=-2; - goto cleanup; - } - - old_output = (float64_t*)LIBOCAS_CALLOC(nData, float64_t); - if(old_output == NULL) - { - ocas.exitflag=-2; - goto cleanup; - } - - /* array of hinge points used in line-serach */ - hpf = (float64_t*) LIBOCAS_CALLOC(nData, float64_t); - if(hpf == NULL) - { - ocas.exitflag=-2; - goto cleanup; - } - - hpb = (float64_t*) LIBOCAS_CALLOC(nData, float64_t); - if(hpb == NULL) - { - ocas.exitflag=-2; - goto cleanup; - } - - /* vectors Ci, Bi are used in the line search procedure */ - Ci = (float64_t*)LIBOCAS_CALLOC(nData, float64_t); - if(Ci == NULL) - { - ocas.exitflag=-2; - goto cleanup; - } - - Bi = (float64_t*)LIBOCAS_CALLOC(nData, float64_t); - if(Bi == NULL) - { - ocas.exitflag=-2; - goto cleanup; - } - - /* initial cutting planes implementing the non-negativity constraints on W*/ - Cs[0]=10000000.0; - Cs[1]=C; - S[0]=1; - S[1]=1; - for(i=0; i < num_nnw; i++) - { - if(add_pw_constr(nnw_idx[i],i, user_data) != 0) - { - ocas.exitflag=-2; - goto cleanup; - } - diag_H[i] = 1.0; - H[LIBOCAS_INDEX(i,i,BufSize)] = 1.0; - I[i] = 1; - } - - max_cp_norm = 1; - max_b = 0; - - /* */ - ocas.nCutPlanes = num_nnw; - ocas.exitflag = 0; - ocas.nIter = 0; - - /* Compute initial value of Q_P assuming that W is zero vector.*/ - sq_norm_W = 0; - xi = nData; - ocas.Q_P = 0.5*sq_norm_W + C*xi; - ocas.Q_D = 0; - - /* Compute the initial cutting plane */ - cut_length = nData; - for(i=0; i < nData; i++) - new_cut[i] = i; - - ocas.trn_err = nData; - ocas.ocas_time = get_time() - ocas_start_time; - /* ocas_print("%4d: tim=%f, Q_P=%f, Q_D=%f, Q_P-Q_D=%f, Q_P-Q_D/abs(Q_P)=%f\n", - ocas.nIter,cur_time, ocas.Q_P,ocas.Q_D,ocas.Q_P-ocas.Q_D,(ocas.Q_P-ocas.Q_D)/LIBOCAS_ABS(ocas.Q_P)); - */ - ocas_print(ocas); - - /* main loop */ - while( ocas.exitflag == 0 ) - { - ocas.nIter++; - - /* append a new cut to the buffer and update H */ - b[ocas.nCutPlanes] = -(float64_t)cut_length; - - max_b = LIBOCAS_MAX(max_b,(float64_t)cut_length); - - start_time = get_time(); - - if(add_new_cut( &H[LIBOCAS_INDEX(0,ocas.nCutPlanes,BufSize)], new_cut, cut_length, ocas.nCutPlanes, user_data ) != 0) - { - ocas.exitflag=-2; - goto cleanup; - } - - ocas.add_time += get_time() - start_time; - - /* copy new added row: H(ocas.nCutPlanes,ocas.nCutPlanes,1:ocas.nCutPlanes-1) = H(1:ocas.nCutPlanes-1:ocas.nCutPlanes)' */ - diag_H[ocas.nCutPlanes] = H[LIBOCAS_INDEX(ocas.nCutPlanes,ocas.nCutPlanes,BufSize)]; - for(i=0; i < ocas.nCutPlanes; i++) { - H[LIBOCAS_INDEX(ocas.nCutPlanes,i,BufSize)] = H[LIBOCAS_INDEX(i,ocas.nCutPlanes,BufSize)]; - } - - max_cp_norm = LIBOCAS_MAX(max_cp_norm, sqrt(diag_H[ocas.nCutPlanes])); - - - ocas.nCutPlanes++; - - /* call inner QP solver */ - start_time = get_time(); - - /* compute upper bound on sum of dual variables associated with the positivity constraints */ - Cs[0] = sqrt((float64_t)nData)*(sqrt(C)*sqrt(max_b) + C*max_cp_norm); - -/* qp_exitflag = libqp_splx_solver(&get_col, diag_H, b, &C, I, &S, alpha, - ocas.nCutPlanes, QPSolverMaxIter, 0.0, QPSolverTolRel, -LIBOCAS_PLUS_INF,0);*/ - qp_exitflag = libqp_splx_solver(&get_col, diag_H, b, Cs, I, S, alpha, - ocas.nCutPlanes, QPSolverMaxIter, 0.0, QPSolverTolRel, -LIBOCAS_PLUS_INF,0); - - ocas.qp_exitflag = qp_exitflag.exitflag; - - ocas.qp_solver_time += get_time() - start_time; - ocas.Q_D = -qp_exitflag.QP; - - ocas.nNZAlpha = 0; - for(i=0; i < ocas.nCutPlanes; i++) { - if( alpha[i] != 0) ocas.nNZAlpha++; - } - - sq_norm_oldW = sq_norm_W; - start_time = get_time(); - compute_W( &sq_norm_W, &dot_prod_WoldW, alpha, ocas.nCutPlanes, user_data ); - clip_neg_W(num_nnw, nnw_idx, user_data); - ocas.w_time += get_time() - start_time; - - /* select a new cut */ - switch( Method ) - { - /* cutting plane algorithm implemented in SVMperf and BMRM */ - case 0: - - start_time = get_time(); - if( compute_output( output, user_data ) != 0) - { - ocas.exitflag=-2; - goto cleanup; - } - ocas.output_time += get_time()-start_time; - - xi = 0; - cut_length = 0; - ocas.trn_err = 0; - for(i=0; i < nData; i++) - { - if(output[i] <= 0) ocas.trn_err++; - - if(output[i] <= 1) { - xi += 1 - output[i]; - new_cut[cut_length] = i; - cut_length++; - } - } - ocas.Q_P = 0.5*sq_norm_W + C*xi; - - ocas.ocas_time = get_time() - ocas_start_time; - - /* ocas_print("%4d: tim=%f, Q_P=%f, Q_D=%f, Q_P-Q_D=%f, 1-Q_D/Q_P=%f, nza=%4d, err=%.2f%%, qpf=%d\n", - ocas.nIter,cur_time, ocas.Q_P,ocas.Q_D,ocas.Q_P-ocas.Q_D,(ocas.Q_P-ocas.Q_D)/LIBOCAS_ABS(ocas.Q_P), - ocas.nNZAlpha, 100*(float64_t)ocas.trn_err/(float64_t)nData, ocas.qp_exitflag ); - */ - - start_time = get_time(); - ocas_print(ocas); - ocas.print_time += get_time() - start_time; - - break; - - - /* Ocas strategy */ - case 1: - - /* Linesearch */ - A0 = sq_norm_W -2*dot_prod_WoldW + sq_norm_oldW; - B0 = dot_prod_WoldW - sq_norm_oldW; - - sg_memcpy( old_output, output, sizeof(float64_t)*nData ); - - start_time = get_time(); - if( compute_output( output, user_data ) != 0) - { - ocas.exitflag=-2; - goto cleanup; - } - ocas.output_time += get_time()-start_time; - - uint32_t num_hp = 0; - GradVal = B0; - for(i=0; i< nData; i++) { - - Ci[i] = C*(1-old_output[i]); - Bi[i] = C*(old_output[i] - output[i]); - - float64_t val; - if(Bi[i] != 0) - val = -Ci[i]/Bi[i]; - else - val = -LIBOCAS_PLUS_INF; - - if (val>0) - { -/* hpi[num_hp] = i;*/ - hpb[num_hp] = Bi[i]; - hpf[num_hp] = val; - num_hp++; - } - - if( (Bi[i] < 0 && val > 0) || (Bi[i] > 0 && val <= 0)) - GradVal += Bi[i]; - - } - - t = 0; - if( GradVal < 0 ) - { - start_time = get_time(); -/* if( sort(hpf, hpi, num_hp) != 0)*/ - if( sort(hpf, hpb, num_hp) != 0 ) - { - ocas.exitflag=-2; - goto cleanup; - } - ocas.sort_time += get_time() - start_time; - - float64_t t_new, GradVal_new; - i = 0; - while( GradVal < 0 && i < num_hp ) - { - t_new = hpf[i]; - GradVal_new = GradVal + LIBOCAS_ABS(hpb[i]) + A0*(t_new-t); - - if( GradVal_new >= 0 ) - { - t = t + GradVal*(t-t_new)/(GradVal_new - GradVal); - } - else - { - t = t_new; - i++; - } - - GradVal = GradVal_new; - } - } - - /* - t = hpf[0] - 1; - i = 0; - GradVal = t*A0 + Bsum; - while( GradVal < 0 && i < num_hp && hpf[i] < LIBOCAS_PLUS_INF ) { - t = hpf[i]; - Bsum = Bsum + LIBOCAS_ABS(Bi[hpi[i]]); - GradVal = t*A0 + Bsum; - i++; - } - */ - t = LIBOCAS_MAX(t,0); /* just sanity check; t < 0 should not ocure */ - - /* this guarantees that the new solution will not violate the positivity constraints on W */ - t = LIBOCAS_MIN(t,1); - - t1 = t; /* new (best so far) W */ - t2 = t+MU*(1.0-t); /* new cutting plane */ - /* t2 = t+(1.0-t)/10.0; */ - - /* update W to be the best so far solution */ - sq_norm_W = update_W( t1, user_data ); - - /* select a new cut */ - xi = 0; - cut_length = 0; - ocas.trn_err = 0; - for(i=0; i < nData; i++ ) { - - if( (old_output[i]*(1-t2) + t2*output[i]) <= 1 ) - { - new_cut[cut_length] = i; - cut_length++; - } - - output[i] = old_output[i]*(1-t1) + t1*output[i]; - - if( output[i] <= 1) xi += 1-output[i]; - if( output[i] <= 0) ocas.trn_err++; - - } - - ocas.Q_P = 0.5*sq_norm_W + C*xi; - - ocas.ocas_time = get_time() - ocas_start_time; - - /* ocas_print("%4d: tim=%f, Q_P=%f, Q_D=%f, Q_P-Q_D=%f, 1-Q_D/Q_P=%f, nza=%4d, err=%.2f%%, qpf=%d\n", - ocas.nIter, cur_time, ocas.Q_P,ocas.Q_D,ocas.Q_P-ocas.Q_D,(ocas.Q_P-ocas.Q_D)/LIBOCAS_ABS(ocas.Q_P), - ocas.nNZAlpha, 100*(float64_t)ocas.trn_err/(float64_t)nData, ocas.qp_exitflag ); - */ - - start_time = get_time(); - ocas_print(ocas); - ocas.print_time += get_time() - start_time; - - break; - } - - /* Stopping conditions */ - if( ocas.Q_P - ocas.Q_D <= TolRel*LIBOCAS_ABS(ocas.Q_P)) ocas.exitflag = 1; - if( ocas.Q_P - ocas.Q_D <= TolAbs) ocas.exitflag = 2; - if( ocas.Q_P <= QPBound) ocas.exitflag = 3; - if( MaxTime > 0 && ocas.ocas_time >= MaxTime) ocas.exitflag = 4; - if(ocas.nCutPlanes >= BufSize) ocas.exitflag = -1; - - } /* end of the main loop */ - -cleanup: - - LIBOCAS_FREE(H); - LIBOCAS_FREE(b); - LIBOCAS_FREE(alpha); - LIBOCAS_FREE(new_cut); - LIBOCAS_FREE(I); - LIBOCAS_FREE(diag_H); - LIBOCAS_FREE(output); - LIBOCAS_FREE(old_output); - LIBOCAS_FREE(hpf); -/* LIBOCAS_FREE(hpi);*/ - LIBOCAS_FREE(hpb); - LIBOCAS_FREE(Ci); - LIBOCAS_FREE(Bi); - - ocas.ocas_time = get_time() - ocas_start_time; - - return(ocas); -} - - - -/*---------------------------------------------------------------------- - Linear binary Ocas-SVM solver. - ----------------------------------------------------------------------*/ -ocas_return_value_T svm_ocas_solver( - float64_t C, - uint32_t nData, - float64_t TolRel, - float64_t TolAbs, - float64_t QPBound, - float64_t MaxTime, - uint32_t _BufSize, - uint8_t Method, - void (*compute_W)(float64_t*, float64_t*, float64_t*, uint32_t, void*), - float64_t (*update_W)(float64_t, void*), - int (*add_new_cut)(float64_t*, uint32_t*, uint32_t, uint32_t, void*), - int (*compute_output)(float64_t*, void* ), - int (*sort)(float64_t*, float64_t*, uint32_t), - void (*ocas_print)(ocas_return_value_T), - void* user_data) -{ - ocas_return_value_T ocas={0,0,0,0,0,0,0,0,0,0,0,0,0,0,0}; - float64_t *b, *alpha, *diag_H; - float64_t *output, *old_output; - float64_t xi, sq_norm_W, QPSolverTolRel, dot_prod_WoldW, sq_norm_oldW; - float64_t A0, B0, GradVal, t, t1, t2, *Ci, *Bi, *hpf, *hpb; - float64_t start_time, ocas_start_time; - uint32_t cut_length; - uint32_t i, *new_cut; - uint32_t *I; - uint8_t S = 1; - libqp_state_T qp_exitflag; - - ocas_start_time = get_time(); - ocas.qp_solver_time = 0; - ocas.output_time = 0; - ocas.sort_time = 0; - ocas.add_time = 0; - ocas.w_time = 0; - ocas.print_time = 0; - float64_t gap; - - BufSize = _BufSize; - - QPSolverTolRel = TolRel*0.5; - - H=NULL; - b=NULL; - alpha=NULL; - new_cut=NULL; - I=NULL; - diag_H=NULL; - output=NULL; - old_output=NULL; - hpf=NULL; - hpb = NULL; - Ci=NULL; - Bi=NULL; - - /* Hessian matrix contains dot product of normal vectors of selected cutting planes */ - H = (float64_t*)LIBOCAS_CALLOC(BufSize*BufSize, float64_t); - if(H == NULL) - { - ocas.exitflag=-2; - goto cleanup; - } - - /* bias of cutting planes */ - b = (float64_t*)LIBOCAS_CALLOC(BufSize, float64_t); - if(b == NULL) - { - ocas.exitflag=-2; - goto cleanup; - } - - alpha = (float64_t*)LIBOCAS_CALLOC(BufSize, float64_t); - if(alpha == NULL) - { - ocas.exitflag=-2; - goto cleanup; - } - - /* indices of examples which define a new cut */ - new_cut = (uint32_t*)LIBOCAS_CALLOC(nData, uint32_t); - if(new_cut == NULL) - { - ocas.exitflag=-2; - goto cleanup; - } - - I = (uint32_t*)LIBOCAS_CALLOC(BufSize, uint32_t); - if(I == NULL) - { - ocas.exitflag=-2; - goto cleanup; - } - - for(i=0; i< BufSize; i++) I[i] = 1; - - diag_H = (float64_t*)LIBOCAS_CALLOC(BufSize, float64_t); - if(diag_H == NULL) - { - ocas.exitflag=-2; - goto cleanup; - } - - output = (float64_t*)LIBOCAS_CALLOC(nData, float64_t); - if(output == NULL) - { - ocas.exitflag=-2; - goto cleanup; - } - - old_output = (float64_t*)LIBOCAS_CALLOC(nData, float64_t); - if(old_output == NULL) - { - ocas.exitflag=-2; - goto cleanup; - } - - /* array of hinge points used in line-serach */ - hpf = (float64_t*) LIBOCAS_CALLOC(nData, float64_t); - if(hpf == NULL) - { - ocas.exitflag=-2; - goto cleanup; - } - - hpb = (float64_t*) LIBOCAS_CALLOC(nData, float64_t); - if(hpb == NULL) - { - ocas.exitflag=-2; - goto cleanup; - } - - /* vectors Ci, Bi are used in the line search procedure */ - Ci = (float64_t*)LIBOCAS_CALLOC(nData, float64_t); - if(Ci == NULL) - { - ocas.exitflag=-2; - goto cleanup; - } - - Bi = (float64_t*)LIBOCAS_CALLOC(nData, float64_t); - if(Bi == NULL) - { - ocas.exitflag=-2; - goto cleanup; - } - - ocas.nCutPlanes = 0; - ocas.exitflag = 0; - ocas.nIter = 0; - - /* Compute initial value of Q_P assuming that W is zero vector.*/ - sq_norm_W = 0; - xi = nData; - ocas.Q_P = 0.5*sq_norm_W + C*xi; - ocas.Q_D = 0; - - /* Compute the initial cutting plane */ - cut_length = nData; - for(i=0; i < nData; i++) - new_cut[i] = i; - - gap=(ocas.Q_P-ocas.Q_D)/CMath::abs(ocas.Q_P); - SG_SABS_PROGRESS(gap, -CMath::log10(gap), -CMath::log10(1), -CMath::log10(TolRel), 6) - - ocas.trn_err = nData; - ocas.ocas_time = get_time() - ocas_start_time; - /* ocas_print("%4d: tim=%f, Q_P=%f, Q_D=%f, Q_P-Q_D=%f, Q_P-Q_D/abs(Q_P)=%f\n", - ocas.nIter,cur_time, ocas.Q_P,ocas.Q_D,ocas.Q_P-ocas.Q_D,(ocas.Q_P-ocas.Q_D)/LIBOCAS_ABS(ocas.Q_P)); - */ - ocas_print(ocas); - - /* main loop */ - while( ocas.exitflag == 0 ) - { - ocas.nIter++; - - /* append a new cut to the buffer and update H */ - b[ocas.nCutPlanes] = -(float64_t)cut_length; - - start_time = get_time(); - - if(add_new_cut( &H[LIBOCAS_INDEX(0,ocas.nCutPlanes,BufSize)], new_cut, cut_length, ocas.nCutPlanes, user_data ) != 0) - { - ocas.exitflag=-2; - goto cleanup; - } - - ocas.add_time += get_time() - start_time; - - /* copy new added row: H(ocas.nCutPlanes,ocas.nCutPlanes,1:ocas.nCutPlanes-1) = H(1:ocas.nCutPlanes-1:ocas.nCutPlanes)' */ - diag_H[ocas.nCutPlanes] = H[LIBOCAS_INDEX(ocas.nCutPlanes,ocas.nCutPlanes,BufSize)]; - for(i=0; i < ocas.nCutPlanes; i++) { - H[LIBOCAS_INDEX(ocas.nCutPlanes,i,BufSize)] = H[LIBOCAS_INDEX(i,ocas.nCutPlanes,BufSize)]; - } - - ocas.nCutPlanes++; - - /* call inner QP solver */ - start_time = get_time(); - - qp_exitflag = libqp_splx_solver(&get_col, diag_H, b, &C, I, &S, alpha, - ocas.nCutPlanes, QPSolverMaxIter, 0.0, QPSolverTolRel, -LIBOCAS_PLUS_INF,0); - - ocas.qp_exitflag = qp_exitflag.exitflag; - - ocas.qp_solver_time += get_time() - start_time; - ocas.Q_D = -qp_exitflag.QP; - - ocas.nNZAlpha = 0; - for(i=0; i < ocas.nCutPlanes; i++) { - if( alpha[i] != 0) ocas.nNZAlpha++; - } - - sq_norm_oldW = sq_norm_W; - start_time = get_time(); - compute_W( &sq_norm_W, &dot_prod_WoldW, alpha, ocas.nCutPlanes, user_data ); - ocas.w_time += get_time() - start_time; - - /* select a new cut */ - switch( Method ) - { - /* cutting plane algorithm implemented in SVMperf and BMRM */ - case 0: - - start_time = get_time(); - if( compute_output( output, user_data ) != 0) - { - ocas.exitflag=-2; - goto cleanup; - } - ocas.output_time += get_time()-start_time; - gap=(ocas.Q_P-ocas.Q_D)/CMath::abs(ocas.Q_P); - SG_SABS_PROGRESS(gap, -CMath::log10(gap), -CMath::log10(1), -CMath::log10(TolRel), 6) - - xi = 0; - cut_length = 0; - ocas.trn_err = 0; - for(i=0; i < nData; i++) - { - if(output[i] <= 0) ocas.trn_err++; - - if(output[i] <= 1) { - xi += 1 - output[i]; - new_cut[cut_length] = i; - cut_length++; - } - } - ocas.Q_P = 0.5*sq_norm_W + C*xi; - - ocas.ocas_time = get_time() - ocas_start_time; - - /* ocas_print("%4d: tim=%f, Q_P=%f, Q_D=%f, Q_P-Q_D=%f, 1-Q_D/Q_P=%f, nza=%4d, err=%.2f%%, qpf=%d\n", - ocas.nIter,cur_time, ocas.Q_P,ocas.Q_D,ocas.Q_P-ocas.Q_D,(ocas.Q_P-ocas.Q_D)/LIBOCAS_ABS(ocas.Q_P), - ocas.nNZAlpha, 100*(float64_t)ocas.trn_err/(float64_t)nData, ocas.qp_exitflag ); - */ - - start_time = get_time(); - ocas_print(ocas); - ocas.print_time += get_time() - start_time; - - break; - - - /* Ocas strategy */ - case 1: - - /* Linesearch */ - A0 = sq_norm_W -2*dot_prod_WoldW + sq_norm_oldW; - B0 = dot_prod_WoldW - sq_norm_oldW; - - sg_memcpy( old_output, output, sizeof(float64_t)*nData ); - - start_time = get_time(); - if( compute_output( output, user_data ) != 0) - { - ocas.exitflag=-2; - goto cleanup; - } - ocas.output_time += get_time()-start_time; - - uint32_t num_hp = 0; - GradVal = B0; - for(i=0; i< nData; i++) { - - Ci[i] = C*(1-old_output[i]); - Bi[i] = C*(old_output[i] - output[i]); - - float64_t val; - if(Bi[i] != 0) - val = -Ci[i]/Bi[i]; - else - val = -LIBOCAS_PLUS_INF; - - if (val>0) - { -/* hpi[num_hp] = i;*/ - hpb[num_hp] = Bi[i]; - hpf[num_hp] = val; - num_hp++; - } - - if( (Bi[i] < 0 && val > 0) || (Bi[i] > 0 && val <= 0)) - GradVal += Bi[i]; - - } - - t = 0; - if( GradVal < 0 ) - { - start_time = get_time(); -/* if( sort(hpf, hpi, num_hp) != 0)*/ - if( sort(hpf, hpb, num_hp) != 0 ) - { - ocas.exitflag=-2; - goto cleanup; - } - ocas.sort_time += get_time() - start_time; - - float64_t t_new, GradVal_new; - i = 0; - while( GradVal < 0 && i < num_hp ) - { - t_new = hpf[i]; - GradVal_new = GradVal + LIBOCAS_ABS(hpb[i]) + A0*(t_new-t); - - if( GradVal_new >= 0 ) - { - t = t + GradVal*(t-t_new)/(GradVal_new - GradVal); - } - else - { - t = t_new; - i++; - } - - GradVal = GradVal_new; - } - } - - /* - t = hpf[0] - 1; - i = 0; - GradVal = t*A0 + Bsum; - while( GradVal < 0 && i < num_hp && hpf[i] < LIBOCAS_PLUS_INF ) { - t = hpf[i]; - Bsum = Bsum + LIBOCAS_ABS(Bi[hpi[i]]); - GradVal = t*A0 + Bsum; - i++; - } - */ - t = LIBOCAS_MAX(t,0); /* just sanity check; t < 0 should not ocure */ - - t1 = t; /* new (best so far) W */ - t2 = t+MU*(1.0-t); /* new cutting plane */ - /* t2 = t+(1.0-t)/10.0; */ - - /* update W to be the best so far solution */ - sq_norm_W = update_W( t1, user_data ); - - /* select a new cut */ - xi = 0; - cut_length = 0; - ocas.trn_err = 0; - for(i=0; i < nData; i++ ) { - - if( (old_output[i]*(1-t2) + t2*output[i]) <= 1 ) - { - new_cut[cut_length] = i; - cut_length++; - } - - output[i] = old_output[i]*(1-t1) + t1*output[i]; - - if( output[i] <= 1) xi += 1-output[i]; - if( output[i] <= 0) ocas.trn_err++; - - } - - ocas.Q_P = 0.5*sq_norm_W + C*xi; - - ocas.ocas_time = get_time() - ocas_start_time; - - /* ocas_print("%4d: tim=%f, Q_P=%f, Q_D=%f, Q_P-Q_D=%f, 1-Q_D/Q_P=%f, nza=%4d, err=%.2f%%, qpf=%d\n", - ocas.nIter, cur_time, ocas.Q_P,ocas.Q_D,ocas.Q_P-ocas.Q_D,(ocas.Q_P-ocas.Q_D)/LIBOCAS_ABS(ocas.Q_P), - ocas.nNZAlpha, 100*(float64_t)ocas.trn_err/(float64_t)nData, ocas.qp_exitflag ); - */ - - start_time = get_time(); - ocas_print(ocas); - ocas.print_time += get_time() - start_time; - - break; - } - - /* Stopping conditions */ - if( ocas.Q_P - ocas.Q_D <= TolRel*LIBOCAS_ABS(ocas.Q_P)) ocas.exitflag = 1; - if( ocas.Q_P - ocas.Q_D <= TolAbs) ocas.exitflag = 2; - if( ocas.Q_P <= QPBound) ocas.exitflag = 3; - if( MaxTime > 0 && ocas.ocas_time >= MaxTime) ocas.exitflag = 4; - if(ocas.nCutPlanes >= BufSize) ocas.exitflag = -1; - - } /* end of the main loop */ - -cleanup: - - LIBOCAS_FREE(H); - LIBOCAS_FREE(b); - LIBOCAS_FREE(alpha); - LIBOCAS_FREE(new_cut); - LIBOCAS_FREE(I); - LIBOCAS_FREE(diag_H); - LIBOCAS_FREE(output); - LIBOCAS_FREE(old_output); - LIBOCAS_FREE(hpf); -/* LIBOCAS_FREE(hpi);*/ - LIBOCAS_FREE(hpb); - LIBOCAS_FREE(Ci); - LIBOCAS_FREE(Bi); - - ocas.ocas_time = get_time() - ocas_start_time; - - return(ocas); -} - - -/*---------------------------------------------------------------------- - Binary linear Ocas-SVM solver which allows using different C for each - training example. - ----------------------------------------------------------------------*/ -ocas_return_value_T svm_ocas_solver_difC( - float64_t *C, - uint32_t nData, - float64_t TolRel, - float64_t TolAbs, - float64_t QPBound, - float64_t MaxTime, - uint32_t _BufSize, - uint8_t Method, - void (*compute_W)(float64_t*, float64_t*, float64_t*, uint32_t, void*), - float64_t (*update_W)(float64_t, void*), - int (*add_new_cut)(float64_t*, uint32_t*, uint32_t, uint32_t, void*), - int (*compute_output)(float64_t*, void* ), - int (*sort)(float64_t*, float64_t*, uint32_t), - void (*ocas_print)(ocas_return_value_T), - void* user_data) -{ - ocas_return_value_T ocas={0,0,0,0,0,0,0,0,0,0,0,0,0,0,0}; - float64_t *b, *alpha, *diag_H; - float64_t *output, *old_output; - float64_t xi, sq_norm_W, QPSolverTolRel, dot_prod_WoldW, sq_norm_oldW; - float64_t A0, B0, GradVal, t, t1, t2, *Ci, *Bi, *hpf, *hpb; - float64_t start_time, ocas_start_time; - float64_t qp_b = 1.0; - float64_t new_b; - uint32_t cut_length; - uint32_t i, *new_cut; - uint32_t *I; - uint8_t S = 1; - libqp_state_T qp_exitflag; - - ocas_start_time = get_time(); - ocas.qp_solver_time = 0; - ocas.output_time = 0; - ocas.sort_time = 0; - ocas.add_time = 0; - ocas.w_time = 0; - ocas.print_time = 0; - - BufSize = _BufSize; - - QPSolverTolRel = TolRel*0.5; - - H=NULL; - b=NULL; - alpha=NULL; - new_cut=NULL; - I=NULL; - diag_H=NULL; - output=NULL; - old_output=NULL; - hpf=NULL; - hpb = NULL; - Ci=NULL; - Bi=NULL; - - /* Hessian matrix contains dot product of normal vectors of selected cutting planes */ - H = (float64_t*)LIBOCAS_CALLOC(BufSize*BufSize, float64_t); - if(H == NULL) - { - ocas.exitflag=-2; - goto cleanup; - } - - /* bias of cutting planes */ - b = (float64_t*)LIBOCAS_CALLOC(BufSize, float64_t); - if(b == NULL) - { - ocas.exitflag=-2; - goto cleanup; - } - - alpha = (float64_t*)LIBOCAS_CALLOC(BufSize, float64_t); - if(alpha == NULL) - { - ocas.exitflag=-2; - goto cleanup; - } - - /* indices of examples which define a new cut */ - new_cut = (uint32_t*)LIBOCAS_CALLOC(nData, uint32_t); - if(new_cut == NULL) - { - ocas.exitflag=-2; - goto cleanup; - } - - I = (uint32_t*)LIBOCAS_CALLOC(BufSize, uint32_t); - if(I == NULL) - { - ocas.exitflag=-2; - goto cleanup; - } - - for(i=0; i< BufSize; i++) I[i] = 1; - - diag_H = (float64_t*)LIBOCAS_CALLOC(BufSize, float64_t); - if(diag_H == NULL) - { - ocas.exitflag=-2; - goto cleanup; - } - - output = (float64_t*)LIBOCAS_CALLOC(nData, float64_t); - if(output == NULL) - { - ocas.exitflag=-2; - goto cleanup; - } - - old_output = (float64_t*)LIBOCAS_CALLOC(nData, float64_t); - if(old_output == NULL) - { - ocas.exitflag=-2; - goto cleanup; - } - - /* array of hinge points used in line-serach */ - hpf = (float64_t*) LIBOCAS_CALLOC(nData, float64_t); - if(hpf == NULL) - { - ocas.exitflag=-2; - goto cleanup; - } - - hpb = (float64_t*) LIBOCAS_CALLOC(nData, float64_t); - if(hpb == NULL) - { - ocas.exitflag=-2; - goto cleanup; - } - - /* vectors Ci, Bi are used in the line search procedure */ - Ci = (float64_t*)LIBOCAS_CALLOC(nData, float64_t); - if(Ci == NULL) - { - ocas.exitflag=-2; - goto cleanup; - } - - Bi = (float64_t*)LIBOCAS_CALLOC(nData, float64_t); - if(Bi == NULL) - { - ocas.exitflag=-2; - goto cleanup; - } - - ocas.nCutPlanes = 0; - ocas.exitflag = 0; - ocas.nIter = 0; - - /* Compute initial value of Q_P assuming that W is zero vector.*/ - sq_norm_W = 0; -/* - xi = nData; - ocas.Q_P = 0.5*sq_norm_W + C*xi; -*/ - ocas.Q_D = 0; - - /* Compute the initial cutting plane */ - cut_length = nData; - new_b = 0; - for(i=0; i < nData; i++) - { - new_cut[i] = i; - new_b += C[i]; - } - - ocas.Q_P = 0.5*sq_norm_W + new_b; - - - ocas.trn_err = nData; - ocas.ocas_time = get_time() - ocas_start_time; - /* ocas_print("%4d: tim=%f, Q_P=%f, Q_D=%f, Q_P-Q_D=%f, Q_P-Q_D/abs(Q_P)=%f\n", - ocas.nIter,cur_time, ocas.Q_P,ocas.Q_D,ocas.Q_P-ocas.Q_D,(ocas.Q_P-ocas.Q_D)/LIBOCAS_ABS(ocas.Q_P)); - */ - ocas_print(ocas); - - /* main loop */ - while( ocas.exitflag == 0 ) - { - ocas.nIter++; - - /* append a new cut to the buffer and update H */ -/* b[ocas.nCutPlanes] = -(float64_t)cut_length*C;*/ - b[ocas.nCutPlanes] = -new_b; - - start_time = get_time(); - - if(add_new_cut( &H[LIBOCAS_INDEX(0,ocas.nCutPlanes,BufSize)], new_cut, cut_length, ocas.nCutPlanes, user_data ) != 0) - { - ocas.exitflag=-2; - goto cleanup; - } - - ocas.add_time += get_time() - start_time; - - /* copy new added row: H(ocas.nCutPlanes,ocas.nCutPlanes,1:ocas.nCutPlanes-1) = H(1:ocas.nCutPlanes-1:ocas.nCutPlanes)' */ - diag_H[ocas.nCutPlanes] = H[LIBOCAS_INDEX(ocas.nCutPlanes,ocas.nCutPlanes,BufSize)]; - for(i=0; i < ocas.nCutPlanes; i++) { - H[LIBOCAS_INDEX(ocas.nCutPlanes,i,BufSize)] = H[LIBOCAS_INDEX(i,ocas.nCutPlanes,BufSize)]; - } - - ocas.nCutPlanes++; - - /* call inner QP solver */ - start_time = get_time(); - -/* qp_exitflag = libqp_splx_solver(&get_col, diag_H, b, &C, I, &S, alpha,*/ -/* ocas.nCutPlanes, QPSolverMaxIter, 0.0, QPSolverTolRel, -LIBOCAS_PLUS_INF,0);*/ - qp_exitflag = libqp_splx_solver(&get_col, diag_H, b, &qp_b, I, &S, alpha, - ocas.nCutPlanes, QPSolverMaxIter, 0.0, QPSolverTolRel, -LIBOCAS_PLUS_INF,0); - - ocas.qp_exitflag = qp_exitflag.exitflag; - - ocas.qp_solver_time += get_time() - start_time; - ocas.Q_D = -qp_exitflag.QP; - - ocas.nNZAlpha = 0; - for(i=0; i < ocas.nCutPlanes; i++) { - if( alpha[i] != 0) ocas.nNZAlpha++; - } - - sq_norm_oldW = sq_norm_W; - start_time = get_time(); - compute_W( &sq_norm_W, &dot_prod_WoldW, alpha, ocas.nCutPlanes, user_data ); - ocas.w_time += get_time() - start_time; - - /* select a new cut */ - switch( Method ) - { - /* cutting plane algorithm implemented in SVMperf and BMRM */ - case 0: - - start_time = get_time(); - if( compute_output( output, user_data ) != 0) - { - ocas.exitflag=-2; - goto cleanup; - } - ocas.output_time += get_time()-start_time; - - xi = 0; - cut_length = 0; - ocas.trn_err = 0; - new_b = 0; - for(i=0; i < nData; i++) - { - if(output[i] <= 0) ocas.trn_err++; - -/* if(output[i] <= 1) {*/ -/* xi += 1 - output[i];*/ - if(output[i] <= C[i]) { - xi += C[i] - output[i]; - new_cut[cut_length] = i; - cut_length++; - new_b += C[i]; - } - } -/* ocas.Q_P = 0.5*sq_norm_W + C*xi;*/ - ocas.Q_P = 0.5*sq_norm_W + xi; - - ocas.ocas_time = get_time() - ocas_start_time; - - /* ocas_print("%4d: tim=%f, Q_P=%f, Q_D=%f, Q_P-Q_D=%f, 1-Q_D/Q_P=%f, nza=%4d, err=%.2f%%, qpf=%d\n", - ocas.nIter,cur_time, ocas.Q_P,ocas.Q_D,ocas.Q_P-ocas.Q_D,(ocas.Q_P-ocas.Q_D)/LIBOCAS_ABS(ocas.Q_P), - ocas.nNZAlpha, 100*(float64_t)ocas.trn_err/(float64_t)nData, ocas.qp_exitflag ); - */ - - start_time = get_time(); - ocas_print(ocas); - ocas.print_time += get_time() - start_time; - - break; - - - /* Ocas strategy */ - case 1: - - /* Linesearch */ - A0 = sq_norm_W -2*dot_prod_WoldW + sq_norm_oldW; - B0 = dot_prod_WoldW - sq_norm_oldW; - - sg_memcpy( old_output, output, sizeof(float64_t)*nData ); - - start_time = get_time(); - if( compute_output( output, user_data ) != 0) - { - ocas.exitflag=-2; - goto cleanup; - } - ocas.output_time += get_time()-start_time; - - uint32_t num_hp = 0; - GradVal = B0; - for(i=0; i< nData; i++) { - -/* Ci[i] = C*(1-old_output[i]);*/ -/* Bi[i] = C*(old_output[i] - output[i]);*/ - Ci[i] = (C[i]-old_output[i]); - Bi[i] = old_output[i] - output[i]; - - float64_t val; - if(Bi[i] != 0) - val = -Ci[i]/Bi[i]; - else - val = -LIBOCAS_PLUS_INF; - - if (val>0) - { -/* hpi[num_hp] = i;*/ - hpb[num_hp] = Bi[i]; - hpf[num_hp] = val; - num_hp++; - } - - if( (Bi[i] < 0 && val > 0) || (Bi[i] > 0 && val <= 0)) - GradVal += Bi[i]; - - } - - t = 0; - if( GradVal < 0 ) - { - start_time = get_time(); -/* if( sort(hpf, hpi, num_hp) != 0)*/ - if( sort(hpf, hpb, num_hp) != 0 ) - { - ocas.exitflag=-2; - goto cleanup; - } - ocas.sort_time += get_time() - start_time; - - float64_t t_new, GradVal_new; - i = 0; - while( GradVal < 0 && i < num_hp ) - { - t_new = hpf[i]; - GradVal_new = GradVal + LIBOCAS_ABS(hpb[i]) + A0*(t_new-t); - - if( GradVal_new >= 0 ) - { - t = t + GradVal*(t-t_new)/(GradVal_new - GradVal); - } - else - { - t = t_new; - i++; - } - - GradVal = GradVal_new; - } - } - - /* - t = hpf[0] - 1; - i = 0; - GradVal = t*A0 + Bsum; - while( GradVal < 0 && i < num_hp && hpf[i] < LIBOCAS_PLUS_INF ) { - t = hpf[i]; - Bsum = Bsum + LIBOCAS_ABS(Bi[hpi[i]]); - GradVal = t*A0 + Bsum; - i++; - } - */ - t = LIBOCAS_MAX(t,0); /* just sanity check; t < 0 should not ocure */ - - t1 = t; /* new (best so far) W */ - t2 = t+(1.0-t)*MU; /* new cutting plane */ - /* t2 = t+(1.0-t)/10.0; new cutting plane */ - - /* update W to be the best so far solution */ - sq_norm_W = update_W( t1, user_data ); - - /* select a new cut */ - xi = 0; - cut_length = 0; - ocas.trn_err = 0; - new_b = 0; - for(i=0; i < nData; i++ ) { - -/* if( (old_output[i]*(1-t2) + t2*output[i]) <= 1 ) */ - if( (old_output[i]*(1-t2) + t2*output[i]) <= C[i] ) - { - new_cut[cut_length] = i; - cut_length++; - new_b += C[i]; - } - - output[i] = old_output[i]*(1-t1) + t1*output[i]; - -/* if( output[i] <= 1) xi += 1-output[i];*/ - if( output[i] <= C[i]) xi += C[i]-output[i]; - if( output[i] <= 0) ocas.trn_err++; - - } - -/* ocas.Q_P = 0.5*sq_norm_W + C*xi;*/ - ocas.Q_P = 0.5*sq_norm_W + xi; - - ocas.ocas_time = get_time() - ocas_start_time; - - /* ocas_print("%4d: tim=%f, Q_P=%f, Q_D=%f, Q_P-Q_D=%f, 1-Q_D/Q_P=%f, nza=%4d, err=%.2f%%, qpf=%d\n", - ocas.nIter, cur_time, ocas.Q_P,ocas.Q_D,ocas.Q_P-ocas.Q_D,(ocas.Q_P-ocas.Q_D)/LIBOCAS_ABS(ocas.Q_P), - ocas.nNZAlpha, 100*(float64_t)ocas.trn_err/(float64_t)nData, ocas.qp_exitflag ); - */ - - start_time = get_time(); - ocas_print(ocas); - ocas.print_time += get_time() - start_time; - - break; - } - - /* Stopping conditions */ - if( ocas.Q_P - ocas.Q_D <= TolRel*LIBOCAS_ABS(ocas.Q_P)) ocas.exitflag = 1; - if( ocas.Q_P - ocas.Q_D <= TolAbs) ocas.exitflag = 2; - if( ocas.Q_P <= QPBound) ocas.exitflag = 3; - if( MaxTime > 0 && ocas.ocas_time >= MaxTime) ocas.exitflag = 4; - if(ocas.nCutPlanes >= BufSize) ocas.exitflag = -1; - - } /* end of the main loop */ - -cleanup: - - LIBOCAS_FREE(H); - LIBOCAS_FREE(b); - LIBOCAS_FREE(alpha); - LIBOCAS_FREE(new_cut); - LIBOCAS_FREE(I); - LIBOCAS_FREE(diag_H); - LIBOCAS_FREE(output); - LIBOCAS_FREE(old_output); - LIBOCAS_FREE(hpf); -/* LIBOCAS_FREE(hpi);*/ - LIBOCAS_FREE(hpb); - LIBOCAS_FREE(Ci); - LIBOCAS_FREE(Bi); - - ocas.ocas_time = get_time() - ocas_start_time; - - return(ocas); -} - - - -/*---------------------------------------------------------------------- - Multiclass SVM-Ocas solver - ----------------------------------------------------------------------*/ - -/* Helper function needed by the multi-class SVM linesearch. - - - This function finds a simplified representation of a piece-wise linear function - by splitting the domain into intervals and fining active terms for these intevals */ -static void findactive(float64_t *Theta, float64_t *SortedA, uint32_t *nSortedA, float64_t *A, float64_t *B, int n, - int (*sort)(float64_t*, float64_t*, uint32_t)) -{ - float64_t tmp, theta; - uint32_t i, j, idx, idx2 = 0, start; - - sort(A,B,n); - - idx = 0; - i = 0; - while( i < (uint32_t)n-1 && A[i] == A[i+1]) - { - if( B[i+1] > B[idx] ) - { - idx = i+1; - } - i++; - } - - (*nSortedA) = 1; - SortedA[0] = A[idx]; - - while(1) - { - start = idx + 1; - while( start < (uint32_t)n && A[idx] == A[start]) - start++; - - theta = LIBOCAS_PLUS_INF; - for(j=start; j < (uint32_t)n; j++) - { - tmp = (B[j] - B[idx])/(A[idx]-A[j]); - if( tmp < theta) - { - theta = tmp; - idx2 = j; - } - } - - if( theta < LIBOCAS_PLUS_INF) - { - Theta[(*nSortedA) - 1] = theta; - SortedA[(*nSortedA)] = A[idx2]; - (*nSortedA)++; - idx = idx2; - } - else - return; - } -} - - -/*---------------------------------------------------------------------- - Multiclass linear OCAS-SVM solver. - ----------------------------------------------------------------------*/ -ocas_return_value_T msvm_ocas_solver( - float64_t C, - float64_t *data_y, - uint32_t nY, - uint32_t nData, - float64_t TolRel, - float64_t TolAbs, - float64_t QPBound, - float64_t MaxTime, - uint32_t _BufSize, - uint8_t Method, - void (*compute_W)(float64_t*, float64_t*, float64_t*, uint32_t, void*), - float64_t (*update_W)(float64_t, void*), - int (*add_new_cut)(float64_t*, uint32_t*, uint32_t, void*), - int (*compute_output)(float64_t*, void* ), - int (*sort)(float64_t*, float64_t*, uint32_t), - void (*ocas_print)(ocas_return_value_T), - void* user_data) -{ - ocas_return_value_T ocas={0,0,0,0,0,0,0,0,0,0,0,0,0,0,0}; - float64_t *b, *alpha, *diag_H; - float64_t *output, *old_output; - float64_t xi, sq_norm_W, QPSolverTolRel, QPSolverTolAbs, dot_prod_WoldW, sq_norm_oldW; - float64_t A0, B0, t, t1, t2, R, tmp, element_b, x; - float64_t *A, *B, *theta, *Theta, *sortedA, *Add; - float64_t start_time, ocas_start_time, grad_sum, grad, min_x = 0, old_x, old_grad; - uint32_t i, y, y2, ypred = 0, *new_cut, cnt1, cnt2, j, nSortedA, idx; - uint32_t *I; - uint8_t S = 1; - libqp_state_T qp_exitflag; - - ocas_start_time = get_time(); - ocas.qp_solver_time = 0; - ocas.output_time = 0; - ocas.sort_time = 0; - ocas.add_time = 0; - ocas.w_time = 0; - ocas.print_time = 0; - - BufSize = _BufSize; - - QPSolverTolRel = TolRel*0.5; - QPSolverTolAbs = TolAbs*0.5; - - H=NULL; - b=NULL; - alpha=NULL; - new_cut=NULL; - I=NULL; - diag_H=NULL; - output=NULL; - old_output=NULL; - A = NULL; - B = NULL; - theta = NULL; - Theta = NULL; - sortedA = NULL; - Add = NULL; - - /* Hessian matrix contains dot product of normal vectors of selected cutting planes */ - H = (float64_t*)LIBOCAS_CALLOC(BufSize*BufSize, float64_t); - if(H == NULL) - { - ocas.exitflag=-2; - goto cleanup; - } - - /* bias of cutting planes */ - b = (float64_t*)LIBOCAS_CALLOC(BufSize, float64_t); - if(b == NULL) - { - ocas.exitflag=-2; - goto cleanup; - } - - alpha = (float64_t*)LIBOCAS_CALLOC(BufSize, float64_t); - if(alpha == NULL) - { - ocas.exitflag=-2; - goto cleanup; - } - - /* indices of examples which define a new cut */ - new_cut = (uint32_t*)LIBOCAS_CALLOC(nData, uint32_t); - if(new_cut == NULL) - { - ocas.exitflag=-2; - goto cleanup; - } - - I = (uint32_t*)LIBOCAS_CALLOC(BufSize, uint32_t); - if(I == NULL) - { - ocas.exitflag=-2; - goto cleanup; - } - - for(i=0; i< BufSize; i++) - I[i] = 1; - - diag_H = (float64_t*)LIBOCAS_CALLOC(BufSize, float64_t); - if(diag_H == NULL) - { - ocas.exitflag=-2; - goto cleanup; - } - - output = (float64_t*)LIBOCAS_CALLOC(nData*nY, float64_t); - if(output == NULL) - { - ocas.exitflag=-2; - goto cleanup; - } - - old_output = (float64_t*)LIBOCAS_CALLOC(nData*nY, float64_t); - if(old_output == NULL) - { - ocas.exitflag=-2; - goto cleanup; - } - - /* auxciliary variables used in the linesearch */ - A = (float64_t*)LIBOCAS_CALLOC(nData*nY, float64_t); - if(A == NULL) - { - ocas.exitflag=-2; - goto cleanup; - } - - B = (float64_t*)LIBOCAS_CALLOC(nData*nY, float64_t); - if(B == NULL) - { - ocas.exitflag=-2; - goto cleanup; - } - - theta = (float64_t*)LIBOCAS_CALLOC(nY, float64_t); - if(theta == NULL) - { - ocas.exitflag=-2; - goto cleanup; - } - - sortedA = (float64_t*)LIBOCAS_CALLOC(nY, float64_t); - if(sortedA == NULL) - { - ocas.exitflag=-2; - goto cleanup; - } - - Theta = (float64_t*)LIBOCAS_CALLOC(nData*nY, float64_t); - if(Theta == NULL) - { - ocas.exitflag=-2; - goto cleanup; - } - - Add = (float64_t*)LIBOCAS_CALLOC(nData*nY, float64_t); - if(Add == NULL) - { - ocas.exitflag=-2; - goto cleanup; - } - - /* Set initial values*/ - ocas.nCutPlanes = 0; - ocas.exitflag = 0; - ocas.nIter = 0; - ocas.Q_D = 0; - ocas.trn_err = nData; - R = (float64_t)nData; - sq_norm_W = 0; - element_b = (float64_t)nData; - ocas.Q_P = 0.5*sq_norm_W + C*R; - - /* initial cutting plane */ - for(i=0; i < nData; i++) - { - y2 = (uint32_t)data_y[i]; - - if(y2 > 0) - new_cut[i] = 0; - else - new_cut[i] = 1; - - } - - ocas.ocas_time = get_time() - ocas_start_time; - - start_time = get_time(); - ocas_print(ocas); - ocas.print_time += get_time() - start_time; - - /* main loop of the OCAS */ - while( ocas.exitflag == 0 ) - { - ocas.nIter++; - - /* append a new cut to the buffer and update H */ - b[ocas.nCutPlanes] = -(float64_t)element_b; - - start_time = get_time(); - - if(add_new_cut( &H[LIBOCAS_INDEX(0,ocas.nCutPlanes,BufSize)], new_cut, ocas.nCutPlanes, user_data ) != 0) - { - ocas.exitflag=-2; - goto cleanup; - } - - ocas.add_time += get_time() - start_time; - - /* copy newly appended row: H(ocas.nCutPlanes,ocas.nCutPlanes,1:ocas.nCutPlanes-1) = H(1:ocas.nCutPlanes-1:ocas.nCutPlanes)' */ - diag_H[ocas.nCutPlanes] = H[LIBOCAS_INDEX(ocas.nCutPlanes,ocas.nCutPlanes,BufSize)]; - for(i=0; i < ocas.nCutPlanes; i++) - { - H[LIBOCAS_INDEX(ocas.nCutPlanes,i,BufSize)] = H[LIBOCAS_INDEX(i,ocas.nCutPlanes,BufSize)]; - } - - ocas.nCutPlanes++; - - /* call inner QP solver */ - start_time = get_time(); - - qp_exitflag = libqp_splx_solver(&get_col, diag_H, b, &C, I, &S, alpha, - ocas.nCutPlanes, QPSolverMaxIter, QPSolverTolAbs, QPSolverTolRel, -LIBOCAS_PLUS_INF,0); - - ocas.qp_exitflag = qp_exitflag.exitflag; - - ocas.qp_solver_time += get_time() - start_time; - ocas.Q_D = -qp_exitflag.QP; - - ocas.nNZAlpha = 0; - for(i=0; i < ocas.nCutPlanes; i++) - if( alpha[i] != 0) ocas.nNZAlpha++; - - sq_norm_oldW = sq_norm_W; - start_time = get_time(); - compute_W( &sq_norm_W, &dot_prod_WoldW, alpha, ocas.nCutPlanes, user_data ); - ocas.w_time += get_time() - start_time; - - /* select a new cut */ - switch( Method ) - { - /* cutting plane algorithm implemented in SVMperf and BMRM */ - case 0: - - start_time = get_time(); - if( compute_output( output, user_data ) != 0) - { - ocas.exitflag=-2; - goto cleanup; - } - ocas.output_time += get_time()-start_time; - - /* the following loop computes: */ - element_b = 0.0; /* element_b = R(old_W) - g'*old_W */ - R = 0; /* R(W) = sum_i max_y ( [[y != y_i]] + (w_y- w_y_i)'*x_i ) */ - ocas.trn_err = 0; /* trn_err = sum_i [[y != y_i ]] */ - /* new_cut[i] = argmax_i ( [[y != y_i]] + (w_y- w_y_i)'*x_i ) */ - for(i=0; i < nData; i++) - { - y2 = (uint32_t)data_y[i]; - - for(xi=-LIBOCAS_PLUS_INF, y=0; y < nY; y++) - { - if(y2 != y && xi < output[LIBOCAS_INDEX(y,i,nY)]) - { - xi = output[LIBOCAS_INDEX(y,i,nY)]; - ypred = y; - } - } - - if(xi >= output[LIBOCAS_INDEX(y2,i,nY)]) - ocas.trn_err ++; - - xi = LIBOCAS_MAX(0,xi+1-output[LIBOCAS_INDEX(y2,i,nY)]); - R += xi; - if(xi > 0) - { - element_b++; - new_cut[i] = ypred; - } - else - new_cut[i] = y2; - } - - ocas.Q_P = 0.5*sq_norm_W + C*R; - - ocas.ocas_time = get_time() - ocas_start_time; - - start_time = get_time(); - ocas_print(ocas); - ocas.print_time += get_time() - start_time; - - break; - - /* The OCAS solver */ - case 1: - sg_memcpy( old_output, output, sizeof(float64_t)*nData*nY ); - - start_time = get_time(); - if( compute_output( output, user_data ) != 0) - { - ocas.exitflag=-2; - goto cleanup; - } - ocas.output_time += get_time()-start_time; - - A0 = sq_norm_W - 2*dot_prod_WoldW + sq_norm_oldW; - B0 = dot_prod_WoldW - sq_norm_oldW; - - for(i=0; i < nData; i++) - { - y2 = (uint32_t)data_y[i]; - - for(y=0; y < nY; y++) - { - A[LIBOCAS_INDEX(y,i,nY)] = C*(output[LIBOCAS_INDEX(y,i,nY)] - old_output[LIBOCAS_INDEX(y,i,nY)] - + old_output[LIBOCAS_INDEX(y2,i,nY)] - output[LIBOCAS_INDEX(y2,i,nY)]); - B[LIBOCAS_INDEX(y,i,nY)] = C*(old_output[LIBOCAS_INDEX(y,i,nY)] - old_output[LIBOCAS_INDEX(y2,i,nY)] - + (float64_t)(y != y2)); - } - } - - /* linesearch */ -/* new_x = msvm_linesearch_mex(A0,B0,AA*C,BB*C);*/ - - grad_sum = B0; - cnt1 = 0; - cnt2 = 0; - for(i=0; i < nData; i++) - { - findactive(theta,sortedA,&nSortedA,&A[i*nY],&B[i*nY],nY,sort); - - idx = 0; - while( idx < nSortedA-1 && theta[idx] < 0 ) - idx++; - - grad_sum += sortedA[idx]; - - for(j=idx; j < nSortedA-1; j++) - { - Theta[cnt1] = theta[j]; - cnt1++; - } - - for(j=idx+1; j < nSortedA; j++) - { - Add[cnt2] = -sortedA[j-1]+sortedA[j]; - cnt2++; - } - } - - start_time = get_time(); - sort(Theta,Add,cnt1); - ocas.sort_time += get_time() - start_time; - - grad = grad_sum; - if(grad >= 0) - { - min_x = 0; - } - else - { - old_x = 0; - old_grad = grad; - - for(i=0; i < cnt1; i++) - { - x = Theta[i]; - - grad = x*A0 + grad_sum; - - if(grad >=0) - { - - min_x = (grad*old_x - old_grad*x)/(grad - old_grad); - - break; - } - else - { - grad_sum = grad_sum + Add[i]; - - grad = x*A0 + grad_sum; - if( grad >= 0) - { - min_x = x; - break; - } - } - - old_grad = grad; - old_x = x; - } - } - /* end of the linesearch which outputs min_x */ - - t = min_x; - t1 = t; /* new (best so far) W */ - t2 = t+(1.0-t)*MU; /* new cutting plane */ - /* t2 = t+(1.0-t)/10.0; */ - - /* update W to be the best so far solution */ - sq_norm_W = update_W( t1, user_data ); - - /* the following code computes a new cutting plane: */ - element_b = 0.0; /* element_b = R(old_W) - g'*old_W */ - /* new_cut[i] = argmax_i ( [[y != y_i]] + (w_y- w_y_i)'*x_i ) */ - for(i=0; i < nData; i++) - { - y2 = (uint32_t)data_y[i]; - - for(xi=-LIBOCAS_PLUS_INF, y=0; y < nY; y++) - { - tmp = old_output[LIBOCAS_INDEX(y,i,nY)]*(1-t2) + t2*output[LIBOCAS_INDEX(y,i,nY)]; - if(y2 != y && xi < tmp) - { - xi = tmp; - ypred = y; - } - } - - tmp = old_output[LIBOCAS_INDEX(y2,i,nY)]*(1-t2) + t2*output[LIBOCAS_INDEX(y2,i,nY)]; - xi = LIBOCAS_MAX(0,xi+1-tmp); - if(xi > 0) - { - element_b++; - new_cut[i] = ypred; - } - else - new_cut[i] = y2; - } - - /* compute Risk, class. error and update outputs to correspond to the new W */ - ocas.trn_err = 0; /* trn_err = sum_i [[y != y_i ]] */ - R = 0; - for(i=0; i < nData; i++) - { - y2 = (uint32_t)data_y[i]; - - for(tmp=-LIBOCAS_PLUS_INF, y=0; y < nY; y++) - { - output[LIBOCAS_INDEX(y,i,nY)] = old_output[LIBOCAS_INDEX(y,i,nY)]*(1-t1) + t1*output[LIBOCAS_INDEX(y,i,nY)]; - - if(y2 != y && tmp < output[LIBOCAS_INDEX(y,i,nY)]) - { - ypred = y; - tmp = output[LIBOCAS_INDEX(y,i,nY)]; - } - } - - R += LIBOCAS_MAX(0,1+tmp - output[LIBOCAS_INDEX(y2,i,nY)]); - if( tmp >= output[LIBOCAS_INDEX(y2,i,nY)]) - ocas.trn_err ++; - } - - ocas.Q_P = 0.5*sq_norm_W + C*R; - - - /* get time and print status */ - ocas.ocas_time = get_time() - ocas_start_time; - - start_time = get_time(); - ocas_print(ocas); - ocas.print_time += get_time() - start_time; - - break; - - } - - /* Stopping conditions */ - if( ocas.Q_P - ocas.Q_D <= TolRel*LIBOCAS_ABS(ocas.Q_P)) ocas.exitflag = 1; - if( ocas.Q_P - ocas.Q_D <= TolAbs) ocas.exitflag = 2; - if( ocas.Q_P <= QPBound) ocas.exitflag = 3; - if( MaxTime > 0 && ocas.ocas_time >= MaxTime) ocas.exitflag = 4; - if(ocas.nCutPlanes >= BufSize) ocas.exitflag = -1; - - } /* end of the main loop */ - -cleanup: - - LIBOCAS_FREE(H); - LIBOCAS_FREE(b); - LIBOCAS_FREE(alpha); - LIBOCAS_FREE(new_cut); - LIBOCAS_FREE(I); - LIBOCAS_FREE(diag_H); - LIBOCAS_FREE(output); - LIBOCAS_FREE(old_output); - LIBOCAS_FREE(A); - LIBOCAS_FREE(B); - LIBOCAS_FREE(theta); - LIBOCAS_FREE(Theta); - LIBOCAS_FREE(sortedA); - LIBOCAS_FREE(Add); - - ocas.ocas_time = get_time() - ocas_start_time; - - return(ocas); -} -} - -#endif //USE_GPL_SHOGUN diff --git a/src/shogun/lib/external/libocas.h b/src/shogun/lib/external/libocas.h deleted file mode 100644 index 1fb247a0849..00000000000 --- a/src/shogun/lib/external/libocas.h +++ /dev/null @@ -1,139 +0,0 @@ -/* - * This program is free software; you can redistribute it and/or modify - * it under the terms of the GNU General Public License as published by - * the Free Software Foundation; either version 3 of the License, or - * (at your option) any later version. - * - * libocas.h: Implementation of the OCAS solver for training - * linear SVM classifiers. - * - * Copyright (C) 2008, 2009 Vojtech Franc, xfrancv@cmp.felk.cvut.cz - * Soeren Sonnenburg, soeren.sonnenburg@first.fraunhofer.de - * Implementation of SVM-Ocas solver. - *-------------------------------------------------------------------- */ - -#include -#ifdef USE_GPL_SHOGUN - -#include - -#ifndef libocas_h -#define libocas_h -#ifndef DOXYGEN_SHOULD_SKIP_THIS -namespace shogun -{ -#define LIBOCAS_PLUS_INF (-log(0.0)) -#define LIBOCAS_CALLOC(x,y) SG_CALLOC(y,x) -#define LIBOCAS_FREE(x) SG_FREE(x) -#define LIBOCAS_INDEX(ROW,COL,NUM_ROWS) ((COL)*(NUM_ROWS)+(ROW)) -#define LIBOCAS_MIN(A,B) ((A) > (B) ? (B) : (A)) -#define LIBOCAS_MAX(A,B) ((A) < (B) ? (B) : (A)) -#define LIBOCAS_ABS(A) ((A) < 0 ? -(A) : (A)) - -typedef struct { - uint32_t nIter; /* number of iterations */ - uint32_t nCutPlanes; /* number of cutitng buffered planes */ - uint32_t nNZAlpha; /* number of non-zero Lagrangeans (effective number of CPs) */ - uint32_t trn_err; /* number of training errors */ - float64_t Q_P; /* primal objective value */ - float64_t Q_D; /* dual objective value */ - float64_t output_time; /* time spent in computing outputs */ - float64_t sort_time; /* time spent in sorting */ - float64_t add_time; /* time spent in adding examples to compute cutting planes */ - float64_t w_time; /* time spent in computing parameter vector */ - float64_t qp_solver_time; /* time spent in inner QP solver */ - float64_t ocas_time; /* total time spent in svm_ocas_solver */ - float64_t print_time; /* time spent in ocas_print function */ - int8_t qp_exitflag; /* exitflag from the last call of the inner QP solver */ - int8_t exitflag; /* 1 .. ocas.Q_P - ocas.Q_D <= TolRel*ABS(ocas.Q_P) - 2 .. ocas.Q_P - ocas.Q_D <= TolAbs - 3 .. ocas.Q_P <= QPBound - 4 .. optimization time >= MaxTime - -1 .. ocas.nCutPlanes >= BufSize - -2 .. not enough memory for the solver */ -} ocas_return_value_T; - -/* binary linear SVM solver */ -ocas_return_value_T svm_ocas_solver( - float64_t C, /* regularizarion constant */ - uint32_t nData, /* number of exmaples */ - float64_t TolRel, /* halts if 1-Q_P/Q_D <= TolRel */ - float64_t TolAbs, /* halts if Q_P-Q_D <= TolRel */ - float64_t QPBound, /* halts if QP <= QPBound */ - float64_t MaxTime, /* maximal time in seconds spent in optmization */ - uint32_t BufSize, /* maximal number of buffered cutting planes */ - uint8_t Method, /* 0..standard CP (SVM-Perf,BMRM), 1..OCAS */ - void (*compute_W)(float64_t*, float64_t*, float64_t*, uint32_t, void*), - float64_t (*update_W)(float64_t, void*), - int (*add_new_cut)(float64_t*, uint32_t*, uint32_t, uint32_t, void*), - int (*compute_output)( float64_t*, void* ), - int (*sort)(float64_t*, float64_t*, uint32_t), - void (*ocas_print)(ocas_return_value_T), - void* user_data); - -/* binary linear SVM solver which allows using different C for each example*/ -ocas_return_value_T svm_ocas_solver_difC( - float64_t *C, /* regularizarion constants for each example */ - uint32_t nData, /* number of exmaples */ - float64_t TolRel, /* halts if 1-Q_P/Q_D <= TolRel */ - float64_t TolAbs, /* halts if Q_P-Q_D <= TolRel */ - float64_t QPBound, /* halts if QP <= QPBound */ - float64_t MaxTime, /* maximal time in seconds spent in optmization */ - uint32_t BufSize, /* maximal number of buffered cutting planes */ - uint8_t Method, /* 0..standard CP (SVM-Perf,BMRM), 1..OCAS */ - void (*compute_W)(float64_t*, float64_t*, float64_t*, uint32_t, void*), - float64_t (*update_W)(float64_t, void*), - int (*add_new_cut)(float64_t*, uint32_t*, uint32_t, uint32_t, void*), - int (*compute_output)( float64_t*, void* ), - int (*sort)(float64_t*, float64_t*, uint32_t), - void (*ocas_print)(ocas_return_value_T), - void* user_data); - -/* multi-class (Singer-Crammer formulation) linear SVM solver */ -ocas_return_value_T msvm_ocas_solver( - float64_t C, - float64_t *data_y, - uint32_t nY, - uint32_t nData, - float64_t TolRel, - float64_t TolAbs, - float64_t QPBound, - float64_t MaxTime, - uint32_t _BufSize, - uint8_t Method, - void (*compute_W)(float64_t*, float64_t*, float64_t*, uint32_t, void*), - float64_t (*update_W)(float64_t, void*), - int (*add_new_cut)(float64_t*, uint32_t*, uint32_t, void*), - int (*compute_output)(float64_t*, void* ), - int (*sort)(float64_t*, float64_t*, uint32_t), - void (*ocas_print)(ocas_return_value_T), - void* user_data); - - -/* binary linear SVM solver */ -ocas_return_value_T svm_ocas_solver_nnw( - float64_t C, /* regularizarion constant */ - uint32_t nData, /* number of exmaples */ - uint32_t num_nnw, /* number of components of W which must non-negative*/ - uint32_t* nnw_idx, /* indices of W which must be non-negative */ - float64_t TolRel, /* halts if 1-Q_P/Q_D <= TolRel */ - float64_t TolAbs, /* halts if Q_P-Q_D <= TolRel */ - float64_t QPBound, /* halts if QP <= QPBound */ - float64_t MaxTime, /* maximal time in seconds spent in optmization */ - uint32_t BufSize, /* maximal number of buffered cutting planes */ - uint8_t Method, /* 0..standard CP (SVM-Perf,BMRM), 1..OCAS */ - int (*add_pw_constr)(uint32_t, uint32_t, void*), - void (*clip_neg_w)(uint32_t, uint32_t*, void*), - void (*compute_W)(float64_t*, float64_t*, float64_t*, uint32_t, void*), - float64_t (*update_W)(float64_t, void*), - int (*add_new_cut)(float64_t*, uint32_t*, uint32_t, uint32_t, void*), - int (*compute_output)( float64_t*, void* ), - int (*sort)(float64_t*, float64_t*, uint32_t), - void (*ocas_print)(ocas_return_value_T), - void* user_data); - -} -#endif // DOXYGEN_SHOULD_SKIP_THIS -#endif /* libocas_h */ -#endif // USE_GPL_SHOGUN - diff --git a/src/shogun/lib/external/libocas_common.h b/src/shogun/lib/external/libocas_common.h deleted file mode 100644 index 4e481fd4c4d..00000000000 --- a/src/shogun/lib/external/libocas_common.h +++ /dev/null @@ -1,16 +0,0 @@ -#include -#ifdef USE_GPL_SHOGUN -#include -#include - - -namespace shogun -{ -#define OCAS_PLUS_INF CMath::INFTY -#define OCAS_CALLOC(...) calloc(__VA_ARGS__) -#define OCAS_FREE(...) SG_FREE(__VA_ARGS__) - -#define INDEX2(ROW,COL,NUM_ROWS) ((COL)*(NUM_ROWS)+(ROW)) -} - -#endif //USE_GPL_SHOGUN diff --git a/src/shogun/lib/external/libqp.h b/src/shogun/lib/external/libqp.h deleted file mode 100644 index cd249236d41..00000000000 --- a/src/shogun/lib/external/libqp.h +++ /dev/null @@ -1,90 +0,0 @@ -/*----------------------------------------------------------------------- - * libqp.h: Library for Quadratic Programming optimization. - * - * The library provides two solvers: - * 1. Solver for QP task with simplex constraints. - * See function ./lib/libqp_splx.c for definition of the QP task. - * - * 2. Solver for QP task with box constraints and a single linear - * equality constraint. - * See function ./lib/libqp_gsmo.c for definiton of the QP task. - * - * Copyright (C) 2006-2008 Vojtech Franc, xfrancv@cmp.felk.cvut.cz - * Center for Machine Perception, CTU FEL Prague - * - * This program is free software; you can redistribute it and/or - * modify it under the terms of the GNU General Public - * License as published by the Free Software Foundation; - * Version 3, 29 June 2007 - *-------------------------------------------------------------------- */ - -#ifndef libqp_h -#define libqp_h - -#include -#ifdef USE_GPL_SHOGUN - -#include - -#include -namespace shogun -{ -#define LIBQP_PLUS_INF (-log(0.0)) -#define LIBQP_CALLOC(x,y) SG_CALLOC(y,x) -#define LIBQP_FREE(x) SG_FREE(x) -#define LIBQP_INDEX(ROW,COL,NUM_ROWS) ((COL)*(NUM_ROWS)+(ROW)) -#define LIBQP_MIN(A,B) ((A) > (B) ? (B) : (A)) -#define LIBQP_MAX(A,B) ((A) < (B) ? (B) : (A)) -#define LIBQP_ABS(A) ((A) < 0 ? -(A) : (A)) - -#ifndef DOXYGEN_SHOULD_SKIP_THIS -/** QP solver return value */ -typedef struct { - /** number of iterations */ - uint32_t nIter; - /** primal objective value */ - float64_t QP; - /** dual objective value */ - float64_t QD; - /** exit flag */ - int8_t exitflag; /* -1 ... not enough memory - 0 ... nIter >= MaxIter - 1 ... QP - QD <= TolRel*ABS(QP) - 2 ... QP - QD <= TolAbs - 3 ... QP <= QP_TH - 4 ... eps-KKT conditions satisfied */ -} libqp_state_T; -#endif - -/** QP solver for tasks with simplex constraints */ -libqp_state_T libqp_splx_solver(const float64_t* (*get_col)(uint32_t), - float64_t *diag_H, - float64_t *f, - float64_t *b, - uint32_t *I, - uint8_t *S, - float64_t *x, - uint32_t n, - uint32_t MaxIter, - float64_t TolAbs, - float64_t TolRel, - float64_t QP_TH, - void (*print_state)(libqp_state_T state)); - -/** Generalized SMO algorithm */ -libqp_state_T libqp_gsmo_solver(const float64_t* (*get_col)(uint32_t), - float64_t *diag_H, - float64_t *f, - float64_t *a, - float64_t b, - float64_t *LB, - float64_t *UB, - float64_t *x, - uint32_t n, - uint32_t MaxIter, - float64_t TolKKT, - void (*print_state)(libqp_state_T state)); - -} -#endif //USE_GPL_SHOGUN -#endif /* libqp_h */ diff --git a/src/shogun/lib/external/libqp_gsmo.cpp b/src/shogun/lib/external/libqp_gsmo.cpp deleted file mode 100644 index aa8cc8f49f9..00000000000 --- a/src/shogun/lib/external/libqp_gsmo.cpp +++ /dev/null @@ -1,257 +0,0 @@ -/*----------------------------------------------------------------------- - * libqp_gsmo.c: implementation of the Generalized SMO algorithm. - * - * DESCRIPTION - * The library provides function which solves the following instance of - * a convex Quadratic Programming task: - * - * min QP(x) := 0.5*x'*H*x + f'*x - * x - * - * s.t. a'*x = b - * LB[i] <= x[i] <= UB[i] for all i=1..n - * - * A precision of the found solution is controlled by the input argument - * TolKKT which defines tightness of the relaxed Karush-Kuhn-Tucker - * stopping conditions. - * - * INPUT ARGUMENTS - * get_col function which returns pointer to the i-th column of H. - * diag_H [float64_t n x 1] vector containing values on the diagonal of H. - * f [float64_t n x 1] vector. - * a [float64_t n x 1] Vector which must not contain zero entries. - * b [float64_t 1 x 1] Scalar. - * LB [float64_t n x 1] Lower bound; -inf is allowed. - * UB [float64_t n x 1] Upper bound; inf is allowed. - * x [float64_t n x 1] solution vector; must be feasible. - * n [uint32_t 1 x 1] dimension of H. - * MaxIter [uint32_t 1 x 1] max number of iterations. - * TolKKT [float64_t 1 x 1] Tightness of KKT stopping conditions. - * print_state print function; if == NULL it is not called. - * - * RETURN VALUE - * structure [libqp_state_T] - * .QP [1x1] Primal objective value. - * .exitflag [1 x 1] Indicates which stopping condition was used: - * -3 ... initial solution vector does not satisfy equality constraint - * -2 ... initial solution vector does not satisfy bounds - * -1 ... not enough memory - * 0 ... Maximal number of iterations reached: nIter >= MaxIter. - * 4 ... Relaxed KKT conditions satisfied. - * .nIter [1x1] Number of iterations. - * - * REFERENCE - * S.S. Keerthi, E.G. Gilbert. Convergence of a generalized SMO algorithm - * for SVM classier design. Technical Report CD-00-01, Control Division, - * Dept. of Mechanical and Production Engineering, National University - * of Singapore, 2000. - * http://citeseer.ist.psu.edu/keerthi00convergence.html - * - * - * Copyright (C) 2006-2008 Vojtech Franc, xfrancv@cmp.felk.cvut.cz - * Center for Machine Perception, CTU FEL Prague - * - * This program is free software; you can redistribute it and/or - * modify it under the terms of the GNU General Public - * License as published by the Free Software Foundation; - * Version 3, 29 June 2007 - *-------------------------------------------------------------------- */ - -#include -#ifdef USE_GPL_SHOGUN - -#include -#include -#include -#include -#include -#include - -#include -#include -#include - -namespace shogun -{ - -libqp_state_T libqp_gsmo_solver(const float64_t* (*get_col)(uint32_t), - float64_t *diag_H, - float64_t *f, - float64_t *a, - float64_t b, - float64_t *LB, - float64_t *UB, - float64_t *x, - uint32_t n, - uint32_t MaxIter, - float64_t TolKKT, - void (*print_state)(libqp_state_T state)) -{ - float64_t *col_u; - float64_t *col_v; - float64_t *Nabla; - float64_t minF_up; - float64_t maxF_low; - float64_t tau; - float64_t F_i; - float64_t tau_ub, tau_lb; - uint32_t i, j; - uint32_t u=0, v=0; - libqp_state_T state; - float64_t atx = 0.0; - - Nabla = NULL; - - /* ------------------------------------------------------------ */ - /* Initialization */ - /* ------------------------------------------------------------ */ - - // check bounds of initial guess - for (i=0; iUB[i]) - { - state.exitflag = -2; - goto cleanup; - } - if (x[i]1e-9) - { - printf("%f \ne %f\n",b,atx); - state.exitflag = -3; - goto cleanup; - } - - /* Nabla = H*x + f is gradient*/ - Nabla = (float64_t*)LIBQP_CALLOC(n, float64_t); - if( Nabla == NULL ) - { - state.exitflag=-1; - goto cleanup; - } - - /* compute gradient */ - for( i=0; i < n; i++ ) - { - Nabla[i] += f[i]; - if( x[i] != 0 ) { - col_u = (float64_t*)get_col(i); - for( j=0; j < n; j++ ) { - Nabla[j] += col_u[j]*x[i]; - } - } - } - - if( print_state != NULL) - { - state.QP = 0; - for(i = 0; i < n; i++ ) - state.QP += 0.5*(x[i]*Nabla[i]+x[i]*f[i]); - - print_state( state ); - } - - - /* ------------------------------------------------------------ */ - /* Main optimization loop */ - /* ------------------------------------------------------------ */ - - state.nIter = 0; - state.exitflag = 100; - while( state.exitflag == 100 ) - { - state.nIter ++; - - /* find the most violating pair of variables */ - minF_up = LIBQP_PLUS_INF; - maxF_low = -LIBQP_PLUS_INF; - for(i = 0; i < n; i++ ) - { - - F_i = Nabla[i]/a[i]; - - if(LB[i] < x[i] && x[i] < UB[i]) - { /* i is from I_0 */ - if( minF_up > F_i) { minF_up = F_i; u = i; } - if( maxF_low < F_i) { maxF_low = F_i; v = i; } - } - else if((a[i] > 0 && x[i] == LB[i]) || (a[i] < 0 && x[i] == UB[i])) - { /* i is from I_1 or I_2 */ - if( minF_up > F_i) { minF_up = F_i; u = i; } - } - else if((a[i] > 0 && x[i] == UB[i]) || (a[i] < 0 && x[i] == LB[i])) - { /* i is from I_3 or I_4 */ - if( maxF_low < F_i) { maxF_low = F_i; v = i; } - } - } - - /* check KKT conditions */ - if( maxF_low - minF_up <= TolKKT ) - state.exitflag = 4; - else - { - /* SMO update of the most violating pair */ - col_u = (float64_t*)get_col(u); - col_v = (float64_t*)get_col(v); - - if( a[u] > 0 ) - { tau_lb = (LB[u]-x[u])*a[u]; tau_ub = (UB[u]-x[u])*a[u]; } - else - { tau_ub = (LB[u]-x[u])*a[u]; tau_lb = (UB[u]-x[u])*a[u]; } - - if( a[v] > 0 ) - { tau_lb = LIBQP_MAX(tau_lb,(x[v]-UB[v])*a[v]); tau_ub = LIBQP_MIN(tau_ub,(x[v]-LB[v])*a[v]); } - else - { tau_lb = LIBQP_MAX(tau_lb,(x[v]-LB[v])*a[v]); tau_ub = LIBQP_MIN(tau_ub,(x[v]-UB[v])*a[v]); } - - tau = (Nabla[v]/a[v]-Nabla[u]/a[u])/ - (diag_H[u]/(a[u]*a[u]) + diag_H[v]/(a[v]*a[v]) - 2*col_u[v]/(a[u]*a[v])); - - tau = LIBQP_MIN(LIBQP_MAX(tau,tau_lb),tau_ub); - - x[u] += tau/a[u]; - x[v] -= tau/a[v]; - - /* update Nabla */ - for(i = 0; i < n; i++ ) - Nabla[i] += col_u[i]*tau/a[u] - col_v[i]*tau/a[v]; - - } - - if( state.nIter >= MaxIter ) - state.exitflag = 0; - - if( print_state != NULL) - { - state.QP = 0; - for(i = 0; i < n; i++ ) - state.QP += 0.5*(x[i]*Nabla[i]+x[i]*f[i]); - - print_state( state ); - } - - } - - /* compute primal objective value */ - state.QP = 0; - for(i = 0; i < n; i++ ) - state.QP += 0.5*(x[i]*Nabla[i]+x[i]*f[i]); - -cleanup: - - LIBQP_FREE(Nabla); - - return( state ); -} - -} /* shogun namespace */ -#endif //USE_GPL_SHOGUN diff --git a/src/shogun/lib/external/libqp_splx.cpp b/src/shogun/lib/external/libqp_splx.cpp deleted file mode 100644 index 06680512fde..00000000000 --- a/src/shogun/lib/external/libqp_splx.cpp +++ /dev/null @@ -1,413 +0,0 @@ -/*----------------------------------------------------------------------- - * libqp_splx.c: solver for Quadratic Programming task with - * simplex constraints. - * - * DESCRIPTION - * The library provides function which solves the following instance of - * a convex Quadratic Programmin task: - * - * min QP(x):= 0.5*x'*H*x + f'*x - * x - * - * subject to: - * sum_{i in I_k} x[i] == b[k] for all k such that S[k] == 0 - * sum_{i in I_k} x[i] <= b[k] for all k such that S[k] == 1 - * x(i) >= 0 for all i=1:n - * - * where I_k = { i | I[i] == k}, k={1,...,m}. - * - * A precision of the found solution is controled by the input argumens - * MaxIter, TolAbs, QP_TH and MaxIter which define the stopping conditions: - * - * nIter >= MaxIter -> exitflag = 0 Number of iterations - * QP-QD <= TolAbs -> exitflag = 1 Abs. tolerance (duality gap) - * QP-QD <= QP*TolRel -> exitflag = 2 Relative tolerance - * QP <= QP_TH -> exitflag = 3 Threshold on objective value - * - * where QP and QD are primal respectively dual objective values. - * - * INPUT ARGUMENTS - * get_col function which returns pointer to the i-th column of H. - * diag_H [float64_t n x 1] vector containing values on the diagonal of H. - * f [float64_t n x 1] vector. - * b [float64_t n x 1] vector of positive numbers. - * I [uint16_T n x 1] vector containing numbers 1...m. - * S [uint8_T n x 1] vector containing numbers 0 and 1. - * x [float64_t n x 1] solution vector; must be feasible. - * n [uint32_t 1 x 1] dimension of H. - * MaxIter [uint32_t 1 x 1] max number of iterations. - * TolAbs [float64_t 1 x 1] Absolute tolerance. - * TolRel [float64_t 1 x 1] Relative tolerance. - * QP_TH [float64_t 1 x 1] Threshold on the primal value. - * print_state print function; if == NULL it is not called. - * - * RETURN VALUE - * structure [libqp_state_T] - * .QP [1 x 1] Primal objective value. - * .QD [1 x 1] Dual objective value. - * .nIter [1 x 1] Number of iterations. - * .exitflag [1 x 1] Indicates which stopping condition was used: - * -1 ... Not enough memory. - * 0 ... Maximal number of iteations reached: nIter >= MaxIter. - * 1 ... Relarive tolerance reached: QP-QD <= abs(QP)*TolRel - * 2 ... Absolute tolerance reached: QP-QD <= TolAbs - * 3 ... Objective value reached threshold: QP <= QP_TH. - * - * REFERENCE - * The algorithm is described in: - * V. Franc, V. Hlavac. A Novel Algorithm for Learning Support Vector Machines - * with Structured Output Spaces. Research Report K333 22/06, CTU-CMP-2006-04. - * May, 2006. ftp://cmp.felk.cvut.cz/pub/cmp/articles/franc/Franc-TR-2006-04.ps - * - * Copyright (C) 2006-2008 Vojtech Franc, xfrancv@cmp.felk.cvut.cz - * Center for Machine Perception, CTU FEL Prague - * - * This program is free software; you can redistribute it and/or - * modify it under the terms of the GNU General Public - * License as published by the Free Software Foundation; - * Version 3, 29 June 2007 - *-------------------------------------------------------------------- */ - -#include -#ifdef USE_GPL_SHOGUN - -#include -#include -#include -#include -#include -#include - -#include -#include -namespace shogun -{ - -libqp_state_T libqp_splx_solver(const float64_t* (*get_col)(uint32_t), - float64_t *diag_H, - float64_t *f, - float64_t *b, - uint32_t *I, - uint8_t *S, - float64_t *x, - uint32_t n, - uint32_t MaxIter, - float64_t TolAbs, - float64_t TolRel, - float64_t QP_TH, - void (*print_state)(libqp_state_T state)) -{ - float64_t *d; - float64_t *col_u, *col_v; - float64_t *x_neq; - float64_t tmp; - float64_t improv; - float64_t tmp_num; - float64_t tmp_den=0; - float64_t tau=0; - float64_t delta; - uint32_t *inx; - uint32_t *nk; - uint32_t m; - int32_t u=0; - int32_t v=0; - uint32_t k; - uint32_t i, j; - libqp_state_T state; - - - /* ------------------------------------------------------------ - Initialization - ------------------------------------------------------------ */ - state.nIter = 0; - state.QP = LIBQP_PLUS_INF; - state.QD = -LIBQP_PLUS_INF; - state.exitflag = 100; - - inx=NULL; - nk=NULL; - d=NULL; - x_neq = NULL; - - /* count number of constraints */ - for( i=0, m=0; i < n; i++ ) - m = LIBQP_MAX(m,I[i]); - - /* auxciliary variables for tranforming equalities to inequalities */ - x_neq = (float64_t*) LIBQP_CALLOC(m, float64_t); - if( x_neq == NULL ) - { - state.exitflag=-1; - goto cleanup; - } - - /* inx is translation table between variable index i and its contraint */ - inx = (uint32_t*) LIBQP_CALLOC(m*n, uint32_t); - if( inx == NULL ) - { - state.exitflag=-1; - goto cleanup; - } - - /* nk is the number of variables coupled by i-th linear constraint */ - nk = (uint32_t*) LIBQP_CALLOC(m, uint32_t); - if( nk == NULL ) - { - state.exitflag=-1; - goto cleanup; - } - - /* setup auxciliary variables */ - for( i=0; i < m; i++ ) - x_neq[i] = b[i]; - - - /* create inx and nk */ - for( i=0; i < n; i++ ) { - k = I[i]-1; - inx[LIBQP_INDEX(nk[k],k,n)] = i; - nk[k]++; - - if(S[k] != 0) - x_neq[k] -= x[i]; - } - - /* d = H*x + f is gradient*/ - d = (float64_t*) LIBQP_CALLOC(n, float64_t); - if( d == NULL ) - { - state.exitflag=-1; - goto cleanup; - } - - /* compute gradient */ - for( i=0; i < n; i++ ) - { - d[i] += f[i]; - if( x[i] > 0 ) { - col_u = (float64_t*)get_col(i); - for( j=0; j < n; j++ ) { - d[j] += col_u[j]*x[i]; - } - } - } - - /* compute state.QP = 0.5*x'*(f+d); - state.QD = 0.5*x'*(f-d); */ - for( i=0, state.QP = 0, state.QD=0; i < n; i++) - { - state.QP += x[i]*(f[i]+d[i]); - state.QD += x[i]*(f[i]-d[i]); - } - state.QP = 0.5*state.QP; - state.QD = 0.5*state.QD; - - for( i=0; i < m; i++ ) - { - for( j=0, tmp = LIBQP_PLUS_INF; j < nk[i]; j++ ) - tmp = LIBQP_MIN(tmp, d[inx[LIBQP_INDEX(j,i,n)]]); - - if(S[i] == 0) - state.QD += b[i]*tmp; - else - state.QD += b[i]*LIBQP_MIN(tmp,0); - } - - /* print initial state */ - if( print_state != NULL) - print_state( state ); - - /* ------------------------------------------------------------ - Main optimization loop - ------------------------------------------------------------ */ - while( state.exitflag == 100 ) - { - state.nIter ++; - - /* go over blocks of variables coupled by lin. constraint */ - for( k=0; k < m; k++ ) - { - - /* compute u = argmin_{i in I_k} d[i] - delta = sum_{i in I_k} x[i]*d[i] - b*min_{i in I_k} */ - for( j=0, tmp = LIBQP_PLUS_INF, delta = 0; j < nk[k]; j++ ) - { - i = inx[LIBQP_INDEX(j,k,n)]; - delta += x[i]*d[i]; - if( tmp > d[i] ) { - tmp = d[i]; - u = i; - } - } - - if(S[k] != 0 && d[u] > 0) - u = -1; - else - delta -= b[k]*d[u]; - - /* if satisfied then k-th block of variables needs update */ - if( delta > TolAbs/m && delta > TolRel*LIBQP_ABS(state.QP)/m) - { - /* for fixed u select v = argmax_{i in I_k} Improvement(i) */ - if( u != -1 ) - { - col_u = (float64_t*)get_col(u); - improv = -LIBQP_PLUS_INF; - for( j=0; j < nk[k]; j++ ) - { - i = inx[LIBQP_INDEX(j,k,n)]; - - if(x[i] > 0 && i != uint32_t(u)) - { - tmp_num = x[i]*(d[i] - d[u]); - tmp_den = x[i]*x[i]*(diag_H[u] - 2*col_u[i] + diag_H[i]); - if( tmp_den > 0 ) - { - if( tmp_num < tmp_den ) - tmp = tmp_num*tmp_num / tmp_den; - else - tmp = tmp_num - 0.5 * tmp_den; - - if( tmp > improv ) - { - improv = tmp; - tau = LIBQP_MIN(1,tmp_num/tmp_den); - v = i; - } - } - } - } - - /* check if virtual variable can be for updated */ - if(x_neq[k] > 0 && S[k] != 0) - { - tmp_num = -x_neq[k]*d[u]; - tmp_den = x_neq[k]*x_neq[k]*diag_H[u]; - if( tmp_den > 0 ) - { - if( tmp_num < tmp_den ) - tmp = tmp_num*tmp_num / tmp_den; - else - tmp = tmp_num - 0.5 * tmp_den; - - if( tmp > improv ) - { - improv = tmp; - tau = LIBQP_MIN(1,tmp_num/tmp_den); - v = -1; - } - } - } - - /* minimize objective w.r.t variable u and v */ - if(v != -1) - { - tmp = x[v]*tau; - x[u] += tmp; - x[v] -= tmp; - - /* update d = H*x + f */ - col_v = (float64_t*)get_col(v); - for(i = 0; i < n; i++ ) - d[i] += tmp*(col_u[i]-col_v[i]); - } - else - { - tmp = x_neq[k]*tau; - x[u] += tmp; - x_neq[k] -= tmp; - - /* update d = H*x + f */ - for(i = 0; i < n; i++ ) - d[i] += tmp*col_u[i]; - } - } - else - { - improv = -LIBQP_PLUS_INF; - for( j=0; j < nk[k]; j++ ) - { - i = inx[LIBQP_INDEX(j,k,n)]; - - if(x[i] > 0) - { - tmp_num = x[i]*d[i]; - tmp_den = x[i]*x[i]*diag_H[i]; - if( tmp_den > 0 ) - { - if( tmp_num < tmp_den ) - tmp = tmp_num*tmp_num / tmp_den; - else - tmp = tmp_num - 0.5 * tmp_den; - - if( tmp > improv ) - { - improv = tmp; - tau = LIBQP_MIN(1,tmp_num/tmp_den); - v = i; - } - } - } - } - - tmp = x[v]*tau; - x_neq[k] += tmp; - x[v] -= tmp; - - /* update d = H*x + f */ - col_v = (float64_t*)get_col(v); - for(i = 0; i < n; i++ ) - d[i] -= tmp*col_v[i]; - } - - /* update objective value */ - state.QP = state.QP - improv; - } - } - - /* Compute primal and dual objectives */ - for( i=0, state.QP = 0, state.QD=0; i < n; i++) - { - state.QP += x[i]*(f[i]+d[i]); - state.QD += x[i]*(f[i]-d[i]); - } - state.QP = 0.5*state.QP; - state.QD = 0.5*state.QD; - - for( k=0; k < m; k++ ) - { - for( j=0,tmp = LIBQP_PLUS_INF; j < nk[k]; j++ ) { - i = inx[LIBQP_INDEX(j,k,n)]; - tmp = LIBQP_MIN(tmp, d[i]); - } - - if(S[k] == 0) - state.QD += b[k]*tmp; - else - state.QD += b[k]*LIBQP_MIN(tmp,0); - } - - /* print state */ - if( print_state != NULL) - print_state( state ); - - /* check stopping conditions */ - if(state.QP-state.QD <= LIBQP_ABS(state.QP)*TolRel ) state.exitflag = 1; - else if( state.QP-state.QD <= TolAbs ) state.exitflag = 2; - else if( state.QP <= QP_TH ) state.exitflag = 3; - else if( state.nIter >= MaxIter) state.exitflag = 0; - } - - /*---------------------------------------------------------- - Clean up - ---------------------------------------------------------- */ -cleanup: - LIBQP_FREE( d ); - LIBQP_FREE( inx ); - LIBQP_FREE( nk ); - LIBQP_FREE( x_neq ); - - return( state ); -} -} -#endif //USE_GPL_SHOGUN diff --git a/src/shogun/lib/external/pr_loqo.cpp b/src/shogun/lib/external/pr_loqo.cpp deleted file mode 100644 index 00153c8ca3f..00000000000 --- a/src/shogun/lib/external/pr_loqo.cpp +++ /dev/null @@ -1,682 +0,0 @@ -/* - * This program is free software; you can redistribute it and/or modify - * it under the terms of the GNU General Public License as published by - * the Free Software Foundation; either version 3 of the License, or - * (at your option) any later version. - * - * Purpose: solves quadratic programming problem for pattern recognition - * for support vectors - * - * Written (W) 1997-1998 Alex J. Smola - * Written (W) 1999-2009 Soeren Sonnenburg - * Written (W) 1999-2008 Gunnar Raetsch - * Copyright (C) 1997-2009 Fraunhofer Institute FIRST and Max-Planck-Society - */ - -#include -#ifdef USE_GPL_SHOGUN - -#include -#include -#include -#include -#include - -#include -#include -#include -#include - -namespace shogun -{ - -#define PREDICTOR 1 -#define CORRECTOR 2 - -/***************************************************************** - replace this by any other function that will exit gracefully - in a larger system - ***************************************************************/ - -void nrerror(char error_text[]) -{ - SG_SDEBUG("terminating optimizer - %s\n", error_text) - // exit(1); -} - -/***************************************************************** - taken from numerical recipes and modified to accept pointers - moreover numerical recipes code seems to be buggy (at least the - ones on the web) - - cholesky solver and backsubstitution - leaves upper right triangle intact (rows first order) - ***************************************************************/ - -#ifdef HAVE_LAPACK -bool choldc(float64_t* a, int32_t n, float64_t* p) -{ - if (n<=0) - return false; - - float64_t* a2=SG_MALLOC(float64_t, n*n); - - for (int32_t i=0; i0) - SG_SDEBUG("Choldc failed, matrix not positive definite\n") - - SG_FREE(a2); - - return result==0; -} -#else -bool choldc(float64_t a[], int32_t n, float64_t p[]) -{ - void nrerror(char error_text[]); - int32_t i, j, k; - float64_t sum; - - for (i = 0; i < n; i++) - { - for (j = i; j < n; j++) - { - sum=a[n*i + j]; - - for (k=i-1; k>=0; k--) - sum -= a[n*i + k]*a[n*j + k]; - - if (i == j) - { - if (sum <= 0.0) - { - SG_SDEBUG("Choldc failed, matrix not positive definite") - sum = 0.0; - return false; - } - - p[i]=sqrt(sum); - - } - else - a[n*j + i] = sum/p[i]; - } - } - - return true; -} -#endif - -void cholsb( - float64_t a[], int32_t n, float64_t p[], float64_t b[], float64_t x[]) -{ - int32_t i, k; - float64_t sum; - - for (i=0; i=0; k--) sum -= a[n*i + k]*x[k]; - x[i]=sum/p[i]; - } - - for (i=n-1; i>=0; i--) { - sum=x[i]; - for (k=i+1; k=0; k--) sum -= a[n*i + k]*x[k]; - x[i]=sum/p[i]; - } -} - -void chol_backward( - float64_t a[], int32_t n, float64_t p[], float64_t b[], float64_t x[]) -{ - int32_t i, k; - float64_t sum; - - for (i=n-1; i>=0; i--) { - sum=b[i]; - for (k=i+1; k 0) { - s[i] = bound; - z[i] = sigma[i] + bound; - } - else { - s[i] = bound - sigma[i]; - z[i] = bound; - } - } - } - else { /* use default start settings */ - for (i=0; i= STATUS) { - SG_SDEBUG("counter | pri_inf | dual_inf | pri_obj | dual_obj | ") - SG_SDEBUG("sigfig | alpha | nu \n") - SG_SDEBUG("-------------------------------------------------------") - SG_SDEBUG("---------------------------\n") - } - - while (status == STILL_RUNNING) { - /* predictor */ - - /* put back original diagonal values */ - for (i=0; i counter_max) status = ITERATION_LIMIT; - if (sigfig > sigfig_max) status = OPTIMAL_SOLUTION; - if (primal_inf > 10e100) status = PRIMAL_INFEASIBLE; - if (dual_inf > 10e100) status = DUAL_INFEASIBLE; - if ((primal_inf > 10e100) & (dual_inf > 10e100)) status = PRIMAL_AND_DUAL_INFEASIBLE; - if (CMath::abs(primal_obj) > 10e100) status = PRIMAL_UNBOUNDED; - if (CMath::abs(dual_obj) > 10e100) status = DUAL_UNBOUNDED; - - /* write some nice routine to enforce the time limit if you - _really_ want, however it's quite useless as you can compute - the time from the maximum number of iterations as every - iteration costs one cholesky decomposition plus a couple of - backsubstitutions */ - - /* generate report */ - if ((verb >= FLOOD) | ((verb == STATUS) & (status != 0))) - SG_SDEBUG("%7i | %.2e | %.2e | % .2e | % .2e | %6.3f | %.4f | %.2e\n", - counter, primal_inf, dual_inf, primal_obj, dual_obj, - sigfig, alfa, mu); - - counter++; - - if (status == 0) { /* we may keep on going, otherwise - it'll cost one loop extra plus a - messed up main diagonal of h_x */ - /* intermediate variables (the ones with hat) */ - for (i=0; i= STATUS)) { - SG_SDEBUG("----------------------------------------------------------------------------------\n") - SG_SDEBUG("optimization converged\n") - } - - /* free memory */ - SG_FREE(workspace); - SG_FREE(diag_h_x); - SG_FREE(h_y); - SG_FREE(c_x); - SG_FREE(c_y); - SG_FREE(h_dot_x); - - SG_FREE(rho); - SG_FREE(nu); - SG_FREE(tau); - SG_FREE(sigma); - SG_FREE(gamma_z); - SG_FREE(gamma_s); - - SG_FREE(hat_nu); - SG_FREE(hat_tau); - - SG_FREE(delta_x); - SG_FREE(delta_y); - SG_FREE(delta_s); - SG_FREE(delta_z); - SG_FREE(delta_g); - SG_FREE(delta_t); - - SG_FREE(d); - - /* and return to sender */ - return status; -} -} -#endif //USE_GPL_SHOGUN diff --git a/src/shogun/lib/external/pr_loqo.h b/src/shogun/lib/external/pr_loqo.h deleted file mode 100644 index 335d7391e62..00000000000 --- a/src/shogun/lib/external/pr_loqo.h +++ /dev/null @@ -1,85 +0,0 @@ -/* - * This program is free software; you can redistribute it and/or modify - * it under the terms of the GNU General Public License as published by - * the Free Software Foundation; either version 3 of the License, or - * (at your option) any later version. - * - * Purpose: solves quadratic programming problem for pattern recognition - * for support vectors - * - * Written (W) 1997-1998 Alex J. Smola - * Written (W) 1999-2009 Soeren Sonnenburg - * Written (W) 1999-2008 Gunnar Raetsch - * Copyright (C) 1997-2009 Fraunhofer Institute FIRST and Max-Planck-Society - */ - -#include -#ifdef USE_GPL_SHOGUN - -namespace shogun -{ -/* verbosity levels */ - -#define QUIET 0 -#define STATUS 1 -#define FLOOD 2 - -/* status outputs */ - -#define STILL_RUNNING 0 -#define OPTIMAL_SOLUTION 1 -#define SUBOPTIMAL_SOLUTION 2 -#define ITERATION_LIMIT 3 -#define PRIMAL_INFEASIBLE 4 -#define DUAL_INFEASIBLE 5 -#define PRIMAL_AND_DUAL_INFEASIBLE 6 -#define INCONSISTENT 7 -#define PRIMAL_UNBOUNDED 8 -#define DUAL_UNBOUNDED 9 -#define TIME_LIMIT 10 - -/* - * solve the quadratic programming problem - * - * minimize c' * x + 1/2 x' * H * x - * subject to A*x = b - * l <= x <= u - * - * for a documentation see R. Vanderbei, LOQO: an Interior Point Code - * for Quadratic Programming - */ - -/** - * n : number of primal variables - * m : number of constraints (typically 1) - * h_x : dot product matrix (n.n) - * a : constraint matrix (n.m) - * b : constant term (m) - * l : lower bound (n) - * u : upper bound (m) - * - * primal : workspace for primal variables, has to be of size 3 n - * - * x = primal; n - * g = x + n; n - * t = g + n; n - * - * dual : workspace for dual variables, has to be of size m + 2 n - * - * y = dual; m - * z = y + m; n - * s = z + n; n - * - * verb : verbosity level - * sigfig_max : number of significant digits - * counter_max: stopping criterion - * restart : 1 if restart desired - * - */ -int32_t pr_loqo( - int32_t n, int32_t m, float64_t c[], float64_t h_x[], float64_t a[], - float64_t b[], float64_t l[], float64_t u[], float64_t primal[], - float64_t dual[], int32_t verb, float64_t sigfig_max, int32_t counter_max, - float64_t margin, float64_t bound, int32_t restart); -} -#endif //USE_GPL_SHOGUN diff --git a/src/shogun/lib/external/shogun_libsvm.cpp b/src/shogun/lib/external/shogun_libsvm.cpp index 07a10dc466f..e6cbb412915 100644 --- a/src/shogun/lib/external/shogun_libsvm.cpp +++ b/src/shogun/lib/external/shogun_libsvm.cpp @@ -35,12 +35,14 @@ #ifndef DOXYGEN_SHOULD_SKIP_THIS -#include -#include +#include +#include #include -#include +#include #include +#include #include +#include #include #include @@ -50,6 +52,8 @@ #include #include +#include + namespace shogun { @@ -289,7 +293,7 @@ LibSVMKernel::~LibSVMKernel() // class Solver { public: - Solver() {}; + Solver() : m_cancel_computation(false){}; virtual ~Solver() {}; struct SolutionInfo { @@ -321,6 +325,7 @@ class Solver { float64_t *G_bar; // gradient, if we treat free variables as 0 int32_t l; bool unshrink; // XXX + std::atomic m_cancel_computation; float64_t get_C(int32_t i) { @@ -342,10 +347,51 @@ class Solver { virtual int32_t select_working_set(int32_t &i, int32_t &j, float64_t &gap); virtual float64_t calculate_rho(); virtual void do_shrinking(); + + /* Custom implementation of signal handling */ + rxcpp::subscription connect_to_signal_handler(); + void reset_computation_variables(); +#ifndef SWIG + /** @return whether the algorithm needs to be stopped */ + SG_FORCED_INLINE bool cancel_computation() const + { + return m_cancel_computation.load(); + } +#endif + void on_pause() + { + } + void on_next() + { + m_cancel_computation.store(false); + } + void on_complete() + { + } + private: bool be_shrunk(int32_t i, float64_t Gmax1, float64_t Gmax2); }; +rxcpp::subscription Solver::connect_to_signal_handler() +{ + // Subscribe this algorithm to the signal handler + auto subscriber = rxcpp::make_subscriber( + [this](int i) { + if (i == SG_PAUSE_COMP) + this->on_pause(); + else + this->on_next(); + }, + [this]() { this->on_complete(); }); + return get_global_signal()->get_observable()->subscribe(subscriber); +} + +void Solver::reset_computation_variables() +{ + m_cancel_computation.store(false); +} + void Solver::swap_index(int32_t i, int32_t j) { Q->swap_index(i,j); @@ -402,6 +448,8 @@ void Solver::Solve( const schar *p_y, float64_t *p_alpha, float64_t p_Cp, float64_t p_Cn, float64_t p_eps, SolutionInfo* p_si, int32_t shrinking, bool use_bias) { + auto sub = connect_to_signal_handler(); + this->l = p_l; this->Q = &p_Q; QD=Q->get_QD(); @@ -429,9 +477,9 @@ void Solver::Solve( } // initialize gradient - CSignal::clear_cancel(); CTime start_time; { + auto pb = progress(range(l)); G = SG_MALLOC(float64_t, l); G_bar = SG_MALLOC(float64_t, l); int32_t i; @@ -442,7 +490,7 @@ void Solver::Solve( } SG_SINFO("Computing gradient for initial set of non-zero alphas\n") //CMath::display_vector(alpha, l, "alphas"); - for(i=0;imax_train_time > 0 && start_time.cur_time_diff() > Q->max_train_time) break; @@ -494,7 +542,8 @@ void Solver::Solve( counter = 1; // do shrinking next iteration } - SG_SABS_PROGRESS(gap, -CMath::log10(gap), -CMath::log10(1), -CMath::log10(eps), 6) + pb.print_absolute( + gap, -CMath::log10(gap), -CMath::log10(1), -CMath::log10(eps)); ++iter; @@ -672,6 +721,7 @@ void Solver::Solve( } #endif } + pb.complete_absolute(); // calculate rho @@ -708,6 +758,9 @@ void Solver::Solve( SG_FREE(active_set); SG_FREE(G); SG_FREE(G_bar); + + sub.unsubscribe(); + reset_computation_variables(); } // return 1 if already optimal, return 0 otherwise @@ -1829,7 +1882,7 @@ void solve_c_svc_weighted( if(prob->y[i] > 0) y[i] = +1; else y[i]=-1; } - WeightedSolver s = WeightedSolver(prob->C); + WeightedSolver s{prob->C}; s.Solve(l, SVC_Q(*prob,*param,y), minus_ones, y, alpha, Cp, Cn, param->eps, si, param->shrinking, param->use_bias); diff --git a/src/shogun/lib/external/ssl.cpp b/src/shogun/lib/external/ssl.cpp deleted file mode 100644 index a24296f8c6a..00000000000 --- a/src/shogun/lib/external/ssl.cpp +++ /dev/null @@ -1,1145 +0,0 @@ -/* Copyright 2006 Vikas Sindhwani (vikass@cs.uchicago.edu) - SVM-lin: Fast SVM Solvers for Supervised and Semi-supervised Learning - - This file is part of SVM-lin. - - SVM-lin is free software; you can redistribute it and/or modify - it under the terms of the GNU General Public License as published by - the Free Software Foundation; either version 2 of the License, or - (at your option) any later version. - - SVM-lin is distributed in the hope that it will be useful, - but WITHOUT ANY WARRANTY; without even the implied warranty of - MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the - GNU General Public License for more details. - - You should have received a copy of the GNU General Public License - along with SVM-lin (see gpl.txt); if not, write to the Free Software - Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA - */ - -#include -#include -#include -#include - -#include -#include -#include -#include - -namespace shogun -{ -void ssl_train(struct data *Data, - struct options *Options, - struct vector_double *Weights, - struct vector_double *Outputs) -{ - // initialize - initialize_ssl(Weights,Data->n,0.0); - initialize_ssl(Outputs,Data->m,0.0); - vector_int *Subset = SG_MALLOC(vector_int, 1); - initialize_ssl(Subset,Data->m); - // call the right algorithm - int32_t optimality = 0; - switch(Options->algo) - { - case -1: - SG_SINFO("Regularized Least Squares Regression (CGLS)\n") - optimality=CGLS(Data,Options,Subset,Weights,Outputs); - break; - case RLS: - SG_SINFO("Regularized Least Squares Classification (CGLS)\n") - optimality=CGLS(Data,Options,Subset,Weights,Outputs); - break; - case SVM: - SG_SINFO("Modified Finite Newton L2-SVM (L2-SVM-MFN)\n") - optimality=L2_SVM_MFN(Data,Options,Weights,Outputs,0); - break; - case TSVM: - SG_SINFO("Transductive L2-SVM (TSVM)\n") - optimality=TSVM_MFN(Data,Options,Weights,Outputs); - break; - case DA_SVM: - SG_SINFO("Deterministic Annealing Semi-supervised L2-SVM (DAS3VM)\n") - optimality=DA_S3VM(Data,Options,Weights,Outputs); - break; - default: - SG_SERROR("Algorithm unspecified\n") - } - - if (!optimality) - SG_SWARNING("SSL-Algorithm terminated without reaching optimum.\n") - - SG_FREE(Subset->vec); - SG_FREE(Subset); - return; -} - -int32_t CGLS( - const struct data *Data, const struct options *Options, - const struct vector_int *Subset, struct vector_double *Weights, - struct vector_double *Outputs) -{ - SG_SDEBUG("CGLS starting...") - - /* Disassemble the structures */ - int32_t active = Subset->d; - int32_t *J = Subset->vec; - CDotFeatures* features=Data->features; - float64_t *Y = Data->Y; - float64_t *C = Data->C; - int32_t n = Data->n; - float64_t lambda = Options->lambda; - int32_t cgitermax = Options->cgitermax; - float64_t epsilon = Options->epsilon; - float64_t *beta = Weights->vec; - float64_t *o = Outputs->vec; - // initialize z - float64_t *z = SG_MALLOC(float64_t, active); - float64_t *q = SG_MALLOC(float64_t, active); - int32_t ii=0; - for (int32_t i = active ; i-- ;){ - ii=J[i]; - z[i] = C[ii]*(Y[ii] - o[ii]); - } - float64_t *r = SG_MALLOC(float64_t, n); - for (int32_t i = n ; i-- ;) - r[i] = 0.0; - for (register int32_t j=0; j < active; j++) - { - features->add_to_dense_vec(z[j], J[j], r, n-1); - r[n-1]+=Options->bias*z[j]; //bias (modelled as last dim) - } - float64_t *p = SG_MALLOC(float64_t, n); - float64_t omega1 = 0.0; - for (int32_t i = n ; i-- ;) - { - r[i] -= lambda*beta[i]; - p[i] = r[i]; - omega1 += r[i]*r[i]; - } - float64_t omega_p = omega1; - float64_t omega_q = 0.0; - float64_t inv_omega2 = 1/omega1; - float64_t scale = 0.0; - float64_t omega_z=0.0; - float64_t gamma = 0.0; - int32_t cgiter = 0; - int32_t optimality = 0; - float64_t epsilon2 = epsilon*epsilon; - // iterate - while(cgiter < cgitermax) - { - cgiter++; - omega_q=0.0; - float64_t t=0.0; - register int32_t i,j; - // #pragma omp parallel for private(i,j) - for (i=0; i < active; i++) - { - ii=J[i]; - t=features->dense_dot(ii, p, n-1); - t+=Options->bias*p[n-1]; //bias (modelled as last dim) - q[i]=t; - omega_q += C[ii]*t*t; - } - gamma = omega1/(lambda*omega_p + omega_q); - inv_omega2 = 1/omega1; - for (i = n ; i-- ;) - { - r[i] = 0.0; - beta[i] += gamma*p[i]; - } - omega_z=0.0; - for (i = active ; i-- ;) - { - ii=J[i]; - o[ii] += gamma*q[i]; - z[i] -= gamma*C[ii]*q[i]; - omega_z+=z[i]*z[i]; - } - for (j=0; j < active; j++) - { - t=z[j]; - - features->add_to_dense_vec(t, J[j], r, n-1); - r[n-1]+=Options->bias*t; //bias (modelled as last dim) - } - omega1 = 0.0; - for (i = n ; i-- ;) - { - r[i] -= lambda*beta[i]; - omega1 += r[i]*r[i]; - } - if(omega1 < epsilon2*omega_z) - { - optimality=1; - break; - } - omega_p=0.0; - scale=omega1*inv_omega2; - for(i = n ; i-- ;) - { - p[i] = r[i] + p[i]*scale; - omega_p += p[i]*p[i]; - } - } - SG_SDEBUG("...Done.") - SG_SINFO("CGLS converged in %d iteration(s)", cgiter) - - SG_FREE(z); - SG_FREE(q); - SG_FREE(r); - SG_FREE(p); - return optimality; -} - -int32_t L2_SVM_MFN( - const struct data *Data, struct options *Options, - struct vector_double *Weights, struct vector_double *Outputs, - int32_t ini) -{ - /* Disassemble the structures */ - CDotFeatures* features=Data->features; - float64_t *Y = Data->Y; - float64_t *C = Data->C; - int32_t n = Data->n; - int32_t m = Data->m; - float64_t lambda = Options->lambda; - float64_t epsilon; - float64_t *w = Weights->vec; - float64_t *o = Outputs->vec; - float64_t F_old = 0.0; - float64_t F = 0.0; - float64_t diff=0.0; - vector_int *ActiveSubset = SG_MALLOC(vector_int, 1); - ActiveSubset->vec = SG_MALLOC(int32_t, m); - ActiveSubset->d = m; - // initialize - if(ini==0) { - epsilon=BIG_EPSILON; - Options->cgitermax=SMALL_CGITERMAX; - Options->epsilon=BIG_EPSILON; - } - else {epsilon = Options->epsilon;} - for (int32_t i=0;i0) - { - ActiveSubset->vec[active]=i; - active++; - F+=0.5*C[i]*diff*diff; - } - else - { - ActiveSubset->vec[inactive]=i; - inactive--; - } - } - ActiveSubset->d=active; - int32_t iter=0; - int32_t opt=0; - int32_t opt2=0; - vector_double *Weights_bar = SG_MALLOC(vector_double, 1); - vector_double *Outputs_bar = SG_MALLOC(vector_double, 1); - float64_t *w_bar = SG_MALLOC(float64_t, n); - float64_t *o_bar = SG_MALLOC(float64_t, m); - Weights_bar->vec=w_bar; - Outputs_bar->vec=o_bar; - Weights_bar->d=n; - Outputs_bar->d=m; - float64_t delta=0.0; - float64_t t=0.0; - int32_t ii = 0; - while(itervec[i]; - - t=features->dense_dot(ii, w_bar, n-1); - t+=Options->bias*w_bar[n-1]; //bias (modelled as last dim) - - o_bar[ii]=t; - } - if(ini==0) {Options->cgitermax=CGITERMAX; ini=1;}; - opt2=1; - for (int32_t i=0;ivec[i]; - if(i=1-epsilon)); - if(opt2==0) break; - } - if(opt && opt2) // l - { - if(epsilon==BIG_EPSILON) - { - epsilon=EPSILON; - Options->epsilon=EPSILON; - SG_SDEBUG("epsilon = %f case converged (speedup heuristic 2). Continuing with epsilon=%f", BIG_EPSILON , EPSILON) - continue; - } - else - { - for (int32_t i=n; i-- ;) - w[i]=w_bar[i]; - for (int32_t i=m; i-- ;) - o[i]=o_bar[i]; - SG_FREE(ActiveSubset->vec); - SG_FREE(ActiveSubset); - SG_FREE(o_bar); - SG_FREE(w_bar); - SG_FREE(Weights_bar); - SG_FREE(Outputs_bar); - SG_SINFO("L2_SVM_MFN converged (optimality) in %d", iter) - return 1; - } - } - delta=line_search(w,w_bar,lambda,o,o_bar,Y,C,n,m); - SG_SDEBUG("LINE_SEARCH delta = %f\n", delta) - F_old=F; - F=0.0; - for (int32_t i=n; i-- ;) { - w[i]+=delta*(w_bar[i]-w[i]); - F+=w[i]*w[i]; - } - F=0.5*lambda*F; - active=0; - inactive=m-1; - for (int32_t i=0; i0) - { - ActiveSubset->vec[active]=i; - active++; - F+=0.5*C[i]*diff*diff; - } - else - { - ActiveSubset->vec[inactive]=i; - inactive--; - } - } - ActiveSubset->d=active; - if(CMath::abs(F-F_old)vec); - SG_FREE(ActiveSubset); - SG_FREE(o_bar); - SG_FREE(w_bar); - SG_FREE(Weights_bar); - SG_FREE(Outputs_bar); - SG_SINFO("L2_SVM_MFN converged (rel. criterion) in %d iterations", iter) - return 2; - } - } - SG_FREE(ActiveSubset->vec); - SG_FREE(ActiveSubset); - SG_FREE(o_bar); - SG_FREE(w_bar); - SG_FREE(Weights_bar); - SG_FREE(Outputs_bar); - SG_SINFO("L2_SVM_MFN converged (max iter exceeded) in %d iterations", iter) - return 0; -} - -float64_t line_search(float64_t *w, - float64_t *w_bar, - float64_t lambda, - float64_t *o, - float64_t *o_bar, - float64_t *Y, - float64_t *C, - int32_t d, /* data dimensionality -- 'n' */ - int32_t l) /* number of examples */ -{ - float64_t omegaL = 0.0; - float64_t omegaR = 0.0; - float64_t diff=0.0; - for(int32_t i=d; i--; ) - { - diff=w_bar[i]-w[i]; - omegaL+=w[i]*diff; - omegaR+=w_bar[i]*diff; - } - omegaL=lambda*omegaL; - omegaR=lambda*omegaR; - float64_t L=0.0; - float64_t R=0.0; - int32_t ii=0; - for (int32_t i=0;i0) - { - deltas[p].delta=(1-Y[i]*o[i])/diff; - deltas[p].index=i; - deltas[p].s=-1; - p++; - } - } - else - { - if(diff<0) - { - deltas[p].delta=(1-Y[i]*o[i])/diff; - deltas[p].index=i; - deltas[p].s=1; - p++; - } - } - } - std::sort(deltas,deltas+p); - float64_t delta_prime=0.0; - for (int32_t i=0;i=0) - break; - ii=deltas[i].index; - diff=(deltas[i].s)*C[ii]*(o_bar[ii]-o[ii]); - L+=diff*(o[ii]-Y[ii]); - R+=diff*(o_bar[ii]-Y[ii]); - } - SG_FREE(deltas); - return (-L/(R-L)); -} - -int32_t TSVM_MFN( - const struct data *Data, struct options *Options, - struct vector_double *Weights, struct vector_double *Outputs) -{ - /* Setup labeled-only examples and train L2_SVM_MFN */ - struct data *Data_Labeled = SG_MALLOC(data, 1); - struct vector_double *Outputs_Labeled = SG_MALLOC(vector_double, 1); - initialize_ssl(Outputs_Labeled,Data->l,0.0); - SG_SDEBUG("Initializing weights, unknown labels") - GetLabeledData(Data_Labeled,Data); /* gets labeled data and sets C=1/l */ - L2_SVM_MFN(Data_Labeled, Options, Weights,Outputs_Labeled,0); - ///FIXME Clear(Data_Labeled); - /* Use this weight vector to classify R*u unlabeled examples as - positive*/ - int32_t p=0,q=0; - float64_t t=0.0; - int32_t *JU = SG_MALLOC(int32_t, Data->u); - float64_t *ou = SG_MALLOC(float64_t, Data->u); - float64_t lambda_0 = TSVM_LAMBDA_SMALL; - for (int32_t i=0;im;i++) - { - if(Data->Y[i]==0.0) - { - t=Data->features->dense_dot(i, Weights->vec, Data->n-1); - t+=Options->bias*Weights->vec[Data->n-1]; //bias (modelled as last dim) - - Outputs->vec[i]=t; - Data->C[i]=lambda_0*1.0/Data->u; - JU[q]=i; - ou[q]=t; - q++; - } - else - { - Outputs->vec[i]=Outputs_Labeled->vec[p]; - Data->C[i]=1.0/Data->l; - p++; - } - } - std::nth_element(ou,ou+int32_t((1-Options->R)*Data->u-1),ou+Data->u); - float64_t thresh=*(ou+int32_t((1-Options->R)*Data->u)-1); - SG_FREE(ou); - for (int32_t i=0;iu;i++) - { - if(Outputs->vec[JU[i]]>thresh) - Data->Y[JU[i]]=1.0; - else - Data->Y[JU[i]]=-1.0; - } - for (int32_t i=0;in;i++) - Weights->vec[i]=0.0; - for (int32_t i=0;im;i++) - Outputs->vec[i]=0.0; - L2_SVM_MFN(Data,Options,Weights,Outputs,0); - int32_t num_switches=0; - int32_t s=0; - int32_t last_round=0; - while(lambda_0 <= Options->lambda_u) - { - int32_t iter2=0; - while(1){ - s=switch_labels(Data->Y,Outputs->vec,JU,Data->u,Options->S); - if(s==0) break; - iter2++; - SG_SDEBUG("****** lambda_0 = %f iteration = %d ************************************\n", lambda_0, iter2) - SG_SDEBUG("Optimizing unknown labels. switched %d labels.\n") - num_switches+=s; - SG_SDEBUG("Optimizing weights\n") - L2_SVM_MFN(Data,Options,Weights,Outputs,1); - } - if(last_round==1) break; - lambda_0=TSVM_ANNEALING_RATE*lambda_0; - if(lambda_0 >= Options->lambda_u) {lambda_0 = Options->lambda_u; last_round=1;} - for (int32_t i=0;iu;i++) - Data->C[JU[i]]=lambda_0*1.0/Data->u; - SG_SDEBUG("****** lambda0 increased to %f%% of lambda_u = %f ************************\n", lambda_0*100/Options->lambda_u, Options->lambda_u) - SG_SDEBUG("Optimizing weights\n") - L2_SVM_MFN(Data,Options,Weights,Outputs,1); - } - SG_SDEBUG("Total Number of Switches = %d\n", num_switches) - /* reset labels */ - for (int32_t i=0;iu;i++) Data->Y[JU[i]] = 0.0; - float64_t F = transductive_cost(norm_square(Weights),Data->Y,Outputs->vec,Outputs->d,Options->lambda,Options->lambda_u); - SG_SDEBUG("Objective Value = %f\n",F) - delete [] JU; - return num_switches; -} - -int32_t switch_labels(float64_t* Y, float64_t* o, int32_t* JU, int32_t u, int32_t S) -{ - int32_t npos=0; - int32_t nneg=0; - for (int32_t i=0;i0) && (o[JU[i]]<1.0)) npos++; - if((Y[JU[i]]<0) && (-o[JU[i]]<1.0)) nneg++; - } - Delta* positive=SG_MALLOC(Delta, npos); - Delta* negative=SG_MALLOC(Delta, nneg); - int32_t p=0; - int32_t n=0; - int32_t ii=0; - for (int32_t i=0;i0.0) && (o[ii]<1.0)) { - positive[p].delta=o[ii]; - positive[p].index=ii; - positive[p].s=0; - p++;}; - if((Y[ii]<0.0) && (-o[ii]<1.0)) - { - negative[n].delta=-o[ii]; - negative[n].index=ii; - negative[n].s=0; - n++;}; - } - std::sort(positive,positive+npos); - std::sort(negative,negative+nneg); - int32_t s=-1; - while(1) - { - s++; - if((s>=S) || (positive[s].delta>=-negative[s].delta) || (s>=npos) || (s>=nneg)) - break; - Y[positive[s].index]=-1.0; - Y[negative[s].index]= 1.0; - } - SG_FREE(positive); - SG_FREE(negative); - return s; -} - -int32_t DA_S3VM( - struct data *Data, struct options *Options, struct vector_double *Weights, - struct vector_double *Outputs) -{ - float64_t T = DA_INIT_TEMP*Options->lambda_u; - int32_t iter1 = 0, iter2 =0; - float64_t *p = SG_MALLOC(float64_t, Data->u); - float64_t *q = SG_MALLOC(float64_t, Data->u); - float64_t *g = SG_MALLOC(float64_t, Data->u); - float64_t F,F_min; - float64_t *w_min = SG_MALLOC(float64_t, Data->n); - float64_t *o_min = SG_MALLOC(float64_t, Data->m); - float64_t *w = Weights->vec; - float64_t *o = Outputs->vec; - float64_t kl_divergence = 1.0; - /*initialize */ - SG_SDEBUG("Initializing weights, p") - for (int32_t i=0;iu; i++) - p[i] = Options->R; - /* record which examples are unlabeled */ - int32_t *JU = SG_MALLOC(int32_t, Data->u); - int32_t j=0; - for(int32_t i=0;im;i++) - { - if(Data->Y[i]==0.0) - {JU[j]=i;j++;} - } - float64_t H = entropy(p,Data->u); - optimize_w(Data,p,Options,Weights,Outputs,0); - F = transductive_cost(norm_square(Weights),Data->Y,Outputs->vec,Outputs->d,Options->lambda,Options->lambda_u); - F_min = F; - for (int32_t i=0;id;i++) - w_min[i]=w[i]; - for (int32_t i=0;id;i++) - o_min[i]=o[i]; - while((iter1 < DA_OUTER_ITERMAX) && (H > Options->epsilon)) - { - iter1++; - iter2=0; - kl_divergence=1.0; - while((iter2 < DA_INNER_ITERMAX) && (kl_divergence > Options->epsilon)) - { - iter2++; - for (int32_t i=0;iu;i++) - { - q[i]=p[i]; - g[i] = Options->lambda_u*((o[JU[i]] > 1 ? 0 : (1 - o[JU[i]])*(1 - o[JU[i]])) - (o[JU[i]]< -1 ? 0 : (1 + o[JU[i]])*(1 + o[JU[i]]))); - } - SG_SDEBUG("Optimizing p.\n") - optimize_p(g,Data->u,T,Options->R,p); - kl_divergence=KL(p,q,Data->u); - SG_SDEBUG("Optimizing weights\n") - optimize_w(Data,p,Options,Weights,Outputs,1); - F = transductive_cost(norm_square(Weights),Data->Y,Outputs->vec,Outputs->d,Options->lambda,Options->lambda_u); - if(F < F_min) - { - F_min = F; - for (int32_t i=0;id;i++) - w_min[i]=w[i]; - for (int32_t i=0;id;i++) - o_min[i]=o[i]; - } - SG_SDEBUG("***** outer_iter = %d T = %g inner_iter = %d kl = %g cost = %g *****\n",iter1,T,iter2,kl_divergence,F) - } - H = entropy(p,Data->u); - SG_SDEBUG("***** Finished outer_iter = %d T = %g Entropy = %g ***\n", iter1,T,H) - T = T/DA_ANNEALING_RATE; - } - for (int32_t i=0;id;i++) - w[i]=w_min[i]; - for (int32_t i=0;id;i++) - o[i]=o_min[i]; - /* may want to reset the original Y */ - SG_FREE(p); - SG_FREE(q); - SG_FREE(g); - SG_FREE(JU); - SG_FREE(w_min); - SG_FREE(o_min); - SG_SINFO("(min) Objective Value = %f", F_min) - return 1; -} - -int32_t optimize_w( - const struct data *Data, const float64_t *p, struct options *Options, - struct vector_double *Weights, struct vector_double *Outputs, int32_t ini) -{ - int32_t i,j; - CDotFeatures* features=Data->features; - int32_t n = Data->n; - int32_t m = Data->m; - int32_t u = Data->u; - float64_t lambda = Options->lambda; - float64_t epsilon; - float64_t *w = Weights->vec; - float64_t *o = SG_MALLOC(float64_t, m+u); - float64_t *Y = SG_MALLOC(float64_t, m+u); - float64_t *C = SG_MALLOC(float64_t, m+u); - int32_t *labeled_indices = SG_MALLOC(int32_t, m); - float64_t F_old = 0.0; - float64_t F = 0.0; - float64_t diff=0.0; - float64_t lambda_u_by_u = Options->lambda_u/u; - vector_int *ActiveSubset = SG_MALLOC(vector_int, 1); - ActiveSubset->vec = SG_MALLOC(int32_t, m); - ActiveSubset->d = m; - // initialize - if(ini==0) - { - epsilon=BIG_EPSILON; - Options->cgitermax=SMALL_CGITERMAX; - Options->epsilon=BIG_EPSILON;} - else {epsilon = Options->epsilon;} - - for(i=0;ivec[i]; - if(Data->Y[i]==0.0) - { - labeled_indices[i]=0; - o[m+j]=o[i]; - Y[i]=1; - Y[m+j]=-1; - C[i]=lambda_u_by_u*p[j]; - C[m+j]=lambda_u_by_u*(1-p[j]); - ActiveSubset->vec[active]=i; - active++; - diff = 1 - CMath::abs(o[i]); - if(diff>0) - { - Data->Y[i] = 2*p[j]-1; - Data->C[i] = lambda_u_by_u; - temp1 = (1 - o[i]); - temp2 = (1 + o[i]); - F+=lambda_u_by_u*(p[j]*temp1*temp1 + (1-p[j])*temp2*temp2); - } - else - { - if(o[i]>0) - { - Data->Y[i] = -1.0; - Data->C[i] = C[m+j]; - } - else - { - Data->Y[i] = 1.0; - Data->C[i] = C[i]; - } - temp1 = (1-Data->Y[i]*o[i]); - F+= Data->C[i]*temp1*temp1; - } - j++; - } - else - { - labeled_indices[i]=1; - Y[i]=Data->Y[i]; - C[i]=1.0/Data->l; - Data->C[i]=1.0/Data->l; - diff=1-Data->Y[i]*o[i]; - if(diff>0) - { - ActiveSubset->vec[active]=i; - active++; - F+=Data->C[i]*diff*diff; - } - else - { - ActiveSubset->vec[inactive]=i; - inactive--; - } - } - } - F=0.5*F; - ActiveSubset->d=active; - int32_t iter=0; - int32_t opt=0; - int32_t opt2=0; - vector_double *Weights_bar = SG_MALLOC(vector_double, 1); - vector_double *Outputs_bar = SG_MALLOC(vector_double, 1); - float64_t *w_bar = SG_MALLOC(float64_t, n); - float64_t *o_bar = SG_MALLOC(float64_t, m+u); - Weights_bar->vec=w_bar; - Outputs_bar->vec=o_bar; - Weights_bar->d=n; - Outputs_bar->d=m; /* read only the top m ; bottom u will be copies */ - float64_t delta=0.0; - float64_t t=0.0; - int32_t ii = 0; - while(itervec[i]; - t=features->dense_dot(ii, w_bar, n-1); - t+=Options->bias*w_bar[n-1]; //bias (modelled as last dim) - - o_bar[ii]=t; - } - // make o_bar consistent in the bottom half - j=0; - for(i=0; icgitermax=CGITERMAX; ini=1;}; - opt2=1; - for(i=0; i < m ;i++) - { - ii=ActiveSubset->vec[i]; - if(iY[ii]*o_bar[ii]<=1+epsilon)); - else - { - if(CMath::abs(o[ii])<1) - opt2=(opt2 && (CMath::abs(o_bar[ii])<=1+epsilon)); - else - opt2=(opt2 && (CMath::abs(o_bar[ii])>=1-epsilon)); - } - } - else - opt2=(opt2 && (Data->Y[ii]*o_bar[ii]>=1-epsilon)); - if(opt2==0) break; - } - if(opt && opt2) // l - { - if(epsilon==BIG_EPSILON) - { - epsilon=EPSILON; - Options->epsilon=EPSILON; - SG_SDEBUG("epsilon = %f case converged (speedup heuristic 2). Continuing with epsilon=%f\n", BIG_EPSILON, EPSILON) - continue; - } - else - { - for(i=n; i-- ;) - w[i]=w_bar[i]; - for(i=m; i-- ;) - Outputs->vec[i]=o_bar[i]; - for(i=m; i-- ;) - { - if(labeled_indices[i]==0) - Data->Y[i]=0.0; - } - SG_FREE(ActiveSubset->vec); - SG_FREE(ActiveSubset); - SG_FREE(o_bar); - SG_FREE(w_bar); - SG_FREE(o); - SG_FREE(Weights_bar); - SG_FREE(Outputs_bar); - SG_FREE(Y); - SG_FREE(C); - SG_FREE(labeled_indices); - SG_SINFO("L2_SVM_MFN converged in %d iteration(s)", iter) - return 1; - } - } - - delta=line_search(w,w_bar,lambda,o,o_bar,Y,C,n,m+u); - SG_SDEBUG("LINE_SEARCH delta = %f", delta) - F_old=F; - F=0.0; - for(i=0;ivec[active]=i; - active++; - diff = 1 - CMath::abs(o[i]); - if(diff>0) - { - Data->Y[i] = 2*p[j]-1; - Data->C[i] = lambda_u_by_u; - temp1 = (1 - o[i]); - temp2 = (1 + o[i]); - F+=lambda_u_by_u*(p[j]*temp1*temp1 + (1-p[j])*temp2*temp2); - } - else - { - if(o[i]>0) - { - Data->Y[i] = -1; - Data->C[i] = C[m+j]; - } - else - { - Data->Y[i] = +1; - Data->C[i] = C[i]; - } - temp1=(1-Data->Y[i]*o[i]); - F+= Data->C[i]*temp1*temp1; - } - j++; - } - else - { - diff=1-Data->Y[i]*o[i]; - if(diff>0) - { - ActiveSubset->vec[active]=i; - active++; - F+=Data->C[i]*diff*diff; - } - else - { - ActiveSubset->vec[inactive]=i; - inactive--; - } - } - } - F=0.5*F; - ActiveSubset->d=active; - if(CMath::abs(F-F_old)vec[i]=o[i]; - if(labeled_indices[i]==0) - Data->Y[i]=0.0; - } - SG_FREE(ActiveSubset->vec); - SG_FREE(labeled_indices); - SG_FREE(ActiveSubset); - SG_FREE(o_bar); - SG_FREE(w_bar); - SG_FREE(o); - SG_FREE(Weights_bar); - SG_FREE(Outputs_bar); - SG_FREE(Y); - SG_FREE(C); - SG_SINFO("L2_SVM_MFN converged in %d iterations", iter) - return 0; -} - -void optimize_p( - const float64_t* g, int32_t u, float64_t T, float64_t r, float64_t* p) -{ - int32_t iter=0; - float64_t epsilon=1e-10; - int32_t maxiter=500; - float64_t nu_minus=g[0]; - float64_t nu_plus=g[0]; - for (int32_t i=0;inu_plus) nu_plus=g[i]; - }; - - float64_t b=T*log((1-r)/r); - nu_minus-=b; - nu_plus-=b; - float64_t nu=(nu_plus+nu_minus)/2; - float64_t Bnu=0.0; - float64_t BnuPrime=0.0; - float64_t s=0.0; - float64_t tmp=0.0; - for (int32_t i=0;iepsilon) && (iter < maxiter)) - { - iter++; - if(CMath::abs(BnuPrime)>0.0) - nuHat=nu-Bnu/BnuPrime; - if((CMath::abs(BnuPrime) > 0.0) | (nuHat>nu_plus) | (nuHat < nu_minus)) - nu=(nu_minus+nu_plus)/2.0; - else - nu=nuHat; - Bnu=0.0; - BnuPrime=0.0; - for(int32_t i=0;iepsilon) - SG_SWARNING("Warning (Root): root not found to required precision\n") - - for (int32_t i=0;i 1 ? 0 : (1 - CMath::abs(o))*(1 - CMath::abs(o)); u++;} - else - {F2 += y*o > 1 ? 0 : (1-y*o)*(1-y*o); l++;} - } - float64_t F; - F = 0.5*(lambda*normWeights + lambda_u*F1/u + F2/l); - return F; -} - -float64_t entropy(const float64_t *p, int32_t u) -{ - float64_t h=0.0; - float64_t q=0.0; - for (int32_t i=0;i0 && q<1) - h+= -(q*CMath::log2(q) + (1-q)*CMath::log2(1-q)); - } - return h/u; -} - -float64_t KL(const float64_t *p, const float64_t *q, int32_t u) -{ - float64_t h=0.0; - float64_t p1=0.0; - float64_t q1=0.0; - float64_t g=0.0; - for (int32_t i=0;i1-1e-8) p1-=1e-8; - if(p1<1-1e-8) p1+=1e-8; - if(q1>1-1e-8) q1-=1e-8; - if(q1<1-1e-8) q1+=1e-8; - g= (p1*CMath::log2(p1/q1) + (1-p1)*CMath::log2((1-p1)/(1-q1))); - if(CMath::abs(g)<1e-12 || CMath::is_nan(g)) g=0.0; - h+=g; - } - return h/u; -} - -/********************** UTILITIES ********************/ -float64_t norm_square(const vector_double *A) -{ - float64_t x=0.0, t=0.0; - for(int32_t i=0;id;i++) - { - t=A->vec[i]; - x+=t*t; - } - return x; -} - -void initialize_ssl(struct vector_double *A, int32_t k, float64_t a) -{ - float64_t *vec = SG_MALLOC(float64_t, k); - for (int32_t i=0;ivec = vec; - A->d = k; - return; -} - -void initialize_ssl(struct vector_int *A, int32_t k) -{ - int32_t *vec = SG_MALLOC(int32_t, k); - for(int32_t i=0;ivec = vec; - A->d = k; - return; -} - -void GetLabeledData(struct data *D, const struct data *Data) -{ - /*FIXME - int32_t *J = SG_MALLOC(int, Data->l); - D->C = SG_MALLOC(float64_t, Data->l); - D->Y = SG_MALLOC(float64_t, Data->l); - int32_t nz=0; - int32_t k=0; - int32_t rowptrs_=Data->l; - for(int32_t i=0;im;i++) - { - if(Data->Y[i]!=0.0) - { - J[k]=i; - D->Y[k]=Data->Y[i]; - D->C[k]=1.0/Data->l; - nz+=(Data->rowptr[i+1] - Data->rowptr[i]); - k++; - } - } - D->val = SG_MALLOC(float64_t, nz); - D->colind = SG_MALLOC(int32_t, nz); - D->rowptr = new int32_trowptrs_+1]; - nz=0; - for(int32_t i=0;il;i++) - { - D->rowptr[i]=nz; - for(int32_t j=Data->rowptr[J[i]];jrowptr[J[i]+1];j++) - { - D->val[nz] = Data->val[j]; - D->colind[nz] = Data->colind[j]; - nz++; - } - } - D->rowptr[rowptrs_]=nz; - D->nz=nz; - D->l=Data->l; - D->m=Data->l; - D->n=Data->n; - D->u=0; - SG_FREE(J);*/ -} -} diff --git a/src/shogun/lib/external/ssl.h b/src/shogun/lib/external/ssl.h deleted file mode 100644 index 19d9b5185f8..00000000000 --- a/src/shogun/lib/external/ssl.h +++ /dev/null @@ -1,236 +0,0 @@ -/* Copyright 2006 Vikas Sindhwani (vikass@cs.uchicago.edu) - SVM-lin: Fast SVM Solvers for Supervised and Semi-supervised Learning - - This file is part of SVM-lin. - - SVM-lin is free software; you can redistribute it and/or modify - it under the terms of the GNU General Public License as published by - the Free Software Foundation; either version 2 of the License, or - (at your option) any later version. - - SVM-lin is distributed in the hope that it will be useful, - but WITHOUT ANY WARRANTY; without even the implied warranty of - MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the - GNU General Public License for more details. - - You should have received a copy of the GNU General Public License - along with SVM-lin (see gpl.txt); if not, write to the Free Software - Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA - */ -#ifndef DOXYGEN_SHOULD_SKIP_THIS - -#ifndef _SSL_H -#define _SSL_H - -/* OPTIMIZATION CONSTANTS */ -#define CGITERMAX 10000 /* maximum number of CGLS iterations */ -#define SMALL_CGITERMAX 10 /* for heuristic 1 in reference [2] */ -#define EPSILON 1e-6 /* most tolerances are set to this value */ -#define BIG_EPSILON 0.01 /* for heuristic 2 in reference [2] */ -#define RELATIVE_STOP_EPS 1e-9 /* for L2-SVM-MFN relative stopping criterion */ -#define MFNITERMAX 50 /* maximum number of MFN iterations */ -#define TSVM_ANNEALING_RATE 1.5 /* rate at which lambda_u is increased in TSVM */ -#define TSVM_LAMBDA_SMALL 1e-5 /* lambda_u starts from this value */ -#define DA_ANNEALING_RATE 1.5 /* annealing rate for DA */ -#define DA_INIT_TEMP 10 /* initial temperature relative to lambda_u */ -#define DA_INNER_ITERMAX 100 /* maximum fixed temperature iterations for DA */ -#define DA_OUTER_ITERMAX 30 /* maximum number of outer loops for DA */ - -#include - -#include -#include - -namespace shogun -{ -/** Data: Input examples are stored in sparse (Compressed Row Storage) format */ -struct data -{ - /** number of examples */ - int32_t m; - /** number of labeled examples */ - int32_t l; - /** number of unlabeled examples l+u = m */ - int32_t u; - /** number of features */ - int32_t n; - /** number of non-zeros */ - int32_t nz; - - /** features */ - shogun::CDotFeatures* features; - /** labels */ - float64_t *Y; - /** cost associated with each example */ - float64_t *C; -}; - -/** defines a vector of doubles */ -struct vector_double -{ - /** number of elements */ - int32_t d; - /** ptr to vector elements*/ - float64_t *vec; -}; - -/** defines a vector of ints for index subsets */ -struct vector_int -{ - /** number of elements */ - int32_t d; - /** ptr to vector elements */ - int32_t *vec; -}; - -enum { RLS, SVM, TSVM, DA_SVM }; /* currently implemented algorithms */ - -/** various options user + internal optimisation */ -struct options -{ - /* user options */ - /** regularization parameter */ - int32_t algo; - /** regularization parameter */ - float64_t lambda; - /** regularization parameter over unlabeled examples */ - float64_t lambda_u; - /** maximum number of TSVM switches per fixed-weight label optimization */ - int32_t S; - /** expected fraction of unlabeled examples in positive class */ - float64_t R; - /** cost for positive examples */ - float64_t Cp; - /** cost for negative examples */ - float64_t Cn; - - /* internal optimization options */ - /** all tolerances */ - float64_t epsilon; - /** max iterations for CGLS */ - int32_t cgitermax; - /** max iterations for L2_SVM_MFN */ - int32_t mfnitermax; - - /** 1.0 if bias is to be used, 0.0 otherwise */ - float64_t bias; -}; - -/** used in line search */ -class Delta { - public: - /** default constructor */ - Delta() { delta=0.0; index=0;s=0; } - - /** delta */ - float64_t delta; - /** index */ - int32_t index; - /** s */ - int32_t s; -}; - -inline bool operator<(const Delta& a , const Delta& b) -{ - return (a.delta < b.delta); -} - -void initialize_ssl(struct vector_double *A, int32_t k, float64_t a); -/* initializes a vector_double to be of length k, all elements set to a */ -void initialize_ssl(struct vector_int *A, int32_t k); -/* initializes a vector_int to be of length k, elements set to 1,2..k. */ -void GetLabeledData(struct data *Data_Labeled, const struct data *Data); -/* extracts labeled data from Data and copies it into Data_Labeled */ -float64_t norm_square(const vector_double *A); /* returns squared length of A */ - -/* ssl_train: takes data, options, uninitialized weight and output - vector_doubles, routes it to the algorithm */ -/* the learnt weight vector and the outputs it gives on the data matrix are saved */ -void ssl_train( - struct data *Data, - struct options *Options, - struct vector_double *W, /* weight vector */ - struct vector_double *O); /* output vector */ - -/* svmlin algorithms and their subroutines */ - -/* Conjugate Gradient for Sparse Linear Least Squares Problems */ -/* Solves: min_w 0.5*Options->lamda*w'*w + 0.5*sum_{i in Subset} Data->C[i] (Y[i]- w' x_i)^2 */ -/* over a subset of examples x_i specified by vector_int Subset */ -int32_t CGLS( - const struct data *Data, - const struct options *Options, - const struct vector_int *Subset, - struct vector_double *Weights, - struct vector_double *Outputs); - -/* Linear Modified Finite Newton L2-SVM*/ -/* Solves: min_w 0.5*Options->lamda*w'*w + 0.5*sum_i Data->C[i] max(0,1 - Y[i] w' x_i)^2 */ -int32_t L2_SVM_MFN( - const struct data *Data, - struct options *Options, - struct vector_double *Weights, - struct vector_double *Outputs, - int32_t ini); /* use ini=0 if no good starting guess for Weights, else 1 */ - -float64_t line_search( - float64_t *w, - float64_t *w_bar, - float64_t lambda, - float64_t *o, - float64_t *o_bar, - float64_t *Y, - float64_t *C, - int32_t d, - int32_t l); - -/* Transductive L2-SVM */ -/* Solves : min_(w, Y[i],i in UNlabeled) 0.5*Options->lamda*w'*w + 0.5*(1/Data->l)*sum_{i in labeled} max(0,1 - Y[i] w' x_i)^2 + 0.5*(Options->lambda_u/Data->u)*sum_{i in UNlabeled} max(0,1 - Y[i] w' x_i)^2 - subject to: (1/Data->u)*sum_{i in UNlabeled} max(0,Y[i]) = Options->R */ -int32_t TSVM_MFN( - const struct data *Data, - struct options *Options, - struct vector_double *Weights, - struct vector_double *Outputs); - -int32_t switch_labels( - float64_t* Y, - float64_t* o, - int32_t* JU, - int32_t u, - int32_t S); - -/* Deterministic Annealing*/ -int32_t DA_S3VM( - struct data *Data, - struct options *Options, - struct vector_double *Weights, - struct vector_double *Outputs); - -void optimize_p( - const float64_t* g, int32_t u, float64_t T, float64_t r, float64_t*p); - -int32_t optimize_w( - const struct data *Data, - const float64_t *p, - struct options *Options, - struct vector_double *Weights, - struct vector_double *Outputs, - int32_t ini); - -float64_t transductive_cost( - float64_t normWeights, - float64_t *Y, - float64_t *Outputs, - int32_t m, - float64_t lambda, - float64_t lambda_u); - -float64_t entropy(const float64_t *p, int32_t u); - -/* KL-divergence */ -float64_t KL(const float64_t *p, const float64_t *q, int32_t u); -} -#endif // _SSL_H - -#endif // DOXYGEN_SHOULD_SKIP_THIS diff --git a/src/shogun/lib/malsar/malsar_clustered.cpp b/src/shogun/lib/malsar/malsar_clustered.cpp deleted file mode 100644 index b2e153ea81f..00000000000 --- a/src/shogun/lib/malsar/malsar_clustered.cpp +++ /dev/null @@ -1,318 +0,0 @@ -/* - * This program is free software; you can redistribute it and/or modify - * it under the terms of the GNU General Public License as published by - * the Free Software Foundation; either version 3 of the License, or - * (at your option) any later version. - * - * Written (W) 2012 Sergey Lisitsyn - * Copyright (C) 2012 Jiayu Zhou and Jieping Ye - */ - -#include -#ifdef USE_GPL_SHOGUN - -#ifndef HAVE_CXX11 -#include -#include -#include -#include - -using namespace Eigen; -using namespace std; - -namespace shogun -{ - -static double* H_diag_matrix; -static int H_diag_matrix_ld; - -static const double* get_col(uint32_t j) -{ - return H_diag_matrix + j*H_diag_matrix_ld; -} - -malsar_result_t malsar_clustered( - CDotFeatures* features, - double* y, - double rho1, - double rho2, - const malsar_options& options) -{ - int task; - int n_feats = features->get_dim_feature_space(); - SG_SDEBUG("n feats = %d\n", n_feats) - int n_vecs = features->get_num_vectors(); - SG_SDEBUG("n vecs = %d\n", n_vecs) - int n_tasks = options.n_tasks; - SG_SDEBUG("n tasks = %d\n", n_tasks) - - H_diag_matrix = SG_CALLOC(double, n_tasks*n_tasks); - for (int i=0; i task_idx = options.tasks_indices[task]; - int n_vecs_task = task_idx.vlen; - for (int i=0; idense_dot(task_idx[i], Ws.col(task).data(), n_feats)+Cs[task]); - double bb = CMath::max(aa,0.0); - - // avoid underflow when computing exponential loss - Fs += (CMath::log(CMath::exp(-bb) + CMath::exp(aa-bb)) + bb)/n_vecs_task; - double b = -y[task_idx[i]]*(1 - 1/(1+CMath::exp(aa)))/n_vecs_task; - gCs[task] += b; - features->add_to_dense_vec(b, task_idx[i], gWs.col(task).data(), n_feats); - } - } - SG_SDEBUG("Computed gradient\n") - - // add regularizer - Fs += c*(Ws*invEtaMWt).trace(); - SG_SDEBUG("Fs = %f \n", Fs) - - double Fzp = 0.0; - - int inner_iter = 0; - // line search, Armijo-Goldstein scheme - while (inner_iter <= 1) - { - Wzp = Ws - gWs/gamma; - Czp = Cs - gCs/gamma; - // compute singular projection of Ms - gMs/gamma with k - //internal::set_is_malloc_allowed(true); - EigenSolver eigensolver; - eigensolver.compute(Ms-gMs/gamma, true); - if (eigensolver.info()!=Eigen::Success) - SG_SERROR("Eigendecomposition failed") - - // solve problem - // min sum_i (s_i - s*_i)^2 s.t. sum_i s_i = k, 0<=s_i<=1 - for (int i=0; i eigenvalue = eigensolver.eigenvalues()[i]; - //cout << "eigenvalue " << eigenvalue << "=" << std::real(eigenvalue) << "+i" << std::imag(eigenvalue) << endl; - f[i] = -2*eigensolver.eigenvalues()[i].real(); - if (f[i]!=f[i]) - SG_SERROR("NaN %d eigenvalue", i) - - SG_SDEBUG("%dth eigenvalue %f\n",i,eigensolver.eigenvalues()[i].real()) - a[i] = 1.0; - lb[i] = 0.0; - ub[i] = 1.0; - x[i] = double(options.n_clusters)/n_tasks; - } - double b = options.n_clusters;//eigensolver.eigenvalues().sum().real(); - SG_SDEBUG("b = %f\n", b) - SG_SDEBUG("Calling libqp\n") - libqp_state_T problem_state = libqp_gsmo_solver(&get_col,diag_H,f,a,b,lb,ub,x,n_tasks,1000,1e-6,NULL); - SG_SDEBUG("Libqp objective = %f\n",problem_state.QP) - SG_SDEBUG("Exit code = %d\n",problem_state.exitflag) - SG_SDEBUG("%d iteration passed\n",problem_state.nIter) - SG_SDEBUG("Solution is \n [ ") - for (int i=0; i Mzp_DiagSigz(x,n_tasks); - Mzp_Pz = eigensolver.eigenvectors().real(); - Mzp = Mzp_Pz*Mzp_DiagSigz.asDiagonal()*Mzp_Pz.transpose(); - //internal::set_is_malloc_allowed(false); - // walk in direction of antigradient - for (int i=0; i task_idx = options.tasks_indices[task]; - int n_vecs_task = task_idx.vlen; - for (int i=0; idense_dot(task_idx[i], Wzp.col(task).data(), n_feats)+Cs[task]); - double bb = CMath::max(aa,0.0); - - Fzp += (CMath::log(CMath::exp(-bb) + CMath::exp(aa-bb)) + bb)/n_vecs_task; - } - } - Fzp += c*(Wzp*invEtaMWt).trace(); - - // compute delta between line search point and search point - delta_Wzp = Wzp - Ws; - delta_Czp = Czp - Cs; - delta_Mzp = Mzp - Ms; - - // norms of delta - double nrm_delta_Wzp = delta_Wzp.squaredNorm(); - double nrm_delta_Czp = delta_Czp.squaredNorm(); - double nrm_delta_Mzp = delta_Mzp.squaredNorm(); - - double r_sum = (nrm_delta_Wzp + nrm_delta_Czp + nrm_delta_Mzp)/3; - - double Fzp_gamma = 0.0; - if (n_feats > n_tasks) - { - Fzp_gamma = Fs + (delta_Wzp.transpose()*gWs).trace() + - (delta_Czp.transpose()*gCs).trace() + - (delta_Mzp.transpose()*gMs).trace() + - (gamma/2)*nrm_delta_Wzp + - (gamma/2)*nrm_delta_Czp + - (gamma/2)*nrm_delta_Mzp; - } - else - { - Fzp_gamma = Fs + (gWs.transpose()*delta_Wzp).trace() + - (gCs.transpose()*delta_Czp).trace() + - (gMs.transpose()*delta_Mzp).trace() + - (gamma/2)*nrm_delta_Wzp + - (gamma/2)*nrm_delta_Czp + - (gamma/2)*nrm_delta_Mzp; - } - - // break if delta is getting too small - if (r_sum <= 1e-20) - { - done = true; - break; - } - - // break if objective at line searc point is smaller than Fzp_gamma - if (Fzp <= Fzp_gamma) - break; - else - gamma *= gamma_inc; - - inner_iter++; - } - - Wz_old = Wz; - Cz_old = Cz; - Mz_old = Mz; - Wz = Wzp; - Cz = Czp; - Mz = Mzp; - - // compute objective value - obj_old = obj; - obj = Fzp; - - // check if process should be terminated - switch (options.termination) - { - case 0: - if (iter>=2) - { - if ( CMath::abs(obj-obj_old) <= options.tolerance ) - done = true; - } - break; - case 1: - if (iter>=2) - { - if ( CMath::abs(obj-obj_old) <= options.tolerance*CMath::abs(obj_old)) - done = true; - } - break; - case 2: - if (CMath::abs(obj) <= options.tolerance) - done = true; - break; - case 3: - if (iter>=options.max_iter) - done = true; - break; - } - - iter++; - t_old = t; - t = 0.5 * (1 + CMath::sqrt(1.0 + 4*t*t)); - } - //internal::set_is_malloc_allowed(true); - SG_SDEBUG("%d iteration passed, objective = %f\n",iter,obj) - - SG_FREE(H_diag_matrix); - SG_FREE(diag_H); - SG_FREE(f); - SG_FREE(a); - SG_FREE(lb); - SG_FREE(ub); - SG_FREE(x); - - SGMatrix tasks_w(n_feats, n_tasks); - for (int i=0; i tasks_c(n_tasks); - for (int i=0; i -#ifdef USE_GPL_SHOGUN -#include -#include - -namespace shogun -{ - -/** - * Routine for learning a linear multitask - * logistic regression model using - * Clustered multitask learning algorithm. - * - */ -malsar_result_t malsar_clustered( - CDotFeatures* features, - double* y, - double rho1, - double rho2, - const malsar_options& options); - -}; -#endif //USE_GPL_SHOGUN -#endif /* ----- #ifndef MALSAR_CLUSTERED_H_ ----- */ diff --git a/src/shogun/lib/malsar/malsar_joint_feature_learning.cpp b/src/shogun/lib/malsar/malsar_joint_feature_learning.cpp deleted file mode 100644 index 02bb4bfadbf..00000000000 --- a/src/shogun/lib/malsar/malsar_joint_feature_learning.cpp +++ /dev/null @@ -1,231 +0,0 @@ -/* - * This program is free software; you can redistribute it and/or modify - * it under the terms of the GNU General Public License as published by - * the Free Software Foundation; either version 3 of the License, or - * (at your option) any later version. - * - * Written (W) 2012 Sergey Lisitsyn - * Copyright (C) 2012 Jiayu Zhou and Jieping Ye - */ - - -#include -#ifdef USE_GPL_SHOGUN -#include -#include -#include -#include - -using namespace Eigen; -using namespace std; - -namespace shogun -{ - -malsar_result_t malsar_joint_feature_learning( - CDotFeatures* features, - double* y, - double rho1, - double rho2, - const malsar_options& options) -{ - int task; - int n_feats = features->get_dim_feature_space(); - SG_SDEBUG("n feats = %d\n", n_feats) - int n_vecs = features->get_num_vectors(); - SG_SDEBUG("n vecs = %d\n", n_vecs) - int n_tasks = options.n_tasks; - SG_SDEBUG("n tasks = %d\n", n_tasks) - - int iter = 0; - - // initialize weight vector and bias for each task - MatrixXd Ws = MatrixXd::Zero(n_feats, n_tasks); - VectorXd Cs = VectorXd::Zero(n_tasks); - MatrixXd Wz=Ws, Wzp=Ws, Wz_old=Ws, delta_Wzp=Ws, gWs=Ws; - VectorXd Cz=Cs, Czp=Cs, Cz_old=Cs, delta_Czp=Cs, gCs=Cs; - - double t=1, t_old=0; - double gamma=1, gamma_inc=2; - double obj=0.0, obj_old=0.0; - - //internal::set_is_malloc_allowed(false); - bool done = false; - while (!done && iter <= options.max_iter && !CSignal::cancel_computations()) - { - double alpha = double(t_old - 1)/t; - - // compute search point - Ws = (1+alpha)*Wz - alpha*Wz_old; - Cs = (1+alpha)*Cz - alpha*Cz_old; - - // zero gradient - gWs.setZero(); - gCs.setZero(); - - // compute gradient and objective at search point - double Fs = 0; - for (task=0; task task_idx = options.tasks_indices[task]; - int n_task_vecs = task_idx.vlen; - for (int i=0; idense_dot(task_idx[i], Ws.col(task).data(), n_feats)+Cs[task]); - double bb = CMath::max(aa,0.0); - - // avoid underflow when computing exponential loss - Fs += (CMath::log(CMath::exp(-bb) + CMath::exp(aa-bb)) + bb)/n_task_vecs; - double b = -y[task_idx[i]]*(1 - 1/(1+CMath::exp(aa)))/n_task_vecs; - - gCs[task] += b; - features->add_to_dense_vec(b, task_idx[i], gWs.col(task).data(), n_feats); - } - } - gWs.noalias() += 2*rho2*Ws; - - // add regularizer - Fs += Ws.squaredNorm(); - - //cout << "gWs" << endl << gWs << endl; - //cout << "gCs" << endl << gCs << endl; - //SG_SPRINT("Fs = %f\n",Fs) - - double Fzp = 0.0; - - int inner_iter = 0; - // line search, Armijo-Goldstein scheme - while (inner_iter <= 1000) - { - // compute lasso projection of Ws - gWs/gamma - for (int i=0; i(); - if (norm == 0.0) - Wzp.row(i).setZero(); - else - { - double threshold = norm - rho1/gamma; - if (threshold < 0.0) - Wzp.row(i).setZero(); - else - Wzp.row(i) *= threshold/norm; - } - } - // walk in direction of antigradient - Czp = Cs - gCs/gamma; - - // compute objective at line search point - Fzp = 0.0; - for (task=0; task task_idx = options.tasks_indices[task]; - int n_task_vecs = task_idx.vlen; - for (int i=0; idense_dot(task_idx[i], Wzp.col(task).data(), n_feats)+Czp[task]); - double bb = CMath::max(aa,0.0); - - Fzp += (CMath::log(CMath::exp(-bb) + CMath::exp(aa-bb)) + bb)/n_task_vecs; - } - } - Fzp += rho2*Wzp.squaredNorm(); - - // compute delta between line search point and search point - delta_Wzp = Wzp - Ws; - delta_Czp = Czp - Cs; - - // norms of delta - double nrm_delta_Wzp = delta_Wzp.squaredNorm(); - double nrm_delta_Czp = delta_Czp.squaredNorm(); - - double r_sum = (nrm_delta_Wzp + nrm_delta_Czp)/2; - - double Fzp_gamma = Fs + (delta_Wzp.transpose()*gWs).trace() + - (delta_Czp.transpose()*gCs).trace() + - (gamma/2)*nrm_delta_Wzp + - (gamma/2)*nrm_delta_Czp; - - // break if delta is getting too small - if (r_sum <= 1e-20) - { - SG_SDEBUG("Line search point is too close to search point\n") - done = true; - break; - } - - // break if objective at line searc point is smaller than Fzp_gamma - if (Fzp <= Fzp_gamma) - break; - else - gamma *= gamma_inc; - - inner_iter++; - } - - Wz_old = Wz; - Cz_old = Cz; - Wz = Wzp; - Cz = Czp; - - // compute objective value - obj_old = obj; - obj = Fzp; - for (int i=0; i()); - //for (task=0; task=2) - { - if ( CMath::abs(obj-obj_old) <= options.tolerance ) - { - SG_SDEBUG("Objective changes less than tolerance\n") - done = true; - } - } - break; - case 1: - if (iter>=2) - { - if ( CMath::abs(obj-obj_old) <= options.tolerance*CMath::abs(obj_old)) - done = true; - } - break; - case 2: - if (CMath::abs(obj) <= options.tolerance) - done = true; - break; - case 3: - if (iter>=options.max_iter) - done = true; - break; - } - - iter++; - t_old = t; - t = 0.5 * (1 + CMath::sqrt(1.0 + 4*t*t)); - } - //internal::set_is_malloc_allowed(true); - SG_SDONE() - SG_SDEBUG("%d iteration passed, objective = %f\n",iter,obj) - - SGMatrix tasks_w(n_feats, n_tasks); - for (int i=0; i tasks_c(n_tasks); - for (int i=0; i -#ifdef USE_GPL_SHOGUN -#include -#include - -namespace shogun -{ - -/** - * Routine for learning a linear multitask - * logistic regression model - * using Joint Feature algorithm. - * - */ -malsar_result_t malsar_joint_feature_learning( - CDotFeatures* features, - double* y, - double rho1, - double rho2, - const malsar_options& options); - -}; -#endif //USE_GPL_SHOGUN -#endif /* ----- #ifndef MALSAR_JOINT_FEATURE_LEARNING_H_ ----- */ - diff --git a/src/shogun/lib/malsar/malsar_low_rank.cpp b/src/shogun/lib/malsar/malsar_low_rank.cpp deleted file mode 100644 index 34ac52ee5f5..00000000000 --- a/src/shogun/lib/malsar/malsar_low_rank.cpp +++ /dev/null @@ -1,214 +0,0 @@ -/* - * This program is free software; you can redistribute it and/or modify - * it under the terms of the GNU General Public License as published by - * the Free Software Foundation; either version 3 of the License, or - * (at your option) any later version. - * - * Written (W) 2012 Sergey Lisitsyn - * Copyright (C) 2012 Jiayu Zhou and Jieping Ye - */ - -#include -#ifdef USE_GPL_SHOGUN -#include -#include -#include - -using namespace Eigen; - -namespace shogun -{ - -malsar_result_t malsar_low_rank( - CDotFeatures* features, - double* y, - double rho, - const malsar_options& options) -{ - int task; - int n_feats = features->get_dim_feature_space(); - SG_SDEBUG("n feats = %d\n", n_feats) - int n_vecs = features->get_num_vectors(); - SG_SDEBUG("n vecs = %d\n", n_vecs) - int n_tasks = options.n_tasks; - SG_SDEBUG("n tasks = %d\n", n_tasks) - - int iter = 0; - - // initialize weight vector and bias for each task - MatrixXd Ws = MatrixXd::Zero(n_feats, n_tasks); - VectorXd Cs = VectorXd::Zero(n_tasks); - MatrixXd Wz=Ws, Wzp=Ws, Wz_old=Ws, delta_Wzp=Ws, gWs=Ws; - VectorXd Cz=Cs, Czp=Cs, Cz_old=Cs, delta_Czp=Cs, gCs=Cs; - - double t=1, t_old=0; - double gamma=1, gamma_inc=2; - double obj=0.0, obj_old=0.0; - - double rho_L2 = 0.0; - - //internal::set_is_malloc_allowed(false); - bool done = false; - while (!done && iter <= options.max_iter) - { - double alpha = double(t_old - 1)/t; - - // compute search point - Ws = (1+alpha)*Wz - alpha*Wz_old; - Cs = (1+alpha)*Cz - alpha*Cz_old; - - // zero gradient - gWs.setZero(); - gCs.setZero(); - - // compute gradient and objective at search point - double Fs = 0; - for (task=0; task task_idx = options.tasks_indices[task]; - int n_task_vecs = task_idx.vlen; - for (int i=0; idense_dot(task_idx[i], Ws.col(task).data(), n_feats)+Cs[task]); - double bb = CMath::max(aa,0.0); - - // avoid underflow when computing exponential loss - Fs += (CMath::log(CMath::exp(-bb) + CMath::exp(aa-bb)) + bb)/n_task_vecs; - double b = -y[task_idx[i]]*(1 - 1/(1+CMath::exp(aa)))/n_task_vecs; - - gCs[task] += b; - features->add_to_dense_vec(b, task_idx[i], gWs.col(task).data(), n_feats); - } - } - gWs.noalias() += 2*rho_L2*Ws; - //SG_SDEBUG("gWs=%f\n",gWs.squaredNorm()) - - // add regularizer - Fs += rho_L2*Ws.squaredNorm(); - - double Fzp = 0.0; - - int inner_iter = 0; - // line search, Armijo-Goldstein scheme - while (inner_iter <= 1000) - { - // compute trace projection of Ws - gWs/gamma with 2*rho/gamma - //internal::set_is_malloc_allowed(true); - Wzp.setZero(); - JacobiSVD svd((Ws - gWs/gamma).transpose(),ComputeThinU | ComputeThinV); - for (int i=0; i rho/gamma) - Wzp += (svd.matrixU().col(i)* - svd.singularValues()[i]* - svd.matrixV().col(i).transpose()).transpose(); - } - //internal::set_is_malloc_allowed(false); - // walk in direction of antigradient - Czp = Cs - gCs/gamma; - - // compute objective at line search point - Fzp = 0.0; - for (task=0; task task_idx = options.tasks_indices[task]; - int n_task_vecs = task_idx.vlen; - for (int i=0; idense_dot(task_idx[i], Wzp.col(task).data(), n_feats)+Czp[task]); - double bb = CMath::max(aa,0.0); - - Fzp += (CMath::log(CMath::exp(-bb) + CMath::exp(aa-bb)) + bb)/n_task_vecs; - } - } - Fzp += rho_L2*Wzp.squaredNorm(); - - // compute delta between line search point and search point - delta_Wzp = Wzp - Ws; - delta_Czp = Czp - Cs; - - // norms of delta - double nrm_delta_Wzp = delta_Wzp.squaredNorm(); - double nrm_delta_Czp = delta_Czp.squaredNorm(); - - double r_sum = (nrm_delta_Wzp + nrm_delta_Czp)/2; - - double Fzp_gamma = Fs + (delta_Wzp.transpose()*gWs).trace() + - (delta_Czp.transpose()*gCs).trace() + - (gamma/2)*nrm_delta_Wzp + - (gamma/2)*nrm_delta_Czp; - - // break if delta is getting too small - if (r_sum <= 1e-20) - { - done = true; - break; - } - - // break if objective at line search point is smaller than Fzp_gamma - if (Fzp <= Fzp_gamma) - break; - else - gamma *= gamma_inc; - } - - Wz_old = Wz; - Cz_old = Cz; - Wz = Wzp; - Cz = Czp; - - // compute objective value - obj_old = obj; - obj = Fzp; - //internal::set_is_malloc_allowed(true); - JacobiSVD svd(Wzp, EigenvaluesOnly); - obj += rho*svd.singularValues().sum(); - //internal::set_is_malloc_allowed(false); - - - // check if process should be terminated - switch (options.termination) - { - case 0: - if (iter>=2) - { - if ( CMath::abs(obj-obj_old) <= options.tolerance ) - done = true; - } - break; - case 1: - if (iter>=2) - { - if ( CMath::abs(obj-obj_old) <= options.tolerance*CMath::abs(obj_old)) - done = true; - } - break; - case 2: - if (CMath::abs(obj) <= options.tolerance) - done = true; - break; - case 3: - if (iter>=options.max_iter) - done = true; - break; - } - - iter++; - t_old = t; - t = 0.5 * (1 + CMath::sqrt(1.0 + 4*t*t)); - } - //internal::set_is_malloc_allowed(true); - SG_SDEBUG("%d iteration passed, objective = %f\n",iter,obj) - - SGMatrix tasks_w(n_feats, n_tasks); - for (int i=0; i tasks_c(n_tasks); - for (int i=0; i -#ifdef USE_GPL_SHOGUN -#include -#include - -namespace shogun -{ - -/** - * Routine for learning a linear multitask - * logistic regression model using - * Low Rank multitask algorithm. - * - */ -malsar_result_t malsar_low_rank( - CDotFeatures* features, - double* y, - double rho, - const malsar_options& options); - -}; -#endif //USE_GPL_SHOGUN -#endif /* ----- #ifndef MALSAR_LOW_RANK_H_ ----- */ diff --git a/src/shogun/lib/malsar/malsar_options.h b/src/shogun/lib/malsar/malsar_options.h deleted file mode 100644 index d76f8dffbee..00000000000 --- a/src/shogun/lib/malsar/malsar_options.h +++ /dev/null @@ -1,69 +0,0 @@ -/* - * This program is free software; you can redistribute it and/or modify - * it under the terms of the GNU General Public License as published by - * the Free Software Foundation; either version 3 of the License, or - * (at your option) any later version. - * - * Copyright (C) 2012 Sergey Lisitsyn - */ - -#ifndef MALSAR_OPTIONS_H_ -#define MALSAR_OPTIONS_H_ - -#include -#ifdef USE_GPL_SHOGUN - -#define IGNORE_IN_CLASSLIST - -#include -#include -#include - -namespace shogun -{ - -#ifndef DOXYGEN_SHOULD_SKIP_THIS -IGNORE_IN_CLASSLIST enum malsar_loss -{ - MALSAR_LOGISTIC, - MALSAR_LEAST_SQUARES -}; - -IGNORE_IN_CLASSLIST struct malsar_options -{ - int termination; - double tolerance; - int max_iter; - int n_tasks; - int n_clusters; - SGVector* tasks_indices; - malsar_loss loss; - - static malsar_options default_options() - { - malsar_options opts; - opts.termination = 2; - opts.tolerance = 1e-3; - opts.max_iter = 1000; - opts.tasks_indices = NULL; - opts.n_clusters = 2; - opts.loss = MALSAR_LOGISTIC; - return opts; - } -}; - -IGNORE_IN_CLASSLIST struct malsar_result_t -{ - SGMatrix w; - SGVector c; - - malsar_result_t(SGMatrix w_, SGVector c_) - { - w = w_; - c = c_; - } -}; -#endif -} -#endif //USE_GPL_SHOGUN -#endif /* ----- #ifndef MALSAR_OPTIONS_H_ ----- */ diff --git a/src/shogun/lib/parameter_observers/ObservedValue.h b/src/shogun/lib/parameter_observers/ObservedValue.h new file mode 100644 index 00000000000..65a622deab9 --- /dev/null +++ b/src/shogun/lib/parameter_observers/ObservedValue.h @@ -0,0 +1,194 @@ +/* +* BSD 3-Clause License +* +* Copyright (c) 2017, Shogun-Toolbox e.V. +* All rights reserved. +* +* Redistribution and use in source and binary forms, with or without +* modification, are permitted provided that the following conditions are met: +* +* * Redistributions of source code must retain the above copyright notice, this +* list of conditions and the following disclaimer. +* +* * Redistributions in binary form must reproduce the above copyright notice, +* this list of conditions and the following disclaimer in the documentation +* and/or other materials provided with the distribution. +* +* * Neither the name of the copyright holder nor the names of its +* contributors may be used to endorse or promote products derived from +* this software without specific prior written permission. +* +* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +* AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE +* DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE +* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL +* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR +* SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER +* CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, +* OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +* OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +* +* Written (W) 2017 Giovanni De Toni +* +*/ + +#ifndef SHOGUN_OBSERVEDVALUE_H +#define SHOGUN_OBSERVEDVALUE_H + +#include +#include +#include +#include + +/** + * Definitions of basic object with are needed by the Parameter + * Observer architecture. + */ +namespace shogun +{ + /* Timepoint */ + typedef std::chrono::steady_clock::time_point time_point; + + /* Type of the observed value */ + enum SG_OBS_VALUE_TYPE + { + TENSORBOARD, + CROSSVALIDATION + }; + + /** + * Observed value which is emitted by algorithms. + */ + class ObservedValue + { + public: + /** + * Constructor + * @param step step + * @param name param's name + * @param value Any-wrapped value of the param + */ + ObservedValue( + int64_t step, std::string& name, Any value, SG_OBS_VALUE_TYPE type) + : m_step(step), m_name(name), m_value(value), m_type(type) + { + } + + ~ObservedValue(){}; + + /** + * Get the step + * @return an integer representing the step + */ + int64_t get_step() const + { + return m_step; + } + + /** + * Set the step + * @param step step + */ + void set_step(int64_t step) + { + m_step = step; + } + + /** + * Get the param's name + * @return param's name + */ + const std::string& get_name() const + { + return m_name; + } + + /** + * Set the param's name + * @param name + */ + void set_name(const std::string& name) + { + m_name = name; + } + + /** + * Get the Any-wrapped value + * @return Any-wrapped value + */ + const Any& get_value() const + { + return m_value; + } + + /** + * Set the param's value + * @param value + */ + void set_value(const Any& value) + { + m_value = value; + } + + /** + * Get the type of this ObservedValue + * @return observed value type + */ + SG_OBS_VALUE_TYPE get_type() const + { + return m_type; + } + + /** + * Set the observed value type + * @param type type of this observed value + */ + void set_type(const SG_OBS_VALUE_TYPE type) + { + m_type = type; + } + + /** + * Helper method to generate an ObservedValue (TensorBoard oriented) + * @param step the step + * @param name the param's name we are observing + * @param value the param's value + * @return an ObservedValue object initialized + */ + static ObservedValue + make_observation(int64_t step, std::string& name, Any value) + { + return ObservedValue(step, name, value, TENSORBOARD); + } + + protected: + /** ObservedValue step (used by Tensorboard to print graphs) */ + int64_t m_step; + /** Parameter's name */ + std::string m_name; + /** Parameter's value */ + Any m_value; + /** ObservedValue type */ + SG_OBS_VALUE_TYPE m_type; + }; + + /** + * Observed value with a timestamp + */ + typedef std::pair TimedObservedValue; + + /** + * Helper method to convert a time_point to milliseconds + * @param value time point we want to convert + * @return the time point converted to milliseconds + */ + SG_FORCED_INLINE double convert_to_millis(const time_point& value) + { + return std::chrono::duration_cast( + value.time_since_epoch()) + .count(); + } +} + +#endif // SHOGUN_OBSERVEDVALUE_H diff --git a/src/shogun/lib/parameter_observers/ParameterObserverCV.cpp b/src/shogun/lib/parameter_observers/ParameterObserverCV.cpp new file mode 100644 index 00000000000..1d2199eddcf --- /dev/null +++ b/src/shogun/lib/parameter_observers/ParameterObserverCV.cpp @@ -0,0 +1,183 @@ +/* +* BSD 3-Clause License +* +* Copyright (c) 2017, Shogun-Toolbox e.V. +* All rights reserved. +* +* Redistribution and use in source and binary forms, with or without +* modification, are permitted provided that the following conditions are met: +* +* * Redistributions of source code must retain the above copyright notice, this +* list of conditions and the following disclaimer. +* +* * Redistributions in binary form must reproduce the above copyright notice, +* this list of conditions and the following disclaimer in the documentation +* and/or other materials provided with the distribution. +* +* * Neither the name of the copyright holder nor the names of its +* contributors may be used to endorse or promote products derived from +* this software without specific prior written permission. +* +* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +* AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE +* DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE +* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL +* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR +* SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER +* CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, +* OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +* OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +* +* Written (W) 2017 Giovanni De Toni +* +*/ + +#include +#include +#include +#include +#include +#include +#include + +using namespace shogun; + +CParameterObserverCV::CParameterObserverCV(bool verbose) + : ParameterObserverInterface(), m_verbose(verbose) +{ + m_type = CROSSVALIDATION; +} + +CParameterObserverCV::~CParameterObserverCV() +{ + for (auto i : m_observations) + SG_UNREF(i) +} + +void CParameterObserverCV::on_next(const shogun::TimedObservedValue& value) +{ + CHECK_OBSERVED_VALUE_TYPE(value.first.get_type()); + + if (value.first.get_value().type_info().hash_code() == + typeid(CrossValidationStorage*).hash_code()) + { + CrossValidationStorage* recalled_value = + recall_type(value.first.get_value()); + SG_REF(recalled_value); + + /* Print information on screen if enabled*/ + if (m_verbose) + print_observed_value(recalled_value); + + m_observations.push_back(recalled_value); + } + else + { + SG_SERROR( + "ParameterObserverCV: The observed value received is not of " + "type CrossValidationStorage\n"); + } +} + +void CParameterObserverCV::on_error(std::exception_ptr ptr) +{ +} + +void CParameterObserverCV::on_complete() +{ +} + +void CParameterObserverCV::clear() +{ + for (auto i : m_observations) + { + SG_UNREF(i) + } + m_observations.clear(); +} + +void CParameterObserverCV::print_observed_value( + CrossValidationStorage* value) const +{ + for (int i = 0; i < value->get_num_folds(); i++) + { + auto f = value->get_fold(i); + SG_SPRINT("\n") + SG_SPRINT("Current run index: %i\n", f->get_current_run_index()) + SG_SPRINT("Current fold index: %i\n", f->get_current_fold_index()) + f->get_train_indices().display_vector("Train Indices "); + f->get_test_indices().display_vector("Test Indices "); + print_machine_information(f->get_trained_machine()); + f->get_test_result()->get_values().display_vector("Test Labels "); + f->get_test_true_result()->get_values().display_vector( + "Test True Label "); + SG_SPRINT("Evaluation result: %f\n", f->get_evaluation_result()); + SG_UNREF(f) + } +} + +void CParameterObserverCV::print_machine_information(CMachine* machine) const +{ + if (dynamic_cast(machine)) + { + CLinearMachine* linear_machine = (CLinearMachine*)machine; + linear_machine->get_w().display_vector("Learned Weights = "); + SG_SPRINT("Learned Bias = %f\n", linear_machine->get_bias()) + } + + if (dynamic_cast(machine)) + { + CKernelMachine* kernel_machine = (CKernelMachine*)machine; + kernel_machine->get_alphas().display_vector("Learned alphas = "); + SG_SPRINT("Learned Bias = %f\n", kernel_machine->get_bias()) + } + + if (dynamic_cast(machine) || + dynamic_cast(machine)) + { + CMulticlassMachine* mc_machine = (CMulticlassMachine*)machine; + for (int i = 0; i < mc_machine->get_num_machines(); i++) + { + CMachine* sub_machine = mc_machine->get_machine(i); + this->print_machine_information(sub_machine); + SG_UNREF(sub_machine); + } + } + + if (dynamic_cast(machine)) + { + CMKL* mkl = (CMKL*)machine; + CCombinedKernel* kernel = + dynamic_cast(mkl->get_kernel()); + kernel->get_subkernel_weights().display_vector( + "MKL sub-kernel weights ="); + SG_UNREF(kernel); + } + + if (dynamic_cast(machine)) + { + CMKLMulticlass* mkl = (CMKLMulticlass*)machine; + CCombinedKernel* kernel = + dynamic_cast(mkl->get_kernel()); + kernel->get_subkernel_weights().display_vector( + "MKL sub-kernel weights ="); + SG_UNREF(kernel); + } +} + +CrossValidationStorage* CParameterObserverCV::get_observation(int run) const +{ + REQUIRE( + run < get_num_observations(), "The run number must be less than %i", + get_num_observations()) + + CrossValidationStorage* obs = m_observations[run]; + SG_REF(obs) + return obs; +} + +const int CParameterObserverCV::get_num_observations() const +{ + return m_observations.size(); +} diff --git a/src/shogun/lib/parameter_observers/ParameterObserverCV.h b/src/shogun/lib/parameter_observers/ParameterObserverCV.h new file mode 100644 index 00000000000..8539b816791 --- /dev/null +++ b/src/shogun/lib/parameter_observers/ParameterObserverCV.h @@ -0,0 +1,110 @@ +/* +* BSD 3-Clause License +* +* Copyright (c) 2017, Shogun-Toolbox e.V. +* All rights reserved. +* +* Redistribution and use in source and binary forms, with or without +* modification, are permitted provided that the following conditions are met: +* +* * Redistributions of source code must retain the above copyright notice, this +* list of conditions and the following disclaimer. +* +* * Redistributions in binary form must reproduce the above copyright notice, +* this list of conditions and the following disclaimer in the documentation +* and/or other materials provided with the distribution. +* +* * Neither the name of the copyright holder nor the names of its +* contributors may be used to endorse or promote products derived from +* this software without specific prior written permission. +* +* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +* AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE +* DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE +* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL +* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR +* SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER +* CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, +* OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +* OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +* +* Written (W) 2017 Giovanni De Toni +* +*/ + +#ifndef SHOGUN_PARAMETEROBSERVERCV_H +#define SHOGUN_PARAMETEROBSERVERCV_H + +#include +#include +#include + +namespace shogun +{ + + /** + * Base ParameterObserver class for CrossValidation. + */ + class CParameterObserverCV : public ParameterObserverInterface, + public CSGObject + { + + public: + CParameterObserverCV(bool verbose = false); + virtual ~CParameterObserverCV(); + + virtual void on_next(const TimedObservedValue& value); + virtual void on_error(std::exception_ptr ptr); + virtual void on_complete(); + + /* Erase all observations done so far */ + virtual void clear(); + + /** + * Get the total number of cross validation runs received + * by this observer. + * @return number of runs. + */ + const int get_num_observations() const; + + /** + * Get a CrossValidationStorage object which will store + * the result of a CrossValidation run. + * @param run index of the run + * @return a CrossValidationStorage object + */ + CrossValidationStorage* get_observation(int run) const; + + /** + * Print data contained into a CrossValidationStorage object. + * @param value CrossValidationStorage object + */ + void print_observed_value(CrossValidationStorage* value) const; + + /** + * Get class name. + * @return class name + */ + virtual const char* get_name() const + { + return "ParameterObserverCV"; + } + + private: + void print_machine_information(CMachine* machine) const; + + protected: + /** + * Observation's vector + */ + std::vector m_observations; + + /** + * enable printing of information + */ + bool m_verbose; + }; +} + +#endif // SHOGUN_PARAMETEROBSERVERCV_H diff --git a/src/shogun/lib/parameter_observers/ParameterObserverHistogram.cpp b/src/shogun/lib/parameter_observers/ParameterObserverHistogram.cpp new file mode 100644 index 00000000000..fe2c365b4f5 --- /dev/null +++ b/src/shogun/lib/parameter_observers/ParameterObserverHistogram.cpp @@ -0,0 +1,82 @@ +/* +* BSD 3-Clause License +* +* Copyright (c) 2017, Shogun-Toolbox e.V. +* All rights reserved. +* +* Redistribution and use in source and binary forms, with or without +* modification, are permitted provided that the following conditions are met: +* +* * Redistributions of source code must retain the above copyright notice, this +* list of conditions and the following disclaimer. +* +* * Redistributions in binary form must reproduce the above copyright notice, +* this list of conditions and the following disclaimer in the documentation +* and/or other materials provided with the distribution. +* +* * Neither the name of the copyright holder nor the names of its +* contributors may be used to endorse or promote products derived from +* this software without specific prior written permission. +* +* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +* AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE +* DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE +* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL +* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR +* SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER +* CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, +* OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +* OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +* +* Written (W) 2017 Giovanni De Toni +* +*/ +#include +#ifdef HAVE_TFLOGGER + +#include +#include + +using namespace shogun; + +ParameterObserverHistogram::ParameterObserverHistogram() + : ParameterObserverTensorBoard() +{ +} + +ParameterObserverHistogram::ParameterObserverHistogram( + std::vector& parameters) + : ParameterObserverTensorBoard(parameters) +{ +} + +ParameterObserverHistogram::ParameterObserverHistogram( + const std::string& filename, std::vector& parameters) + : ParameterObserverTensorBoard(filename, parameters) +{ +} + +ParameterObserverHistogram::~ParameterObserverHistogram() +{ +} + +void ParameterObserverHistogram::on_next(const TimedObservedValue& value) +{ + CHECK_OBSERVED_VALUE_TYPE(value.first.get_type()) + + auto node_name = std::string("node"); + auto format = TBOutputFormat(); + auto event_value = format.convert_vector(value, node_name); + m_writer.writeEvent(event_value); +} + +void ParameterObserverHistogram::on_error(std::exception_ptr) +{ +} + +void ParameterObserverHistogram::on_complete() +{ +} + +#endif // HAVE_TFLOGGER diff --git a/src/shogun/lib/parameter_observers/ParameterObserverHistogram.h b/src/shogun/lib/parameter_observers/ParameterObserverHistogram.h new file mode 100644 index 00000000000..8670c7b4078 --- /dev/null +++ b/src/shogun/lib/parameter_observers/ParameterObserverHistogram.h @@ -0,0 +1,77 @@ +/* +* BSD 3-Clause License +* +* Copyright (c) 2017, Shogun-Toolbox e.V. +* All rights reserved. +* +* Redistribution and use in source and binary forms, with or without +* modification, are permitted provided that the following conditions are met: +* +* * Redistributions of source code must retain the above copyright notice, this +* list of conditions and the following disclaimer. +* +* * Redistributions in binary form must reproduce the above copyright notice, +* this list of conditions and the following disclaimer in the documentation +* and/or other materials provided with the distribution. +* +* * Neither the name of the copyright holder nor the names of its +* contributors may be used to endorse or promote products derived from +* this software without specific prior written permission. +* +* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +* AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE +* DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE +* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL +* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR +* SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER +* CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, +* OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +* OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +* +* Written (W) 2017 Giovanni De Toni +* +*/ +#include +#ifdef HAVE_TFLOGGER + +#ifndef SHOGUN_PARAMETEROBSERVERHISTOGRAM_H +#define SHOGUN_PARAMETEROBSERVERHISTOGRAM_H + +#include +#include + +namespace shogun +{ + /** + * Implementation of a ParameterObserver which write to file + * histograms, given object emitted from a parameter observable. + */ + class ParameterObserverHistogram : public ParameterObserverTensorBoard, + public CSGObject + { + + public: + ParameterObserverHistogram(); + ParameterObserverHistogram(std::vector& parameters); + ParameterObserverHistogram( + const std::string& filename, std::vector& parameters); + ~ParameterObserverHistogram(); + + virtual void on_next(const TimedObservedValue& value); + virtual void on_error(std::exception_ptr); + virtual void on_complete(); + + /** + * Get class name. + * @return class name + */ + virtual const char* get_name() const + { + return "ParameterObserverHistogram"; + } + }; +} + +#endif // SHOGUN_PARAMETEROBSERVERHISTOGRAM_H +#endif // HAVE_TFLOGGER diff --git a/src/shogun/lib/parameter_observers/ParameterObserverInterface.cpp b/src/shogun/lib/parameter_observers/ParameterObserverInterface.cpp new file mode 100644 index 00000000000..482c1c1f4f0 --- /dev/null +++ b/src/shogun/lib/parameter_observers/ParameterObserverInterface.cpp @@ -0,0 +1,70 @@ +/* +* BSD 3-Clause License +* +* Copyright (c) 2017, Shogun-Toolbox e.V. +* All rights reserved. +* +* Redistribution and use in source and binary forms, with or without +* modification, are permitted provided that the following conditions are met: +* +* * Redistributions of source code must retain the above copyright notice, this +* list of conditions and the following disclaimer. +* +* * Redistributions in binary form must reproduce the above copyright notice, +* this list of conditions and the following disclaimer in the documentation +* and/or other materials provided with the distribution. +* +* * Neither the name of the copyright holder nor the names of its +* contributors may be used to endorse or promote products derived from +* this software without specific prior written permission. +* +* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +* AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE +* DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE +* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL +* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR +* SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER +* CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, +* OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +* OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +* +* Written (W) 2017 Giovanni De Toni +* +*/ +#include +#include + +using namespace shogun; + +ParameterObserverInterface::ParameterObserverInterface() : m_parameters() +{ +} + +ParameterObserverInterface::ParameterObserverInterface( + std::vector& parameters) + : m_parameters(parameters) +{ +} + +ParameterObserverInterface::ParameterObserverInterface( + const std::string& filename, std::vector& parameters) + : m_parameters(parameters) +{ +} + +ParameterObserverInterface::~ParameterObserverInterface() +{ +} + +bool ParameterObserverInterface::filter(const std::string& param) +{ + // If there are no specified parameters, then watch everything + if (m_parameters.size() == 0) + return true; + + for (auto v : m_parameters) + if (v == param) + return true; + return false; +} \ No newline at end of file diff --git a/src/shogun/lib/parameter_observers/ParameterObserverInterface.h b/src/shogun/lib/parameter_observers/ParameterObserverInterface.h new file mode 100644 index 00000000000..baee4b9fc3e --- /dev/null +++ b/src/shogun/lib/parameter_observers/ParameterObserverInterface.h @@ -0,0 +1,124 @@ +/* +* BSD 3-Clause License +* +* Copyright (c) 2017, Shogun-Toolbox e.V. +* All rights reserved. +* +* Redistribution and use in source and binary forms, with or without +* modification, are permitted provided that the following conditions are met: +* +* * Redistributions of source code must retain the above copyright notice, this +* list of conditions and the following disclaimer. +* +* * Redistributions in binary form must reproduce the above copyright notice, +* this list of conditions and the following disclaimer in the documentation +* and/or other materials provided with the distribution. +* +* * Neither the name of the copyright holder nor the names of its +* contributors may be used to endorse or promote products derived from +* this software without specific prior written permission. +* +* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +* AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE +* DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE +* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL +* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR +* SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER +* CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, +* OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +* OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +* +* Written (W) 2017 Giovanni De Toni +* +*/ +#ifndef SHOGUN_PARAMETEROBSERVERINTERFACE_H +#define SHOGUN_PARAMETEROBSERVERINTERFACE_H + +#include +#include + +#include +#include + +/* Used to check if the observed value type */ +#define CHECK_OBSERVED_VALUE_TYPE(type) \ + if (type != m_type) \ + return; + +namespace shogun +{ + + /** + * Interface for the parameter observer classes + */ + class ParameterObserverInterface + { + + public: + /** + * Default constructor + */ + ParameterObserverInterface(); + + /** + * Constructor + * @param parameters list of parameters which we want to watch over + */ + ParameterObserverInterface(std::vector& parameters); + + /** + * Constructor + * @param filename name of the generated output file + * @param parameters list of parameters which we want to watch over + */ + ParameterObserverInterface( + const std::string& filename, std::vector& parameters); + /** + * Virtual destructor + */ + virtual ~ParameterObserverInterface(); + + /** + * Filter function, check if the parameter name supplied is what + * we want to monitor + * @param param the param name + * @return true if param is found inside of m_parameters list + */ + virtual bool filter(const std::string& param); + + /** + * Method which will be called when the parameter observable emits a + * value. + * @param value the value emitted by the parameter observable + */ + virtual void on_next(const TimedObservedValue& value) = 0; + /** + * Method which will be called on errors + */ + virtual void on_error(std::exception_ptr) = 0; + /** + * Method which will be called on completion + */ + virtual void on_complete() = 0; + + /** + * Method useful to empty the observer from + * obseverd value it may have stored. + */ + virtual void clear(){}; + + protected: + /** + * List of parameter's names we want to monitor + */ + std::vector m_parameters; + + /** + * The type of params this observers accept + */ + SG_OBS_VALUE_TYPE m_type; + }; +} + +#endif // SHOGUN_PARAMETEROBSERVER_H diff --git a/src/shogun/lib/parameter_observers/ParameterObserverScalar.cpp b/src/shogun/lib/parameter_observers/ParameterObserverScalar.cpp new file mode 100644 index 00000000000..7db0b150bb7 --- /dev/null +++ b/src/shogun/lib/parameter_observers/ParameterObserverScalar.cpp @@ -0,0 +1,82 @@ +/* +* BSD 3-Clause License +* +* Copyright (c) 2017, Shogun-Toolbox e.V. +* All rights reserved. +* +* Redistribution and use in source and binary forms, with or without +* modification, are permitted provided that the following conditions are met: +* +* * Redistributions of source code must retain the above copyright notice, this +* list of conditions and the following disclaimer. +* +* * Redistributions in binary form must reproduce the above copyright notice, +* this list of conditions and the following disclaimer in the documentation +* and/or other materials provided with the distribution. +* +* * Neither the name of the copyright holder nor the names of its +* contributors may be used to endorse or promote products derived from +* this software without specific prior written permission. +* +* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +* AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE +* DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE +* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL +* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR +* SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER +* CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, +* OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +* OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +* +* Written (W) 2017 Giovanni De Toni +* +*/ +#include +#ifdef HAVE_TFLOGGER + +#include +#include + +using namespace shogun; + +ParameterObserverScalar::ParameterObserverScalar() + : ParameterObserverTensorBoard() +{ +} + +ParameterObserverScalar::ParameterObserverScalar( + std::vector& parameters) + : ParameterObserverTensorBoard(parameters) +{ +} + +ParameterObserverScalar::ParameterObserverScalar( + const std::string& filename, std::vector& parameters) + : ParameterObserverTensorBoard(filename, parameters) +{ +} + +ParameterObserverScalar::~ParameterObserverScalar() +{ +} + +void ParameterObserverScalar::on_next(const TimedObservedValue& value) +{ + CHECK_OBSERVED_VALUE_TYPE(value.first.get_type()) + + auto node_name = std::string("node"); + auto format = TBOutputFormat(); + auto event_value = format.convert_scalar(value, node_name); + m_writer.writeEvent(event_value); +} + +void ParameterObserverScalar::on_error(std::exception_ptr) +{ +} + +void ParameterObserverScalar::on_complete() +{ +} + +#endif // HAVE_TFLOGGER diff --git a/src/shogun/lib/parameter_observers/ParameterObserverScalar.h b/src/shogun/lib/parameter_observers/ParameterObserverScalar.h new file mode 100644 index 00000000000..e16eea6225e --- /dev/null +++ b/src/shogun/lib/parameter_observers/ParameterObserverScalar.h @@ -0,0 +1,77 @@ +/* +* BSD 3-Clause License +* +* Copyright (c) 2017, Shogun-Toolbox e.V. +* All rights reserved. +* +* Redistribution and use in source and binary forms, with or without +* modification, are permitted provided that the following conditions are met: +* +* * Redistributions of source code must retain the above copyright notice, this +* list of conditions and the following disclaimer. +* +* * Redistributions in binary form must reproduce the above copyright notice, +* this list of conditions and the following disclaimer in the documentation +* and/or other materials provided with the distribution. +* +* * Neither the name of the copyright holder nor the names of its +* contributors may be used to endorse or promote products derived from +* this software without specific prior written permission. +* +* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +* AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE +* DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE +* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL +* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR +* SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER +* CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, +* OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +* OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +* +* Written (W) 2017 Giovanni De Toni +* +*/ +#include +#ifdef HAVE_TFLOGGER + +#ifndef SHOGUN_PARAMETEROBSERVERSCALAR_H +#define SHOGUN_PARAMETEROBSERVERSCALAR_H + +#include +#include + +namespace shogun +{ + /** + * Implementation of a ParameterObserver which write to file + * scalar values, given object emitted from a parameter observable. + */ + class ParameterObserverScalar : public ParameterObserverTensorBoard, + public CSGObject + { + + public: + ParameterObserverScalar(); + ParameterObserverScalar(std::vector& parameters); + ParameterObserverScalar( + const std::string& filename, std::vector& parameters); + ~ParameterObserverScalar(); + + virtual void on_next(const TimedObservedValue& value); + virtual void on_error(std::exception_ptr); + virtual void on_complete(); + + /** + * Get class name. + * @return class name + */ + virtual const char* get_name() const + { + return "ParameterObserverScalar"; + } + }; +} + +#endif // SHOGUN_PARAMETEROBSERVERSCALAR_H +#endif // HAVE_TFLOGGER diff --git a/src/shogun/lib/parameter_observers/ParameterObserverTensorBoard.cpp b/src/shogun/lib/parameter_observers/ParameterObserverTensorBoard.cpp new file mode 100644 index 00000000000..0448ee8c05b --- /dev/null +++ b/src/shogun/lib/parameter_observers/ParameterObserverTensorBoard.cpp @@ -0,0 +1,68 @@ +/* +* BSD 3-Clause License +* +* Copyright (c) 2017, Shogun-Toolbox e.V. +* All rights reserved. +* +* Redistribution and use in source and binary forms, with or without +* modification, are permitted provided that the following conditions are met: +* +* * Redistributions of source code must retain the above copyright notice, this +* list of conditions and the following disclaimer. +* +* * Redistributions in binary form must reproduce the above copyright notice, +* this list of conditions and the following disclaimer in the documentation +* and/or other materials provided with the distribution. +* +* * Neither the name of the copyright holder nor the names of its +* contributors may be used to endorse or promote products derived from +* this software without specific prior written permission. +* +* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +* AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE +* DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE +* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL +* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR +* SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER +* CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, +* OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +* OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +* +* Written (W) 2017 Giovanni De Toni +* +*/ +#include +#ifdef HAVE_TFLOGGER + +#include "ParameterObserverTensorBoard.h" + +using namespace shogun; + +ParameterObserverTensorBoard::ParameterObserverTensorBoard() + : ParameterObserverInterface(), m_writer("shogun") +{ + m_writer.init(); +} + +ParameterObserverTensorBoard::ParameterObserverTensorBoard( + std::vector& parameters) + : ParameterObserverInterface(parameters), m_writer("shogun") +{ + m_writer.init(); +} + +ParameterObserverTensorBoard::ParameterObserverTensorBoard( + const std::string& filename, std::vector& parameters) + : ParameterObserverInterface(parameters), m_writer(filename.c_str()) +{ + m_writer.init(); +} + +ParameterObserverTensorBoard::~ParameterObserverTensorBoard() +{ + m_writer.flush(); + m_writer.close(); +} + +#endif // HAVE_TFLOGGER diff --git a/src/shogun/lib/parameter_observers/ParameterObserverTensorBoard.h b/src/shogun/lib/parameter_observers/ParameterObserverTensorBoard.h new file mode 100644 index 00000000000..e83fb9b596c --- /dev/null +++ b/src/shogun/lib/parameter_observers/ParameterObserverTensorBoard.h @@ -0,0 +1,83 @@ +/* +* BSD 3-Clause License +* +* Copyright (c) 2017, Shogun-Toolbox e.V. +* All rights reserved. +* +* Redistribution and use in source and binary forms, with or without +* modification, are permitted provided that the following conditions are met: +* +* * Redistributions of source code must retain the above copyright notice, this +* list of conditions and the following disclaimer. +* +* * Redistributions in binary form must reproduce the above copyright notice, +* this list of conditions and the following disclaimer in the documentation +* and/or other materials provided with the distribution. +* +* * Neither the name of the copyright holder nor the names of its +* contributors may be used to endorse or promote products derived from +* this software without specific prior written permission. +* +* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +* AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE +* DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE +* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL +* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR +* SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER +* CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, +* OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE +* OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +* +* Written (W) 2017 Giovanni De Toni +* +*/ +#include +#ifdef HAVE_TFLOGGER + +#ifndef SHOGUN_PARAMETEROBSERVERTENSORBOARD_H +#define SHOGUN_PARAMETEROBSERVERTENSORBOARD_H + +#include + +#include + +namespace shogun +{ + class ParameterObserverTensorBoard : public ParameterObserverInterface + { + + public: + /** + * Default constructor + */ + ParameterObserverTensorBoard(); + + /** + * Constructor + * @param parameters list of parameters which we want to watch over + */ + ParameterObserverTensorBoard(std::vector& parameters); + + /** + * Constructor + * @param filename name of the generated output file + * @param parameters list of parameters which we want to watch over + */ + ParameterObserverTensorBoard( + const std::string& filename, std::vector& parameters); + /** + * Virtual destructor + */ + virtual ~ParameterObserverTensorBoard(); + + protected: + /** + * Writer object which will be used to write tensorflow::Event files + */ + tflogger::EventLogger m_writer; + }; +} + +#endif // SHOGUN_PARAMETEROBSERVERTENSORBOARD_H +#endif // HAVE_TFLOGGER diff --git a/src/shogun/lib/slep/SpInvCoVa/invCov.cpp b/src/shogun/lib/slep/SpInvCoVa/invCov.cpp deleted file mode 100644 index 3a2e46130b4..00000000000 --- a/src/shogun/lib/slep/SpInvCoVa/invCov.cpp +++ /dev/null @@ -1,231 +0,0 @@ -/* This program is free software: you can redistribute it and/or modify - * it under the terms of the GNU General Public License as published by - * the Free Software Foundation, either version 3 of the License, or - * (at your option) any later version. - * - * This program is distributed in the hope that it will be useful, - * but WITHOUT ANY WARRANTY; without even the implied warranty of - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the - * GNU General Public License for more details. - * - * You should have received a copy of the GNU General Public License - * along with this program. If not, see . - * - * Copyright (C) 2009 - 2012 Jun Liu and Jieping Ye - */ - - -#include -#ifdef USE_GPL_SHOGUN -#include - -#include -#include - -void m_Ax(double *Ax, double *A, double *x, int n, int ith) -{ - int i, j; - double t; - for(i=0;i lambda) - x_new= (s_v-lambda) / u; - else{ - if(s_v < -lambda) - x_new= (s_v + lambda) / u; - else - x_new=0; - } - if (x[i]!=x_new){ - for(j=0;j. - * - * Copyright (C) 2009 - 2012 Jun Liu and Jieping Ye - */ - - -#ifndef INVCOV_SLEP -#define INVCOV_SLEP - -#include -#ifdef USE_GPL_SHOGUN - -/* - * A: n x n - * x: n x 1 - * Ax: n x 1 - * - * Perform the task of Ax= A* x, - * where the ith row and column in A, and ith row in x - * are undefined, so that in Ax, the ith row has no meaning - */ -void m_Ax(double *Ax, double *A, double *x, int n, int ith); - -int lassoCD(double *Theta, double *W, double *S, double lambda, int n, - int ith, int flag, int maxIter, double fGap, double xGap); - -void invCov(double *Theta, double *W, double *S, double lambda, - double sum_S, int n, - int LassoMaxIter, double fGap, - double xGap, /*for the Lasso (inner iteration)*/ - int maxIter, double xtol); /*for the outer iteration*/ - -#endif //USE_GPL_SHOGUN -#endif /* ----- #ifndef INVCOV_SLEP ----- */ - diff --git a/src/shogun/lib/slep/flsa/flsa.cpp b/src/shogun/lib/slep/flsa/flsa.cpp deleted file mode 100644 index c0b1bc4d5dc..00000000000 --- a/src/shogun/lib/slep/flsa/flsa.cpp +++ /dev/null @@ -1,251 +0,0 @@ -/* This program is free software: you can redistribute it and/or modify - * it under the terms of the GNU General Public License as published by - * the Free Software Foundation, either version 3 of the License, or - * (at your option) any later version. - * - * This program is distributed in the hope that it will be useful, - * but WITHOUT ANY WARRANTY; without even the implied warranty of - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the - * GNU General Public License for more details. - * - * You should have received a copy of the GNU General Public License - * along with this program. If not, see . - * - * Copyright (C) 2009 - 2012 Jun Liu and Jieping Ye - */ - -#include -#ifdef USE_GPL_SHOGUN - -#include -#include -#include -#include -#include - -void flsa(double *x, double *z, double *infor, - double * v, double *z0, - double lambda1, double lambda2, int n, - int maxStep, double tol, int tau, int flag) -{ - - int i, nn=n-1, m; - double zMax, temp; - double *Av, *g, *s; - int iterStep, numS; - double gap; - double *zz = NULL; /*to replace z0, so that z0 shall not revised after */ - - - Av=(double *) malloc(sizeof(double)*nn); - - /* - Compute Av= A*v (n=4, nn=3) - A= [ -1 1 0 0; - 0 -1 1 0; - 0 0 -1 1] - */ - - for (i=0;i= zMax, which leads to a solution with same entry values - 2) lambda2 < zMax, which needs to first run sfa, and then perform soft thresholding - */ - - /* - First case: lambda2 >= zMax - */ - if (lambda2 >= zMax){ - - temp=0; - m=n%5; - if (m!=0){ - for (i=0;i lambda1) - temp= temp-lambda1; - else - if (temp < -lambda1) - temp= temp+lambda1; - else - temp=0; - - m=n%7; - if (m!=0){ - for (i=0;i lambda2) - z[i]=lambda2; - else - if (z0[i]<-lambda2) - z[i]=-lambda2; - else - z[i]=z0[i]; - } - } - else{ - if (lambda2 >= 0.5 * zMax){ - for (i=0;i lambda2) - z[i]=lambda2; - else - if (z[i]<-lambda2) - z[i]=-lambda2; - } - } - else{ - for (i=0;i=1 && flag<=4){ - zz =(double *) malloc(sizeof(double)*nn); - - for (i=0;i lambda1) - x[i]-=lambda1; - else - if (x[i]<-lambda1) - x[i]+=lambda1; - else - x[i]=0; - - - free(Av); - free(g); - free(s); - - if (infor) - { - infor[0]=gap; - infor[1]=iterStep; - infor[2]=zMax; - infor[3]=numS; - } -} - -#endif //USE_GPL_SHOGUN diff --git a/src/shogun/lib/slep/flsa/flsa.h b/src/shogun/lib/slep/flsa/flsa.h deleted file mode 100644 index eb78b76a53b..00000000000 --- a/src/shogun/lib/slep/flsa/flsa.h +++ /dev/null @@ -1,123 +0,0 @@ -/* This program is free software: you can redistribute it and/or modify - * it under the terms of the GNU General Public License as published by - * the Free Software Foundation, either version 3 of the License, or - * (at your option) any later version. - * - * This program is distributed in the hope that it will be useful, - * but WITHOUT ANY WARRANTY; without even the implied warranty of - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the - * GNU General Public License for more details. - * - * You should have received a copy of the GNU General Public License - * along with this program. If not, see . - * - * Copyright (C) 2009 - 2012 Jun Liu and Jieping Ye - */ - - -#ifndef FLSA_SLEP -#define FLSA_SLEP - -#include -#ifdef USE_GPL_SHOGUN - -/* - - In this file, we solve the Fused Lasso Signal Approximator (FLSA) problem: - - min_x 1/2 \|x-v\|^2 + lambda1 * \|x\|_1 + lambda2 * \|A x\|_1, (1) - - It can be shown that, if x* is the solution to - - min_x 1/2 \|x-v\|^2 + lambda2 \|A x\|_1, (2) - - then - x**= sgn(x*) max(|x*|-lambda_1, 0) (3) - - is the solution to (1). - - By some derivation (see the description in sfa.h), (2) can be solved by - - x*= v - A^T z*, - - where z* is the optimal solution to - - min_z 1/2 z^T A AT z - < z, A v>, - subject to \|z\|_{infty} \leq lambda2 (4) - */ - - - -/* - - In flsa, we solve (1) corresponding to a given (lambda1, lambda2) - - void flsa(double *x, double *z, double *gap, - double * v, double *z0, - double lambda1, double lambda2, int n, - int maxStep, double tol, int flag) - - Output parameters: -x: the solution to problem (1) -z: the solution to problem (4) -infor: the information about running the subgradient finding algorithm -infor[0] = gap: the computed gap (either the duality gap -or the summation of the absolute change of the adjacent solutions) -infor[1] = steps: the number of iterations -infor[2] = lambad2_max: the maximal value of lambda2_max -infor[3] = numS: the number of elements in the support set - -Input parameters: -v: the input vector to be projected -z0: a guess of the solution of z - -lambad1: the regularization parameter -labmda2: the regularization parameter -n: the length of v and x - -maxStep: the maximal allowed iteration steps -tol: the tolerance parameter -tau: the program sfa is checked every tau iterations for termination -flag: the flag for initialization and deciding calling sfa -switch ( flag ) -1-4, 11-14: sfa - -switch ( flag ) -case 1, 2, 3, or 4: -z0 is a "good" starting point -(such as the warm-start of the previous solution, -or the user want to test the performance of this starting point; -the starting point shall be further projected to the L_{infty} ball, -to make sure that it is feasible) - -case 11, 12, 13, or 14: z0 is a "random" guess, and thus not used -(we shall initialize z as follows: -if lambda2 >= 0.5 * lambda_2^max, we initialize the solution of the linear system; -if lambda2 < 0.5 * lambda_2^max, we initialize with zero -this solution is projected to the L_{infty} ball) - -switch( flag ) -5, 15: sfa_special - -switch( flag ) -5: z0 is a good starting point -15: z0 is a bad starting point, use the solution of the linear system - - -switch( flag ) -6, 16: sfa_one - -switch( flag ) -6: z0 is a good starting point -16: z0 is a bad starting point, use the solution of the linear system - -Revision made on October 31, 2009. -The input variable z0 is not modified after calling sfa. For this sake, we allocate a new variable zz to replace z0. -*/ -void flsa(double *x, double *z, double *infor, - double * v, double *z0, - double lambda1, double lambda2, int n, - int maxStep, double tol, int tau, int flag); -#endif //USE_GPL_SHOGUN -#endif /* ----- #ifndef FLSA_SLEP ----- */ - diff --git a/src/shogun/lib/slep/flsa/sfa.cpp b/src/shogun/lib/slep/flsa/sfa.cpp deleted file mode 100644 index e30dc348ddb..00000000000 --- a/src/shogun/lib/slep/flsa/sfa.cpp +++ /dev/null @@ -1,1892 +0,0 @@ -/* This program is free software: you can redistribute it and/or modify - * it under the terms of the GNU General Public License as published by - * the Free Software Foundation, either version 3 of the License, or - * (at your option) any later version. - * - * This program is distributed in the hope that it will be useful, - * but WITHOUT ANY WARRANTY; without even the implied warranty of - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the - * GNU General Public License for more details. - * - * You should have received a copy of the GNU General Public License - * along with this program. If not, see . - * - * Copyright (C) 2009 - 2012 Jun Liu and Jieping Ye - */ - -#include -#ifdef USE_GPL_SHOGUN - -#include -#include -#include -#include - -#define delta 1e-10 - -/* - Revision History - - First Version available on October 10, 2009 - - A runnable version on October 15, 2009 - - Major revision on October 29, 2009 - (Some functions appearing in a previous version have deleted, please refer to the previous version for the old functions. - Some new functions have been added as well) - -*/ - -/* - - Files contained in this header file sfa.h: - - 1. Algorithms for solving the linear system A A^T z0 = Av (see the description of A from the following context) - - void Thomas(double *zMax, double *z0, - double * Av, int nn) - - void Rose(double *zMax, double *z0, - double * Av, int nn) - - int supportSet(double *x, double *v, double *z, - double *g, int * S, double lambda, int nn) - - void dualityGap(double *gap, double *z, - double *g, double *s, double *Av, - double lambda, int nn) - - void dualityGap2(double *gap, double *z, - double *g, double *s, double *Av, - double lambda, int nn) - - - 2. The Subgraident Finding Algorithm (SFA) for solving problem (4) (refer to the description of the problem for detail) - - int sfa(double *x, double *gap, - double *z, double *z0, double * v, double * Av, - double lambda, int nn, int maxStep, - double *s, double *g, - double tol, int tau, int flag) - - int sfa_special(double *x, double *gap, - double *z, double * v, double * Av, - double lambda, int nn, int maxStep, - double *s, double *g, - double tol, int tau) - - int sfa_one(double *x, double *gap, - double *z, double * v, double * Av, - double lambda, int nn, int maxStep, - double *s, double *g, - double tol, int tau) - - -*/ - - -/* - - Some mathematical background. - - In this file, we discuss how to solve the following subproblem, - - min_x 1/2 \|x-v\|^2 + lambda \|A x\|_1, (1) - - which is a key problem used in the Fused Lasso Signal Approximator (FLSA). - - Also, note that, FLSA is a building block for solving the optimation problmes with fused Lasso penalty. - - In (1), x and v are n-dimensional vectors, - and A is a matrix with size (n-1) x n, and is defined as follows (e.g., n=4): - A= [ -1 1 0 0; - 0 -1 1 0; - 0 0 -1 1] - - The above problem can be reformulated as the following equivalent min-max optimization problem - - min_x max_z 1/2 \|x-v\|^2 + - subject to \|z\|_{infty} \leq lambda (2) - - - It is easy to get that, at the optimal point - - x = v - AT z, (3) - - where z is the optimal solution to the following optimization problem - - min_z 1/2 z^T A AT z - < z, A v>, - subject to \|z\|_{infty} \leq lambda (4) - - - - Let B=A A^T. It is easy to get that B is a (n-1) x (n-1) tridiagonal matrix. - When n=5, B is defined as: - B= [ 2 -1 0 0; - -1 2 -1 0; - 0 -1 2 -1; - 0 0 -1 2] - - Let z0 be the solution to the linear system: - - A A^T * z0 = A * v (5) - - The problem (5) can be solve by the Thomas Algorithm, in about 5n multiplications and 4n additions. - - It can also be solved by the Rose's Algorithm, in about 2n multiplications and 2n additions. - - Moreover, considering the special structure of the matrix A (and B), - it can be solved in about n multiplications and 3n additions - - If lambda \geq \|z0\|_{infty}, x_i= mean(v), for all i, - the problem (1) admits near analytical solution - - - We have also added the restart technique, please refer to our paper for detail! - -*/ - - -/* -/////////////// Solving the Linear System via Thomas's Algorithm \\\\\\\\\\\\\\\\\\ -*/ - -void Thomas(double *zMax, double *z0, double * Av, int nn){ - - /* - - We apply the Tomas algorithm for solving the following linear system - B * z0 = Av - Thomas algorithm is also called the tridiagonal matrix algorithm - - B=[ 2 -1 0 0; - -1 2 -1 0; - 0 -1 2 -1; - 0 0 -1 2] - - z0 is the result, Av is unchanged after the computation - - - c is a precomputed nn dimensional vector - c=[-1/2, -2/3, -3/4, -4/5, ..., -nn/(nn+1)] - - c[i]=- (i+1) / (i+2) - c[i-1]=- i / (i+1) - - z0 is an nn dimensional vector - -*/ - - int i; - double tt, z_max; - - /* - Modify the coefficients in Av (copy to z0) - */ - z0[0]=Av[0]/2; - for (i=1;i < nn; i++){ - tt=Av[i] + z0[i-1]; - z0[i]=tt - tt / (i+2); - } - - /*z0[i]=(Av[i] + z0[i-1]) * (i+1) / (i+2);*/ - - /*z0[i]=(Av[i] + z0[i-1])/ ( 2 - i / (i+1));*/ - - - /* - Back substitute (obtain the result in z0) - */ - z_max= fabs(z0[nn-1]); - - for (i=nn-2; i>=0; i--){ - - z0[i]+= z0[i+1] - z0[i+1]/ (i+2); - - /*z0[i]+= z0[i+1] * (i+1) / (i+2);*/ - - tt=fabs(z0[i]); - - if (tt > z_max) - z_max=tt; - - } - *zMax=z_max; - -} - - - - -/* -/////////////// Solving the Linear System via Rose's Algorithm \\\\\\\\\\\\\\\\\\ -*/ - -void Rose(double *zMax, double *z0, double * Av, int nn){ - - /* - We use the Rose algorithm for solving the following linear system - B * z0 = Av - - - B=[ 2 -1 0 0; - -1 2 -1 0; - 0 -1 2 -1; - 0 0 -1 2] - - z0 is the result, Av is unchanged after the computation - - z0 is an nn dimensional vector - -*/ - - int i, m; - double s=0, z_max; - - - /* - We follow the style in CLAPACK - */ - m= nn % 5; - if (m!=0){ - for (i=0;i=0; i--){ - z0[i]=Av[i] + z0[i+1]; - } - - /* - from 0 to nn-1 - */ - z_max= fabs(z0[0]); - for (i=0; i z_max) - z_max=s; - - } - *zMax=z_max; - -} - - - -/* -//////////////// compute x for restarting \\\\\\\\\\\\\\\\\\\\\\\\\ - -x=omega(z) - -v: the vector to be projected -z: the approximate solution -g: the gradient at z (g should be computed before calling this function - -nn: the length of z, g, and S (maximal length for S) - -n: the length of x and v - -S: records the indices of the elements in the support set -*/ - -int supportSet(double *x, double *v, double *z, double *g, int * S, double lambda, int nn){ - - int i, j, n=nn+1, numS=0; - double temp; - - - /* - we first scan z and g to obtain the support set S - */ - - /*numS: number of the elements in the support set S*/ - for(i=0;idelta) )){ - S[numS]=i; - numS++; - } - } - - /* - printf("\n %d",numS); - */ - - if (numS==0){ /*this shows that S is empty*/ - temp=0; - for (i=0;i=1 - */ - - /*process the first block - - j=0 - */ - temp=0; - for (i=0;i<=S[0]; i++) - temp+=v[i]; - /*temp =sum (v [0: s[0] ]*/ - temp=( temp + z[ S[0] ] ) / (S[0] +1); - for (i=0;i<=S[0]; i++) - x[i]=temp; - - - /*process the middle blocks - - If numS=1, it belongs the last block - */ - for (j=1; j < numS; j++){ - temp=0; - for (i= S[j-1] +1; i<= S[j]; i++){ - temp+=v[i]; - } - - /*temp =sum (v [ S[j-1] +1: s[j] ]*/ - - temp=(temp - z[ S[j-1] ] + z[ S[j] ])/ (S[j]- S[j-1]); - - for (i= S[j-1] +1; i<= S[j]; i++){ - x[i]=temp; - } - } - - /*process the last block - j=numS-1; - */ - temp=0; - for (i=S[numS-1] +1 ;i< n; i++) - temp+=v[i]; - /*temp =sum (v [ (S[numS-1] +1): (n-1) ]*/ - - temp=( temp - z[ S[numS-1] ] ) / (nn - S[numS-1]); /*S[numS-1] <= nn-1*/ - - for (i=S[numS-1] +1 ;i< n; i++) - x[i]=temp; - - return numS; - -} - - - -/* - -//////////// Computing the duality gap \\\\\\\\\\\\\\\\\\\\\\\\\\ - -we compute the duality corresponding the solution z - -z: the approximate solution -g: the gradient at z (we recompute the gradient) -s: an auxiliary variable -Av: A*v - -nn: the lenght for z, g, s, and Av - -The variables g and s shall be revised. - -The variables z and Av remain unchanged. -*/ - -void dualityGap(double *gap, double *z, double *g, double *s, double *Av, double lambda, int nn){ - - int i, m; - double temp; - - - g[0]=z[0] + z[0] - z[1] - Av[0]; - for (i=1;i0) - s[i]=lambda + z[i]; - else - s[i]=-lambda + z[i]; - - - temp=0; - m=nn%5; - - if (m!=0){ - for(i=0;i0) - s[i]=lambda + z[i]; - else - s[i]=-lambda + z[i]; - - - temp=0; - m=nn%5; - - if (m!=0){ - for(i=0;i=3, and thus nn>=2 - - We have two ways for recovering x. - The first way is x = v - A^T z - The second way is x =omega(z) - */ - - temp=0; - m=nn%5; - if (m!=0){ - for (i=0;i funVal1){ /* - we compute the solution by the first way - */ - x[0]=v[0] + z[0]; - for(i=1;ilambda) - g[i]=lambda; - else - if (g[i]<-lambda) - g[i]=-lambda; - } - - - free (S); - free (x); - free (s); - free (Av); - -} - - - -/* - -/////////////////////////////////////// Explanation for the function sfa \\\\\\\\\\\\\\\\\\\\\\\\\\\\ - -Our objective is to solve the fused Lasso signal approximator (flsa) problem: - -min_x g(x) 1/2 \|x-v\|^2 + lambda \|A x\|_1, (1) - -Let x* be the solution (which is unique), it satisfies - -0 in x* - v + A^T * lambda *SGN(Ax*) (2) - -To solve x*, it suffices to find - -y* in A^T * lambda *SGN(Ax*) (3) -that satisfies - -x* - v + y* =0 (4) -which leads to -x*= v - y* (5) - -Due to the uniqueness of x*, we conclude that y* is unique. - -As y* is a subgradient of lambda \|A x*\|_1, -we name our method as Subgradient Finding Algorithm (sfa). - -y* in (3) can be further written as - -y*= A^T * z* (6) -where - -z* in lambda* SGN (Ax*) (7) - -From (6), we have -z* = (A A^T)^{-1} A * y* (8) - -Therefore, from the uqniueness of y*, we conclude that z* is also unique. -Next, we discuss how to solve this unique z*. - -The problem (1) can reformulated as the following equivalent problem: - -min_x max_z f(x, z)= 1/2 \|x-v\|^2 + -subject to \|z\|_{infty} \leq lambda (9) - -At the saddle point, we have - -x = v - AT z, (10) - -which somehow concides with (5) and (6) - -Plugging (10) into (9), we obtain the problem - -min_z 1/2 z^T A AT z - < z, A v>, -subject to \|z\|_{infty} \leq lambda, (11) - -In this program, we apply the Nesterov's method for solving (11). - - -Duality gap: - -At a given point z0, we compute x0= v - A^T z0. -It is easy to show that -min_x f(x, z0) = f(x0, z0) <= max_z f(x0, z) (12) - -Moreover, we have -max_z f(x0, z) - min_x f(x, z0) -<= lambda * \|A x0\|_1 - < z0, Av - A A^T z0> (13) - -It is also to get that - -f(x0, z0) <= f(x*, z*) <= max_z f(x0, z) (14) - -g(x*)=f(x*, z*) (15) - -g(x0)=max_z f(x0, z) (17) - - Therefore, we have - -g(x0)-g(x*) <= lambda * \|A x0\|_1 - < z0, Av - A A^T z0> (18) - - - We have applied a restarting technique, which is quite involved; and thus, we do not explain here. - - /////////////////////////////////////// Explanation for the function sfa \\\\\\\\\\\\\\\\\\\\\\\\\\\\ - */ - - - /* - //////////// sfa \\\\\\\\\\\\\\\\\\\\\ - - For sfa, the stepsize of the Nesterov's method is fixed to 1/4, so that no line search is needed. - - - - Explanation of the parameters: - - Output parameters - x: the solution to the primal problem - gap: the duality gap (pointer) - - Input parameters - z: the solution to the dual problem (before calling this function, z contains a starting point) - !!!!we assume that the starting point has been successfully initialized in z !!!! - z0: a variable used for multiple purposes: - 1) the previous solution z0 - 2) the difference between z and z0, i.e., z0=z- z0 - - lambda: the regularization parameter (and the radius of the infity ball, see (11)). - nn: the length of z, z0, Av, g, and s - maxStep: the maximal number of iterations - - v: the point to be projected (not changed after the program) - Av: A*v (not changed after the program) - - s: the search point (used for multiple purposes) - g: the gradient at g (and it is also used for multiple purposes) - - tol: the tolerance of the gap - tau: the duality gap or the restarting technique is done every tau steps - flag: if flag=1, we apply the resart technique - flag=2, just run the SFA algorithm, terminate it when the absolution change is less than tol - flag=3, just run the SFA algorithm, terminate it when the duality gap is less than tol - flag=4, just run the SFA algorithm, terminate it when the relative duality gap is less than tol - - - We would like to emphasis that the following assumptions - have been checked in the functions that call this function: - 1) 0< lambda < z_max - 2) nn >=2 - 3) z has been initialized with a starting point - 4) z0 has been initialized with all zeros - - The termination condition is checked every tau iterations. - - For the duality gap, please refer to (12-18) - */ - - int sfa(double *x, double *gap, int * activeS, - double *z, double *z0, double * v, double * Av, - double lambda, int nn, int maxStep, - double *s, double *g, - double tol, int tau, int flag){ - - int i, iterStep, m, tFlag=0, n=nn+1; - double alphap=0, alpha=1, beta=0, temp; - int* S=(int *) malloc(sizeof(int)*nn); - double gapp=-1, gappp=-1; /*gapp denotes the previous gap*/ - int numS=-1, numSp=-2, numSpp=-3;; - /* - numS denotes the number of elements in the Support Set S - numSp denotes the number of elements in the previous Support Set S - */ - - *gap=-1; /*initial a value -1*/ - - /* - The main algorithm by Nesterov's method - - B is an nn x nn tridiagonal matrix. - - The nn eigenvalues of B are 2- 2 cos (i * PI/ n), i=1, 2, ..., nn - */ - - for (iterStep=1; iterStep<=maxStep; iterStep++){ - - - /*------------- Step 1 ---------------------*/ - - beta=(alphap -1 ) / alpha; - /* - compute search point - - s= z + beta * z0 - - We follow the style of CLAPACK - */ - m=nn % 5; - if (m!=0){ - for (i=0;i=3, which leads to nn>=2 - */ - g[0]=s[0] + s[0] - s[1] - Av[0]; - for (i=1;ilambda) - z[i]=lambda; - else - if (z[i]<-lambda) - z[i]=-lambda; - } - - /* - compute the difference between the new solution - and the previous solution (stored in z0=-z_p) - - the difference is written to z0 - */ - - m=nn%5; - if (m!=0){ - for (i=0;i - = lambda * \|g\|_1 + - - In fact, gap=0 indicates that, - if g_i >0, then z_i=-lambda - if g_i <0, then z_i=lambda - */ - - gappp=gapp; - gapp=*gap; /*record the previous gap*/ - numSpp=numSp; - numSp=numS; /*record the previous numS*/ - - dualityGap(gap, z, g, s, Av, lambda, nn); - /*g is computed as the gradient of z in this function*/ - - - /* - printf("\n Iteration: %d, gap=%e, numS=%d", iterStep, *gap, numS); - */ - - /* - If *gap <=tol, we terminate the iteration - Otherwise, we restart the algorithm - */ - - if (*gap <=tol){ - tFlag=1; - break; - - } /* end of *gap <=tol */ - else{ - - /* we apply the restarting technique*/ - - /* - we compute the solution by the second way - */ - numS = supportSet(x, v, z, g, S, lambda, nn); - /*g, the gradient of z should be computed before calling this function*/ - - /*With x, we compute z via - AA^T z = Av - Ax - */ - - /* - printf("\n iterStep=%d, numS=%d, gap=%e",iterStep, numS, *gap); - */ - - - m=1; - if (nn > 1000000) - m=10; - else - if (nn > 100000) - m=5; - - if ( abs(numS-numSp) < m){ - - numS=generateSolution(x, z, gap, v, Av, - g, s, S, lambda, nn); - /*g, the gradient of z should be computed before calling this function*/ - - - if (*gap lambda) - z[i]=lambda; - else - if (z[i]<-lambda) - z[i]=-lambda; - } - - - - m=nn%7; - if (m!=0){ - for (i=0;i - = lambda * \|g\|_1 + - - In fact, gap=0 indicates that, - if g_i >0, then z_i=-lambda - if g_i <0, then z_i=lambda - */ - - - g[0]=z[0] + z[0] - z[1] - Av[0]; - for (i=1;i0) - s[i]=lambda + z[i]; - else - s[i]=-lambda + z[i]; - - temp=0; - m=nn%5; - if (m!=0){ - for(i=0;i - = lambda * \|g\|_1 + - - In fact, gap=0 indicates that, - if g_i >0, then z_i=-lambda - if g_i <0, then z_i=lambda - - - Here, the "relative duality gap" is defined as: - duality gap / - 1/2 \|A^T z\|^2 + < z, Av> - - We efficiently compute - 1/2 \|A^T z\|^2 + < z, Av> using the following relationship - - - 1/2 \|A^T z\|^2 + < z, Av> - = -1/2 - = -1/2 - */ - - - g[0]=z[0] + z[0] - z[1] - Av[0]; - for (i=1;i0) - s[i]=lambda + z[i]; - else - s[i]=-lambda + z[i]; - - temp=0; - m=nn%5; - if (m!=0){ - for(i=0;i - =-1/2 - */ - - temp=0; - m=nn%5; - if (m!=0){ - for(i=0;i=3, and thus nn>=2 - - We have two ways for recovering x. - The first way is x = v - A^T z - The second way is x =omega(z) - */ - - /* - We first compute the objective function value of the first choice in terms f(x), see our paper - */ - - /* - for numerical reason, we do a gradient descent step - */ - - /* - --------------------------------------------------- - A gradient step begins - */ - g[0]=z[0] + z[0] - z[1] - Av[0]; - for (i=1;ilambda) - z[i]=lambda; - else - if (z[i]<-lambda) - z[i]=-lambda; - } - - /* - --------------------------------------------------- - A gradient descent step ends - */ - - /*compute the gradient at z*/ - - g[0]=z[0] + z[0] - z[1] - Av[0]; - for (i=1;ilambda) - z[i]=lambda; - else - if (z[i]<-lambda) - z[i]=-lambda; - } - - - if (iterStep%tau==0){ - gapp=*gap; /*record the previous gap*/ - - dualityGap(gap, z, g, s, Av, lambda, nn); - - /* - printf("\n iterStep=%d, numS=%d, gap=%e, diff=%e",iterStep, numS, *gap, *gap -gapp); - -*/ - - if (*gap <=tol){ - //tFlag=1; - break; - } - - if ( (*gap <1) && (numS==numSp) && (gapp == *gap) ){ - //tFlag=1; - break; - /*we terminate the program is *gap <1 - numS==numSP - and gapp==*gap - */ - } - - }/*end of if tau*/ - - }/*end for */ - - free (S); - - * activeS=numS; - return(iterStep); - -} - - -/* - - We do one gradient descent, and then restart the program - */ - - -int sfa_one(double *x, double *gap, int * activeS, - double *z, double * v, double * Av, - double lambda, int nn, int maxStep, - double *s, double *g, - double tol, int tau){ - - int i, iterStep, m; - int tFlag=0; - //int n=nn+1; - double temp; - int* S=(int *) malloc(sizeof(int)*nn); - double gapp=-1, gappp=-2; /*gapp denotes the previous gap*/ - int numS=-100, numSp=-200, numSpp=-300; - /* - numS denotes the number of elements in the Support Set S - numSp denotes the number of elements in the previous Support Set S - */ - - *gap=-1; /*initialize *gap a value*/ - - /* - The main algorithm by Nesterov's method - - B is an nn x nn tridiagonal matrix. - - The nn eigenvalues of B are 2- 2 cos (i * PI/ n), i=1, 2, ..., nn - */ - - - /* - we first do a gradient step based on z - */ - - - /* - --------------------------------------------------- - A gradient step begins - */ - g[0]=z[0] + z[0] - z[1] - Av[0]; - for (i=1;ilambda) - z[i]=lambda; - else - if (z[i]<-lambda) - z[i]=-lambda; - } - - /* - --------------------------------------------------- - A gradient descent step ends - */ - - - /*compute the gradient at z*/ - - g[0]=z[0] + z[0] - z[1] - Av[0]; - for (i=1;ilambda) - z[i]=lambda; - else - if (z[i]<-lambda) - z[i]=-lambda; - } - - /* - --------------------------------------------------- - restart the algorithm with x=omega(z) - - we have computed a new z, based on the above relationship - */ - - - /* - --------------------------------------------------- - A gradient step begins - */ - g[0]=z[0] + z[0] - z[1] - Av[0]; - for (i=1;ilambda) - z[i]=lambda; - else - if (z[i]<-lambda) - z[i]=-lambda; - } - - /* - --------------------------------------------------- - A gradient descent step ends - */ - - /*compute the gradient at z*/ - - g[0]=z[0] + z[0] - z[1] - Av[0]; - for (i=1;i 1000000) - m=5; - else - if (nn > 100000) - m=3; - - if ( abs( numS-numSp) . - * - * Copyright (C) 2009 - 2012 Jun Liu and Jieping Ye - */ - - -#ifndef SFA_SLEP -#define SFA_SLEP - -#include -#ifdef USE_GPL_SHOGUN - -/* - Revision History - - First Version available on October 10, 2009 - - A runnable version on October 15, 2009 - - Major revision on October 29, 2009 - (Some functions appearing in a previous version have deleted, please refer to the previous version for the old functions. - Some new functions have been added as well) - -*/ - -/* - - Files contained in this header file sfa.h: - - 1. Algorithms for solving the linear system A A^T z0 = Av (see the description of A from the following context) - - void Thomas(double *zMax, double *z0, - double * Av, int nn) - - void Rose(double *zMax, double *z0, - double * Av, int nn) - - int supportSet(double *x, double *v, double *z, - double *g, int * S, double lambda, int nn) - - void dualityGap(double *gap, double *z, - double *g, double *s, double *Av, - double lambda, int nn) - - void dualityGap2(double *gap, double *z, - double *g, double *s, double *Av, - double lambda, int nn) - - - 2. The Subgraident Finding Algorithm (SFA) for solving problem (4) (refer to the description of the problem for detail) - - int sfa(double *x, double *gap, - double *z, double *z0, double * v, double * Av, - double lambda, int nn, int maxStep, - double *s, double *g, - double tol, int tau, int flag) - - int sfa_special(double *x, double *gap, - double *z, double * v, double * Av, - double lambda, int nn, int maxStep, - double *s, double *g, - double tol, int tau) - - int sfa_one(double *x, double *gap, - double *z, double * v, double * Av, - double lambda, int nn, int maxStep, - double *s, double *g, - double tol, int tau) - - -*/ - - -/* - - Some mathematical background. - - In this file, we discuss how to solve the following subproblem, - - min_x 1/2 \|x-v\|^2 + lambda \|A x\|_1, (1) - - which is a key problem used in the Fused Lasso Signal Approximator (FLSA). - - Also, note that, FLSA is a building block for solving the optimation problmes with fused Lasso penalty. - - In (1), x and v are n-dimensional vectors, - and A is a matrix with size (n-1) x n, and is defined as follows (e.g., n=4): - A= [ -1 1 0 0; - 0 -1 1 0; - 0 0 -1 1] - - The above problem can be reformulated as the following equivalent min-max optimization problem - - min_x max_z 1/2 \|x-v\|^2 + - subject to \|z\|_{infty} \leq lambda (2) - - - It is easy to get that, at the optimal point - - x = v - AT z, (3) - - where z is the optimal solution to the following optimization problem - - min_z 1/2 z^T A AT z - < z, A v>, - subject to \|z\|_{infty} \leq lambda (4) - - - - Let B=A A^T. It is easy to get that B is a (n-1) x (n-1) tridiagonal matrix. - When n=5, B is defined as: - B= [ 2 -1 0 0; - -1 2 -1 0; - 0 -1 2 -1; - 0 0 -1 2] - - Let z0 be the solution to the linear system: - - A A^T * z0 = A * v (5) - - The problem (5) can be solve by the Thomas Algorithm, in about 5n multiplications and 4n additions. - - It can also be solved by the Rose's Algorithm, in about 2n multiplications and 2n additions. - - Moreover, considering the special structure of the matrix A (and B), - it can be solved in about n multiplications and 3n additions - - If lambda \geq \|z0\|_{infty}, x_i= mean(v), for all i, - the problem (1) admits near analytical solution - - - We have also added the restart technique, please refer to our paper for detail! - -*/ - - -void Thomas(double *zMax, double *z0, double * Av, int nn); - -void Rose(double *zMax, double *z0, double * Av, int nn); - -/* -//////////////// compute x for restarting \\\\\\\\\\\\\\\\\\\\\\\\\ - -x=omega(z) - -v: the vector to be projected -z: the approximate solution -g: the gradient at z (g should be computed before calling this function - -nn: the length of z, g, and S (maximal length for S) - -n: the length of x and v - -S: records the indices of the elements in the support set -*/ -int supportSet(double *x, double *v, double *z, double *g, int * S, double lambda, int nn); - -/* -//////////// Computing the duality gap \\\\\\\\\\\\\\\\\\\\\\\\\\ - -we compute the duality corresponding the solution z - -z: the approximate solution -g: the gradient at z (we recompute the gradient) -s: an auxiliary variable -Av: A*v - -nn: the lenght for z, g, s, and Av - -The variables g and s shall be revised. - -The variables z and Av remain unchanged. -*/ -void dualityGap(double *gap, double *z, double *g, double *s, double *Av, double lambda, int nn); - -/* - Similar to dualityGap, - - The difference is that, we assume that g has been computed. - */ -void dualityGap2(double *gap, double *z, double *g, double *s, double *Av, double lambda, int nn); - -/* -generateSolution: - -generate the solution x based on the information of z and g -(!!!!we assume that g has been computed as the gradient of z!!!!) - -*/ -int generateSolution(double *x, double *z, double *gap, - double *v, double *Av, - double *g, double *s, int *S, - double lambda, int nn); - -void restartMapping(double *g, double *z, double * v, - double lambda, int nn); - -/* -/////////////////////////////////////// Explanation for the function sfa \\\\\\\\\\\\\\\\\\\\\\\\\\\\ - -Our objective is to solve the fused Lasso signal approximator (flsa) problem: - -min_x g(x) 1/2 \|x-v\|^2 + lambda \|A x\|_1, (1) - -Let x* be the solution (which is unique), it satisfies - -0 in x* - v + A^T * lambda *SGN(Ax*) (2) - -To solve x*, it suffices to find - -y* in A^T * lambda *SGN(Ax*) (3) -that satisfies - -x* - v + y* =0 (4) -which leads to -x*= v - y* (5) - -Due to the uniqueness of x*, we conclude that y* is unique. - -As y* is a subgradient of lambda \|A x*\|_1, -we name our method as Subgradient Finding Algorithm (sfa). - -y* in (3) can be further written as - -y*= A^T * z* (6) -where - -z* in lambda* SGN (Ax*) (7) - -From (6), we have -z* = (A A^T)^{-1} A * y* (8) - -Therefore, from the uqniueness of y*, we conclude that z* is also unique. -Next, we discuss how to solve this unique z*. - -The problem (1) can reformulated as the following equivalent problem: - -min_x max_z f(x, z)= 1/2 \|x-v\|^2 + -subject to \|z\|_{infty} \leq lambda (9) - -At the saddle point, we have - -x = v - AT z, (10) - -which somehow concides with (5) and (6) - -Plugging (10) into (9), we obtain the problem - -min_z 1/2 z^T A AT z - < z, A v>, -subject to \|z\|_{infty} \leq lambda, (11) - -In this program, we apply the Nesterov's method for solving (11). - - -Duality gap: - -At a given point z0, we compute x0= v - A^T z0. -It is easy to show that -min_x f(x, z0) = f(x0, z0) <= max_z f(x0, z) (12) - -Moreover, we have -max_z f(x0, z) - min_x f(x, z0) -<= lambda * \|A x0\|_1 - < z0, Av - A A^T z0> (13) - -It is also to get that - -f(x0, z0) <= f(x*, z*) <= max_z f(x0, z) (14) - -g(x*)=f(x*, z*) (15) - -g(x0)=max_z f(x0, z) (17) - - Therefore, we have - -g(x0)-g(x*) <= lambda * \|A x0\|_1 - < z0, Av - A A^T z0> (18) - - - We have applied a restarting technique, which is quite involved; and thus, we do not explain here. - - /////////////////////////////////////// Explanation for the function sfa \\\\\\\\\\\\\\\\\\\\\\\\\\\\ - */ - - - /* - //////////// sfa \\\\\\\\\\\\\\\\\\\\\ - - For sfa, the stepsize of the Nesterov's method is fixed to 1/4, so that no line search is needed. - - - - Explanation of the parameters: - - Output parameters - x: the solution to the primal problem - gap: the duality gap (pointer) - - Input parameters - z: the solution to the dual problem (before calling this function, z contains a starting point) - !!!!we assume that the starting point has been successfully initialized in z !!!! - z0: a variable used for multiple purposes: - 1) the previous solution z0 - 2) the difference between z and z0, i.e., z0=z- z0 - - lambda: the regularization parameter (and the radius of the infity ball, see (11)). - nn: the length of z, z0, Av, g, and s - maxStep: the maximal number of iterations - - v: the point to be projected (not changed after the program) - Av: A*v (not changed after the program) - - s: the search point (used for multiple purposes) - g: the gradient at g (and it is also used for multiple purposes) - - tol: the tolerance of the gap - tau: the duality gap or the restarting technique is done every tau steps - flag: if flag=1, we apply the resart technique - flag=2, just run the SFA algorithm, terminate it when the absolution change is less than tol - flag=3, just run the SFA algorithm, terminate it when the duality gap is less than tol - flag=4, just run the SFA algorithm, terminate it when the relative duality gap is less than tol - - - We would like to emphasis that the following assumptions - have been checked in the functions that call this function: - 1) 0< lambda < z_max - 2) nn >=2 - 3) z has been initialized with a starting point - 4) z0 has been initialized with all zeros - - The termination condition is checked every tau iterations. - - For the duality gap, please refer to (12-18) - */ -int sfa(double *x, double *gap, int * activeS, - double *z, double *z0, double * v, double * Av, - double lambda, int nn, int maxStep, - double *s, double *g, - double tol, int tau, int flag); - -/* - - Refer to sfa for the defintions of the variables - - In this file, we restart the program every step, and neglect the gradient step. - - It seems that, this program does not converge. - - This function shows that the gradient step is necessary. - */ -int sfa_special(double *x, double *gap, int * activeS, - double *z, double * v, double * Av, - double lambda, int nn, int maxStep, - double *s, double *g, - double tol, int tau); - -/* - We do one gradient descent, and then restart the program - */ -int sfa_one(double *x, double *gap, int * activeS, - double *z, double * v, double * Av, - double lambda, int nn, int maxStep, - double *s, double *g, - double tol, int tau); -#endif //USE_GPL_SHOGUN -#endif /* ----- #ifndef SFA_SLEP ----- */ - diff --git a/src/shogun/lib/slep/flsa/tesla_proj.cpp b/src/shogun/lib/slep/flsa/tesla_proj.cpp deleted file mode 100644 index ec4ecfb5940..00000000000 --- a/src/shogun/lib/slep/flsa/tesla_proj.cpp +++ /dev/null @@ -1,192 +0,0 @@ -/* This program is free software: you can redistribute it and/or modify - * it under the terms of the GNU General Public License as published by - * the Free Software Foundation, either version 3 of the License, or - * (at your option) any later version. - * - * This program is distributed in the hope that it will be useful, - * but WITHOUT ANY WARRANTY; without even the implied warranty of - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the - * GNU General Public License for more details. - * - * You should have received a copy of the GNU General Public License - * along with this program. If not, see . - * - * Copyright (C) 2009 - 2012 Jun Liu and Jieping Ye - */ - -#ifdef USE_GPL_SHOGUN - -#ifndef TESLA_PROJ_SLEP -#define TESLA_PROJ_SLEP - -#include -#include -#include -#include -#include - - -/* - - Functions contained in "flsa.h" - - void flsa(double *x, double *z, double *infor, - double * v, double *z0, - double lambda1, double lambda2, int n, - int maxStep, double tol, int tau, int flag) - -*/ - -/* - In this file, we need to make use of the function flsa for solving the following problem - - min 1/2 \|X - V\|_2^2 + lambda1 * \|X\|_1 + lambda2 \|X A^T\|_1 (1) - - where X and V are of size p x n - - For the definition of A, please refer to "flsa.h" and the included "sfa.h". - - The problem can be decoupled into the following - - min_x 1/2 \|x-v\|^2 + lambda1 * \|x\|_1 + lambda2 * \|A x\|_1, (2) - - where x and v correspond to a row of of X and V, respectively. - - The problem (2) is essentially the flsa problem, and can be solved by the function flsa. - - - void tesla_proj(double *X, double *Z, double *gap, - double *V, double *Z0, - double lambda1, double lambda2, int p, int n, - int maxStep, double tol, int flag) - - Output parameters: -X: the solution (of size p x n) -Z: the auxiliary variable (result by subgradient finding), -Z can be used as a warm start for the next "tesla_proj", -size: p x (n-1) -gap: the gap for each decoupled flsa problem (of size p x 1) - -Input parameters: -V: the one to be projected -Z0: the starting point (see flag for whether it is used as the starting point) -size: p x (n-1) - -lambda1: the regularization parameter -lambda2: the regularization parameter -p: the number of rows in X and V -n: the number of columns in X and V - -maxStep: the maximal allowed iteration steps -tol: the tolerance parameter -flag: the flag for initialization and deciding calling sfa -switch ( flag ) -1-4, 11-14: sfa - -switch ( flag ) -case 1, 2, 3, or 4: -z0 is a "good" starting point -(such as the warm-start of the previous solution, -or the user want to test the performance of this starting point; -the starting point shall be further projected to the L_{infty} ball, -to make sure that it is feasible) - -case 11, 12, 13, or 14: z0 is a "random" guess, and thus not used -(we shall initialize z as follows: -if lambda2 >= 0.5 * lambda_2^max, we initialize the solution of the linear system; -if lambda2 < 0.5 * lambda_2^max, we initialize with zero -this solution is projected to the L_{infty} ball) - -switch( flag ) -5, 15: sfa_special - -switch( flag ) -5: z0 is a good starting point -15: z0 is a bad starting point, use the solution of the linear system - - -switch( flag ) -6, 16: sfa_one - -switch( flag ) - 6: z0 is a good starting point - 16: z0 is a bad starting point, use the solution of the linear system - - */ - - void tesla_proj(double *X, double *Z, double *gap, - double *V, double *Z0, - double lambda1, double lambda2, int p, int n, - int maxStep, double tol, int tau, int flag){ - /* - We assume that X and V are of size p x n - */ - - int i, j; - int nn=n-1; - double - *x =(double *) malloc(sizeof(double)*n), - *v =(double *) malloc(sizeof(double)*n), - *z =(double *) malloc(sizeof(double)*nn), - *z0 =(double *) malloc(sizeof(double)*nn), - *infor=(double *) malloc(sizeof(double)*4); - //double temp; - - - - if (n<3){ - printf("\n n should be equal to or larger than 3"); - exit(-1); - } - - - for(i=0;i. - * - * Copyright (C) 2009 - 2012 Jun Liu and Jieping Ye - */ - - -#ifndef ORDERTREE_SLEP -#define ORDERTREE_SLEP - -#define IGNORE_IN_CLASSLIST - -#include -#ifdef USE_GPL_SHOGUN - -#include -#include -#include -#include - - -/* - * In this file, we propose an O(n^2) algorithm for solving the problem: - * - * min 1/2 \|x - u\|^2 - * s.t. x_i \ge x_j \ge 0, (i,j) \in I, - * - * where I is the edge set of the tree - * - * - */ - -/* - * Last updated on January, 21, 2011 - * - * 1) the function merge is a non-recursive process for merging one tree with the other - * - * 2) we follow the writeup to revise the function computeMaximalMean - * - */ - -#ifndef DOXYGEN_SHOULD_SKIP_THIS -IGNORE_IN_CLASSLIST struct NodeNum -{ - int node_num; - struct NodeNum *next; -}; - -IGNORE_IN_CLASSLIST struct ChildrenNum -{ - int children_num; - int *children; -}; - -IGNORE_IN_CLASSLIST struct Node -{ - int flag; /*if the maximal root-tree of the subtree rooted at this node has been computed, flag=1, otherwise 0*/ - double m; /*During the computation, it stores the maximal mean from this node to (grandson) child node - *The number of nodes on this path is stored in num - * - *It is intialized with the value of u(node_num) - */ - int num; /*the number of nodes, whose avarage gives the maximal mean---x*/ - struct Node *brother; /*the pointer to the brother node(s)*/ - struct Node *child; /*the pointer to the child node(s)*/ - struct NodeNum *firstNode; /*the first node in the "maximal mean" group*/ - struct NodeNum *lastNode; /*the last node in the "maximal mean" group*/ -}; -#endif - -/* - * We build a tree with the input from a file - * - * The file has n rows represented in the following format - * - | parent | number of children | children - 18 3 10 13 17 - 10 3 5 8 9 - 13 2 11 12 - 17 3 13 14 15 - 5 2 1 4 - 8 2 6 7 - 9 0 - 11 0 - 12 0 - 14 0 - 15 0 - 16 0 - 1 0 - 4 2 2 3 - 6 0 - 7 0 - 2 0 - 3 0 - * - * - * Each row provides the information of one parent node and its children - * - * If a parent node is not included in any row, it is regarded that it is the leaf node. - * For example, it is valid that the rows with zero children can be deleted. - * - * Node number is numbered from 1 to n, where n denotes the number of nodes. - * - * In the program, we deduct the number by 1, as C starts from 0, instead of 1. - * - */ - -void readFromFile(char * FileName, struct ChildrenNum ** TreeInfo, int n){ - FILE *fp; - struct ChildrenNum * treeInfo; - int i, j, num, nodeId; - - - fp=fopen(FileName, "r"); - - if(!fp){ - printf("\n\n Fatal Error!!!"); - printf("\n\n Failure in reading the file:%s!", FileName); - printf("\n\n The program does not check the correctness of the tree provided in the file: %s!", FileName); - return; - } - - treeInfo=(struct ChildrenNum *)malloc(sizeof(struct ChildrenNum)*n); - - if(!treeInfo){ - printf("\n Allocation of treeInfo failure!"); - return; - } - - for(i=0;in || i<1){ - printf("\n The node number should be between [1, %d]!",n); - return; - } - - i=i-1; - /*i=i-1, as C starts from 0 instead of 1*/ - if (num>0){ - treeInfo[i].children_num=num; - - treeInfo[i].children=(int *)malloc(sizeof(int)*num); - - if(!treeInfo[i].children){ - printf("\n Allocation of treeInfo failure!"); - return; - } - - for(j=0;jn || nodeId<1){ - printf("\n The node number should be between [1, %d]!", n); - return; - } - - treeInfo[i].children[j]=nodeId-1; - /*add -1, as C starts from 0 instead of 1*/ - } - - } - } - } - - fclose(fp); - - /* - printf("\n In readFromFile!"); - for(i=0;ifirstNode->node_num; - int numberOfChildren=treeInfo[currentRoot].children_num; - int i; - - /* insert the children nodes of the current root - */ - for(i=0;iflag=0; - newNode->m=u[treeInfo[currentRoot].children[i]]; - newNode->num=1; - newNode->child=NULL; - - currentNode->node_num=treeInfo[currentRoot].children[i]; - currentNode->next=NULL; - newNode->firstNode=newNode->lastNode=currentNode; - - /* - * insert newnode to be the children nodes of root - * - */ - newNode->brother=root->child; - root->child=newNode; - - /* - * treat newNode as the root, and add its children - * - */ - - buildTree(newNode, treeInfo, u); - } -} - -/* - * initilize the root, which means that the tree is built by this function. - * as the root is the starting point of a tree - * - * we use the input file for building the tree - */ - -void initializeRoot(struct Node ** Root, char * FileName, double *u, int rootNum, int n){ - - struct NodeNum * currentNode; - struct Node *root; - struct ChildrenNum * treeInfo; - int i; - - /*read the from the file to construct treeInfo*/ - readFromFile(FileName, &treeInfo, n); - - if(rootNum>n || rootNum <1){ - printf("\n The node number of the root should be between [1, %d]!", n); - return; - } - - rootNum=rootNum-1; - /*add -1, as C starts from 0 instead of 1*/ - - root=(struct Node *)malloc(sizeof(struct Node)); - currentNode=(struct NodeNum *)malloc(sizeof(struct NodeNum)); - - if(!root){ - printf("\n Allocation in computeGroups failure!"); - return; - } - - if(!currentNode){ - printf("\n Allocation in computeGroups failure!"); - return; - } - - - root->flag=0; - root->m=u[rootNum]; - root->num=1; - root->brother=root->child=NULL; - - currentNode->node_num=rootNum; - currentNode->next=NULL; - root->firstNode=root->lastNode=currentNode; - - /*build the tree using buildTree*/ - buildTree(root, treeInfo, u); - - /*free treeInfo*/ - for(i=0;iflag=0; - root->m=u[rootNum]; - root->num=1; - root->brother=root->child=NULL; - - currentNode->node_num=rootNum; - currentNode->next=NULL; - root->firstNode=root->lastNode=currentNode; - - /*build the tree using buildTree*/ - buildTree(root, treeInfo, u); - - /*free treeInfo*/ - for(i=0;i1){ - treeInfo[0].children_num=n-1; - treeInfo[0].children=(int *)malloc(sizeof(int)*(n-1)); - for(i=1;iflag=0; - root->m=u[rootNum]; - root->num=1; - root->brother=root->child=NULL; - - currentNode->node_num=rootNum; - currentNode->next=NULL; - root->firstNode=root->lastNode=currentNode; - - /*build the tree using buildTree*/ - buildTree(root, treeInfo, u); - - /*free treeInfo*/ - if(n>1) - free(treeInfo[0].children); - free(treeInfo); - - *Root=root; -} - - - -/* - * merge root with maxNode - */ -void merge(struct Node * root, struct Node * maxNode ){ - struct Node * childrenNode, *maxNodeChild; - - root->m= (root->m* root->num + maxNode->m *maxNode->num)/(root->num + maxNode->num); - root->num+=maxNode->num; - root->lastNode->next=maxNode->firstNode; - root->lastNode=maxNode->lastNode; - - /* - * update the brother list of maxNode (when removing maxNode) - * - */ - if (root->child==maxNode){ - root->child=maxNode->brother; - } - else{ - childrenNode=root->child; - - while(childrenNode->brother!=maxNode){ - childrenNode=childrenNode->brother; - } - /*childrenNode's brother is maxNode*/ - childrenNode->brother=maxNode->brother; - } - - - /* - * change the children of maxNode to the children of root - */ - maxNodeChild=maxNode->child; - if (maxNodeChild){ - /*if maxNode has at least a child*/ - - while(maxNodeChild->brother) - maxNodeChild=maxNodeChild->brother; - /*maxNodeChild points to the last child of maxNode*/ - - maxNodeChild->brother=root->child; - root->child=maxNode->child; - } - - /* - * remove maxNode from the children list of root - */ - free(maxNode); - -} - - - -/* - * compute the maximal mean for each node - * - */ - -void computeMaximalMean(struct Node * root){ - struct Node * childrenNode, *maxNode; - double mean; - - /*if root already points to a leaf node, we do nothing*/ - if(!root->child){ - - /*the value of a maximal root-tree is non-negative*/ - if (root->m <0) - root->m =0; - - root->flag=1; - return; - } - - /*the following loop corresponds to line 5-20 of the algorithm*/ - while(1){ - - childrenNode=root->child; - if(!childrenNode){ - - if (root->m <0) - root->m =0; - - root->flag=1; - return; - } - - /*we note that, childrenNode->m >=0*/ - - mean=0; - - /*visit all its children nodes, to get the maximal "mean" and corresponding maxNode*/ - while(childrenNode){ - - /*if the maximal root-tree at childrenNode is not computed, we compute it*/ - if (!childrenNode->flag) - computeMaximalMean(childrenNode); - - if (childrenNode->m >= mean){ - mean=childrenNode->m; - maxNode=childrenNode; - } - - childrenNode=childrenNode->brother; - } - - if ( (root->m <= 0) && (mean==0) ){ - /* merge root with all its children, in this case, - * its children is a super-node - * (thus does not has any other children, due to merge)*/ - - childrenNode=root->child; - while(childrenNode){ - merge(root, childrenNode); - childrenNode=root->child; - } - - root->m =0; - root->flag=1; - return; - } - - if (root->m > mean){ - - root->flag=1; - return; - } - - merge(root,maxNode); - } - -} - - - -/* - * compute the maximal mean for each node, without the non-negative constraint - * - * Composed on November 23, 2011. - * - */ - -void computeMaximalMean_without_nonnegative(struct Node * root){ - struct Node * childrenNode, *maxNode; - double mean; - int mean_flag; - - /*if root already points to a leaf node, we do nothing*/ - if(!root->child){ - - /*the value of a maximal root-tree is not necessarily non-negative, when the non-negative constraint is not imposed*/ - - /* - The following is removed - if (root->m <0) - root->m =0; - */ - - - root->flag=1; - return; - } - - /*the following loop corresponds to line 5-20 of the algorithm */ - while(1){ - - childrenNode=root->child; - if(!childrenNode){ - - /*the value of a maximal root-tree is not necessarily non-negative, when the non-negative constraint is not imposed*/ - - /* - The following is removed - - if (root->m <0) - root->m =0; - */ - - root->flag=1; - return; - } - - /*we note that, childrenNode->m >=0 does not necesarily hold. - Therefore, for mean, we need to initialize with a small value. We initialize it with the value of its first child node - */ - - mean_flag=0; /*0 denotes that "mean" has not been really specified*/ - - /*visit all its children nodes, to get the maximal "mean" and corresponding maxNode*/ - while(childrenNode){ - - /*if the maximal root-tree at childrenNode is not computed, we compute it*/ - if (!childrenNode->flag) - computeMaximalMean_without_nonnegative(childrenNode); - - /*if mean has not been specified, let us specify it, and set mean_flag to 1*/ - if (!mean_flag){ - mean=childrenNode->m; - mean_flag=1; - } - - if (childrenNode->m >= mean){ - mean=childrenNode->m; - maxNode=childrenNode; - } - - childrenNode=childrenNode->brother; - } - - if (root->m > mean){ - - root->flag=1; - return; - } - - merge(root,maxNode); - } - -} - - -/* - * computeSolution - * - */ - - -void computeSolution(double *x, struct Node *root){ - struct Node * child; - struct NodeNum *currentNode; - double mean; - - if (root){ - /* - * process the root - * - * set the value for x - */ - - mean=root->m; - - currentNode=root->firstNode; - while(currentNode){ - x[currentNode->node_num]=mean; - currentNode=currentNode->next; - } - - /*process the children of root*/ - child=root->child; - while(child){ - computeSolution(x, child); - - child=child->brother; - } - } -} - -/* - * traverse the tree - * used for debugging the correctness of the code - */ - -void traversalTree(struct Node *root){ - struct Node * child; - struct NodeNum *currentNode; - - if (root){ - printf("\n\n root->m =%2.5f, num:%d \n Nodes:",root->m,root->num); - - currentNode=root->firstNode; - while(currentNode){ - printf(" %d ", currentNode->node_num); - currentNode=currentNode->next; - } - - printf("\n root: %d, child:", root->m); - - /*print out the children of root*/ - child=root->child; - while(child){ - printf(" %d", child->m); - child=child->brother; - } - - /*print out the children of children*/ - child=root->child; - while(child){ - traversalTree(child); - - child=child->brother; - } - } -} - - - - - -/* - * free the dynamic space generated by alloc - */ - -void deleteTree(struct Node *root){ - struct Node *child, *temp; - struct NodeNum *currentNode; - - if (root){ - - child=root->child; - - while(child){ - - temp=child->brother; - /*point to its brother*/ - - deleteTree(child); - /*free its chlidren*/ - - child=temp; - } - - /* - * free root - * - * 1. free NodeNum pointed by firstNode and lastNode - * 2. free Node - */ - currentNode=root->firstNode; - while(currentNode){ - root->firstNode=currentNode->next; - free(currentNode); - - currentNode=root->firstNode; - } - root->lastNode=NULL; - free(root); - } -} - -/* - * This is the main function for the general tree - * - */ - -void orderTree(double *x, char * FileName, double *u, int rootNum, int n){ - struct Node * root; - - /* - * build the tree using initializeRoot - */ - initializeRoot(&root, FileName, u, rootNum, n); - - /* - printf("\n\n Before computation"); - traversalTree(root); - */ - - - /* - * compute the maximal average for each node - */ - - computeMaximalMean(root); - - - /*compute the solution from the tree*/ - - computeSolution(x, root); - - - /* - printf("\n\n After computation"); - traversalTree(root); - */ - - - /* delete the tree - */ - deleteTree(root); -} - - -/* - * This is the main function for the general tree, without the non-negative constraint - * - */ - -void orderTree_without_nonnegative(double *x, char * FileName, double *u, int rootNum, int n){ - struct Node * root; - - /* - * build the tree using initializeRoot - */ - initializeRoot(&root, FileName, u, rootNum, n); - - /* - printf("\n\n Before computation"); - traversalTree(root); - */ - - - /* - * compute the maximal average for each node - */ - - computeMaximalMean_without_nonnegative(root); - - - /*compute the solution from the tree*/ - - computeSolution(x, root); - - - /* - printf("\n\n After computation"); - traversalTree(root); - */ - - - /* delete the tree - */ - deleteTree(root); -} - - - -/* - * This is the main function for the full binary tree - * - */ - -void orderTreeBinary(double *x, double *u, int n){ - struct Node * root; - - /* - * build the tree using initializeRootBinary for the binary tree - * - * please make sure that n=2^{depth +1} -1 - * - */ - - initializeRootBinary(&root, u, n); - - /* - printf("\n\n Before computation"); - traversalTree(root); - */ - - - /* - * compute the maximal average for each node - */ - - computeMaximalMean(root); - - - /*compute the solution from the tree*/ - - computeSolution(x, root); - - - /* - printf("\n\n After computation"); - traversalTree(root); - */ - - - /* delete the tree - */ - deleteTree(root); -} - - -/* - * This is the main function for the tree with depth 1 - * - */ - -void orderTreeDepth1(double *x, double *u, int n){ - struct Node * root; - - /* - * build the tree using initializeRootDepth1 for the tree with depth 1 - * - */ - - initializeRootDepth1(&root, u, n); - - /* - printf("\n\n Before computation"); - traversalTree(root); - */ - - - /* - * compute the maximal average for each node - */ - - computeMaximalMean(root); - - - /*compute the solution from the tree*/ - - computeSolution(x, root); - - - /* - printf("\n\n After computation"); - traversalTree(root); - */ - - - /* delete the tree - */ - deleteTree(root); -} -#endif //USE_GPL_SHOGUN -#endif /* ----- #ifndef ORDERTREE_SLEP ----- */ diff --git a/src/shogun/lib/slep/order/sequence.h b/src/shogun/lib/slep/order/sequence.h deleted file mode 100644 index fc706e67f46..00000000000 --- a/src/shogun/lib/slep/order/sequence.h +++ /dev/null @@ -1,225 +0,0 @@ -/* This program is free software: you can redistribute it and/or modify - * it under the terms of the GNU General Public License as published by - * the Free Software Foundation, either version 3 of the License, or - * (at your option) any later version. - * - * This program is distributed in the hope that it will be useful, - * but WITHOUT ANY WARRANTY; without even the implied warranty of - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the - * GNU General Public License for more details. - * - * You should have received a copy of the GNU General Public License - * along with this program. If not, see . - * - * Copyright (C) 2009 - 2012 Jun Liu and Jieping Ye - */ - -#ifndef SEQUENCE_SLEP -#define SEQUENCE_SLEP - -#include - -#ifdef USE_GPL_SHOGUN - -#include -#include -#include -#include - - -/* - * In this file, we propose the algorithms for solving the problem: - * - * min 1/2 \|x - u\|^2 - * s.t. x1 \ge x2 \ge x3 \ge ... \ge xn \ge 0 - * - * - */ - -/* - * - * x= sequence_bottomup(u,n) - * - * we compute using a bottom up order - * - */ - -void sequence_bottomup(double *x, double *u, int n){ - int i, j; - int *location=(int *)malloc(sizeof(int)*n); - int num; - - if(!location){ - printf("\n Allocation of array failure!"); - exit(1); - } - - - /* - * compute the maximal mean from the root to the i-th point: - * - * x[i]: the maximal mean - * location[i]: the ending index of the mean - * - */ - - - /* process the last element*/ - if (n<1){ - printf("\n n=%d should be an integer over 1!",n); - exit(1); - } - else{ - i=n-1; - x[i]=u[i]; - location[i]=i; - } - - /*process the remaining elements in a bottom-up recursive manner*/ - for(i=n-2;i>=0;i--){ - - - if (u[i]>x[i+1]){ - x[i]=u[i]; - location[i]=i; - } - else{ - /*make use of x[i: (n-1)] and location[i: (n-1)] for update*/ - - /*merge with the first group*/ - num=location[i+1]-i; - x[i]=(u[i] + x[i+1]*num)/(num+1); - location[i]=location[i+1]; - j=location[i+1]+1; - - /*If necessary, we need to further merge with the remainig groups */ - for(;j0){ - for(j=i+1;j<=location[i];j++){ - x[j]=x[i]; - } - - i=location[i]+1; - } - else{ - for(j=i;j= max){ - max=mean; - location[i]=j; - } - } - - if (max>0){ - x[i]=max; /*record the maximal mean*/ - i=location[i]+1; /*the next i*/ - } - else{ - x[i]=-1; /* any value less or equal to 0 - * - * This shows that the remaining elements should be zero - * - */ - break; - } - } - - - /* - * compute the solution x with the mean and location - * - */ - - for(i=0;i0){ - for(j=i+1;j<=location[i];j++){ - x[j]=x[i]; - } - - i=location[i]+1; - } - else{ - for(j=i;j. - * - * Copyright (C) 2009 - 2012 Jun Liu and Jieping Ye - */ - -#ifdef USE_GPL_SHOGUN - -#include -#include -#include -#include - -void identifySomeZeroEntries(double * u, int * zeroGroupFlag, int *entrySignFlag, - int *pp, int *gg, - double *v, double lambda1, double lambda2, - int p, int g, double * w, double *G){ - - int i, j, newZeroNum, iterStep=0; - double twoNorm, temp; - - /* - * process the L1 norm - * - * generate the u>=0, and assign values to entrySignFlag - * - */ - for(i=0;i lambda1){ - u[i]=v[i]-lambda1; - - entrySignFlag[i]=1; - } - else{ - if (v[i] < -lambda1){ - u[i]= -v[i] -lambda1; - - entrySignFlag[i]=-1; - } - else{ - u[i]=0; - - entrySignFlag[i]=0; - } - } - } - - /* - * Applying Algorithm 1 for identifying some sparse groups - * - */ - - /* zeroGroupFlag denotes whether the corresponding group is zero */ - for(i=0;ig+1){ - - printf("\n Identify Zero Group: iterStep= %d. The code might have a bug! Check it!", iterStep); - return; - } - - /*record the number of newly detected sparse groups*/ - newZeroNum=0; - - for (i=0;i=0){ - y[i]=0; - } - else{ - y[i]=x[i]; - x[i]=0; - } - } -} - -void YFromx(double *Y, - double *xnew, double *Ynew, - double lambda2, int g, int *zeroGroupFlag, - double *G, double *w){ - - int i, j; - double twoNorm, temp; - - for(i=0;i 0 ){ /*if x_{G_i} is non-zero*/ - temp=lambda2 * w[3*i+2] / twoNorm; - - for(j=(int) w[3*i] ; j<= (int) w[3*i +1]; j++) - Y[j] *= temp; - } - else /*if x_{G_j} =0, we let Y^i=Ynew^i*/ - { - for(j=(int) w[3*i] ; j<= (int) w[3*i +1]; j++) - Y[j]=Ynew[j]; - } - } - }/*end of for(i=0;i innerProduct) - *gap+=twoNorm-innerProduct; - } - }/*end of for(i=0;i lambda2 * w[3*i+2] ){ - temp=lambda2 * w[3*i+2] / twoNorm; - - for(j=(int) w[3*i] ; j<= (int) w[3*i +1]; j++) - Y[j]*=temp; - } - } - else{ /*this group is zero*/ - for(j=(int) w[3*i] ; j<= (int) w[3*i +1]; j++) - Y[j]=0; - } - } - - /* - * set Ynew to zero - * - * in the following processing, we only operator Y and Ynew in the - * possibly non-zero groups by "if(zeroGroupFlag[i])" - * - */ - for(i=0;i lambda2 * w[3*i+2] ){ - temp=lambda2 * w[3*i+2] / twoNorm; - - for(j=(int) w[3*i] ; j<= (int) w[3*i +1]; j++) - Ynew[j]*=temp; - } - } - }/*end of for(i=0;i5) - L=L*0.8; - - break; - } - else{ - temp=leftValue / rightValue; - - if (L*2 <= temp) - L=temp; - else - L=2*L; - - - if ( L / g - 2* g ){ - - if (rightValue < 1e-16){ - break; - } - else{ - - printf("\n GD: leftValue=%e, rightValue=%e, ratio=%e", leftValue, rightValue, temp); - - printf("\n L=%e > 2 * %d * %d. There might be a bug here. Otherwise, it is due to numerical issue.", L, g, g); - - break; - } - } - } - } - - /* compute the duality gap at (xnew, Ynew) - * - * void dualityGap(double *gap, double *penalty2, - * double *x, double *Y, int g, int *zeroGroupFlag, - * double *G, double *w, double lambda2) - * - */ - dualityGap(gap, penalty2, xnew, Ynew, g, zeroGroupFlag, G, w, lambda2); - - /* - * flag =1 means restart - * - * flag =0 means with restart - * - * nextRestartStep denotes the next "step number" for - * initializing the restart process. - * - * This is based on the fact that, the result is only beneficial when - * xnew is good. In other words, - * if xnew is not good, then the - * restart might not be helpful. - */ - - if ( (flag==0) || (flag==1 && iterStep < nextRestartStep )){ - - /* copy Ynew to Y, and xnew to x */ - sg_memcpy(x, xnew, sizeof(double) * p); - sg_memcpy(Y, Ynew, sizeof(double) * YSize); - - /* - printf("\n iterStep=%d, L=%2.5f, gap=%e", iterStep, L, *gap); - */ - - } - else{ - /* - * flag=1 - * - * We allow the restart of the program. - * - * Here, Y is constructed as a subgradient of xnew, based on the - * assumption that Y might be a better choice than Ynew, provided - * that xnew is good enough. - * - */ - - /* - * compute the restarting point Y with xnew and Ynew - * - *void YFromx(double *Y, - * double *xnew, double *Ynew, - * double lambda2, int g, int *zeroGroupFlag, - * double *G, double *w) - */ - YFromx(Y, xnew, Ynew, lambda2, g, zeroGroupFlag, G, w); - - /*compute the solution with the starting point Y - * - *void xFromY(double *x, double *y, - * double *u, double *Y, - * int p, int g, int *zeroGroupFlag, - * double *G, double *w) - * - */ - xFromY(x, y, u, Y, p, g, zeroGroupFlag, G, w); - - /*compute the duality at (x, Y) - * - * void dualityGap(double *gap, double *penalty2, - * double *x, double *Y, int g, int *zeroGroupFlag, - * double *G, double *w, double lambda2) - * - */ - dualityGap(&gapR, &penalty2R, x, Y, g, zeroGroupFlag, G, w, lambda2); - - if (*gap< gapR){ - /*(xnew, Ynew) is better in terms of duality gap*/ - /* copy Ynew to Y, and xnew to x */ - sg_memcpy(x, xnew, sizeof(double) * p); - sg_memcpy(Y, Ynew, sizeof(double) * YSize); - - /*In this case, we do not apply restart, as (x,Y) is not better - * - * We postpone the "restart" by giving a - * "nextRestartStep" - */ - - /* - * we test *gap here, in case *gap=0 - */ - if (*gap <=tol) - break; - else{ - nextRestartStep=iterStep+ (int) sqrt(gapR / *gap); - } - } - else{ /*we use (x, Y), as it is better in terms of duality gap*/ - *gap=gapR; - *penalty2=penalty2R; - } - - /* - printf("\n iterStep=%d, L=%2.5f, gap=%e, gapR=%e", iterStep, L, *gap, gapR); - */ - - } - - /* - * if the duality gap is within pre-specified parameter tol - * - * we terminate the algorithm - */ - if (*gap <=tol) - break; - } - - penalty2[3]=iterStep; - - penalty2[4]=0; - for(i=0;i(int) w[3*i +1]) - penalty2[4]=penalty2[4]+1; - } - } - - /* - * assign sign to the solution x - */ - for(i=0;i lambda2 * w[3*i+2] ){ - temp=lambda2 * w[3*i+2] / twoNorm; - - for(j=(int) w[3*i] ; j<= (int) w[3*i +1]; j++) - Ynew[j]*=temp; - } - } - }/*end of for(i=0;i5) - L=L*0.8; - - break; - } - else{ - temp=leftValue / rightValue; - - if (L*2 <= temp) - L=temp; - else - L=2*L; - - if ( L / g - 2* g >0 ){ - - if (rightValue < 1e-16){ - break; - } - else{ - - printf("\n One Gradient Step: leftValue=%e, rightValue=%e, ratio=%e", leftValue, rightValue, temp); - - printf("\n L=%e > 2 * %d * %d. There might be a bug here. Otherwise, it is due to numerical issue.", L, g, g); - - break; - } - } - } - } - - *LL=L; -} - -void overlapping_agd(double *x, double *gap, double *penalty2, - double *v, int p, int g, double lambda1, double lambda2, - double *w, double *G, double *Y, int maxIter, int flag, double tol){ - - int YSize=(int) w[3*(g-1) +1]+1; - double *u=(double *)malloc(sizeof(double)*p); - double *y=(double *)malloc(sizeof(double)*p); - - double *xnew=(double *)malloc(sizeof(double)*p); - double *Ynew=(double *)malloc(sizeof(double)* YSize ); - - double *xS=(double *)malloc(sizeof(double)*p); - double *YS=(double *)malloc(sizeof(double)* YSize ); - - /*double *xp=(double *)malloc(sizeof(double)*p);*/ - double *Yp=(double *)malloc(sizeof(double)* YSize ); - - int *zeroGroupFlag=(int *)malloc(sizeof(int)*g); - int *entrySignFlag=(int *)malloc(sizeof(int)*p); - - int pp, gg; - int i, j, iterStep; - double twoNorm,temp, L=1, leftValue, rightValue, gapR, penalty2R; - int nextRestartStep=0; - - double alpha, alphap=0.5, beta, gamma; - - /* - * call the function to identify some zero entries - * - * entrySignFlag[i]=0 denotes that the corresponding entry is definitely zero - * - * zeroGroupFlag[i]=0 denotes that the corresponding group is definitely zero - * - */ - - identifySomeZeroEntries(u, zeroGroupFlag, entrySignFlag, - &pp, &gg, - v, lambda1, lambda2, - p, g, w, G); - - penalty2[1]=pp; - penalty2[2]=gg; - /*store pp and gg to penalty2[1] and penalty2[2]*/ - - /* - *------------------- - * Process Y - *------------------- - * We make sure that Y is feasible - * and if x_i=0, then set Y_{ij}=0 - */ - for(i=0;i lambda2 * w[3*i+2] ){ - temp=lambda2 * w[3*i+2] / twoNorm; - - for(j=(int) w[3*i] ; j<= (int) w[3*i +1]; j++) - Y[j]*=temp; - } - } - else{ /*this group is zero*/ - for(j=(int) w[3*i] ; j<= (int) w[3*i +1]; j++) - Y[j]=0; - } - } - - /* - * set Ynew and Yp to zero - * - * in the following processing, we only operate, Yp, Y and Ynew in the - * possibly non-zero groups by "if(zeroGroupFlag[i])" - * - */ - for(i=0;i lambda2 * w[3*i+2] ){ - temp=lambda2 * w[3*i+2] / twoNorm; - - for(j=(int) w[3*i] ; j<= (int) w[3*i +1]; j++) - Ynew[j]*=temp; - } - } - }/*end of for(i=0;i5) - L=L*0.8; - - break; - } - else{ - temp=leftValue / rightValue; - - if (L*2 <= temp) - L=temp; - else - L=2*L; - - - - if ( L / g - 2* g >0 ){ - - if (rightValue < 1e-16){ - break; - } - else{ - - printf("\n AGD: leftValue=%e, rightValue=%e, ratio=%e", leftValue, rightValue, temp); - - printf("\n L=%e > 2 * %d * %d. There might be a bug here. Otherwise, it is due to numerical issue.", L, g, g); - - break; - } - } - } - } - - /* compute the duality gap at (xnew, Ynew) - * - * void dualityGap(double *gap, double *penalty2, - * double *x, double *Y, int g, int *zeroGroupFlag, - * double *G, double *w, double lambda2) - * - */ - dualityGap(gap, penalty2, - xnew, Ynew, g, zeroGroupFlag, - G, w, lambda2); - - - /* - * if the duality gap is within pre-specified parameter tol - * - * we terminate the algorithm - */ - if (*gap <=tol){ - - sg_memcpy(x, xnew, sizeof(double) * p); - sg_memcpy(Y, Ynew, sizeof(double) * YSize); - - break; - } - - - - /* - * flag =1 means restart - * - * flag =0 means with restart - * - * nextRestartStep denotes the next "step number" for - * initializing the restart process. - * - * This is based on the fact that, the result is only beneficial when - * xnew is good. In other words, - * if xnew is not good, then the - * restart might not be helpful. - */ - - if ( (flag==0) || (flag==1 && iterStep < nextRestartStep )){ - - - /*sg_memcpy(xp, x, sizeof(double) * p);*/ - sg_memcpy(Yp, Y, sizeof(double) * YSize); - - /*sg_memcpy(x, xnew, sizeof(double) * p);*/ - sg_memcpy(Y, Ynew, sizeof(double) * YSize); - - gamma=gamma * (1-alpha); - - alphap=alpha; - - /* - printf("\n iterStep=%d, L=%2.5f, gap=%e", iterStep, L, *gap); - */ - - } - else{ - /* - * flag=1 - * - * We allow the restart of the program. - * - * Here, Y is constructed as a subgradient of xnew, based on the - * assumption that Y might be a better choice than Ynew, provided - * that xnew is good enough. - * - */ - - /* - * compute the restarting point YS with xnew and Ynew - * - *void YFromx(double *Y, - * double *xnew, double *Ynew, - * double lambda2, int g, int *zeroGroupFlag, - * double *G, double *w) - */ - YFromx(YS, xnew, Ynew, lambda2, g, zeroGroupFlag, G, w); - - /*compute the solution with the starting point YS - * - *void xFromY(double *x, double *y, - * double *u, double *Y, - * int p, int g, int *zeroGroupFlag, - * double *G, double *w) - * - */ - xFromY(xS, y, u, YS, p, g, zeroGroupFlag, G, w); - - /*compute the duality at (xS, YS) - * - * void dualityGap(double *gap, double *penalty2, - * double *x, double *Y, int g, int *zeroGroupFlag, - * double *G, double *w, double lambda2) - * - */ - dualityGap(&gapR, &penalty2R, xS, YS, g, zeroGroupFlag, G, w, lambda2); - - if (*gap< gapR){ - /*(xnew, Ynew) is better in terms of duality gap*/ - - /*In this case, we do not apply restart, as (xS,YS) is not better - * - * We postpone the "restart" by giving a - * "nextRestartStep" - */ - - /*sg_memcpy(xp, x, sizeof(double) * p);*/ - sg_memcpy(Yp, Y, sizeof(double) * YSize); - - /*sg_memcpy(x, xnew, sizeof(double) * p);*/ - sg_memcpy(Y, Ynew, sizeof(double) * YSize); - - gamma=gamma * (1-alpha); - - alphap=alpha; - - nextRestartStep=iterStep+ (int) sqrt(gapR / *gap); - } - else{ - /*we use (xS, YS), as it is better in terms of duality gap*/ - - *gap=gapR; - *penalty2=penalty2R; - - if (*gap <=tol){ - - sg_memcpy(x, xS, sizeof(double) * p); - sg_memcpy(Y, YS, sizeof(double) * YSize); - - break; - }else{ - /* - * we do a gradient descent based on (xS, YS) - * - */ - - /* - * compute (x, Y) from (xS, YS) - * - * - * gradientDescentStep(double *xnew, double *Ynew, - * double *LL, double *u, double *y, int *entrySignFlag, double lambda2, - * double *x, double *Y, int p, int g, int * zeroGroupFlag, - * double *G, double *w) - */ - gradientDescentStep(x, Y, - &L, u, y, entrySignFlag,lambda2, - xS, YS, p, g, zeroGroupFlag, - G, w); - - /*sg_memcpy(xp, xS, sizeof(double) * p);*/ - sg_memcpy(Yp, YS, sizeof(double) * YSize); - - gamma=L; - - alphap=0.5; - - } - - - } - - /* - * printf("\n iterStep=%d, L=%2.5f, gap=%e, gapR=%e", iterStep, L, *gap, gapR); - */ - - }/* flag =1*/ - - } /* main loop */ - - - - penalty2[3]=iterStep+1; - - /* - * get the number of nonzero groups - */ - - penalty2[4]=0; - for(i=0;i(int) w[3*i +1]) - penalty2[4]=penalty2[4]+1; - } - } - - - /* - * assign sign to the solution x - */ - for(i=0;i. - * - * Copyright (C) 2009 - 2012 Jun Liu and Jieping Ye - */ - - -#ifndef OVERLAPPING_SLEP -#define OVERLAPPING_SLEP - -#include -#ifdef USE_GPL_SHOGUN - - - -/* - * ------------------------------------------------------------------- - * Function and parameter - * ------------------------------------------------------------------- - * - * In this file, we focus solving the following problem - * - * 1/2 \|x-v\|^2 + \lambda_1 \|x\|_1 + \lambda_2 \sum w_i \|x_{G_i}\|, - * - * where x and v are of dimension p, - * w >0, and G_i contains the indices for the i-th group - * - * The file is implemented in the following in Matlab: - * - * x=overlapping(v, p, g, lambda1, lambda2, w, G, Y, maxIter, flag, tol); - * - * x and v are vectors of dimension p - * - * g denotes the number of groups - * - * lambda1 and labmda2 are non-negative regularization paramter - * - * G is a vector containing the indices for the groups - * G_1, G_2, ..., G_g - * - * w is a 3xg matrix - * w(1,i) contains the starting index of the i-th group in G - * w(2,i) contains the ending index of the i-th group in G - * w(3,i) contains the weight for the i-th group - * - * Y is the dual of \|x_{G_i}\|, it is of the same size as G - * - * maxIter is the maximal number of iteration - * - * flag=0, we apply the pure projected gradient descent - * (forward and backward line search is used) - * - * flag=1, we apply the projected gradient descent with restart - * - * in the future, we may apply the accelerated gradient descent - * with adaptive line search (see our KDD'09 paper) with other "flag" - * - * - * Note: - * - * 1. One should ensure w(2,i)-w(1,i)+1=|G_i|. - * !! The program does not check w(2,i)-w(1,i)+1=|G_i|.!! - * - * 2. The index in G and w starts from 0 - * - * ------------------------------------------------------------------- - * History: - * ------------------------------------------------------------------- - * - * Composed by Jun Liu on May 17, 2010 - * - * For any question or suggestion, please email j.liu@asu.edu or - * jun.liu.80@gmail.com - * - */ - - -/* - * -------------------------------------------------------------------- - * Identifying some zero Entries - * -------------------------------------------------------------------- - * - * lambda1, lambda2 should be non-negative - * - * v is the vector of size p to be projected - * - * - * zeroGroupFlag is a vector of size g - * - * zeroGroupFlag[i]=0 denotes that the corresponding group is definitely zero - * zeroGroupFlag[i]=1 denotes that the corresponding group is (possibly) nonzero - * - * - * u is a vector of size p - * - * - * entrySignFlag is a vector of size p - * - * entrySignFlag[i]=0 denotes that the corresponding entry is definitely zero - * entrySignFlag[i]=1 denotes that the corresponding entry is (possibly) positive - * entrySignFlag[i]=-1 denotes that the corresponding entry is (possibly) negative - * - */ -void identifySomeZeroEntries(double * u, int * zeroGroupFlag, int *entrySignFlag, - int *pp, int *gg, - double *v, double lambda1, double lambda2, - int p, int g, double * w, double *G); - -/* - * - * function: xFromY - * - * compute x=max(u-Y * e, 0); - * - * xFromY(x, y, u, Y, p, g, zeroGroupFlag, G, w); - * - * y=u-Y * e - max( u - Y * e, 0) - * - */ -void xFromY(double *x, double *y, - double *u, double *Y, - int p, int g, int *zeroGroupFlag, - double *G, double *w); - -/* - * - * function: YFromx - * - * compute Y=subgradient(x) - * - * YFromx(Y, xnew, Ynew, lambda2, g, zeroGroupFlag, G, w); - * - * The idea is that, if x_{G_i} is nonzero, - * we compute Y^i as x_{G_i}/ \|x_{G_i}\| * lambda2 * w[3*i+2] - * otherwise - * Y^i=Ynew^i - * - */ -void YFromx(double *Y, - double *xnew, double *Ynew, - double lambda2, int g, int *zeroGroupFlag, - double *G, double *w); - -/* - * function: dualityGap - * - * compute the duality gap for the approximate solution (x, Y) - * - * Meanwhile, we compute - * - * penalty2=\sum_{i=1}^g w_i \|x_{G_i}\| - * - */ -void dualityGap(double *gap, double *penalty2, - double *x, double *Y, int g, int *zeroGroupFlag, - double *G, double *w, double lambda2); - -/* - * we solve the proximal opeartor: - * - * 1/2 \|x-v\|^2 + \lambda_1 \|x\|_1 + \lambda_2 \sum w_i \|x_{G_i}\| - * - * See the description of the variables in the beginning of this file - * - * x is the primal variable, each of its entry is non-negative - * - * Y is the dual variable, each of its entry should be non-negative - * - * flag =0: no restart - * flag =1; restart - * - * tol: the precision parameter - * - * The following code apply the projected gradient descent method - * - */ -void overlapping_gd(double *x, double *gap, double *penalty2, - double *v, int p, int g, double lambda1, double lambda2, - double *w, double *G, double *Y, int maxIter, int flag, double tol); - -/* - * - * do a gradient descent step based (x, Y) to get (xnew, Ynew) - * - * (x, Y) is known. Here we do a line search for determining the value of L - * - * gradientDescentStep(double *xnew, double *Ynew, - double *LL, double *u, - double *x, double *Y, int p, int g, int * zeroGroupFlag, - double *G, double *w) - * - */ -void gradientDescentStep(double *xnew, double *Ynew, - double *LL, double *u, double *y, int *entrySignFlag, double lambda2, - double *x, double *Y, int p, int g, int * zeroGroupFlag, - double *G, double *w); - -/* - * - * we use the accelerated gradient descent - * - */ -void overlapping_agd(double *x, double *gap, double *penalty2, - double *v, int p, int g, double lambda1, double lambda2, - double *w, double *G, double *Y, int maxIter, int flag, double tol); - -/* - * This is main function for the projection - * - * It calls overlapping_gd and overlapping_agd based on flag - * - * - */ -void overlapping(double *x, double *gap, double *penalty2, - double *v, int p, int g, double lambda1, double lambda2, - double *w, double *G, double *Y, int maxIter, int flag, double tol); - -#endif //USE_GPL_SHOGUN -#endif /* ----- #ifndef OVERLAPPING_SLEP ----- */ diff --git a/src/shogun/lib/slep/q1/ep1R.h b/src/shogun/lib/slep/q1/ep1R.h deleted file mode 100644 index d7447c5dd1f..00000000000 --- a/src/shogun/lib/slep/q1/ep1R.h +++ /dev/null @@ -1,71 +0,0 @@ -/* This program is free software: you can redistribute it and/or modify - * it under the terms of the GNU General Public License as published by - * the Free Software Foundation, either version 3 of the License, or - * (at your option) any later version. - * - * This program is distributed in the hope that it will be useful, - * but WITHOUT ANY WARRANTY; without even the implied warranty of - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the - * GNU General Public License for more details. - * - * You should have received a copy of the GNU General Public License - * along with this program. If not, see . - * - * Copyright (C) 2009 - 2012 Jun Liu and Jieping Ye - */ - - -#ifndef EP1R_SLEP -#define EP1R_SLEP - -#include -#ifdef USE_GPL_SHOGUN - -#include -#include -#include -#include - - -/* - Euclidean Projection onto l_{2,1} Ball - - min 1/2 ||x- u||_2^2 + 1/2 ||t- v||_2^2 - s.t. |x|<=t - - -Usage: -[x, t]=ep1R(u, v, n); - -*/ - - -void ep1R(double * x, double *t, double * u, double * v, int n) -{ - int j; - - - for(j=0;j fabs(v[j])){ - t[j]=(fabs(u[j]) + v[j])/2; - - if (u[j] >0) - x[j]=t[j]; - else - x[j]=-t[j]; - } - else - if(fabs(u[j]) <= v[j]){ - t[j]=v[j]; - x[j]=u[j]; - } - else{ - t[j]=x[j]=0; - } - - } -} -#endif //USE_GPL_SHOGUN -#endif /* ----- #ifndef EP1R_SLEP ----- */ - diff --git a/src/shogun/lib/slep/q1/ep21R.h b/src/shogun/lib/slep/q1/ep21R.h deleted file mode 100644 index f53a9abbc86..00000000000 --- a/src/shogun/lib/slep/q1/ep21R.h +++ /dev/null @@ -1,81 +0,0 @@ -/* This program is free software: you can redistribute it and/or modify - * it under the terms of the GNU General Public License as published by - * the Free Software Foundation, either version 3 of the License, or - * (at your option) any later version. - * - * This program is distributed in the hope that it will be useful, - * but WITHOUT ANY WARRANTY; without even the implied warranty of - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the - * GNU General Public License for more details. - * - * You should have received a copy of the GNU General Public License - * along with this program. If not, see . - * - * Copyright (C) 2009 - 2012 Jun Liu and Jieping Ye - */ - - -#ifndef EP21R_SLEP -#define EP21R_SLEP - -#include -#ifdef USE_GPL_SHOGUN - -#include -#include -#include -#include - - -/* - Euclidean Projection onto l_{2,1} Ball - - min 1/2 ||x- u||_2^2 + 1/2 ||t- v||_2^2 - s.t. ||x^j||_{2,1} <= t^j - - -Usage: -[x, t]=ep21R(u, v, n, k); - -*/ - - -void ep21R(double * x, double *t, double * u, double * v, int n, int k) -{ - int i, j, tn=n*k; - double temp; - - /* compute the 2 norm of each group - */ - - for(j=0;j fabs(v[j])){ - t[j]=(temp + v[j])/2; - for (i=j; i. - * - * Copyright (C) 2009 - 2012 Jun Liu and Jieping Ye - */ - - -#ifndef EP21D_SLEP -#define EP21D_SLEP - -#include -#ifdef USE_GPL_SHOGUN - -#include -#include -#include -#include -#include /* This is the head file that contains the implementation of the used functions*/ - -/* - Euclidean Projection onto l_{2,1} Ball - - min 1/2 ||X- V||_2^2 - s.t. ||X||_{2,1} <= z - - which is converted to the following zero finding problem - - f(lambda)= \sum_i ( max( |v^i|-lambda,0) )-z=0 - - v^i denotes the i-th row of V - -Usage: -[x, lambda, iter_step]=ep21d(y, n, k, z, lambda0); - -*/ - - -void ep21d(double * x, double *root, int * steps, double * v, int n, int k, double z, double lambda0) -{ - int i, j, tn=n*k; - double *vnorm=(double *)malloc(sizeof(double)*n); - double *vproj=(double *)malloc(sizeof(double)*n); - double t; - - /* compute the 2 norm of each group - */ - - for(j=0;j. - * - * Copyright (C) 2009 - 2012 Jun Liu and Jieping Ye - */ - -#include -#ifdef USE_GPL_SHOGUN - -#include -#include -#include -#include - -void eppMatrix(double *X, double * V, int k, int n, double rho, double p) -{ - int i, j, *iter_step; - double *v, *x; - double c0, c; - - v=(double *)malloc(sizeof(double)*n); - x=(double *)malloc(sizeof(double)*n); - iter_step=(int *)malloc(sizeof(int)*2); - - /* - *X and V are k x n matrices in matlab, stored in column priority manner - *x corresponds a row of X - */ - - - c0=0; - for(i=0; i. - * - * Copyright (C) 2009 - 2012 Jun Liu and Jieping Ye - */ - - -#ifndef EPPMATRIXQ1_SLEP -#define EPPMATRIXQ1_SLEP - -#include -#ifdef USE_GPL_SHOGUN - -#include /* This is the head file that contains the implementation of the used functions*/ - -/* - Lp Norm Regularized Euclidean Projection - - min 1/2 ||x- v||_2^2 + rho * ||x||_p - - Usage (in Matlab): - [x, c, iter_step]=epp(v, n, rho, p, c0); - - Usage in C: - epp(x, c, iter_step, v, n, rho, p, c0); - - The function epp implements the following three functions - epp1(x, v, n, rho) for p=1 - epp2(x, v, n, rho) for p=2 - eppInf(x, c, iter_step, v, n, rho, c0) for p=inf - eppO(x, c, iter_step, v, n, rho, p) for other p - ------------------------------------------------------------- - - Here, the input and output are of matrix form. Each row corresponds a group - - - Written by Jun Liu, May 18th, 2009 - For any problem, please contact: j.liu@asu.edu - - */ -void eppMatrix(double *X, double * V, int k, int n, double rho, double p); -#endif //USE_GPL_SHOGUN -#endif /* ----- #ifndef EPPMATRIXQ1_SLEP ----- */ - diff --git a/src/shogun/lib/slep/q1/eppVector.h b/src/shogun/lib/slep/q1/eppVector.h deleted file mode 100644 index 94a6a98a793..00000000000 --- a/src/shogun/lib/slep/q1/eppVector.h +++ /dev/null @@ -1,78 +0,0 @@ -/* This program is free software: you can redistribute it and/or modify - * it under the terms of the GNU General Public License as published by - * the Free Software Foundation, either version 3 of the License, or - * (at your option) any later version. - * - * This program is distributed in the hope that it will be useful, - * but WITHOUT ANY WARRANTY; without even the implied warranty of - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the - * GNU General Public License for more details. - * - * You should have received a copy of the GNU General Public License - * along with this program. If not, see . - * - * Copyright (C) 2009 - 2012 Jun Liu and Jieping Ye - */ - - -#ifndef EPPVECTOR_SLEP -#define EPPVECTOR_SLEP - -#include -#ifdef USE_GPL_SHOGUN - -#include -#include -#include -#include -#include /* This is the head file that contains the implementation of the used functions*/ - - -/* - Lp Norm Regularized Euclidean Projection - - min 1/2 ||x- v||_2^2 + rho * ||x||_p - - Usage (in Matlab): - [x, c, iter_step]=epp(v, n, rho, p, c0); - - Usage in C: - epp(x, c, iter_step, v, n, rho, p, c0); - - The function epp implements the following three functions - epp1(x, v, n, rho) for p=1 - epp2(x, v, n, rho) for p=2 - eppInf(x, c, iter_step, v, n, rho, c0) for p=inf - eppO(x, c, iter_step, v, n, rho, p) for other p - - ------------------------------------------------------------ - - Here, the input and output are of Vector form. - - - Written by Jun Liu, May 18th, 2009 - For any problem, please contact: j.liu@asu.edu - -*/ - -void eppVector(double *x, double * v, int* ind, int k, int n, double * rho, double rho_multiplier, double p){ - int i, *iter_step; - double c0, c; - double *px, *pv; - - iter_step=(int *)malloc(sizeof(int)*2); - - c0=0; - for(i=0; i. - * - * Copyright (C) 2009 - 2012 Jun Liu and Jieping Ye - */ - - -#ifndef EPPVECTORR_SLEP -#define EPPVECTORR_SLEP - -#include -#ifdef USE_GPL_SHOGUN - -#include -#include -#include -#include - -/* - min 1/2 ( ||x- u||_2^2 + ||t-v||_2^2 ) - s.t. ||x_j||_2 <= t_j - - */ - -void eppVectorR(double *x, double * t, double * u, double * v, double * ind, int n, int k){ - int i, j; - double temp; - - /* compute the 2 norm of each group - */ - - for(j=0;j fabs(v[j])){ - t[j]=(temp + v[j])/2; - - for(i=(int) (ind[j]); i< (int) (ind[j+1]); i++) - x[i]= t[j] / temp * u[i]; - } - else - if(temp <= v[j]){ - t[j]=v[j]; - - for(i=(int) (ind[j]); i< (int) (ind[j+1]); i++) - x[i]= u[i]; - } - else{ - t[j]=0; - - for(i=(int) (ind[j]); i< (int) (ind[j+1]); i++) - x[i]=0; - } - - } -} -#endif //USE_GPL_SHOGUN -#endif /* ----- #ifndef EPPVECTORR_SLEP ----- */ diff --git a/src/shogun/lib/slep/q1/epph.cpp b/src/shogun/lib/slep/q1/epph.cpp deleted file mode 100644 index 07b986c5a2f..00000000000 --- a/src/shogun/lib/slep/q1/epph.cpp +++ /dev/null @@ -1,690 +0,0 @@ -/* This program is free software: you can redistribute it and/or modify - * it under the terms of the GNU General Public License as published by - * the Free Software Foundation, either version 3 of the License, or - * (at your option) any later version. - * - * This program is distributed in the hope that it will be useful, - * but WITHOUT ANY WARRANTY; without even the implied warranty of - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the - * GNU General Public License for more details. - * - * You should have received a copy of the GNU General Public License - * along with this program. If not, see . - * - * Copyright (C) 2009 - 2012 Jun Liu and Jieping Ye - */ - -#include -#ifdef USE_GPL_SHOGUN - -#include -#include -#include -#include - -#define delta 1e-8 - -#define innerIter 1000 -#define outerIter 1000 - -void eplb(double * x, double *root, int * steps, double * v,int n, double z, double lambda0) -{ - - int i, j, flag=0; - int rho_1, rho_2, rho, rho_T, rho_S; - int V_i_b, V_i_e, V_i; - double lambda_1, lambda_2, lambda_T, lambda_S, lambda; - double s_1, s_2, s, s_T, s_S, v_max, temp; - double f_lambda_1, f_lambda_2, f_lambda, f_lambda_T, f_lambda_S; - int iter_step=0; - - /* find the maximal absolute value in v - * and copy the (absolute) values from v to x - */ - - if (z< 0){ - printf("\n z should be nonnegative!"); - return; - } - - V_i=0; - if (v[0] !=0){ - rho_1=1; - s_1=x[V_i]=v_max=fabs(v[0]); - V_i++; - } - else{ - rho_1=0; - s_1=v_max=0; - } - - for (i=1;i v_max) - v_max=x[V_i]; - V_i++; - } - } - - /* If ||v||_1 <= z, then v is the solution */ - if (s_1 <= z){ - flag=1; lambda=0; - for(i=0;i lambda_1) ){ - /*------------------------------------------------------------------- - Initialization with the root - *------------------------------------------------------------------- - */ - - i=V_i_b; j=V_i_e; rho=0; s=0; - while (i <= j){ - while( (i <= V_i_e) && (x[i] <= lambda) ){ - i++; - } - while( (j>=V_i_b) && (x[j] > lambda) ){ - s+=x[j]; - j--; - } - if (i lambda_T) - lambda_T=lambda_2 + f_lambda_2 /rho_2; - } - - /* compute lambda_S */ - lambda_S=lambda_2 - f_lambda_2 *(lambda_2-lambda_1)/(f_lambda_2-f_lambda_1); - - if (fabs(lambda_T-lambda_S) <= delta){ - lambda=lambda_T; flag=1; - break; - } - - /* set lambda as the middle point of lambda_T and lambda_S */ - lambda=(lambda_T+lambda_S)/2; - - s_T=s_S=s=0; - rho_T=rho_S=rho=0; - i=V_i_b; j=V_i_e; - while (i <= j){ - while( (i <= V_i_e) && (x[i] <= lambda) ){ - if (x[i]> lambda_T){ - s_T+=x[i]; rho_T++; - } - i++; - } - while( (j>=V_i_b) && (x[j] > lambda) ){ - if (x[j] > lambda_S){ - s_S+=x[j]; rho_S++; - } - else{ - s+=x[j]; rho++; - } - j--; - } - if (i lambda_S){ - s_S+=x[i]; rho_S++; - } - else{ - s+=x[i]; rho++; - } - - if (x[j]> lambda_T){ - s_T+=x[j]; rho_T++; - } - - temp=x[i]; x[i]=x[j]; x[j]=temp; - i++; j--; - } - } - - s_S+=s_2; rho_S+=rho_2; - s+=s_S; rho+=rho_S; - s_T+=s; rho_T+=rho; - f_lambda_S=s_S-rho_S*lambda_S-z; - f_lambda=s-rho*lambda-z; - f_lambda_T=s_T-rho_T*lambda_T-z; - - /*printf("\n %d & %d & %5.6f & %5.6f & %5.6f & %5.6f & %5.6f \\\\ \n \\hline ", iter_step, V_i, lambda_1, lambda_T, lambda, lambda_S, lambda_2);*/ - - if ( fabs(f_lambda)< delta ){ - /*printf("\n lambda");*/ - flag=1; - break; - } - if ( fabs(f_lambda_S)< delta ){ - /* printf("\n lambda_S");*/ - lambda=lambda_S; flag=1; - break; - } - if ( fabs(f_lambda_T)< delta ){ - /* printf("\n lambda_T");*/ - lambda=lambda_T; flag=1; - break; - } - - /* - printf("\n\n f_lambda_1=%5.6f, f_lambda_2=%5.6f, f_lambda=%5.6f",f_lambda_1,f_lambda_2, f_lambda); - printf("\n lambda_1=%5.6f, lambda_2=%5.6f, lambda=%5.6f",lambda_1, lambda_2, lambda); - printf("\n rho_1=%d, rho_2=%d, rho=%d ",rho_1, rho_2, rho); - */ - - if (f_lambda <0){ - lambda_2=lambda; s_2=s; rho_2=rho; - f_lambda_2=f_lambda; - - lambda_1=lambda_T; s_1=s_T; rho_1=rho_T; - f_lambda_1=f_lambda_T; - - V_i_e=j; i=V_i_b; - while (i <= j){ - while( (i <= V_i_e) && (x[i] <= lambda_T) ){ - i++; - } - while( (j>=V_i_b) && (x[j] > lambda_T) ){ - j--; - } - if (i=V_i_b) && (x[j] > lambda_S) ){ - j--; - } - if (i lambda) - x[i]=v[i]-lambda; - else - if (v[i]< -lambda) - x[i]=v[i]+lambda; - else - x[i]=0; - } - *root=lambda; - *steps=iter_step; -} - -void epp1(double *x, double *v, int n, double rho) -{ - int i; - - /* - we assume rho>=0 - */ - - for(i=0;i=0 - */ - - for(i=0; i< n; i++){ - v2+=v[i]*v[i]; - } - v2=sqrt(v2); - - if (rho >= v2) - for(i=0;i=0 - */ - - eplb(x, c, &steps, v, n, rho, c0); - - for(i=0; i< n; i++){ - x[i]=v[i]-x[i]; - } - iter_step[0]=steps; - iter_step[1]=0; -} - -void zerofind(double *root, int * iterStep, double v, double p, double c, double x0) -{ - double x, f, fprime, p1=p-1, pp; - int step=0; - - - if (v==0){ - *root=0; *iterStep=0; return; - } - - if (c==0){ - *root=v; * iterStep=0; return; - } - - - if ( (x0 0) ) - x=x0; - else - x=v; - - - pp=pow(x, p1); - f= x + c* pp -v; - - - /* - We apply the Newton's method for solving the root - */ - while (1){ - step++; - - fprime=1 + c* p1 * pp / x; - /* - The derivative at the current solution x - */ - - x = x- f/fprime; /* - The new solution is computed by the Newton method - */ - - - - if (p>2){ - if (x>v){ - x=v; - } - } - else{ - if ( (x<0) || (x>v)){ - x=1e-30; - - f= x+c* pow(x,p1)-v; - - if (f>0){ /* - If f(x) = x + c x^{p-1} - v <0 at x=1e-30, - this shows that the real root is between (0, 1e-30). - For numerical reason, we just set x=0 - */ - - *root=x; - * iterStep=step; - - break; - } - } - } - /* - This makes sure that x lies in the interval [0, v] - */ - - pp=pow(x, p1); - f= x + c* pp -v; - /* - The function value at the new solution - */ - - if ( fabs(f) <= delta){ - *root=x; - * iterStep=step; - break; - } - - if (step>=innerIter){ - printf("\n The number of steps exceed %d, in finding the root for f(x)= x + c x^{p-1} - v, 0< x< v.", innerIter); - printf("\n If you meet with this problem, please contact Jun Liu (j.liu@asu.edu). Thanks!"); - return; - } - - } - - /* - printf("\n x=%e, f=%e, step=%d\n",x, f, step); - */ -} - -double norm(double * v, double p, int n) -{ - int i; - double t=0; - - - /* - we assume that v[i]>=0 - p>1 - */ - - for(i=0;i0) - flag[i]=0; - else - { - flag[i]=1; - v[i]=-v[i];/* - we set v[i] to its absolute value - */ - } - - vq+=pow(v[i], q); - - - if (v[i]>vmax) - vmax=v[i]; - - if (v[i]= vq){ - *cc=0; - iter_step[0]=iter_step[1]=0; - - - for(i=0;i2){ - - if ( log((1-epsilon) * vmin) - (p-1) * log( epsilon* vmin ) >= 709 ) - { - /* If this contition holds, we have c2 >= 1e308, exceeding the machine precision. - - In this case, the reason is that p is too large - and meanwhile epsilon * vmin is typically small. - - For numerical stablity, we just regard p=inf, and run eppInf - */ - - - for(i=0;i0){ - if ( (x_diff <=delta) && (p_n==0) ) - break; - - c1=c; p_n=1; - } - else{ - - if ( (x_diff <=delta) && (p_n==1) ) - break; - - c2=c; p_n=0; - } - } - c=(c1+c2)/2; - - if (bisStep>=outerIter){ - - - if ( fabs(c1-c2) <=delta * c2 ) - break; - else{ - printf("\n The number of bisection steps exceed %d.", outerIter); - printf("\n c1=%e, c2=%e, x_diff=%e, f=%e",c1,c2,x_diff,f); - printf("\n If you meet with this problem, please contact Jun Liu (j.liu@asu.edu). Thanks!"); - free(flag); - - return; - } - } - - /* - printf("\n c1=%e, c2=%e, f=%e, newtonStep=%d", c1, c2, f, newtonStep); - */ - } - - /* - printf("\n c1=%e, c2=%e, x_diff=%e, f=%e, bisStep=%d, totoalStep=%d",c1,c2, x_diff, f,bisStep,totoalStep); - */ - - for(i=0;i=1e6) /* when p >=1e6, we treat it as infity*/ - eppInf(x, c, iter_step, v, n, rho, c0); - else - eppO(x, c, iter_step, v, n, rho, p); -} - -#endif //USE_GPL_SHOGUN diff --git a/src/shogun/lib/slep/q1/epph.h b/src/shogun/lib/slep/q1/epph.h deleted file mode 100644 index cd3d36e25d4..00000000000 --- a/src/shogun/lib/slep/q1/epph.h +++ /dev/null @@ -1,157 +0,0 @@ -/* This program is free software: you can redistribute it and/or modify - * it under the terms of the GNU General Public License as published by - * the Free Software Foundation, either version 3 of the License, or - * (at your option) any later version. - * - * This program is distributed in the hope that it will be useful, - * but WITHOUT ANY WARRANTY; without even the implied warranty of - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the - * GNU General Public License for more details. - * - * You should have received a copy of the GNU General Public License - * along with this program. If not, see . - * - * Copyright (C) 2009 - 2012 Jun Liu and Jieping Ye - */ - - -#ifndef EPPHQ1_SLEP -#define EPPHQ1_SLEP - -#include -#ifdef USE_GPL_SHOGUN - -/* -------------------------- Function eplb ----------------------------- - - Euclidean Projection onto l1 Ball (eplb) - - min 1/2 ||x- v||_2^2 - s.t. ||x||_1 <= z - - which is converted to the following zero finding problem - - f(lambda)= sum( max( |v|-lambda,0) )-z=0 - - For detail, please refer to our paper: - - Jun Liu and Jieping Ye. Efficient Euclidean Projections in Linear Time, - ICML 2009. - - Usage (in matlab): - [x, lambda, iter_step]=eplb(v, n, z, lambda0); - - -------------------------- Function eplb ----------------------------- - */ -void eplb(double * x, double *root, int * steps, double * v,int n, double z, double lambda0); - -/* -------------------------- Function epp1 ----------------------------- - - The L1-norm Regularized Euclidean Projection (epp1) - - min 1/2 ||x- v||_2^2 + rho ||x||_1 - - which has the closed form solution - - x= sign(v) max( |v|- rho, 0) - - Usage (in matlab) - x=epp1(v, n, rho); - - -------------------------- Function epp1 ----------------------------- - */ -void epp1(double *x, double *v, int n, double rho); - -/* -------------------------- Function epp2 ----------------------------- - - The L2-norm Regularized Euclidean Projection (epp2) - - min 1/2 ||x- v||_2^2 + rho ||x||_2 - - which has the closed form solution - - x= max( ||v||_2- rho, 0) / ||v||_2 * v - - Usage (in matlab) - x=epp2(v, n, rho); - - -------------------------- Function epp2 ----------------------------- - */ -void epp2(double *x, double *v, int n, double rho); - -/* -------------------------- Function eppInf ----------------------------- - - The LInf-norm Regularized Euclidean Projection (eppInf) - - min 1/2 ||x- v||_2^2 + rho ||x||_Inf - - which is can be solved by using eplb - - Usage (in matlab) - [x, lambda, iter_step]=eppInf(v, n, rho, rho0); - - -------------------------- Function eppInf ----------------------------- - */ -void eppInf(double *x, double * c, int * iter_step, double *v, int n, double rho, double c0); - -/* -------------------------- Function zerofind ----------------------------- - - Find the root for the function: f(x) = x + c x^{p-1} - v, - 0 <= x <= v, v>=0 - 1< p < infty, p \neq 2 - - Property: when p>2, f(x) is a convex function - when 1=1 - - min 1/2 ||x- v||_2^2 + rho ||x||_p - - This function uses the previously defined functions. - - Usage (in matlab) - [x, c, iter_step]=eppO(v, n, rho, p, c0); - - -------------------------- Function epp ----------------------------- - */ -void epp(double *x, double * c, int * iter_step, double * v, int n, double rho, double p, double c0); -#endif //USE_GPL_SHOGUN -#endif /* ----- #ifndef EPPHQ1_SLEP ----- */ - diff --git a/src/shogun/lib/slep/q1/epsgLasso.h b/src/shogun/lib/slep/q1/epsgLasso.h deleted file mode 100644 index 30fc0e3f98a..00000000000 --- a/src/shogun/lib/slep/q1/epsgLasso.h +++ /dev/null @@ -1,203 +0,0 @@ -/* This program is free software: you can redistribute it and/or modify - * it under the terms of the GNU General Public License as published by - * the Free Software Foundation, either version 3 of the License, or - * (at your option) any later version. - * - * This program is distributed in the hope that it will be useful, - * but WITHOUT ANY WARRANTY; without even the implied warranty of - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the - * GNU General Public License for more details. - * - * You should have received a copy of the GNU General Public License - * along with this program. If not, see . - * - * Copyright (C) 2009 - 2012 Jun Liu and Jieping Ye - */ - - -#ifndef EPSGLASSO_SLEP -#define EPSGLASSO_SLEP - -#include -#ifdef USE_GPL_SHOGUN - -#include -#include -#include -#include -#include /* This is the head file that contains the implementation of the used functions*/ - - -/* - Projection for sgLasso - - min 1/2 \|X - V\|_F^2 + \lambda_1 \|X\|_1 + \lambda_2 \|X\|_{p,1} - - Written by Jun Liu, January 15, 2010 - For any problem, please contact: j.liu@asu.edu - - */ - -void epsgLasso(double *X, double * normx, double * V, int k, int n, double lambda1, double lambda2, int pflag){ - int i, j, *iter_step, nn=n*k, m; - double *v, *x; - double normValue,c0=0, c; - - v=(double *)malloc(sizeof(double)*n); - x=(double *)malloc(sizeof(double)*n); - iter_step=(int *)malloc(sizeof(int)*2); - - /* - initialize normx - */ - normx[0]=normx[1]=0; - - - /* - X and V are k x n matrices in matlab, stored in column priority manner - x corresponds a row of X - - pflag=2: p=2 - pflag=0: p=inf - */ - - /* - soft thresholding - by lambda1 - - the results are stored in X - */ - for (i=0;i lambda1) - X[i]=V[i] - lambda1; - else - X[i]=0; - } - - /* - Shrinkage or Truncating - by lambda2 - */ - if (pflag==2){ - for(i=0; i c) - X[i + j*k]=c; - else - if (X[i + j*k]<-c) - X[i + j*k]=-c; - } - } - } - } - - - free(v); - free(x); - free(iter_step); -} -#endif //USE_GPL_SHOGUN -#endif /* ----- #ifndef EPSGLASSO_SLEP ----- */ - diff --git a/src/shogun/lib/slep/q1/epsp.h b/src/shogun/lib/slep/q1/epsp.h deleted file mode 100644 index 7a14b0981fe..00000000000 --- a/src/shogun/lib/slep/q1/epsp.h +++ /dev/null @@ -1,305 +0,0 @@ -/* This program is free software: you can redistribute it and/or modify - * it under the terms of the GNU General Public License as published by - * the Free Software Foundation, either version 3 of the License, or - * (at your option) any later version. - * - * This program is distributed in the hope that it will be useful, - * but WITHOUT ANY WARRANTY; without even the implied warranty of - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the - * GNU General Public License for more details. - * - * You should have received a copy of the GNU General Public License - * along with this program. If not, see . - * - * Copyright (C) 2009 - 2012 Jun Liu and Jieping Ye - */ - -#ifdef USE_GPL_SHOGUN - -#ifndef EPSP_SLEP -#define EPSP_SLEP - -#include - -#include -#include -#include -#include - -#define delta 1e-12 - -/* - Euclidean Projection onto the simplex (epsp) - - min 1/2 ||x- y||_2^2 - s.t. ||x||_1 = z, x >=0 - -which is converted to the following zero finding problem - - f(lambda)= sum( max( x-lambda,0) )-z=0 - - Usage: - [x, lambda, iter_step]=epsp(y, n, z, lambda0); - - */ - -void epsp(double * x, double *root, int * stepsp, double * v, -int n, double z, double lambda0) -{ - - int i, j, flag=0; - int rho_1, rho_2, rho, rho_T, rho_S; - int V_i_b, V_i_e, V_i; - double lambda_1, lambda_2, lambda_T, lambda_S, lambda; - double s_1, s_2, s, s_T, s_S, v_max, temp; - double f_lambda_1, f_lambda_2, f_lambda, f_lambda_T, f_lambda_S; - int iter_step=0; - - - - if (z< 0){ - printf("\n z should be nonnegative!"); - exit(-1); - } - - - /* - * find the maximal value in v - */ - v_max=v[0]; - for (i=1;i v_max) - v_max=v[i]; - } - - - lambda_1=v_max - z; lambda_2=v_max; - /* - * copy v to x - * compute f_lambda_1, rho_1, s_1 - */ - V_i=0;s_1=0; rho_1=0; - for (i=0;i lambda_1){ - x[V_i]=v[i]; - - s_1+=x[V_i]; rho_1++; - - V_i++; - } - } - f_lambda_1=s_1-rho_1* lambda_1 -z; - - rho_2=0; s_2=0; f_lambda_2=-z; - V_i_b=0; V_i_e=V_i-1; - - lambda=lambda0; - if ( (lambda lambda_1) ){ - /*------------------------------------------------------------------- - Initialization with the root - *------------------------------------------------------------------- - */ - - i=V_i_b; j=V_i_e; rho=0; s=0; - while (i <= j){ - while( (i <= V_i_e) && (x[i] <= lambda) ){ - i++; - } - while( (j>=V_i_b) && (x[j] > lambda) ){ - s+=x[j]; - j--; - } - if (i lambda_T) - lambda_T=lambda_2 + f_lambda_2 /rho_2; - } - - /* compute lambda_S */ - lambda_S=lambda_2 - f_lambda_2 *(lambda_2-lambda_1)/(f_lambda_2-f_lambda_1); - - if (fabs(lambda_T-lambda_S) <= delta){ - lambda=lambda_T; flag=1; - break; - } - - /* set lambda as the middle point of lambda_T and lambda_S */ - lambda=(lambda_T+lambda_S)/2; - - s_T=s_S=s=0; - rho_T=rho_S=rho=0; - i=V_i_b; j=V_i_e; - while (i <= j){ - while( (i <= V_i_e) && (x[i] <= lambda) ){ - if (x[i]> lambda_T){ - s_T+=x[i]; rho_T++; - } - i++; - } - while( (j>=V_i_b) && (x[j] > lambda) ){ - if (x[j] > lambda_S){ - s_S+=x[j]; rho_S++; - } - else{ - s+=x[j]; rho++; - } - j--; - } - if (i lambda_S){ - s_S+=x[i]; rho_S++; - } - else{ - s+=x[i]; rho++; - } - - if (x[j]> lambda_T){ - s_T+=x[j]; rho_T++; - } - - temp=x[i]; x[i]=x[j]; x[j]=temp; - i++; j--; - } - } - - s_S+=s_2; rho_S+=rho_2; - s+=s_S; rho+=rho_S; - s_T+=s; rho_T+=rho; - f_lambda_S=s_S-rho_S*lambda_S-z; - f_lambda=s-rho*lambda-z; - f_lambda_T=s_T-rho_T*lambda_T-z; - - /*printf("\n %d & %d & %5.6f & %5.6f & %5.6f & %5.6f & %5.6f \\\\ \n \\hline ", iter_step, V_i, lambda_1, lambda_T, lambda, lambda_S, lambda_2);*/ - - if ( fabs(f_lambda)< delta ){ - /*printf("\n lambda");*/ - flag=1; - break; - } - if ( fabs(f_lambda_S)< delta ){ - /* printf("\n lambda_S");*/ - lambda=lambda_S; flag=1; - break; - } - if ( fabs(f_lambda_T)< delta ){ - /* printf("\n lambda_T");*/ - lambda=lambda_T; flag=1; - break; - } - - /* - printf("\n\n f_lambda_1=%5.6f, f_lambda_2=%5.6f, f_lambda=%5.6f",f_lambda_1,f_lambda_2, f_lambda); - printf("\n lambda_1=%5.6f, lambda_2=%5.6f, lambda=%5.6f",lambda_1, lambda_2, lambda); - printf("\n rho_1=%d, rho_2=%d, rho=%d ",rho_1, rho_2, rho); - */ - - if (f_lambda <0){ - lambda_2=lambda; s_2=s; rho_2=rho; - f_lambda_2=f_lambda; - - lambda_1=lambda_T; s_1=s_T; rho_1=rho_T; - f_lambda_1=f_lambda_T; - - V_i_e=j; i=V_i_b; - while (i <= j){ - while( (i <= V_i_e) && (x[i] <= lambda_T) ){ - i++; - } - while( (j>=V_i_b) && (x[j] > lambda_T) ){ - j--; - } - if (i=V_i_b) && (x[j] > lambda_S) ){ - j--; - } - if (i lambda) - x[i]=v[i]-lambda; - else - x[i]=0; - } - *root=lambda; - *stepsp=iter_step; -} -#endif /* ----- #ifndef EPSP_SLEP ----- */ - -#endif //USE_GPL_SHOGUN diff --git a/src/shogun/lib/slep/slep_mc_plain_lr.cpp b/src/shogun/lib/slep/slep_mc_plain_lr.cpp deleted file mode 100644 index f10592f2e48..00000000000 --- a/src/shogun/lib/slep/slep_mc_plain_lr.cpp +++ /dev/null @@ -1,235 +0,0 @@ -/* - * This program is free software; you can redistribute it and/or modify - * it under the terms of the GNU General Public License as published by - * the Free Software Foundation; either version 3 of the License, or - * (at your option) any later version. - * - * Written (W) 2012 Sergey Lisitsyn - * Copyright (C) 2010-2012 Jun Liu, Jieping Ye - */ - -#include -#ifdef USE_GPL_SHOGUN - -#include -#include -#include -#include -#include -#include - -using namespace shogun; -using namespace Eigen; -using namespace std; - -namespace shogun -{ - -slep_result_t slep_mc_plain_lr( - CDotFeatures* features, - CMulticlassLabels* labels, - float64_t z, - const slep_options& options) -{ - int i,j; - // obtain problem parameters - int n_feats = features->get_dim_feature_space(); - int n_vecs = features->get_num_vectors(); - int n_classes = labels->get_num_classes(); - - // labels vector containing values in range (0 .. n_classes) - SGVector labels_vector = labels->get_labels(); - - // initialize matrices and vectors to be used - // weight vector - MatrixXd w = MatrixXd::Zero(n_feats, n_classes); - // intercepts (biases) - VectorXd c = VectorXd::Zero(n_classes); - - if (options.last_result) - { - SGMatrix last_w = options.last_result->w; - SGVector last_c = options.last_result->c; - for (i=0; idense_dot_range(Aw.col(j).data(), 0, n_vecs, NULL, w.col(j).data(), n_feats, 0.0); - MatrixXd As = MatrixXd::Zero(n_vecs, n_classes); - MatrixXd Awp = MatrixXd::Zero(n_vecs, n_classes); - // gradients - MatrixXd g = MatrixXd::Zero(n_feats, n_classes); - VectorXd gc = VectorXd::Zero(n_classes); - // projection - MatrixXd v = MatrixXd::Zero(n_feats, n_classes); - - // Lipschitz continuous gradient parameter for line search - double L = 1.0/(n_vecs*n_classes); - // coefficients for search point computation - double alphap = 0, alpha = 1; - - // lambda regularization parameter - double lambda = z; - // objective values - double objective = 0.0; - double objective_p = 0.0; - - int iter = 0; - bool done = false; - CTime time; - //internal::set_is_malloc_allowed(false); - while ((!done) && (iter 0.0 ? aa : 0.0; - // avoid underflow via log-sum-exp trick - fun_s += CMath::log(CMath::exp(-bb) + CMath::exp(aa-bb)) + bb; - double prob = 1.0/(1+CMath::exp(aa)); - double b = ((vec_class == j) ? -1.0 : 1.0)*(1-prob);///(n_vecs*n_classes); - // update gradient of intercepts - gc[j] += b; - // update gradient of weight vectors - features->add_to_dense_vec(b, i, g.col(j).data(), n_feats); - } - } - //fun_s /= (n_vecs*n_classes); - - wp = w; - Awp = Aw; - cp = c; - - int inner_iter = 0; - double fun_x = 0; - - // line search process - while (inner_iter<5000) - { - // compute line search point - v = search_w - g/L; - c = search_c - gc/L; - - // compute projection of gradient - eppMatrix(w.data(),v.data(),n_feats,n_classes,lambda/L,options.q); - - v = w - search_w; - - // update dot products - for (j=0; jdense_dot_range(Aw.col(j).data(), 0, n_vecs, NULL, w.col(j).data(), n_feats, 0.0); - - // compute objective at search point - fun_x = 0; - for (i=0; i 0.0 ? aa : 0.0; - fun_x += CMath::log(CMath::exp(-bb) + CMath::exp(aa-bb)) + bb; - } - } - //fun_x /= (n_vecs*n_classes); - - // check for termination of line search - double r_sum = (v.squaredNorm() + (c-search_c).squaredNorm())/2; - double l_sum = fun_x - fun_s - v.cwiseProduct(g).sum() - (c-search_c).dot(gc); - - // stop if projected gradient is less than 1e-20 - if (r_sum <= 1e-20) - { - SG_SINFO("Gradient step makes little improvement (%f)\n",r_sum) - done = true; - break; - } - - if (l_sum <= r_sum*L) - break; - else - L = CMath::max(2*L, l_sum/r_sum); - - inner_iter++; - } - - // update alpha coefficients - alphap = alpha; - alpha = (1+CMath::sqrt(4*alpha*alpha+1))/2; - - // update wwp and ccp - wwp = w - wp; - ccp = c - cp; - - // update objectives - objective_p = objective; - objective = fun_x; - - // regularize objective with tree norm - double L1q_norm = 0.0; - for (int m=0; m r_w(n_feats,n_classes); - for (j=0; j r_c(n_classes); - for (j=0; j -#ifdef USE_GPL_SHOGUN -#include -#include -#include - -namespace shogun -{ - -/** Accelerated projected gradient solver for multiclass - * logistic regression problem with feature tree regularization. - * - * @param features features to be used - * @param labels labels to be used - * @param z regularization ratio - * @param options options of solver - */ -slep_result_t slep_mc_plain_lr( - CDotFeatures* features, - CMulticlassLabels* labels, - float64_t z, - const slep_options& options); - -}; -#endif //USE_GPL_SHOGUN -#endif /* SLEP_MC_PLAIN_LR_H_ */ - diff --git a/src/shogun/lib/slep/slep_mc_tree_lr.cpp b/src/shogun/lib/slep/slep_mc_tree_lr.cpp deleted file mode 100644 index fc8bb4c0878..00000000000 --- a/src/shogun/lib/slep/slep_mc_tree_lr.cpp +++ /dev/null @@ -1,249 +0,0 @@ -/* - * This program is free software; you can redistribute it and/or modify - * it under the terms of the GNU General Public License as published by - * the Free Software Foundation; either version 3 of the License, or - * (at your option) any later version. - * - * Written (W) 2012 Sergey Lisitsyn - * Copyright (C) 2010-2012 Jun Liu, Jieping Ye - */ - -#include -#ifdef USE_GPL_SHOGUN - -#include -#include -#include -#include -#include -#include -#include -#include - -using namespace shogun; -using namespace Eigen; -using namespace std; - -namespace shogun -{ - -slep_result_t slep_mc_tree_lr( - CDotFeatures* features, - CMulticlassLabels* labels, - float64_t z, - const slep_options& options) -{ - int i,j; - // obtain problem parameters - int n_feats = features->get_dim_feature_space(); - int n_vecs = features->get_num_vectors(); - int n_classes = labels->get_num_classes(); - - // labels vector containing values in range (0 .. n_classes) - SGVector labels_vector = labels->get_labels(); - - // initialize matrices and vectors to be used - // weight vector - MatrixXd w = MatrixXd::Zero(n_feats, n_classes); - // intercepts (biases) - VectorXd c = VectorXd::Zero(n_classes); - - if (options.last_result) - { - SGMatrix last_w = options.last_result->w; - SGVector last_c = options.last_result->c; - for (i=0; idense_dot_range(Aw.col(j).data(), 0, n_vecs, NULL, w.col(j).data(), n_feats, 0.0); - MatrixXd As = MatrixXd::Zero(n_vecs, n_classes); - MatrixXd Awp = MatrixXd::Zero(n_vecs, n_classes); - // gradients - MatrixXd g = MatrixXd::Zero(n_feats, n_classes); - VectorXd gc = VectorXd::Zero(n_classes); - // projection - MatrixXd v = MatrixXd::Zero(n_feats, n_classes); - - // Lipschitz continuous gradient parameter for line search - double L = 1.0/(n_vecs*n_classes); - // coefficients for search point computation - double alphap = 0, alpha = 1; - - // lambda regularization parameter - double lambda = z; - // objective values - double objective = 0.0; - double objective_p = 0.0; - - int iter = 0; - bool done = false; - CTime time; - //internal::set_is_malloc_allowed(false); - while ((!done) && (iter 0.0 ? aa : 0.0; - // avoid underflow via log-sum-exp trick - fun_s += CMath::log(CMath::exp(-bb) + CMath::exp(aa-bb)) + bb; - double prob = 1.0/(1+CMath::exp(aa)); - double b = ((vec_class == j) ? -1.0 : 1.0)*(1-prob);///(n_vecs*n_classes); - // update gradient of intercepts - gc[j] += b; - // update gradient of weight vectors - features->add_to_dense_vec(b, i, g.col(j).data(), n_feats); - } - } - //fun_s /= (n_vecs*n_classes); - - wp = w; - Awp = Aw; - cp = c; - - int inner_iter = 0; - double fun_x = 0; - - // line search process - while (inner_iter<5000) - { - // compute line search point - v = search_w - g/L; - c = search_c - gc/L; - - // compute projection of gradient - if (options.general) - general_altra_mt(w.data(),v.data(),n_classes,n_feats,options.G,options.ind_t,options.n_nodes,lambda/L); - else - altra_mt(w.data(),v.data(),n_classes,n_feats,options.ind_t,options.n_nodes,lambda/L); - v = w - search_w; - - // update dot products - for (j=0; jdense_dot_range(Aw.col(j).data(), 0, n_vecs, NULL, w.col(j).data(), n_feats, 0.0); - - // compute objective at search point - fun_x = 0; - for (i=0; i 0.0 ? aa : 0.0; - fun_x += CMath::log(CMath::exp(-bb) + CMath::exp(aa-bb)) + bb; - } - } - //fun_x /= (n_vecs*n_classes); - - // check for termination of line search - double r_sum = (v.squaredNorm() + (c-search_c).squaredNorm())/2; - double l_sum = fun_x - fun_s - v.cwiseProduct(g).sum() - (c-search_c).dot(gc); - - // stop if projected gradient is less than 1e-20 - if (r_sum <= 1e-20) - { - SG_SINFO("Gradient step makes little improvement (%f)\n",r_sum) - done = true; - break; - } - - if (l_sum <= r_sum*L) - break; - else - L = CMath::max(2*L, l_sum/r_sum); - - inner_iter++; - } - - // update alpha coefficients - alphap = alpha; - alpha = (1+CMath::sqrt(4*alpha*alpha+1))/2; - - // update wwp and ccp - wwp = w - wp; - ccp = c - cp; - - // update objectives - objective_p = objective; - objective = fun_x; - - // compute tree norm - double tree_norm = 0.0; - if (options.general) - { - for (i=0; i r_w(n_feats,n_classes); - for (j=0; j r_c(n_classes); - for (j=0; j -#ifdef USE_GPL_SHOGUN -#include -#include -#include - -namespace shogun -{ - -/** Accelerated projected gradient solver for multiclass - * logistic regression problem with feature tree regularization. - * - * @param features features to be used - * @param labels labels to be used - * @param z regularization ratio - * @param options options of solver - */ -slep_result_t slep_mc_tree_lr( - CDotFeatures* features, - CMulticlassLabels* labels, - float64_t z, - const slep_options& options); - -}; -#endif //USE_GPL_SHOGUN -#endif /* SLEP_MC_TREE_LR_H_ */ - diff --git a/src/shogun/lib/slep/slep_options.h b/src/shogun/lib/slep/slep_options.h deleted file mode 100644 index 0db0cb975c7..00000000000 --- a/src/shogun/lib/slep/slep_options.h +++ /dev/null @@ -1,104 +0,0 @@ -/* - * This program is free software; you can redistribute it and/or modify - * it under the terms of the GNU General Public License as published by - * the Free Software Foundation; either version 3 of the License, or - * (at your option) any later version. - * - * Written (W) 2012 Sergey Lisitsyn - * Copyright (C) 2012 Sergey Lisitsyn - */ - -#ifndef SLEP_OPTIONS_H_ -#define SLEP_OPTIONS_H_ - - -#define IGNORE_IN_CLASSLIST - -#include -#ifdef USE_GPL_SHOGUN - -#include -#include -#include - -namespace shogun -{ - -#ifndef DOXYGEN_SHOULD_SKIP_THIS -IGNORE_IN_CLASSLIST enum slep_mode -{ - MULTITASK_GROUP, - MULTITASK_TREE, - FEATURE_GROUP, - FEATURE_TREE, - PLAIN, - FUSED -}; - -IGNORE_IN_CLASSLIST enum slep_loss -{ - LOGISTIC, - LEAST_SQUARES -}; - -IGNORE_IN_CLASSLIST struct slep_result_t -{ - SGMatrix w; - SGVector c; - - slep_result_t(SGMatrix w_, SGVector c_) - { - w = w_; - c = c_; - } -}; - -IGNORE_IN_CLASSLIST struct slep_options -{ - bool general; - int termination; - double tolerance; - int max_iter; - int restart_num; - int n_nodes; - int n_tasks; - int regularization; - int n_feature_blocks; - int* ind; - double rsL2; - double* ind_t; - double* G; - double* gWeight; - double q; - SGVector* tasks_indices; - slep_loss loss; - slep_mode mode; - slep_result_t* last_result; - - static slep_options default_options() - { - slep_options opts; - opts.general = false; - opts.termination = 0; - opts.tolerance = 1e-3; - opts.max_iter = 1000; - opts.restart_num = 100; - opts.regularization = 0; - opts.q = 2.0; - opts.gWeight = NULL; - opts.ind = NULL; - opts.ind_t = NULL; - opts.G = NULL; - opts.rsL2 = 0.0; - opts.last_result = NULL; - opts.tasks_indices = NULL; - opts.loss = LOGISTIC; - opts.mode = MULTITASK_GROUP; - return opts; - } -}; -#endif -} -#endif //USE_GPL_SHOGUN -#endif /* ----- #ifndef SLEP_OPTIONS_H_ ----- */ - diff --git a/src/shogun/lib/slep/slep_solver.cpp b/src/shogun/lib/slep/slep_solver.cpp deleted file mode 100644 index 47adda59485..00000000000 --- a/src/shogun/lib/slep/slep_solver.cpp +++ /dev/null @@ -1,746 +0,0 @@ -/* - * This program is free software; you can redistribute it and/or modify - * it under the terms of the GNU General Public License as published by - * the Free Software Foundation; either version 3 of the License, or - * (at your option) any later version. - * - * Written (W) 2012 Sergey Lisitsyn - * Copyright (C) 2010-2012 Jun Liu, Jieping Ye - */ - - -#include -#ifdef USE_GPL_SHOGUN -#include -#include -#include -#include -#include -#include -#include - -namespace shogun -{ - -double compute_regularizer(double* w, double lambda, double lambda2, int n_vecs, int n_feats, - int n_blocks, const slep_options& options) -{ - double regularizer = 0.0; - switch (options.mode) - { - case MULTITASK_GROUP: - { - for (int i=0; i1) - SG_SERROR("z is not in range [0,1]") - - double q_bar = 0.0; - if (options.q==1) - q_bar = CMath::ALMOST_INFTY; - else if (options.q>1e6) - q_bar = 1; - else - q_bar = options.q/(options.q-1); - - SG_SINFO("q bar = %f \n",q_bar) - - switch (options.mode) - { - case MULTITASK_GROUP: - case MULTITASK_TREE: - { - for (int t=0; t task_idx = options.tasks_indices[t]; - int n_vecs_task = task_idx.vlen; - - switch (options.loss) - { - case LOGISTIC: - { - double b = 0.0; - int m1 = 0, m2 = 0; - for (int i=0; i0) - m1++; - else - m2++; - } - for (int i=0; i0) - b = double(m1)/(m1+m2); - else - b = -double(m2)/(m1+m2); - - features->add_to_dense_vec(b,task_idx[i],ATx+t*n_feats,n_feats); - } - } - break; - case LEAST_SQUARES: - { - for (int i=0; iadd_to_dense_vec(y[task_idx[i]],task_idx[i],ATx+t*n_feats,n_feats); - } - } - } - } - break; - case FEATURE_GROUP: - case FEATURE_TREE: - case PLAIN: - case FUSED: - { - switch (options.loss) - { - case LOGISTIC: - { - int m1 = 0, m2 = 0; - double b = 0.0; - for (int i=0; i0 ? m1++ : m2++; - - SG_SDEBUG("# pos = %d , # neg = %d\n",m1,m2) - - for (int i=0; i0 ? b=double(m2) / CMath::sq(n_vecs) : b=-double(m1) / CMath::sq(n_vecs); - features->add_to_dense_vec(b,i,ATx,n_feats); - } - } - break; - case LEAST_SQUARES: - { - for (int i=0; iadd_to_dense_vec(y[i],i,ATx,n_feats); - } - break; - } - } - break; - } - - switch (options.mode) - { - case MULTITASK_GROUP: - { - for (int i=0; ilambda_max) - lambda_max = sum; - } - } - break; - case FEATURE_TREE: - { - if (options.general) - lambda_max = general_findLambdaMax(ATx, n_feats, options.G, options.ind_t, options.n_nodes); - else - lambda_max = findLambdaMax(ATx, n_feats, options.ind_t, options.n_nodes); - } - break; - case PLAIN: - case FUSED: - { - double max = 0.0; - for (int i=0; i max) - max = CMath::abs(ATx[i]); - } - lambda_max = max; - } - break; - } - - SG_SINFO("Computed lambda = %f * %f = %f\n",z,lambda_max,z*lambda_max) - return z*lambda_max; -} - -void projection(double* w, double* v, int n_feats, int n_blocks, double lambda, double lambda2, - double L, double* z, double* z0, const slep_options& options) -{ - switch (options.mode) - { - case MULTITASK_GROUP: - eppMatrix(w, v, n_feats, n_blocks, lambda/L, options.q); - break; - case MULTITASK_TREE: - if (options.general) - general_altra_mt(w, v, n_feats, n_blocks, options.G, options.ind_t, options.n_nodes, lambda/L); - else - altra_mt(w, v, n_feats, n_blocks, options.ind_t, options.n_nodes, lambda/L); - break; - case FEATURE_GROUP: - eppVector(w, v, options.ind, n_blocks, n_feats, options.gWeight, lambda/L, options.q > 1e6 ? 1e6 : options.q); - break; - case FEATURE_TREE: - if (options.general) - general_altra(w, v, n_feats, options.G, options.ind_t, options.n_nodes, lambda/L); - else - altra(w, v, n_feats, options.ind_t, options.n_nodes, lambda/L); - break; - case PLAIN: - for (int i=0; i task_idx = options.tasks_indices[t]; - int n_vecs_task = task_idx.vlen; - switch (options.loss) - { - case LOGISTIC: - gc[t] = 0.0; - for (int i=0; iadd_to_dense_vec(b,task_idx[i],g+t*n_feats,n_feats); - } - break; - case LEAST_SQUARES: - for (int i=0; iadd_to_dense_vec(As[task_idx[i]],task_idx[i],g+t*n_feats,n_feats); - break; - } - } - break; - case FEATURE_GROUP: - case FEATURE_TREE: - case PLAIN: - case FUSED: - switch (options.loss) - { - case LOGISTIC: - gc[0] = 0.0; - - for (int i=0; i0) - fun_s += (CMath::log(CMath::exp(-bb) + CMath::exp(aa-bb)) + bb)*pos_weight; - else - fun_s += (CMath::log(CMath::exp(-bb) + CMath::exp(aa-bb)) + bb)*neg_weight; - */ - double prob = 1.0/(1.0+CMath::exp(aa)); - //double b = 0; - double b = -y[i]*(1.0-prob)/n_vecs; - /* - if (y[i]>0) - b = -y[i]*(1.0-prob)*pos_weight; - else - b = -y[i]*(1.0-prob)*neg_weight; - */ - gc[0] += b; - features->add_to_dense_vec(b,i,g,n_feats); - } - fun_s /= n_vecs; - break; - case LEAST_SQUARES: - for (int i=0; iadd_to_dense_vec(As[i],i,g,n_feats); - break; - } - break; - } - SG_SDEBUG("G=%f\n", CMath::dot(g,g,n_feats*n_tasks)) - - return fun_s; -} - -slep_result_t slep_solver( - CDotFeatures* features, - double* y, - double z, - const slep_options& options) -{ - int i,t; - int n_feats = features->get_dim_feature_space(); - int n_vecs = features->get_num_vectors(); - double lambda, beta; - double funcp = 0.0, func = 0.0; - - int n_blocks = 0; - int n_tasks = 0; - - switch (options.mode) - { - case MULTITASK_GROUP: - case MULTITASK_TREE: - n_tasks = options.n_tasks; - n_blocks = options.n_tasks; - break; - case FEATURE_GROUP: - case FEATURE_TREE: - n_tasks = 1; - n_blocks = options.n_feature_blocks; - break; - case PLAIN: - case FUSED: - n_tasks = 1; - n_blocks = 1; - break; - } - SG_SDEBUG("n_tasks = %d, n_blocks = %d\n",n_tasks,n_blocks) - SG_SDEBUG("n_nodes = %d\n",options.n_nodes) - - int iter = 1; - bool done = false; - bool gradient_break = false; - - double rsL2 = options.rsL2; - - double* ATx = SG_CALLOC(double, n_feats*n_tasks); - if (options.regularization!=0) - { - lambda = compute_lambda(ATx, z, features, y, n_vecs, n_feats, n_blocks, options); - rsL2*= lambda; - } - else - lambda = z; - - double lambda2 = 0.0; - - SGMatrix w(n_feats,n_tasks); - w.zero(); - SGVector c(n_tasks); - c.zero(); - - if (options.last_result) - { - w = options.last_result->w; - c = options.last_result->c; - } - - double* s = SG_CALLOC(double, n_feats*n_tasks); - double* sc = SG_CALLOC(double, n_tasks); - double* g = SG_CALLOC(double, n_feats*n_tasks); - double* v = SG_CALLOC(double, n_feats*n_tasks); - double* z_flsa = SG_CALLOC(double, n_feats); - double* z0_flsa = SG_CALLOC(double, n_feats); - - double* Aw = SG_CALLOC(double, n_vecs); - switch (options.mode) - { - case MULTITASK_GROUP: - case MULTITASK_TREE: - { - for (t=0; t task_idx = options.tasks_indices[t]; - //task_idx.display_vector("task"); - int n_vecs_task = task_idx.vlen; - for (i=0; idense_dot(task_idx[i],w.matrix+t*n_feats,n_feats); - } - } - break; - case FEATURE_GROUP: - case FEATURE_TREE: - case PLAIN: - case FUSED: - { - for (i=0; idense_dot(i,w.matrix,n_feats); - } - break; - } - - double* Av = SG_MALLOC(double, n_vecs); - double* As = SG_MALLOC(double, n_vecs); - - double L = 1.0/n_vecs; - - if (options.mode==FUSED) - L += rsL2; - - double* wp = SG_CALLOC(double, n_feats*n_tasks); - for (i=0; i task_idx = options.tasks_indices[t]; - int n_vecs_task = task_idx.vlen; - for (i=0; idense_dot(task_idx[i],w.matrix+t*n_feats,n_feats); - if (options.loss==LOGISTIC) - { - double aa = -y[task_idx[i]]*(Aw[task_idx[i]]+c[t]); - double bb = CMath::max(aa,0.0); - fun_x += (CMath::log(CMath::exp(-bb) + CMath::exp(aa-bb)) + bb); - } - } - } - break; - case FEATURE_GROUP: - case FEATURE_TREE: - case PLAIN: - case FUSED: - for (i=0; idense_dot(i, w.matrix, n_feats); - if (options.loss==LOGISTIC) - { - double aa = -y[i]*(Aw[i]+c[0]); - double bb = CMath::max(aa,0.0); - if (y[i]>0) - fun_x += (CMath::log(CMath::exp(-bb) + CMath::exp(aa-bb)) + bb);//*pos_weight; - else - fun_x += (CMath::log(CMath::exp(-bb) + CMath::exp(aa-bb)) + bb);//*neg_weight; - } - } - break; - } - if (options.loss==LOGISTIC) - fun_x /= n_vecs; - if (options.mode==PLAIN || options.mode==FUSED) - fun_x += rsL2/2 * CMath::dot(w.matrix,w.matrix,n_feats); - - double l_sum = 0.0, r_sum = 0.0; - switch (options.loss) - { - case LOGISTIC: - r_sum = CMath::dot(v,v,n_feats*n_tasks); - l_sum = fun_x - fun_s - CMath::dot(v,g,n_feats*n_tasks); - for (t=0; t=2) - { - step = CMath::abs(func-funcp); - if (step <= options.tolerance) - { - SG_SINFO("Objective changes less than tolerance\n") - done = true; - } - } - break; - case 1: - if (iter>=2) - { - step = CMath::abs(func-funcp); - if (step <= step*options.tolerance) - { - SG_SINFO("Objective changes relatively less than tolerance\n") - done = true; - } - } - break; - case 2: - if (func <= options.tolerance) - { - SG_SINFO("Objective is less than tolerance\n") - done = true; - } - break; - case 3: - norm_wwp = CMath::sqrt(CMath::dot(wwp,wwp,n_feats*n_tasks)); - if (norm_wwp <= options.tolerance) - done = true; - break; - case 4: - norm_wp = CMath::sqrt(CMath::dot(wp,wp,n_feats*n_tasks)); - norm_wwp = CMath::sqrt(CMath::dot(wwp,wwp,n_feats*n_tasks)); - if (norm_wwp <= options.tolerance*CMath::max(norm_wp,1.0)) - done = true; - break; - default: - done = true; - } - - iter++; - } - SG_SINFO("Finished %d iterations, objective = %f\n", iter, func) - - SG_FREE(ATx); - SG_FREE(wp); - SG_FREE(wwp); - SG_FREE(s); - SG_FREE(sc); - SG_FREE(cp); - SG_FREE(ccp); - SG_FREE(g); - SG_FREE(v); - SG_FREE(Aw); - SG_FREE(Awp); - SG_FREE(Av); - SG_FREE(As); - SG_FREE(gc); - SG_FREE(z_flsa); - SG_FREE(z0_flsa); - - return slep_result_t(w,c); -}; -}; - -#endif //USE_GPL_SHOGUN diff --git a/src/shogun/lib/slep/slep_solver.h b/src/shogun/lib/slep/slep_solver.h deleted file mode 100644 index cf77ffa0670..00000000000 --- a/src/shogun/lib/slep/slep_solver.h +++ /dev/null @@ -1,47 +0,0 @@ -/* - * This program is free software; you can redistribute it and/or modify - * it under the terms of the GNU General Public License as published by - * the Free Software Foundation; either version 3 of the License, or - * (at your option) any later version. - * - * Written (W) 2012 Sergey Lisitsyn - * Copyright (C) 2010-2012 Jun Liu, Jieping Ye - */ - -#ifndef SLEP_MT_LOGISTIC_H_ -#define SLEP_MT_LOGISTIC_H_ - - -#include -#ifdef USE_GPL_SHOGUN - -#include -#include - -namespace shogun -{ - -/** Learning optimization task solver ported from the - * SLEP (Sparse LEarning Package) library. - * - * Based on accelerated projected gradient method. - * - * Supports two types of losses: logistic and least squares. - * - * Supports multitask problems (task group [MULTITASK_GROUP] - * and task tree [MULTITASK_TREE] relations), - * problems with feature relations (feature group [FEATURE_GROUP] - * and feature tree [FEATURE_TREE]), - * basic regularized problems [PLAIN] and fused formulation. - * - */ -slep_result_t slep_solver( - CDotFeatures* features, - double* y, - double z, - const slep_options& options); - -}; -#endif //USE_GPL_SHOGUN -#endif /* ----- #ifndef SLEP_LOGISTIC_H_ ----- */ - diff --git a/src/shogun/lib/slep/tree/altra.cpp b/src/shogun/lib/slep/tree/altra.cpp deleted file mode 100644 index 5c4e5453233..00000000000 --- a/src/shogun/lib/slep/tree/altra.cpp +++ /dev/null @@ -1,402 +0,0 @@ -/* This program is free software: you can redistribute it and/or modify - * it under the terms of the GNU General Public License as published by - * the Free Software Foundation, either version 3 of the License, or - * (at your option) any later version. - * - * This program is distributed in the hope that it will be useful, - * but WITHOUT ANY WARRANTY; without even the implied warranty of - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the - * GNU General Public License for more details. - * - * You should have received a copy of the GNU General Public License - * along with this program. If not, see . - * - * Copyright (C) 2009 - 2012 Jun Liu and Jieping Ye - */ - - -#include -#ifdef USE_GPL_SHOGUN -#include -#include -#include - -void altra(double *x, double *v, int n, double *ind, int nodes, double mult) -{ - int i, j; - double lambda,twoNorm, ratio; - - /* - * test whether the first node is special - */ - if ((int) ind[0]==-1){ - - /* - *Recheck whether ind[1] equals to zero - */ - if ((int) ind[1]!=-1){ - printf("\n Error! \n Check ind"); - exit(1); - } - - lambda=mult*ind[2]; - - for(j=0;jlambda) - x[j]=v[j]-lambda; - else - if (v[j]<-lambda) - x[j]=v[j]+lambda; - else - x[j]=0; - } - - i=1; - } - else{ - shogun::sg_memcpy(x, v, sizeof(double) * n); - i=0; - } - - /* - * sequentially process each node - * - */ - for(;i < nodes; i++){ - /* - * compute the L2 norm of this group - */ - twoNorm=0; - for(j=(int) ind[3*i]-1;j< (int) ind[3*i+1];j++) - twoNorm += x[j] * x[j]; - twoNorm=sqrt(twoNorm); - - lambda=mult*ind[3*i+2]; - if (twoNorm>lambda){ - ratio=(twoNorm-lambda)/twoNorm; - - /* - * shrinkage this group by ratio - */ - for(j=(int) ind[3*i]-1;j<(int) ind[3*i+1];j++) - x[j]*=ratio; - } - else{ - /* - * threshold this group to zero - */ - for(j=(int) ind[3*i]-1;j<(int) ind[3*i+1];j++) - x[j]=0; - } - } -} - -void altra_mt(double *X, double *V, int n, int k, double *ind, int nodes, double mult) -{ - int i, j; - - double *x=(double *)malloc(sizeof(double)*k); - double *v=(double *)malloc(sizeof(double)*k); - - for (i=0;i*lambda2_max ) - *lambda2_max=twoNorm; - } -} - -double treeNorm(double *x, int ldx, int n, double *ind, int nodes){ - - int i, j; - double twoNorm, lambda; - - double tree_norm = 0; - - /* - * test whether the first node is special - */ - if ((int) ind[0]==-1){ - - /* - *Recheck whether ind[1] equals to zero - */ - if ((int) ind[1]!=-1){ - printf("\n Error! \n Check ind"); - exit(1); - } - - lambda=ind[2]; - - for(j=0;j=n) { - /*x is a zero vector*/ - lambda2=lambda; - lambda1=lambda; - - num=0; - - while(1){ - num++; - - lambda2=lambda; - lambda1=lambda1/2; - /* update ind2 - */ - for(i=0;i=n){ - break; - /*x is a zero vector - *we have found lambda2 - */ - } - } - } - - /* - printf("\n num=%d, lambda1=%2.5f, lambda2=%2.5f",num, lambda1,lambda2); - */ - - while ( fabs(lambda2-lambda1) > lambda2 * 1e-10 ){ - - num++; - - lambda=(lambda1+lambda2)/2; - - /* update ind2 - */ - for(i=0;i=n){ - lambda2=lambda; - } - else{ - lambda1=lambda; - } - - /* - printf("\n lambda1=%2.5f, lambda2=%2.5f",lambda1,lambda2); - */ - } - - /* - printf("\n num=%d",num); - - printf(" lambda1=%2.5f, lambda2=%2.5f",lambda1,lambda2); - -*/ - - free(x); - free(ind2); - - return lambda2; -} - -double findLambdaMax_mt(double *V, int n, int k, double *ind, int nodes) -{ - int i, j; - - double *v=(double *)malloc(sizeof(double)*k); - double lambda; - - double lambdaMax=0; - - for (i=0;ilambdaMax) - lambdaMax=lambda; - } - - /* - printf("\n *lambdaMax=%5.2f",*lambdaMax); - */ - - free(v); - return lambdaMax; -} - -#endif //USE_GPL_SHOGUN diff --git a/src/shogun/lib/slep/tree/altra.h b/src/shogun/lib/slep/tree/altra.h deleted file mode 100644 index 904b8bbde81..00000000000 --- a/src/shogun/lib/slep/tree/altra.h +++ /dev/null @@ -1,183 +0,0 @@ -/* This program is free software: you can redistribute it and/or modify - * it under the terms of the GNU General Public License as published by - * the Free Software Foundation, either version 3 of the License, or - * (at your option) any later version. - * - * This program is distributed in the hope that it will be useful, - * but WITHOUT ANY WARRANTY; without even the implied warranty of - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the - * GNU General Public License for more details. - * - * You should have received a copy of the GNU General Public License - * along with this program. If not, see . - * - * Copyright (C) 2009 - 2012 Jun Liu and Jieping Ye - */ - - -#ifndef ALTRA_SLEP -#define ALTRA_SLEP - -#include -#ifdef USE_GPL_SHOGUN - - - -/* - * Important Notice: September 20, 2010 - * - * In this head file, we assume that the features in the tree strucutre - * are well ordered. That is to say, the indices of the left nodes is always less - * than the right nodes. Ideally, this can be achieved by reordering the features. - * - * The advantage of this ordered features is that, we donot need to use an explicit - * variable for recording the indices. - * - * To deal with the more general case when the features might not be well ordered, - * we provide the functions in the head file "general_altra.h". Compared with the files in this head file, - * we need an additional parameter G, which contains the indices of the nodes. - * - * - */ - -/* - * ------------------------------------------------------------------- - * Functions and parameter - * ------------------------------------------------------------------- - * - * altra solves the following problem - * - * 1/2 \|x-v\|^2 + \sum \lambda_i \|x_{G_i}\|, - * - * where x and v are of dimension n, - * \lambda_i >=0, and G_i's follow the tree structure - * - * It is implemented in Matlab as follows: - * - * x=altra(v, n, ind, nodes); - * - * ind is a 3 x nodes matrix. - * Each column corresponds to a node. - * - * The first element of each column is the starting index, - * the second element of each column is the ending index - * the third element of each column corrreponds to \lambbda_i. - * - * ------------------------------------------------------------------- - * Notices: - * ------------------------------------------------------------------- - * - * 1. The nodes in the parameter "ind" should be given in the - * either - * the postordering of depth-first traversal - * or - * the reverse breadth-first traversal. - * - * 2. When each elements of x are penalized via the same L1 - * (equivalent to the L2 norm) parameter, one can simplify the input - * by specifying - * the "first" column of ind as (-1, -1, lambda) - * - * In this case, we treat it as a single "super" node. Thus in the value - * nodes, we only count it once. - * - * 3. The values in "ind" are in [1,n]. - * - * 4. The third element of each column should be positive. The program does - * not check the validity of the parameter. - * - * It is still valid to use the zero regularization parameter. - * In this case, the program does not change the values of - * correponding indices. - * - * - * ------------------------------------------------------------------- - * History: - * ------------------------------------------------------------------- - * - * Composed by Jun Liu on April 20, 2010 - * - * For any question or suggestion, please email j.liu@asu.edu. - * - */ -void altra(double *x, double *v, int n, double *ind, int nodes, double mult=1.0); - -/* - * altra_mt is a generalization of altra to the - * - * multi-task learning scenario (or equivalently the multi-class case) - * - * altra_mt(X, V, n, k, ind, nodes); - * - * It applies altra for each row (1xk) of X and V - * - */ -void altra_mt(double *X, double *V, int n, int k, double *ind, int nodes, double mult=1.0); - -/* - * compute - * lambda2_max=computeLambda2Max(x,n,ind,nodes); - * - * compute the 2 norm of each group, which is divided by the ind(3,:), - * then the maximum value is returned - */ -/* - *This function does not consider the case ind={[-1, -1, 100]',...} - * - *This functions is not used currently. - */ -void computeLambda2Max(double *lambda2_max, double *x, int n, double *ind, int nodes); - -/* - * ------------------------------------------------------------------- - * Function and parameter - * ------------------------------------------------------------------- - * - * treeNorm compute - * - * \sum \lambda_i \|x_{G_i}\|, - * - * where x is of dimension n, - * \lambda_i >=0, and G_i's follow the tree structure - * - * The file is implemented in the following in Matlab: - * - * tree_norm=treeNorm(x, n, ind,nodes); - */ -double treeNorm(double *x, int ldx, int n, double *ind, int nodes); - -/* - * ------------------------------------------------------------------- - * Function and parameter - * ------------------------------------------------------------------- - * - * findLambdaMax compute - * - * the lambda_{max} that achieves a zero solution for - * - * min 1/2 \|x-v\|^2 + \lambda_{\max} * \sum w_i \|x_{G_i}\|, - * - * where x is of dimension n, - * w_i >=0, and G_i's follow the tree structure - * - * The file is implemented in the following in Matlab: - * - * lambdaMax=findLambdaMax(v, n, ind,nodes); - */ -double findLambdaMax(double *v, int n, double *ind, int nodes); - -/* - * findLambdaMax_mt is a generalization of findLambdaMax to the - * - * multi-task learning scenario (or equivalently the multi-class case) - * - * lambdaMax=findLambdaMax_mt(X, V, n, k, ind, nodes); - * - * It applies findLambdaMax for each row (1xk) of X and V - * - */ -double findLambdaMax_mt(double *V, int n, int k, double *ind, int nodes); -#endif //USE_GPL_SHOGUN -#endif /* ----- #ifndef ALTRA_SLEP ----- */ - - diff --git a/src/shogun/lib/slep/tree/general_altra.cpp b/src/shogun/lib/slep/tree/general_altra.cpp deleted file mode 100644 index feab1ae62bf..00000000000 --- a/src/shogun/lib/slep/tree/general_altra.cpp +++ /dev/null @@ -1,405 +0,0 @@ -/* This program is free software: you can redistribute it and/or modify - * it under the terms of the GNU General Public License as published by - * the Free Software Foundation, either version 3 of the License, or - * (at your option) any later version. - * - * This program is distributed in the hope that it will be useful, - * but WITHOUT ANY WARRANTY; without even the implied warranty of - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the - * GNU General Public License for more details. - * - * You should have received a copy of the GNU General Public License - * along with this program. If not, see . - * - * Copyright (C) 2009 - 2012 Jun Liu and Jieping Ye - */ - -#include -#ifdef USE_GPL_SHOGUN - -#include -#include -#include - -void general_altra(double *x, double *v, int n, double *G, double *ind, int nodes, double mult) -{ - - int i, j; - double lambda,twoNorm, ratio; - - /* - * test whether the first node is special - */ - if ((int) ind[0]==-1){ - - /* - *Recheck whether ind[1] equals to zero - */ - if ((int) ind[1]!=-1){ - printf("\n Error! \n Check ind"); - exit(1); - } - - lambda=mult*ind[2]; - - for(j=0;jlambda) - x[j]=v[j]-lambda; - else - if (v[j]<-lambda) - x[j]=v[j]+lambda; - else - x[j]=0; - } - - i=1; - } - else{ - shogun::sg_memcpy(x, v, sizeof(double) * n); - i=0; - } - - /* - * sequentially process each node - * - */ - for(;i < nodes; i++){ - /* - * compute the L2 norm of this group - */ - twoNorm=0; - for(j=(int) ind[3*i]-1;j< (int) ind[3*i+1];j++) - twoNorm += x[(int) G[j]-1 ] * x[(int) G[j]-1 ]; - twoNorm=sqrt(twoNorm); - - lambda=mult*ind[3*i+2]; - if (twoNorm>lambda){ - ratio=(twoNorm-lambda)/twoNorm; - - /* - * shrinkage this group by ratio - */ - for(j=(int) ind[3*i]-1;j<(int) ind[3*i+1];j++) - x[(int) G[j]-1 ]*=ratio; - } - else{ - /* - * threshold this group to zero - */ - for(j=(int) ind[3*i]-1;j<(int) ind[3*i+1];j++) - x[(int) G[j]-1 ]=0; - } - } -} - -void general_altra_mt(double *X, double *V, int n, int k, double *G, double *ind, int nodes, double mult) -{ - int i, j; - - double *x=(double *)malloc(sizeof(double)*k); - double *v=(double *)malloc(sizeof(double)*k); - - for (i=0;i*lambda2_max ) - *lambda2_max=twoNorm; - } -} - -double general_treeNorm(double *x, int ldx, int n, double *G, double *ind, int nodes) -{ - - int i, j; - double twoNorm, lambda; - - double tree_norm=0; - - /* - * test whether the first node is special - */ - if ((int) ind[0]==-1){ - - /* - *Recheck whether ind[1] equals to zero - */ - if ((int) ind[1]!=-1){ - printf("\n Error! \n Check ind"); - exit(1); - } - - lambda=ind[2]; - - for(j=0;j=n) { - /*x is a zero vector*/ - lambda2=lambda; - lambda1=lambda; - - num=0; - - while(1){ - num++; - - lambda2=lambda; - lambda1=lambda1/2; - /* update ind2 - */ - for(i=0;i=n){ - break; - /*x is a zero vector - *we have found lambda2 - */ - } - } - } - - /* - printf("\n num=%d, lambda1=%2.5f, lambda2=%2.5f",num, lambda1,lambda2); - */ - - while ( fabs(lambda2-lambda1) > lambda2 * 1e-10 ){ - - num++; - - lambda=(lambda1+lambda2)/2; - - /* update ind2 - */ - for(i=0;i=n){ - lambda2=lambda; - } - else{ - lambda1=lambda; - } - - /* - printf("\n lambda1=%2.5f, lambda2=%2.5f",lambda1,lambda2); - */ - } - - /* - printf("\n num=%d",num); - - printf(" lambda1=%2.5f, lambda2=%2.5f",lambda1,lambda2); - */ - - free(x); - free(ind2); - - return lambda2; -} - -double general_findLambdaMax_mt(double *V, int n, int k, double *G, double *ind, int nodes) -{ - int i, j; - - double *v=(double *)malloc(sizeof(double)*k); - double lambda; - - double lambdaMax=0; - - for (i=0;ilambdaMax) - lambdaMax=lambda; - } - - /* - printf("\n *lambdaMax=%5.2f",*lambdaMax); - */ - - free(v); - return lambdaMax; -} - -#endif //USE_GPL_SHOGUN diff --git a/src/shogun/lib/slep/tree/general_altra.h b/src/shogun/lib/slep/tree/general_altra.h deleted file mode 100644 index 75970bcb685..00000000000 --- a/src/shogun/lib/slep/tree/general_altra.h +++ /dev/null @@ -1,197 +0,0 @@ -/* This program is free software: you can redistribute it and/or modify - * it under the terms of the GNU General Public License as published by - * the Free Software Foundation, either version 3 of the License, or - * (at your option) any later version. - * - * This program is distributed in the hope that it will be useful, - * but WITHOUT ANY WARRANTY; without even the implied warranty of - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the - * GNU General Public License for more details. - * - * You should have received a copy of the GNU General Public License - * along with this program. If not, see . - * - * Copyright (C) 2009 - 2012 Jun Liu and Jieping Ye - */ - - -#ifndef GENERAL_ALTRA_SLEP -#define GENERAL_ALTRA_SLEP - -#include -#ifdef USE_GPL_SHOGUN - - -/* - * Important Notice: September 20, 2010 - * - * In this head file, we deal with the case that the features might not be well ordered. - * - * If the features in the tree strucutre are well ordered, i.e., the indices of the left nodes is always less - * than the right nodes, please refer to "altra.h". - * - * The advantage of "altra.h" is that, we donot need to use an explicit - * variable for recording the indices. - * - * - */ - -/* - * ------------------------------------------------------------------- - * Functions and parameter - * ------------------------------------------------------------------- - * - * general_altra solves the following problem - * - * 1/2 \|x-v\|^2 + \sum \lambda_i \|x_{G_i}\|, - * - * where x and v are of dimension n, - * \lambda_i >=0, and G_i's follow the tree structure - * - * It is implemented in Matlab as follows: - * - * x=general_altra(v, n, G, ind, nodes); - * - * G contains the indices of the groups. - * It is a row vector. Its length equals to \sum_i \|G_i\|. - * If all the entries are penalized with L1 norm, - * its length is \sum_i \|G_i\| - n. - * - * ind is a 3 x nodes matrix. - * Each column corresponds to a node. - * - * The first element of each column is the starting index, - * the second element of each column is the ending index - * the third element of each column corrreponds to \lambbda_i. - * - * - * - * The following example shows how G and ind works: - * - * G={ {1, 2}, {4, 5}, {3, 6}, {7, 8}, - * {1, 2, 3, 6}, {4, 5, 7, 8}, - * {1, 2, 3, 4, 5, 6, 7, 8} }. - * - * ind={ [1, 2, 100]', [3, 4, 100]', [5, 6, 100]', [7, 8, 100]', - * [9, 12, 100]', [13, 16, 100]', [17, 24, 100]' } - * - * where "100" denotes the weight for the nodes. - * - * - * - * ------------------------------------------------------------------- - * Notices: - * ------------------------------------------------------------------- - * - * 1. The features in the tree might not be well ordered. Otherwise, you are - * suggested to use "altra.h". - * - * 2. When each elements of x are penalized via the same L1 - * (equivalent to the L2 norm) parameter, one can simplify the input - * by specifying - * the "first" column of ind as (-1, -1, lambda) - * - * In this case, we treat it as a single "super" node. Thus in the value - * nodes, we only count it once. - * - * 3. The values in "ind" are in [1,length(G)]. - * - * 4. The third element of each column should be positive. The program does - * not check the validity of the parameter. - * - * 5. The values in G should be within [1, n]. - * - * It is still valid to use the zero regularization parameter. - * In this case, the program does not change the values of - * correponding indices. - * - * - * ------------------------------------------------------------------- -* History: -* ------------------------------------------------------------------- -* -* Composed by Jun Liu on April 20, 2010 -* -* For any question or suggestion, please email j.liu@asu.edu. -* -*/ -void general_altra(double *x, double *v, int n, double *G, double *ind, int nodes, double mult=1.0); - -/* - * altra_mt is a generalization of altra to the - * - * multi-task learning scenario (or equivalently the multi-class case) - * - * altra_mt(X, V, n, k, G, ind, nodes); - * - * It applies altra for each row (1xk) of X and V - * - */ -void general_altra_mt(double *X, double *V, int n, int k, double *G, double *ind, int nodes, double mult=1.0); - -/* - * compute - * lambda2_max=general_computeLambda2Max(x,n,G, ind,nodes); - * - * compute the 2 norm of each group, which is divided by the ind(3,:), - * then the maximum value is returned - */ -/* - *This function does not consider the case ind={[-1, -1, 100]',...} - * - *This functions is not used currently. - */ -void general_computeLambda2Max(double *lambda2_max, double *x, int n, double *G, double *ind, int nodes); - -/* - * ------------------------------------------------------------------- - * Function and parameter - * ------------------------------------------------------------------- - * - * treeNorm compute - * - * \sum \lambda_i \|x_{G_i}\|, - * - * where x is of dimension n, - * \lambda_i >=0, and G_i's follow the tree structure - * - * The file is implemented in the following in Matlab: - * - * tree_norm=general_treeNorm(x, n, G, ind,nodes); - */ -double general_treeNorm(double *x, int ldx, int n, double *G, double *ind, int nodes); - -/* - * ------------------------------------------------------------------- - * Function and parameter - * ------------------------------------------------------------------- - * - * findLambdaMax compute - * - * the lambda_{max} that achieves a zero solution for - * - * min 1/2 \|x-v\|^2 + \lambda_{\max} * \sum w_i \|x_{G_i}\|, - * - * where x is of dimension n, - * w_i >=0, and G_i's follow the tree structure - * - * The file is implemented in the following in Matlab: - * - * lambdaMax=general_findLambdaMax(v, n, G, ind,nodes); - */ -double general_findLambdaMax(double *v, int n, double *G, double *ind, int nodes); - -/* - * findLambdaMax_mt is a generalization of findLambdaMax to the - * - * multi-task learning scenario (or equivalently the multi-class case) - * - * lambdaMax=general_findLambdaMax_mt(X, V, n, k, G, ind, nodes); - * - * It applies findLambdaMax for each row (1xk) of X and V - * - */ -double general_findLambdaMax_mt(double *V, int n, int k, double *G, double *ind, int nodes); -#endif //USE_GPL_SHOGUN -#endif /* ----- #ifndef GENERAL_ALTRA_SLEP ----- */ - diff --git a/src/shogun/lib/tapkee/neighbors/covertree.hpp b/src/shogun/lib/tapkee/neighbors/covertree.hpp deleted file mode 100644 index 640a3fdb081..00000000000 --- a/src/shogun/lib/tapkee/neighbors/covertree.hpp +++ /dev/null @@ -1,873 +0,0 @@ -/* - * This program is free software; you can redistribute it and/or modify - * it under the terms of the GNU General Public License as published by - * the Free Software Foundation; either version 3 of the License, or - * (at your option) any later version. - * - * Copyright (c) 2009-2013 John Langford, Dinoj Surendran, Fernando José Iglesias García - */ - -#ifndef COVERTREE_H_ -#define COVERTREE_H_ - -#ifdef USE_GPL_SHOGUN - -/* Tapkee includes */ -#include -/* End of Tapkee includes */ - -#include -#include -#include -#include - -#ifdef _WIN32 -#undef far -#undef near -#endif -/* First written by John Langford jl@hunch.net - Templatization by Dinoj Surendran dinojs@gmail.com - Adaptation to Shogun by Fernando José Iglesias García - */ -namespace tapkee -{ -namespace tapkee_internal -{ - -/** - * Cover tree node TODO better doc - */ -template -struct node -{ - node() : p(), max_dist(0.0), parent_dist(0.0), - children(NULL), num_children(0), scale(0) - { - } - - node(P _p, ScalarType _max_dist, ScalarType _parent_dist, node

    * _children, - unsigned short int _num_children, short int _scale) : p(_p), - max_dist(_max_dist), parent_dist(_parent_dist), children(_children), - num_children(_num_children), scale(_scale) - { - } - - /** Point */ - P p; - - /** The maximum distance to any grandchild */ - ScalarType max_dist; - - /** The distance to the parent */ - ScalarType parent_dist; - - /** Pointer to the list of children of this node */ - node

    * children; - - /** The number of children nodes of this node */ - unsigned short int num_children; - - /** Essentially, an upper bound on the distance to any child */ - short int scale; -}; - -template -void free_children(const node

    & n) -{ - for (int i=0; i(n.children[i]); - n.children[i].~node

    (); - } - free(n.children); -} - - -/** - * Cover tree node with an associated set of distances TODO better doc - */ -template -struct ds_node { - - ds_node() : dist(), p() {} - - /** Vector of distances TODO better doc*/ - v_array dist; - - /** Point TODO better doc */ - P p; -}; - -static ScalarType base = COVERTREE_BASE; -static ScalarType il2 = 1. / log(base); - -inline ScalarType dist_of_scale (int s) -{ - return pow(base, s); -} - -inline int get_scale(ScalarType d) -{ - return (int)ceil(il2 * log(d)); -} - - template -node

    new_node(const P &p) -{ - node

    new_node; - new_node.p = p; - return new_node; -} - - template -node

    new_leaf(const P &p) -{ - node

    new_leaf(p,0.,0.,NULL,0,100); - return new_leaf; -} - - template -ScalarType max_set(v_array > &v) -{ - ScalarType max = 0.; - for (int i = 0; i < v.index; i++) - if ( max < v[i].dist.last()) - max = v[i].dist.last(); - return max; -} - -void print_space(int s) -{ - for (int i = 0; i < s; i++) - printf(" "); -} - -template -void print(int depth, node

    &top_node) -{ - print_space(depth); - print(top_node.p); - if ( top_node.num_children > 0 ) - { - print_space(depth); - printf("scale = %i\n",top_node.scale); - print_space(depth); - printf("max_dist = %f\n",top_node.max_dist); - print_space(depth); - printf("num children = %i\n",top_node.num_children); - for (int i = 0; i < top_node.num_children;i++) - print(depth+1, top_node.children[i]); - } -} - -template -void split(v_array >& point_set, v_array >& far_set, int max_scale) -{ - IndexType new_index = 0; - ScalarType fmax = dist_of_scale(max_scale); - for (int i = 0; i < point_set.index; i++) - { - if (point_set[i].dist.last() <= fmax) - { - point_set[new_index++] = point_set[i]; - } - else - push(far_set,point_set[i]); - } - point_set.index=new_index; -} - -template -void dist_split(DistanceCallback& dcb, v_array >& point_set, - v_array >& new_point_set, - P new_point, - int max_scale) -{ - IndexType new_index = 0; - ScalarType fmax = dist_of_scale(max_scale); - for(int i = 0; i < point_set.index; i++) - { - ScalarType new_d; - new_d = distance(dcb, new_point, point_set[i].p, fmax); - if (new_d <= fmax ) - { - push(point_set[i].dist, new_d); - push(new_point_set,point_set[i]); - } - else - point_set[new_index++] = point_set[i]; - } - point_set.index = new_index; -} - -/* - max_scale is the maximum scale of the node we might create here. - point_set contains points which are 2*max_scale or less away. - */ -template -node

    batch_insert(DistanceCallback& dcb, const P& p, - int max_scale, - int top_scale, - v_array >& point_set, - v_array >& consumed_set, - v_array > >& stack) -{ - if (point_set.index == 0) - return new_leaf(p); - else { - ScalarType max_dist = max_set(point_set); //O(|point_set|) - int next_scale = std::min(max_scale - 1, get_scale(max_dist)); - if (next_scale == -2147483647-1) // We have points with distance 0. - { - v_array > children; - push(children,new_leaf(p)); - while (point_set.index > 0) - { - push(children,new_leaf(point_set.last().p)); - push(consumed_set,point_set.last()); - point_set.decr(); - } - node

    n = new_node(p); - n.scale = 100; // A magic number meant to be larger than all scales. - n.max_dist = 0; - alloc(children,children.index); - n.num_children = children.index; - n.children = children.elements; - return n; - } - else - { - v_array > far = pop(stack); - split(point_set,far,max_scale); //O(|point_set|) - - node

    child = batch_insert(dcb, p, next_scale, top_scale, point_set, consumed_set, stack); - - if (point_set.index == 0) - { - push(stack,point_set); - point_set=far; - return child; - } - else { - node

    n = new_node(p); - v_array > children; - push(children, child); - v_array > new_point_set = pop(stack); - v_array > new_consumed_set = pop(stack); - while (point_set.index != 0) { //O(|point_set| * num_children) - P new_point = point_set.last().p; - ScalarType new_dist = point_set.last().dist.last(); - push(consumed_set, point_set.last()); - point_set.decr(); - - dist_split(dcb,point_set,new_point_set,new_point,max_scale); //O(|point_saet|) - dist_split(dcb,far,new_point_set,new_point,max_scale); //O(|far|) - - node

    new_child = - batch_insert(dcb, new_point, next_scale, top_scale, new_point_set, new_consumed_set, stack); - new_child.parent_dist = new_dist; - - push(children, new_child); - - ScalarType fmax = dist_of_scale(max_scale); - for(int i = 0; i< new_point_set.index; i++) //O(|new_point_set|) - { - new_point_set[i].dist.decr(); - if (new_point_set[i].dist.last() <= fmax) - push(point_set, new_point_set[i]); - else - push(far, new_point_set[i]); - } - for(int i = 0; i< new_consumed_set.index; i++) //O(|new_point_set|) - { - new_consumed_set[i].dist.decr(); - push(consumed_set, new_consumed_set[i]); - } - new_point_set.index = 0; - new_consumed_set.index = 0; - } - push(stack,new_point_set); - push(stack,new_consumed_set); - push(stack,point_set); - point_set=far; - n.scale = top_scale - max_scale; - n.max_dist = max_set(consumed_set); - alloc(children,children.index); - n.num_children = children.index; - n.children = children.elements; - return n; - } - } - } -} - -template -node

    batch_create(DistanceCallback& dcb, v_array

    points) -{ - assert(points.index > 0); - v_array > point_set; - v_array > > stack; - - for (int i = 1; i < points.index; i++) { - ds_node

    temp; - push(temp.dist, distance(dcb, points[0], points[i], std::numeric_limits::max())); - temp.p = points[i]; - push(point_set,temp); - } - - v_array > consumed_set; - - ScalarType max_dist = max_set(point_set); - - node

    top = batch_insert (dcb, points[0], - get_scale(max_dist), - get_scale(max_dist), - point_set, - consumed_set, - stack); - for (int i = 0; i &heights) -{ - if (heights.index <= d) - for(;heights.index <= d;) - push(heights,0); - heights[d] = heights[d] + 1; -} - -template -int height_dist(const node

    top_node,v_array &heights) -{ - if (top_node.num_children == 0) - { - add_height(0,heights); - return 0; - } - else - { - int max_v=0; - for (int i = 0; i max_v) - max_v = d; - } - add_height(1 + max_v, heights); - return (1 + max_v); - } -} - -template -void depth_dist(int top_scale, const node

    top_node,v_array &depths) -{ - if (top_node.num_children > 0) - for (int i = 0; i -void breadth_dist(const node

    top_node,v_array &breadths) -{ - if (top_node.num_children == 0) - add_height(0,breadths); - else - { - for (int i = 0; i -struct d_node -{ - /** Distance TODO better doc*/ - ScalarType dist; - - /** List of nodes TODO better doc*/ - const node

    *n; -}; - -template -inline ScalarType compare(const d_node

    *p1, const d_node

    * p2) -{ - return p1 -> dist - p2 -> dist; -} - -template -void halfsort (v_array > cover_set) -{ - if (cover_set.index <= 1) - return; - register d_node

    *base_ptr = cover_set.elements; - - d_node

    *hi = &base_ptr[cover_set.index - 1]; - d_node

    *right_ptr = hi; - d_node

    *left_ptr; - - while (right_ptr > base_ptr) - { - d_node

    *mid = base_ptr + ((hi - base_ptr) >> 1); - - if (compare ( mid, base_ptr) < 0.) - std::swap(*mid, *base_ptr); - if (compare ( hi, mid) < 0.) - std::swap(*mid, *hi); - else - goto jump_over; - if (compare ( mid, base_ptr) < 0.) - std::swap(*mid, *base_ptr); -jump_over:; - - left_ptr = base_ptr + 1; - right_ptr = hi - 1; - - do - { - while (compare (left_ptr, mid) < 0.) - left_ptr++; - - while (compare (mid, right_ptr) < 0.) - right_ptr--; - - if (left_ptr < right_ptr) - { - std::swap(*left_ptr, *right_ptr); - if (mid == left_ptr) - mid = right_ptr; - else if (mid == right_ptr) - mid = left_ptr; - left_ptr++; - right_ptr--; - } - else if (left_ptr == right_ptr) - { - left_ptr ++; - right_ptr --; - break; - } - } - while (left_ptr <= right_ptr); - hi = right_ptr; - } -} - -template -v_array > > get_cover_sets(v_array > > > &spare_cover_sets) -{ - v_array > > ret = pop(spare_cover_sets); - while (ret.index < 101) - { - v_array > temp; - push(ret, temp); - } - return ret; -} - -inline bool shell(ScalarType parent_query_dist, ScalarType child_parent_dist, ScalarType upper_bound) -{ - return parent_query_dist - child_parent_dist <= upper_bound; - // && child_parent_dist - parent_query_dist <= upper_bound; -} - -int internal_k =1; -void update_k(ScalarType *k_upper_bound, ScalarType upper_bound) -{ - ScalarType *end = k_upper_bound + internal_k-1; - ScalarType *begin = k_upper_bound; - for (;end != begin; begin++) - { - if (upper_bound < *(begin+1)) - *begin = *(begin+1); - else { - *begin = upper_bound; - break; - } - } - if (end == begin) - *begin = upper_bound; -} -ScalarType *alloc_k() -{ - return (ScalarType*)malloc(sizeof(ScalarType) * internal_k); -} -void set_k(ScalarType* begin, ScalarType max) -{ - for(ScalarType *end = begin+internal_k;end != begin; begin++) - *begin = max; -} - -ScalarType internal_epsilon =0.; -//void update_epsilon(ScalarType *upper_bound, ScalarType new_dist) {} -ScalarType *alloc_epsilon() -{ - return (ScalarType *)malloc(sizeof(ScalarType)); -} -void set_epsilon(ScalarType* begin) -{ - *begin = internal_epsilon; -} - -void update_unequal(ScalarType *upper_bound, ScalarType new_dist) -{ - if (new_dist != 0.) - *upper_bound = new_dist; -} -ScalarType* (*alloc_unequal)() = alloc_epsilon; -void set_unequal(ScalarType* begin, ScalarType max) -{ - *begin = max; -} - -void (*update)(ScalarType *foo, ScalarType bar) = update_k; -void (*setter)(ScalarType *foo, ScalarType bar) = set_k; -ScalarType* (*alloc_upper)() = alloc_k; - -template -inline void copy_zero_set(DistanceCallback& dcb, node

    * query_chi, - ScalarType* new_upper_bound, v_array > &zero_set, - v_array > &new_zero_set) -{ - new_zero_set.index = 0; - d_node

    *end = zero_set.elements + zero_set.index; - for (d_node

    *ele = zero_set.elements; ele != end ; ele++) - { - ScalarType upper_dist = *new_upper_bound + query_chi->max_dist; - if (shell(ele->dist, query_chi->parent_dist, upper_dist)) - { - ScalarType d = distance(dcb, query_chi->p, ele->n->p, upper_dist); - - if (d <= upper_dist) - { - if (d < *new_upper_bound) - update(new_upper_bound, d); - d_node

    temp = {d, ele->n}; - push(new_zero_set,temp); - } - } - } -} - -template -inline void copy_cover_sets(DistanceCallback& dcb, node

    * query_chi, - ScalarType* new_upper_bound, - v_array > > &cover_sets, - v_array > > &new_cover_sets, - int current_scale, int max_scale) -{ - for (; current_scale <= max_scale; current_scale++) - { - d_node

    * ele = cover_sets[current_scale].elements; - d_node

    * end = cover_sets[current_scale].elements + cover_sets[current_scale].index; - for (; ele != end; ele++) - { - ScalarType upper_dist = *new_upper_bound + query_chi->max_dist + ele->n->max_dist; - if (shell(ele->dist, query_chi->parent_dist, upper_dist)) - { - ScalarType d = distance(dcb, query_chi->p, ele->n->p, upper_dist); - - if (d <= upper_dist) - { - if (d < *new_upper_bound) - update(new_upper_bound,d); - d_node

    temp = {d, ele->n}; - push(new_cover_sets[current_scale],temp); - } - } - } - } -} - -template -void print_query(const node

    *top_node) -{ - printf("query = \n"); - print(top_node->p); - if ( top_node->num_children > 0 ) { - printf("scale = %i\n",top_node->scale); - printf("max_dist = %f\n",top_node->max_dist); - printf("num children = %i\n",top_node->num_children); - } -} - -template -void print_cover_sets(v_array > > &cover_sets, - v_array > &zero_set, - int current_scale, int max_scale) -{ - printf("cover set = \n"); - for (; current_scale <= max_scale; current_scale++) - { - d_node

    *ele = cover_sets[current_scale].elements; - d_node

    *end = cover_sets[current_scale].elements + cover_sets[current_scale].index; - printf("%i\n", current_scale); - for (; ele != end; ele++) - { - node

    *n = (node

    *)ele->n; - print(n->p); - } - } - d_node

    *end = zero_set.elements + zero_set.index; - printf("infinity\n"); - for (d_node

    *ele = zero_set.elements; ele != end ; ele++) - { - node

    *n = (node

    *)ele->n; - print(n->p); - } -} - -/* - An optimization to consider: - Make all distance evaluations occur in descend. - - Instead of passing a cover_set, pass a stack of cover sets. The - last element holds d_nodes with your distance. The next lower - element holds a d_node with the distance to your query parent, - next = query grand parent, etc.. - - Compute distances in the presence of the tighter upper bound. - */ -template -inline -void descend(DistanceCallback& dcb, const node

    * query, ScalarType* upper_bound, - int current_scale,int &max_scale, v_array > > &cover_sets, - v_array > &zero_set) -{ - d_node

    *end = cover_sets[current_scale].elements + cover_sets[current_scale].index; - for (d_node

    *parent = cover_sets[current_scale].elements; parent != end; parent++) - { - const node

    *par = parent->n; - ScalarType upper_dist = *upper_bound + query->max_dist + query->max_dist; - if (parent->dist <= upper_dist + par->max_dist) - { - node

    *chi = par->children; - if (parent->dist <= upper_dist + chi->max_dist) - { - if (chi->num_children > 0) - { - if (max_scale < chi->scale) - max_scale = chi->scale; - d_node

    temp = {parent->dist, chi}; - push(cover_sets[chi->scale], temp); - } - else if (parent->dist <= upper_dist) - { - d_node

    temp = {parent->dist, chi}; - push(zero_set, temp); - } - } - node

    *child_end = par->children + par->num_children; - for (chi++; chi != child_end; chi++) - { - ScalarType upper_chi = *upper_bound + chi->max_dist + query->max_dist + query->max_dist; - if (shell(parent->dist, chi->parent_dist, upper_chi)) - { - ScalarType d = distance(dcb, query->p, chi->p, upper_chi); - if (d <= upper_chi) - { - if (d < *upper_bound) - update(upper_bound, d); - if (chi->num_children > 0) - { - if (max_scale < chi->scale) - max_scale = chi->scale; - d_node

    temp = {d, chi}; - push(cover_sets[chi->scale],temp); - } - else - if (d <= upper_chi - chi->max_dist) - { - d_node

    temp = {d, chi}; - push(zero_set, temp); - } - } - } - } - } - } -} - -template -void brute_nearest(DistanceCallback& dcb, const node

    * query, - v_array > zero_set, ScalarType* upper_bound, - v_array > &results, - v_array > > &spare_zero_sets) -{ - if (query->num_children > 0) - { - v_array > new_zero_set = pop(spare_zero_sets); - node

    * query_chi = query->children; - brute_nearest(dcb, query_chi, zero_set, upper_bound, results, spare_zero_sets); - ScalarType* new_upper_bound = alloc_upper(); - - node

    *child_end = query->children + query->num_children; - for (query_chi++;query_chi != child_end; query_chi++) - { - setter(new_upper_bound,*upper_bound + query_chi->parent_dist); - copy_zero_set(dcb, query_chi, new_upper_bound, zero_set, new_zero_set); - brute_nearest(dcb, query_chi, new_zero_set, new_upper_bound, results, spare_zero_sets); - } - free (new_upper_bound); - new_zero_set.index = 0; - push(spare_zero_sets, new_zero_set); - } - else - { - v_array

    temp; - push(temp, query->p); - d_node

    *end = zero_set.elements + zero_set.index; - for (d_node

    *ele = zero_set.elements; ele != end ; ele++) - if (ele->dist <= *upper_bound) - push(temp, ele->n->p); - push(results,temp); - } -} - -template -void internal_batch_nearest_neighbor(DistanceCallback& dcb, const node

    *query, - v_array > > &cover_sets, - v_array > &zero_set, - int current_scale, - int max_scale, - ScalarType* upper_bound, - v_array > &results, - v_array > > > &spare_cover_sets, - v_array > > &spare_zero_sets) -{ - if (current_scale > max_scale) // All remaining points are in the zero set. - brute_nearest(dcb, query, zero_set, upper_bound, results, spare_zero_sets); - else - if (query->scale <= current_scale && query->scale != 100) - // Our query has too much scale. Reduce. - { - node

    *query_chi = query->children; - v_array > new_zero_set = pop(spare_zero_sets); - v_array > > new_cover_sets = get_cover_sets(spare_cover_sets); - ScalarType* new_upper_bound = alloc_upper(); - - node

    *child_end = query->children + query->num_children; - for (query_chi++; query_chi != child_end; query_chi++) - { - setter(new_upper_bound,*upper_bound + query_chi->parent_dist); - copy_zero_set(dcb, query_chi, new_upper_bound, zero_set, new_zero_set); - copy_cover_sets(dcb, query_chi, new_upper_bound, cover_sets, new_cover_sets, - current_scale, max_scale); - internal_batch_nearest_neighbor(dcb, query_chi, new_cover_sets, new_zero_set, - current_scale, max_scale, new_upper_bound, - results, spare_cover_sets, spare_zero_sets); - } - free (new_upper_bound); - new_zero_set.index = 0; - push(spare_zero_sets, new_zero_set); - push(spare_cover_sets, new_cover_sets); - internal_batch_nearest_neighbor(dcb, query->children, cover_sets, zero_set, - current_scale, max_scale, upper_bound, results, - spare_cover_sets, spare_zero_sets); - } - else // reduce cover set scale - { - halfsort(cover_sets[current_scale]); - descend(dcb, query, upper_bound, current_scale, max_scale,cover_sets, zero_set); - cover_sets[current_scale++].index = 0; - internal_batch_nearest_neighbor(dcb, query, cover_sets, zero_set, - current_scale, max_scale, upper_bound, results, - spare_cover_sets, spare_zero_sets); - } -} - -template -void batch_nearest_neighbor(DistanceCallback &dcb, const node

    &top_node, - const node

    &query, v_array > &results) -{ - v_array > > > spare_cover_sets; - v_array > > spare_zero_sets; - - v_array > > cover_sets = get_cover_sets(spare_cover_sets); - v_array > zero_set = pop(spare_zero_sets); - - ScalarType* upper_bound = alloc_upper(); - setter(upper_bound, std::numeric_limits::max()); - - ScalarType top_dist = distance(dcb, query.p, top_node.p, std::numeric_limits::max()); - update(upper_bound, top_dist); - - d_node

    temp = {top_dist, &top_node}; - push(cover_sets[0], temp); - - internal_batch_nearest_neighbor(dcb, &query,cover_sets,zero_set,0,0,upper_bound,results, - spare_cover_sets,spare_zero_sets); - - free(upper_bound); - push(spare_cover_sets, cover_sets); - - for (int i = 0; i < spare_cover_sets.index; i++) - { - v_array > > cover_sets2 = spare_cover_sets[i]; - for (int j = 0; j < cover_sets2.index; j++) - free (cover_sets2[j].elements); - free(cover_sets2.elements); - } - free(spare_cover_sets.elements); - - push(spare_zero_sets, zero_set); - - for (int i = 0; i < spare_zero_sets.index; i++) - free(spare_zero_sets[i].elements); - free(spare_zero_sets.elements); -} - -template -void k_nearest_neighbor(DistanceCallback &dcb, const node

    &top_node, - const node

    &query, v_array > &results, int k) -{ - internal_k = k; - update = update_k; - setter = set_k; - alloc_upper = alloc_k; - - batch_nearest_neighbor(dcb, top_node, query, results); -} -/* -template -void epsilon_nearest_neighbor(DistanceCallback &dcb, const node

    &top_node, - const node

    &query, v_array > &results, - ScalarType epsilon) -{ - internal_epsilon = epsilon; - // update = update_epsilon; - setter = set_epsilon; - alloc_upper = alloc_epsilon; - - batch_nearest_neighbor(dcb, top_node, query, results); -} - -template -void unequal_nearest_neighbor(DistanceCallback &dcb, const node

    &top_node, - const node

    &query, v_array > &results) -{ - update = update_unequal; - setter = set_unequal; - alloc_upper = alloc_unequal; - - batch_nearest_neighbor(dcb, top_node, query, results); -} -*/ - -} -} -#endif //USE_GPL_SHOGUN -#endif diff --git a/src/shogun/lib/tapkee/neighbors/covertree_point.hpp b/src/shogun/lib/tapkee/neighbors/covertree_point.hpp deleted file mode 100644 index a346b9a76d1..00000000000 --- a/src/shogun/lib/tapkee/neighbors/covertree_point.hpp +++ /dev/null @@ -1,177 +0,0 @@ -/* * This program is free software; you can redistribute it and/or modify - * it under the terms of the GNU General Public License as published by - * the Free Software Foundation; either version 3 of the License, or - * (at your option) any later version. - * - * Written (W) 2012 Fernando José Iglesias García - * Written (W) John Langford and Dinoj Surendran, v_array and its templatization - * Copyright (C) 2012 Fernando José Iglesias García - */ - -#ifndef _JL_COVERTREE_POINT_H_ -#define _JL_COVERTREE_POINT_H_ - -#ifdef USE_GPL_SHOGUN - -/* Tapkee includes */ -#include -/* End of Tapkee includes */ - -#include -#include - -namespace tapkee -{ -namespace tapkee_internal -{ - -/** @brief Class v_array taken directly from JL's implementation */ -template -class v_array{ - - public: - /** Getter for the the last element of the v_array - * @return the last element of the array */ - T last() { return elements[index-1];} - - /** Decrement the pointer to the last element */ - void decr() { index--;} - - /** Create an empty v_array */ - v_array() : index(0), length(0), elements(NULL) {} - - /** Element access operator - * @param i of the element to be read - * @return the corresponding element */ - T& operator[](IndexType i) { return elements[i]; } - - public: - /** Pointer to the last element of the v_array */ - int index; - - /** Length of the v_array */ - int length; - - /** Pointer to the beginning of the v_array elements */ - T* elements; -}; - -/** - * Insert a new element at the end of the vector - * - * @param v vector - * @param new_ele element to insert - */ -template -void push(v_array& v, const T &new_ele) -{ - while(v.index >= v.length) - { - v.length = 2*v.length + 3; - v.elements = (T *)realloc(v.elements,sizeof(T) * v.length); - } - v[v.index++] = new_ele; -} - -/** - * Used to modify the capacity of the vector - * - * @param v vector - * @param length the new length of the vector - */ -template -void alloc(v_array& v, int length) -{ - v.elements = (T *)realloc(v.elements, sizeof(T) * length); - v.length = length; -} - -/** - * Returns the vector previous to the pointed one in the stack of - * vectors and decrements the index of the stack. No memory is - * freed here. If there are no vectors stored in the stack, create - * and return a new empty vector - * - * @param stack of vectors - * @return the adequate vector according to the previous conditions - */ -template -v_array pop(v_array > &stack) -{ - if (stack.index > 0) - return stack[--stack.index]; - else - return v_array(); -} - -/** @brief Class Point to use with John Langford's CoverTree. This - * class must have some associated functions defined (distance, - * and print, see below) so it can be used with the CoverTree - * implementation. - */ -template -struct CoverTreePoint -{ - CoverTreePoint() : iter_(), norm_(0.0) - { - }; - CoverTreePoint(const RandomAccessIterator& iter, ScalarType norm) : - iter_(iter), norm_(norm) - { - }; - - RandomAccessIterator iter_; - ScalarType norm_; -}; /* struct JLCoverTreePoint */ - -template -struct distance_impl; - -/** Functions declared out of the class definition to respect CoverTree - * structure */ -template -inline ScalarType distance(Callback& cb, const CoverTreePoint& l, - const CoverTreePoint& r, ScalarType upper_bound) -{ - //assert(upper_bound>=0); - - if (l.iter_==r.iter_) - return 0.0; - - return distance_impl()(cb,l,r,upper_bound); -} - -struct KernelType; - -template -struct distance_impl -{ - inline ScalarType operator()(Callback& cb, const CoverTreePoint& l, - const CoverTreePoint& r, ScalarType /*upper_bound*/) - { - return std::sqrt(l.norm_ + r.norm_ - 2*cb(r.iter_,l.iter_)); - } -}; - -struct DistanceType; - -template -struct distance_impl -{ - inline ScalarType operator()(Callback& cb, const CoverTreePoint& l, - const CoverTreePoint& r, ScalarType /*upper_bound*/) - { - return cb(l.iter_,r.iter_); - } -}; - -/** Print the information of the CoverTree point */ -template -void print(const CoverTreePoint&) -{ -} - -} -} -#endif //USE_GPL_SHOGUN -#endif /* _JL_COVERTREE_POINT_H_*/ diff --git a/src/shogun/lib/tfhistogram/histogram.cpp b/src/shogun/lib/tfhistogram/histogram.cpp new file mode 100644 index 00000000000..1b43b77e242 --- /dev/null +++ b/src/shogun/lib/tfhistogram/histogram.cpp @@ -0,0 +1,230 @@ +/* Copyright 2015 The TensorFlow Authors. All Rights Reserved. + +Licensed under the Apache License, Version 2.0 (the "License"); +you may not use this file except in compliance with the License. +You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + +Unless required by applicable law or agreed to in writing, software +distributed under the License is distributed on an "AS IS" BASIS, +WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +See the License for the specific language governing permissions and +limitations under the License. +==============================================================================*/ + +// clang-format off +#include +#ifdef HAVE_TFLOGGER + +#include "histogram.h" +#include +#include +#include +#include +#include + +namespace tensorflow { +namespace histogram { + +static std::vector* InitDefaultBucketsInner() { + std::vector buckets; + std::vector neg_buckets; + // Make buckets whose range grows by 10% starting at 1.0e-12 up to 1.0e20 + double v = 1.0e-12; + while (v < 1.0e20) { + buckets.push_back(v); + neg_buckets.push_back(-v); + v *= 1.1; + } + buckets.push_back(DBL_MAX); + neg_buckets.push_back(-DBL_MAX); + std::reverse(neg_buckets.begin(), neg_buckets.end()); + std::vector* result = new std::vector; + result->insert(result->end(), neg_buckets.begin(), neg_buckets.end()); + result->push_back(0.0); + result->insert(result->end(), buckets.begin(), buckets.end()); + return result; +} + +static std::vector InitDefaultBuckets() { + static std::vector* default_bucket_limits = InitDefaultBucketsInner(); + return *default_bucket_limits; +} + +Histogram::Histogram() : bucket_limits_(InitDefaultBuckets()) { Clear(); } + +// Create a histogram with a custom set of bucket limits, +// specified in "custom_buckets[0..custom_buckets.size()-1]" +Histogram::Histogram(std::vector custom_bucket_limits) + : custom_bucket_limits_(custom_bucket_limits.begin(), + custom_bucket_limits.end()), + bucket_limits_(custom_bucket_limits_) { + Clear(); +} + +bool Histogram::DecodeFromProto(const HistogramProto& proto) { + if ((proto.bucket_size() != proto.bucket_limit_size()) || + (proto.bucket_size() == 0)) { + return false; + } + min_ = proto.min(); + max_ = proto.max(); + num_ = proto.num(); + sum_ = proto.sum(); + sum_squares_ = proto.sum_squares(); + custom_bucket_limits_.clear(); + custom_bucket_limits_.insert(custom_bucket_limits_.end(), + proto.bucket_limit().begin(), + proto.bucket_limit().end()); + bucket_limits_ = custom_bucket_limits_; + buckets_.clear(); + buckets_.insert(buckets_.end(), proto.bucket().begin(), proto.bucket().end()); + return true; +} + +void Histogram::Clear() { + min_ = bucket_limits_[bucket_limits_.size() - 1]; + max_ = -DBL_MAX; + num_ = 0; + sum_ = 0; + sum_squares_ = 0; + buckets_.resize(bucket_limits_.size()); + for (size_t i = 0; i < bucket_limits_.size(); i++) { + buckets_[i] = 0; + } +} + +void Histogram::Add(double value) { + int b = + std::upper_bound(bucket_limits_.begin(), bucket_limits_.end(), value) - + bucket_limits_.begin(); + + buckets_[b] += 1.0; + if (min_ > value) min_ = value; + if (max_ < value) max_ = value; + num_++; + sum_ += value; + sum_squares_ += (value * value); +} + +double Histogram::Median() const { return Percentile(50.0); } + +// Linearly map the variable x from [x0, x1] unto [y0, y1] +double Histogram::Remap(double x, double x0, double x1, double y0, + double y1) const { + return y0 + (x - x0) / (x1 - x0) * (y1 - y0); +} + +// Pick tight left-hand-side and right-hand-side bounds and then +// interpolate a histogram value at percentile p +double Histogram::Percentile(double p) const { + if (num_ == 0.0) return 0.0; + + double threshold = num_ * (p / 100.0); + double cumsum_prev = 0; + for (size_t i = 0; i < buckets_.size(); i++) { + double cumsum = cumsum_prev + buckets_[i]; + + // Find the first bucket whose cumsum >= threshold + if (cumsum >= threshold) { + // Prevent divide by 0 in remap which happens if cumsum == cumsum_prev + // This should only get hit when p == 0, cumsum == 0, and cumsum_prev == 0 + if (cumsum == cumsum_prev) { + continue; + } + + // Calculate the lower bound of interpolation + double lhs = (i == 0 || cumsum_prev == 0) ? min_ : bucket_limits_[i - 1]; + lhs = std::max(lhs, min_); + + // Calculate the upper bound of interpolation + double rhs = bucket_limits_[i]; + rhs = std::min(rhs, max_); + + double weight = Remap(threshold, cumsum_prev, cumsum, lhs, rhs); + return weight; + } + + cumsum_prev = cumsum; + } + return max_; +} + +double Histogram::Average() const { + if (num_ == 0.0) return 0; + return sum_ / num_; +} + +double Histogram::StandardDeviation() const { + if (num_ == 0.0) return 0; + double variance = (sum_squares_ * num_ - sum_ * sum_) / (num_ * num_); + return sqrt(variance); +} + +std::string Histogram::ToString() const { + std::string r; + char buf[200]; + snprintf(buf, sizeof(buf), "Count: %.0f Average: %.4f StdDev: %.2f\n", num_, + Average(), StandardDeviation()); + r.append(buf); + snprintf(buf, sizeof(buf), "Min: %.4f Median: %.4f Max: %.4f\n", + (num_ == 0.0 ? 0.0 : min_), Median(), max_); + r.append(buf); + r.append("------------------------------------------------------\n"); + const double mult = num_ > 0 ? 100.0 / num_ : 0.0; + double sum = 0; + for (size_t b = 0; b < buckets_.size(); b++) { + if (buckets_[b] <= 0.0) continue; + sum += buckets_[b]; + snprintf(buf, sizeof(buf), "[ %10.2g, %10.2g ) %7.0f %7.3f%% %7.3f%% ", + ((b == 0) ? -DBL_MAX : bucket_limits_[b - 1]), // left + bucket_limits_[b], // right + buckets_[b], // count + mult * buckets_[b], // percentage + mult * sum); // cum percentage + r.append(buf); + + // Add hash marks based on percentage; 20 marks for 100%. + int marks = static_cast(20 * (buckets_[b] / num_) + 0.5); + r.append(marks, '#'); + r.push_back('\n'); + } + return r; +} + +void Histogram::EncodeToProto(HistogramProto* proto, + bool preserve_zero_buckets) const { + proto->Clear(); + proto->set_min(min_); + proto->set_max(max_); + proto->set_num(num_); + proto->set_sum(sum_); + proto->set_sum_squares(sum_squares_); + for (size_t i = 0; i < buckets_.size();) { + double end = bucket_limits_[i]; + double count = buckets_[i]; + i++; + if (!preserve_zero_buckets && count <= 0.0) { + // Find run of empty buckets and collapse them into one + while (i < buckets_.size() && buckets_[i] <= 0.0) { + end = bucket_limits_[i]; + count = buckets_[i]; + i++; + } + } + proto->add_bucket_limit(end); + proto->add_bucket(count); + } + if (proto->bucket_size() == 0.0) { + // It's easier when we restore if we always have at least one bucket entry + proto->add_bucket_limit(DBL_MAX); + proto->add_bucket(0.0); + } +} + +} // namespace histogram +} // namespace tensorflow + +#endif //HAVE_TFLOGGER +// clang-format on diff --git a/src/shogun/lib/tfhistogram/histogram.h b/src/shogun/lib/tfhistogram/histogram.h new file mode 100644 index 00000000000..7fddd2e7eeb --- /dev/null +++ b/src/shogun/lib/tfhistogram/histogram.h @@ -0,0 +1,107 @@ +/* Copyright 2015 The TensorFlow Authors. All Rights Reserved. + +Licensed under the Apache License, Version 2.0 (the "License"); +you may not use this file except in compliance with the License. +You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + +Unless required by applicable law or agreed to in writing, software +distributed under the License is distributed on an "AS IS" BASIS, +WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +See the License for the specific language governing permissions and +limitations under the License. +==============================================================================*/ +// clang-format off +#include +#ifdef HAVE_TFLOGGER + +#ifndef TENSORFLOW_LIB_HISTOGRAM_HISTOGRAM_H_ +#define TENSORFLOW_LIB_HISTOGRAM_HISTOGRAM_H_ + +#include +#include + +namespace tensorflow { + +class HistogramProto; + +namespace histogram { + +class Histogram { + public: + // Create a histogram with a default set of bucket boundaries. + // Buckets near zero cover very small ranges (e.g. 10^-12), and each + // bucket range grows by ~10% as we head away from zero. The + // buckets cover the range from -DBL_MAX to DBL_MAX. + Histogram(); + + // Create a histogram with a custom set of bucket boundaries, + // specified in "custom_bucket_limits[0..custom_bucket_limits.size()-1]" + // REQUIRES: custom_bucket_limits[i] values are monotonically increasing. + // REQUIRES: custom_bucket_limits is not empty() + explicit Histogram(std::vector custom_bucket_limits); + + // Restore the state of a histogram that was previously encoded + // via Histogram::EncodeToProto. Note that only the bucket boundaries + // generated by EncodeToProto will be restored. + bool DecodeFromProto(const HistogramProto& proto); + + ~Histogram() {} + + void Clear(); + void Add(double value); + + // Save the current state of the histogram to "*proto". If + // "preserve_zero_buckets" is false, only non-zero bucket values and + // ranges are saved, and the bucket boundaries of zero-valued buckets + // are lost. + void EncodeToProto(HistogramProto* proto, bool preserve_zero_buckets) const; + + // Return the median of the values in the histogram + double Median() const; + + // Return the "p"th percentile [0.0..100.0] of the values in the + // distribution + double Percentile(double p) const; + + // Return the average value of the distribution + double Average() const; + + // Return the standard deviation of values in the distribution + double StandardDeviation() const; + + // Returns a multi-line human-readable string representing the histogram + // contents. Example output: + // Count: 4 Average: 251.7475 StdDev: 432.02 + // Min: -3.0000 Median: 5.0000 Max: 1000.0000 + // ------------------------------------------------------ + // [ -5, 0 ) 1 25.000% 25.000% ##### + // [ 0, 5 ) 1 25.000% 50.000% ##### + // [ 5, 10 ) 1 25.000% 75.000% ##### + // [ 1000, 10000 ) 1 25.000% 100.000% ##### + std::string ToString() const; + + private: + double min_; + double max_; + double num_; + double sum_; + double sum_squares_; + + std::vector custom_bucket_limits_; + std::vector bucket_limits_; + std::vector buckets_; + + double Remap(double x, double x0, double x1, double y0, double y1) const; + + Histogram(const Histogram&) = delete; + void operator=(const Histogram&) = delete; +}; + +} // namespace histogram +} // namespace tensorflow + +#endif // TENSORFLOW_LIB_HISTOGRAM_HISTOGRAM_H_ +#endif // HAVE_TFLOGGER +// clang-format on diff --git a/src/shogun/machine/BaggingMachine.cpp b/src/shogun/machine/BaggingMachine.cpp index 4a9802680bc..0f81fbbf8e9 100644 --- a/src/shogun/machine/BaggingMachine.cpp +++ b/src/shogun/machine/BaggingMachine.cpp @@ -8,8 +8,11 @@ * Copyright (C) 2013 Viktor Gal */ -#include #include +#include +#include +#include + #include using namespace shogun; @@ -44,39 +47,86 @@ CBaggingMachine::~CBaggingMachine() CBinaryLabels* CBaggingMachine::apply_binary(CFeatures* data) { - SGVector combined_vector = apply_get_outputs(data); + SGMatrix output = apply_outputs_without_combination(data); + + CMeanRule* mean_rule = new CMeanRule(); + + SGVector labels = m_combination_rule->combine(output); + SGVector probabilities = mean_rule->combine(output); + + float64_t threshold = 0.5; + CBinaryLabels* pred = new CBinaryLabels(probabilities, threshold); + + SG_UNREF(mean_rule); - CBinaryLabels* pred = new CBinaryLabels(combined_vector); return pred; } CMulticlassLabels* CBaggingMachine::apply_multiclass(CFeatures* data) { - SGVector combined_vector = apply_get_outputs(data); + SGMatrix bagged_outputs = + apply_outputs_without_combination(data); + + REQUIRE(m_labels, "Labels not set.\n"); + REQUIRE( + m_labels->get_label_type() == LT_MULTICLASS, + "Labels (%s) are not compatible with multiclass.\n", + m_labels->get_name()); + + auto labels_multiclass = dynamic_cast(m_labels); + auto num_samples = bagged_outputs.size() / m_num_bags; + auto num_classes = labels_multiclass->get_num_classes(); + + CMulticlassLabels* pred = new CMulticlassLabels(num_samples); + pred->allocate_confidences_for(num_classes); + + SGMatrix class_probabilities(num_classes, num_samples); + class_probabilities.zero(); + + for (auto i = 0; i < num_samples; ++i) + { + for (auto j = 0; j < m_num_bags; ++j) + { + int32_t class_idx = bagged_outputs(i, j); + class_probabilities(class_idx, i) += 1; + } + } + + class_probabilities = linalg::scale(class_probabilities, 1.0 / m_num_bags); + + for (auto i = 0; i < num_samples; ++i) + pred->set_multiclass_confidences(i, class_probabilities.get_column(i)); + + SGVector combined = m_combination_rule->combine(bagged_outputs); + pred->set_labels(combined); - CMulticlassLabels* pred = new CMulticlassLabels(combined_vector); return pred; } CRegressionLabels* CBaggingMachine::apply_regression(CFeatures* data) { - SGVector combined_vector = apply_get_outputs(data); - - CRegressionLabels* pred = new CRegressionLabels(combined_vector); - - return pred; + return new CRegressionLabels(apply_get_outputs(data)); } SGVector CBaggingMachine::apply_get_outputs(CFeatures* data) { ASSERT(data != NULL); REQUIRE(m_combination_rule != NULL, "Combination rule is not set!"); + + SGMatrix output = apply_outputs_without_combination(data); + SGVector combined = m_combination_rule->combine(output); + + return combined; +} + +SGMatrix +CBaggingMachine::apply_outputs_without_combination(CFeatures* data) +{ ASSERT(m_num_bags == m_bags->get_num_elements()); SGMatrix output(data->get_num_vectors(), m_num_bags); output.zero(); - #pragma omp parallel for for (int32_t i = 0; i < m_num_bags; ++i) { @@ -95,9 +145,7 @@ SGVector CBaggingMachine::apply_get_outputs(CFeatures* data) SG_UNREF(m); } - SGVector combined = m_combination_rule->combine(output); - - return combined; + return output; } bool CBaggingMachine::train_machine(CFeatures* data) @@ -318,7 +366,7 @@ float64_t CBaggingMachine::get_oob_error(CEvaluation* eval) const SG_UNREF(l); } - DynArray idx; + std::vector idx; for (index_t i = 0; i < m_features->get_num_vectors(); i++) { if (m_all_oob_idx[i]) @@ -326,9 +374,9 @@ float64_t CBaggingMachine::get_oob_error(CEvaluation* eval) const } SGVector combined = m_combination_rule->combine(output); - SGVector lab(idx.get_num_elements()); + SGVector lab(idx.size()); for (int32_t i=0;iget_label_type()) @@ -349,7 +397,7 @@ float64_t CBaggingMachine::get_oob_error(CEvaluation* eval) const SG_ERROR("Unsupported label type\n"); } - m_labels->add_subset(SGVector(idx.get_array(), idx.get_num_elements(), false)); + m_labels->add_subset(SGVector(idx.data(), idx.size(), false)); float64_t res = eval->evaluate(predicted, m_labels); m_labels->remove_subset(); @@ -363,15 +411,8 @@ CDynamicArray* CBaggingMachine::get_oob_indices(const SGVector out_of_bag.set_const(true); // mark the ones that are in_bag - index_t oob_count = m_features->get_num_vectors(); for (index_t i = 0; i < in_bag.vlen; i++) - { - if (out_of_bag[in_bag[i]]) - { - out_of_bag[in_bag[i]] = false; - oob_count--; - } - } + out_of_bag[in_bag[i]] &= false; CDynamicArray* oob = new CDynamicArray(); // store the indicies of vectors that are out of the bag diff --git a/src/shogun/machine/BaggingMachine.h b/src/shogun/machine/BaggingMachine.h index 8a1d3ee25ff..25388a4f974 100644 --- a/src/shogun/machine/BaggingMachine.h +++ b/src/shogun/machine/BaggingMachine.h @@ -139,20 +139,30 @@ namespace shogun */ SGVector apply_get_outputs(CFeatures* data); - /** Register paramaters */ - void register_parameters(); - - /** Initialize the members with default values */ - void init(); - - /** - * get the vector of indices for feature vectors that are out of bag - * - * @param in_bag vector of indices that are in bag. - * NOTE: in_bag is a randomly generated with replacement - * @return the vector of indices - */ - CDynamicArray* get_oob_indices(const SGVector& in_bag); + /** helper function for the apply_{binary,..} functions that + * computes the output probabilities without combination rules + * + * @param data the data to compute the output for + * @return predictions + */ + SGMatrix + apply_outputs_without_combination(CFeatures* data); + + /** Register paramaters */ + void register_parameters(); + + /** Initialize the members with default values */ + void init(); + + /** + * get the vector of indices for feature vectors that are out of bag + * + * @param in_bag vector of indices that are in bag. + * NOTE: in_bag is a randomly generated with replacement + * @return the vector of indices + */ + CDynamicArray* + get_oob_indices(const SGVector& in_bag); protected: /** bags array */ diff --git a/src/shogun/machine/KernelMachine.cpp b/src/shogun/machine/KernelMachine.cpp index 06335587a5e..65f4d0d9178 100644 --- a/src/shogun/machine/KernelMachine.cpp +++ b/src/shogun/machine/KernelMachine.cpp @@ -9,15 +9,20 @@ * Copyright (C) 1999-2009 Fraunhofer Institute FIRST and Max-Planck-Society */ -#include -#include -#include +#include #include +#include +#include #include #include #include +#ifdef HAVE_OPENMP +#include + +#endif + using namespace shogun; #ifndef DOXYGEN_SHOULD_SKIP_THIS @@ -292,8 +297,6 @@ SGVector CKernelMachine::apply_get_outputs(CFeatures* data) { SG_DEBUG("computing output on %d test examples\n", num_vectors) - CSignal::clear_cancel(); - if (io->get_show_progress()) io->enable_progress(); else @@ -333,72 +336,64 @@ SGVector CKernelMachine::apply_get_outputs(CFeatures* data) } else { - // TODO: port to use OpenMP backend instead of pthread -#ifdef HAVE_PTHREAD - int32_t num_threads=parallel->get_num_threads(); -#else - int32_t num_threads=1; -#endif - ASSERT(num_threads>0) - - if (num_threads < 2) - { - S_THREAD_PARAM_KERNEL_MACHINE params; - params.kernel_machine=this; - params.result = output.vector; - params.start=0; - params.end=num_vectors; - params.verbose=true; - params.indices = NULL; - params.indices_len = 0; - apply_helper((void*) ¶ms); - } -#ifdef HAVE_PTHREAD - else + auto pb = progress(range(num_vectors)); + int32_t num_threads; + int64_t step; +#pragma omp parallel shared(num_threads, step) { - pthread_t* threads = SG_MALLOC(pthread_t, num_threads-1); - S_THREAD_PARAM_KERNEL_MACHINE* params = SG_MALLOC(S_THREAD_PARAM_KERNEL_MACHINE, num_threads); - int32_t step= num_vectors/num_threads; - - int32_t t; - for (t=0; thas_property(KP_LINADD) && + (kernel->get_is_initialized())) + { + float64_t score = kernel->compute_optimized(vec); + output[vec] = score + get_bias(); + } + else + { + float64_t score = 0; + for (int32_t i = 0; i < get_num_support_vectors(); i++) + score += + kernel->kernel(get_support_vector(i), vec) * + get_alpha(i); + output[vec] = score + get_bias(); + } + } + } + pb.complete(); } #ifndef WIN32 - if ( CSignal::cancel_computations() ) + if (cancel_computation()) SG_INFO("prematurely stopped. \n") - else #endif - SG_DONE() } SG_DEBUG("leaving %s::apply_get_outputs(%s at %p)\n", @@ -407,54 +402,6 @@ SGVector CKernelMachine::apply_get_outputs(CFeatures* data) return output; } -float64_t CKernelMachine::apply_one(int32_t num) -{ - ASSERT(kernel) - - if (kernel->has_property(KP_LINADD) && (kernel->get_is_initialized())) - { - float64_t score = kernel->compute_optimized(num); - return score+get_bias(); - } - else - { - float64_t score=0; - for(int32_t i=0; ikernel(get_support_vector(i), num)*get_alpha(i); - - return score+get_bias(); - } -} - -void* CKernelMachine::apply_helper(void* p) -{ - S_THREAD_PARAM_KERNEL_MACHINE* params = (S_THREAD_PARAM_KERNEL_MACHINE*) p; - float64_t* result = params->result; - CKernelMachine* kernel_machine = params->kernel_machine; - -#ifdef WIN32 - for (int32_t vec=params->start; vecend; vec++) -#else - for (int32_t vec=params->start; vecend && - !CSignal::cancel_computations(); vec++) -#endif - { - if (params->verbose) - { - int32_t num_vectors=params->end - params->start; - int32_t v=vec-params->start; - if ( (v% (num_vectors/100+1))== 0) - SG_SPROGRESS(v, 0.0, num_vectors-1) - } - - /* eventually use index mapping if exists */ - index_t idx=params->indices ? params->indices[vec] : vec; - result[vec] = kernel_machine->apply_one(idx); - } - - return NULL; -} - void CKernelMachine::store_model_features() { if (!kernel) @@ -545,91 +492,89 @@ SGVector CKernelMachine::apply_locked_get_output( int32_t num_inds=indices.vlen; SGVector output(num_inds); - CSignal::clear_cancel(); - if (io->get_show_progress()) io->enable_progress(); else io->disable_progress(); /* custom kernel never has batch evaluation property so dont do this here */ - // TODO: port to use OpenMP backend instead of pthread -#ifdef HAVE_PTHREAD - int32_t num_threads=parallel->get_num_threads(); -#else - int32_t num_threads=1; -#endif - ASSERT(num_threads>0) - - if (num_threads<2) + auto pb = progress(range(0, num_inds)); + int32_t num_threads; + int64_t step; +#pragma omp parallel shared(num_threads, step) { - S_THREAD_PARAM_KERNEL_MACHINE params; - params.kernel_machine=this; - params.result=output.vector; - - /* use the parameter index vector */ - params.start=0; - params.end=num_inds; - params.indices=indices.vector; - params.indices_len=indices.vlen; - - params.verbose=true; - apply_helper((void*) ¶ms); - } -#ifdef HAVE_PTHREAD - else - { - pthread_t* threads = SG_MALLOC(pthread_t, num_threads-1); - S_THREAD_PARAM_KERNEL_MACHINE* params=SG_MALLOC(S_THREAD_PARAM_KERNEL_MACHINE, num_threads); - int32_t step= num_inds/num_threads; - - int32_t t; - for (t=0; thas_property(KP_LINADD) && + (kernel->get_is_initialized())) + { + float64_t score = kernel->compute_optimized(index); + output[vec] = score + get_bias(); + } + else + { + float64_t score = 0; + for (int32_t i = 0; i < get_num_support_vectors(); i++) + score += kernel->kernel(get_support_vector(i), index) * + get_alpha(i); - params[t].kernel_machine=this; - params[t].result=output.vector; - - /* use the parameter index vector */ - params[t].start=t*step; - params[t].end=num_inds; - params[t].indices=indices.vector; - params[t].indices_len=indices.vlen; - - params[t].verbose=true; - apply_helper((void*) ¶ms[t]); - - for (t=0; thas_property(KP_LINADD) && (kernel->get_is_initialized())) + { + float64_t score = kernel->compute_optimized(num); + return score + get_bias(); + } + else + { + float64_t score = 0; + for (int32_t i = 0; i < get_num_support_vectors(); i++) + score += kernel->kernel(get_support_vector(i), num) * get_alpha(i); + + return score + get_bias(); + } +} + void CKernelMachine::data_lock(CLabels* labs, CFeatures* features) { if ( !kernel ) @@ -710,4 +655,3 @@ bool CKernelMachine::supports_locking() const { return true; } - diff --git a/src/shogun/machine/KernelMachine.h b/src/shogun/machine/KernelMachine.h index f2e2017e277..d4e5e1f293f 100644 --- a/src/shogun/machine/KernelMachine.h +++ b/src/shogun/machine/KernelMachine.h @@ -228,13 +228,6 @@ class CKernelMachine : public CMachine */ virtual float64_t apply_one(int32_t num); - /** apply example helper, used in threads - * - * @param p params of the thread - * @return nothing really - */ - static void* apply_helper(void* p); - #ifndef SWIG // SWIG should skip this part /** Trains a locked machine on a set of indices. Error if machine is * not locked @@ -305,7 +298,7 @@ class CKernelMachine : public CMachine */ virtual void store_model_features(); - private: + private: /** register parameters and do misc init */ void init(); diff --git a/src/shogun/machine/LinearMachine.cpp b/src/shogun/machine/LinearMachine.cpp index eee71ba8ba1..be95ea6cdc1 100644 --- a/src/shogun/machine/LinearMachine.cpp +++ b/src/shogun/machine/LinearMachine.cpp @@ -44,7 +44,7 @@ void CLinearMachine::init() features = NULL; m_compute_bias = true; - SG_ADD(&w, "w", "Parameter vector w.", MS_NOT_AVAILABLE); + SG_ADD(&m_w, "w", "Parameter vector w.", MS_NOT_AVAILABLE); SG_ADD(&bias, "bias", "Bias b.", MS_NOT_AVAILABLE); SG_ADD((CSGObject**) &features, "features", "Feature object.", MS_NOT_AVAILABLE); @@ -58,7 +58,7 @@ CLinearMachine::~CLinearMachine() float64_t CLinearMachine::apply_one(int32_t vec_idx) { - return features->dense_dot(vec_idx, w.vector, w.vlen) + bias; + return features->dense_dot(vec_idx, m_w.vector, m_w.vlen) + bias; } CRegressionLabels* CLinearMachine::apply_regression(CFeatures* data) @@ -88,21 +88,21 @@ SGVector CLinearMachine::apply_get_outputs(CFeatures* data) int32_t num=features->get_num_vectors(); ASSERT(num>0) - ASSERT(w.vlen==features->get_dim_feature_space()) + ASSERT(m_w.vlen==features->get_dim_feature_space()) float64_t* out=SG_MALLOC(float64_t, num); - features->dense_dot_range(out, 0, num, NULL, w.vector, w.vlen, bias); + features->dense_dot_range(out, 0, num, NULL, m_w.vector, m_w.vlen, bias); return SGVector(out,num); } SGVector CLinearMachine::get_w() const { - return w; + return m_w; } -void CLinearMachine::set_w(const SGVector src_w) +void CLinearMachine::set_w(const SGVector w) { - w = src_w; + m_w = w; } void CLinearMachine::set_bias(float64_t b) diff --git a/src/shogun/machine/LinearMachine.h b/src/shogun/machine/LinearMachine.h index a9371a18516..457b4e76b1a 100644 --- a/src/shogun/machine/LinearMachine.h +++ b/src/shogun/machine/LinearMachine.h @@ -188,7 +188,7 @@ class CLinearMachine : public CMachine private: /** w */ - SGVector w; + SGVector m_w; protected: /** bias */ diff --git a/src/shogun/machine/Machine.cpp b/src/shogun/machine/Machine.cpp index 44362db11c9..9cbc218bbcf 100644 --- a/src/shogun/machine/Machine.cpp +++ b/src/shogun/machine/Machine.cpp @@ -9,12 +9,16 @@ * Copyright (C) 1999-2009 Fraunhofer Institute FIRST and Max-Planck-Society */ +#include +#include +#include #include using namespace shogun; -CMachine::CMachine() : CSGObject(), m_max_train_time(0), m_labels(NULL), - m_solver_type(ST_AUTO) +CMachine::CMachine() + : CSGObject(), m_max_train_time(0), m_labels(NULL), m_solver_type(ST_AUTO), + m_cancel_computation(false), m_pause_computation_flag(false) { m_data_locked=false; m_store_model_features=false; @@ -54,7 +58,10 @@ bool CMachine::train(CFeatures* data) m_labels->ensure_valid(get_name()); } + auto sub = connect_to_signal_handler(); bool result = train_machine(data); + sub.unsubscribe(); + reset_computation_variables(); if (m_store_model_features) store_model_features(); @@ -269,3 +276,17 @@ CLatentLabels* CMachine::apply_locked_latent(SGVector indices) "for %s\n", get_name()); return NULL; } + +rxcpp::subscription CMachine::connect_to_signal_handler() +{ + // Subscribe this algorithm to the signal handler + auto subscriber = rxcpp::make_subscriber( + [this](int i) { + if (i == SG_PAUSE_COMP) + this->on_pause(); + else + this->on_next(); + }, + [this]() { this->on_complete(); }); + return get_global_signal()->get_observable()->subscribe(subscriber); +} diff --git a/src/shogun/machine/Machine.h b/src/shogun/machine/Machine.h index aa8a1d3b940..b53785a9833 100644 --- a/src/shogun/machine/Machine.h +++ b/src/shogun/machine/Machine.h @@ -23,6 +23,9 @@ #include #include +#include +#include + namespace shogun { @@ -123,6 +126,11 @@ enum EProblemType */ \ virtual EProblemType get_machine_problem_type() const { return PT; } +#define COMPUTATION_CONTROLLERS \ + if (cancel_computation()) \ + continue; \ + pause_computation(); + /** @brief A generic learning machine interface. * * A machine takes as input CFeatures and CLabels (by default). @@ -306,6 +314,37 @@ class CMachine : public CSGObject return PT_BINARY; } +#ifndef SWIG + /** @return whether the algorithm needs to be stopped */ + SG_FORCED_INLINE bool cancel_computation() const + { + return m_cancel_computation.load(); + } +#endif + +#ifndef SWIG + /** Pause the algorithm if the flag is set */ + SG_FORCED_INLINE void pause_computation() + { + if (m_pause_computation_flag.load()) + { + std::unique_lock lck(m_mutex); + while (m_pause_computation_flag.load()) + m_pause_computation.wait(lck); + } + } +#endif + +#ifndef SWIG + /** Resume current computation (sets the flag) */ + SG_FORCED_INLINE void resume_computation() + { + std::unique_lock lck(m_mutex); + m_pause_computation_flag = false; + m_pause_computation.notify_all(); + } +#endif + virtual const char* get_name() const { return "Machine"; } protected: @@ -357,6 +396,38 @@ class CMachine : public CSGObject /** returns whether machine require labels for training */ virtual bool train_require_labels() const { return true; } + /** connect the machine instance to the signal handler */ + rxcpp::subscription connect_to_signal_handler(); + + /** reset the computation variables */ + void reset_computation_variables() + { + m_cancel_computation = false; + m_pause_computation_flag = false; + } + + /** The action which will be done when the user decides to + * premature stop the CMachine execution */ + virtual void on_next() + { + m_cancel_computation.store(true); + } + + /** The action which will be done when the user decides to + * pause the CMachine execution */ + virtual void on_pause() + { + m_pause_computation_flag.store(true); + /* Here there should be the actual code*/ + resume_computation(); + } + + /** The action which will be done when the user decides to + * return to prompt and terminate the program execution */ + virtual void on_complete() + { + } + protected: /** maximum training time */ float64_t m_max_train_time; @@ -372,6 +443,18 @@ class CMachine : public CSGObject /** whether data is locked */ bool m_data_locked; + + /** Cancel computation */ + std::atomic m_cancel_computation; + + /** Pause computation flag */ + std::atomic m_pause_computation_flag; + + /** Conditional variable to make threads wait */ + std::condition_variable m_pause_computation; + + /** Mutex used to pause threads */ + std::mutex m_mutex; }; } #endif // _MACHINE_H__ diff --git a/src/shogun/machine/OnlineLinearMachine.cpp b/src/shogun/machine/OnlineLinearMachine.cpp index 47688e49e44..cb7c2a0d37d 100644 --- a/src/shogun/machine/OnlineLinearMachine.cpp +++ b/src/shogun/machine/OnlineLinearMachine.cpp @@ -12,13 +12,16 @@ #include #include #include +#include + +#include using namespace shogun; COnlineLinearMachine::COnlineLinearMachine() -: CMachine(), w_dim(0), w(NULL), bias(0), features(NULL) +: CMachine(), bias(0), features(NULL) { - m_parameters->add_vector(&w, &w_dim, "w", "Parameter vector w."); + SG_ADD(&m_w, "m_w", "Parameter vector w.", MS_NOT_AVAILABLE); SG_ADD(&bias, "bias", "Bias b.", MS_NOT_AVAILABLE); SG_ADD((CSGObject**) &features, "features", "Feature object.", MS_NOT_AVAILABLE); @@ -26,10 +29,6 @@ COnlineLinearMachine::COnlineLinearMachine() COnlineLinearMachine::~COnlineLinearMachine() { - // It is possible that a derived class may have already - // called SG_FREE() on the weight vector - if (w != NULL) - SG_FREE(w); SG_UNREF(features); } @@ -58,37 +57,32 @@ SGVector COnlineLinearMachine::apply_get_outputs(CFeatures* data) ASSERT(features) ASSERT(features->has_property(FP_STREAMING_DOT)) - DynArray* labels_dynarray=new DynArray(); - int32_t num_labels=0; - + std::vector labels; features->start_parser(); while (features->get_next_example()) { - float64_t current_lab=features->dense_dot(w, w_dim) + bias; - - labels_dynarray->append_element(current_lab); - num_labels++; + float64_t current_lab=features->dense_dot(m_w.vector, m_w.vlen) + bias; + labels.push_back(current_lab); features->release_example(); } features->end_parser(); - SGVector labels_array(num_labels); - for (int32_t i=0; i labels_array(labels.size()); + sg_memcpy(labels_array.vector, labels.data(), sizeof(float64_t)*labels.size()); - delete labels_dynarray; return labels_array; } float32_t COnlineLinearMachine::apply_one(float32_t* vec, int32_t len) { - return CMath::dot(vec, w, len)+bias; + SGVector wrap(vec, len, false); + return linalg::dot(wrap, m_w)+bias; } float32_t COnlineLinearMachine::apply_to_current_example() { - return features->dense_dot(w, w_dim)+bias; + return features->dense_dot(m_w.vector, m_w.vlen)+bias; } bool COnlineLinearMachine::train_machine(CFeatures *data) diff --git a/src/shogun/machine/OnlineLinearMachine.h b/src/shogun/machine/OnlineLinearMachine.h index c8f3e625017..6cc6d6f0f94 100644 --- a/src/shogun/machine/OnlineLinearMachine.h +++ b/src/shogun/machine/OnlineLinearMachine.h @@ -57,18 +57,6 @@ class COnlineLinearMachine : public CMachine COnlineLinearMachine(); virtual ~COnlineLinearMachine(); - /** get w - * - * @param dst_w store w in this argument - * @param dst_dims dimension of w - */ - virtual void get_w(float32_t*& dst_w, int32_t& dst_dims) - { - ASSERT(w && w_dim>0) - dst_w=w; - dst_dims=w_dim; - } - /** * Get w as a _new_ float64_t array * @@ -77,23 +65,20 @@ class COnlineLinearMachine : public CMachine */ virtual void get_w(float64_t*& dst_w, int32_t& dst_dims) { - ASSERT(w && w_dim>0) - dst_w=SG_MALLOC(float64_t, w_dim); - for (int32_t i=0; i 0) + dst_w=SG_MALLOC(float64_t, m_w.vlen); + for (int32_t i=0; i get_w() + virtual SGVector get_w() const { - float32_t * dst_w = SG_MALLOC(float32_t, w_dim); - for (int32_t i=0; i(dst_w, w_dim); + return m_w; } /** set w @@ -101,12 +86,9 @@ class COnlineLinearMachine : public CMachine * @param src_w new w * @param src_w_dim dimension of new w */ - virtual void set_w(float32_t* src_w, int32_t src_w_dim) + virtual void set_w(const SGVector w) { - SG_FREE(w); - w=SG_MALLOC(float32_t, src_w_dim); - sg_memcpy(w, src_w, size_t(src_w_dim)*sizeof(float32_t)); - w_dim=src_w_dim; + m_w = w; } /** @@ -117,11 +99,9 @@ class COnlineLinearMachine : public CMachine */ virtual void set_w(float64_t* src_w, int32_t src_w_dim) { - SG_FREE(w); - w=SG_MALLOC(float32_t, src_w_dim); + m_w = SGVector(src_w_dim); for (int32_t i=0; i m_w; /** bias */ float32_t bias; /** features */ diff --git a/src/shogun/machine/StructuredOutputMachine.cpp b/src/shogun/machine/StructuredOutputMachine.cpp index 1604cb99dff..cf3426440e2 100644 --- a/src/shogun/machine/StructuredOutputMachine.cpp +++ b/src/shogun/machine/StructuredOutputMachine.cpp @@ -15,6 +15,8 @@ #include #include +#include + using namespace shogun; CStructuredOutputMachine::CStructuredOutputMachine() @@ -94,7 +96,7 @@ CLossFunction* CStructuredOutputMachine::get_surrogate_loss() const return m_surrogate_loss; } -float64_t CStructuredOutputMachine::risk_nslack_margin_rescale(float64_t* subgrad, float64_t* W, TMultipleCPinfo* info) +float64_t CStructuredOutputMachine::risk_nslack_margin_rescale(SGVector& subgrad, SGVector& W, TMultipleCPinfo* info) { int32_t dim = m_model->get_dim(); @@ -113,16 +115,15 @@ float64_t CStructuredOutputMachine::risk_nslack_margin_rescale(float64_t* subgra SG_UNREF(features); float64_t R = 0.0; - for (int32_t i=0; iargmax(SGVector(W,dim,false), i, true); + CResultSet* result = m_model->argmax(SGVector(W.vector,dim,false), i, true); SGVector psi_pred = result->psi_pred; SGVector psi_truth = result->psi_truth; - SGVector::vec1_plus_scalar_times_vec2(subgrad, 1.0, psi_pred.vector, dim); - SGVector::vec1_plus_scalar_times_vec2(subgrad, -1.0, psi_truth.vector, dim); + SGVector::vec1_plus_scalar_times_vec2(subgrad.vector, 1.0, psi_pred.vector, dim); + SGVector::vec1_plus_scalar_times_vec2(subgrad.vector, -1.0, psi_truth.vector, dim); R += result->score; SG_UNREF(result); } @@ -130,31 +131,31 @@ float64_t CStructuredOutputMachine::risk_nslack_margin_rescale(float64_t* subgra return R; } -float64_t CStructuredOutputMachine::risk_nslack_slack_rescale(float64_t* subgrad, float64_t* W, TMultipleCPinfo* info) +float64_t CStructuredOutputMachine::risk_nslack_slack_rescale(SGVector& subgrad, SGVector& W, TMultipleCPinfo* info) { SG_ERROR("%s::risk_nslack_slack_rescale() has not been implemented!\n", get_name()); return 0.0; } -float64_t CStructuredOutputMachine::risk_1slack_margin_rescale(float64_t* subgrad, float64_t* W, TMultipleCPinfo* info) +float64_t CStructuredOutputMachine::risk_1slack_margin_rescale(SGVector& subgrad, SGVector& W, TMultipleCPinfo* info) { SG_ERROR("%s::risk_1slack_margin_rescale() has not been implemented!\n", get_name()); return 0.0; } -float64_t CStructuredOutputMachine::risk_1slack_slack_rescale(float64_t* subgrad, float64_t* W, TMultipleCPinfo* info) +float64_t CStructuredOutputMachine::risk_1slack_slack_rescale(SGVector& subgrad, SGVector& W, TMultipleCPinfo* info) { SG_ERROR("%s::risk_1slack_slack_rescale() has not been implemented!\n", get_name()); return 0.0; } -float64_t CStructuredOutputMachine::risk_customized_formulation(float64_t* subgrad, float64_t* W, TMultipleCPinfo* info) +float64_t CStructuredOutputMachine::risk_customized_formulation(SGVector& subgrad, SGVector& W, TMultipleCPinfo* info) { SG_ERROR("%s::risk_customized_formulation() has not been implemented!\n", get_name()); return 0.0; } -float64_t CStructuredOutputMachine::risk(float64_t* subgrad, float64_t* W, +float64_t CStructuredOutputMachine::risk(SGVector& subgrad, SGVector& W, TMultipleCPinfo* info, EStructRiskType rtype) { float64_t ret = 0.0; diff --git a/src/shogun/machine/StructuredOutputMachine.h b/src/shogun/machine/StructuredOutputMachine.h index 308f549407e..37f0b8b432a 100644 --- a/src/shogun/machine/StructuredOutputMachine.h +++ b/src/shogun/machine/StructuredOutputMachine.h @@ -119,7 +119,7 @@ class CStructuredOutputMachine : public CMachine * @param rtype The type of structured risk * @return Value of the computed risk at given point W */ - virtual float64_t risk(float64_t* subgrad, float64_t* W, + virtual float64_t risk(SGVector& subgrad, SGVector& W, TMultipleCPinfo* info=0, EStructRiskType rtype = N_SLACK_MARGIN_RESCALING); /** @return training progress helper */ @@ -167,7 +167,7 @@ class CStructuredOutputMachine : public CMachine * @param info Helper info for multiple cutting plane models algorithm * @return Value of the computed risk at given point W */ - virtual float64_t risk_nslack_margin_rescale(float64_t* subgrad, float64_t* W, TMultipleCPinfo* info=0); + virtual float64_t risk_nslack_margin_rescale(SGVector& subgrad, SGVector& W, TMultipleCPinfo* info=0); /** n-slack formulation and slack rescaling * @@ -176,7 +176,7 @@ class CStructuredOutputMachine : public CMachine * @param info Helper info for multiple cutting plane models algorithm * @return Value of the computed risk at given point W */ - virtual float64_t risk_nslack_slack_rescale(float64_t* subgrad, float64_t* W, TMultipleCPinfo* info=0); + virtual float64_t risk_nslack_slack_rescale(SGVector& subgrad, SGVector& W, TMultipleCPinfo* info=0); /** 1-slack formulation and margin rescaling * @@ -185,7 +185,7 @@ class CStructuredOutputMachine : public CMachine * @param info Helper info for multiple cutting plane models algorithm * @return Value of the computed risk at given point W */ - virtual float64_t risk_1slack_margin_rescale(float64_t* subgrad, float64_t* W, TMultipleCPinfo* info=0); + virtual float64_t risk_1slack_margin_rescale(SGVector& subgrad, SGVector& W, TMultipleCPinfo* info=0); /** 1-slack formulation and slack rescaling * @@ -194,7 +194,7 @@ class CStructuredOutputMachine : public CMachine * @param info Helper info for multiple cutting plane models algorithm * @return Value of the computed risk at given point W */ - virtual float64_t risk_1slack_slack_rescale(float64_t* subgrad, float64_t* W, TMultipleCPinfo* info=0); + virtual float64_t risk_1slack_slack_rescale(SGVector& subgrad, SGVector& W, TMultipleCPinfo* info=0); /** customized risk type * @@ -203,7 +203,7 @@ class CStructuredOutputMachine : public CMachine * @param info Helper info for multiple cutting plane models algorithm * @return Value of the computed risk at given point W */ - virtual float64_t risk_customized_formulation(float64_t* subgrad, float64_t* W, TMultipleCPinfo* info=0); + virtual float64_t risk_customized_formulation(SGVector& subgrad, SGVector& W, TMultipleCPinfo* info=0); private: /** register class members */ diff --git a/src/shogun/machine/gp/KLDualInferenceMethod.cpp b/src/shogun/machine/gp/KLDualInferenceMethod.cpp index f126d96804d..3f9ab520f4c 100644 --- a/src/shogun/machine/gp/KLDualInferenceMethod.cpp +++ b/src/shogun/machine/gp/KLDualInferenceMethod.cpp @@ -540,10 +540,8 @@ float64_t CKLDualInferenceMethod::optimization() cost_fun->set_target(this); bool cleanup=false; -#ifdef USE_REFERENCE_COUNTING if(this->ref_count()>1) cleanup=true; -#endif minimizer->set_cost_function(cost_fun); float64_t nlml_opt = minimizer->minimize(); diff --git a/src/shogun/machine/gp/KLInference.cpp b/src/shogun/machine/gp/KLInference.cpp index c8cad41481c..b820d7a96d7 100644 --- a/src/shogun/machine/gp/KLInference.cpp +++ b/src/shogun/machine/gp/KLInference.cpp @@ -344,10 +344,9 @@ float64_t CKLInference::optimization() KLInferenceCostFunction *cost_fun=new KLInferenceCostFunction(); cost_fun->set_target(this); bool cleanup=false; -#ifdef USE_REFERENCE_COUNTING if(this->ref_count()>1) cleanup=true; -#endif + FirstOrderMinimizer* opt= dynamic_cast(m_minimizer); REQUIRE(opt, "FirstOrderMinimizer is required\n") diff --git a/src/shogun/machine/gp/LaplaceInference.cpp b/src/shogun/machine/gp/LaplaceInference.cpp index ddec6e77971..14a2fcec333 100644 --- a/src/shogun/machine/gp/LaplaceInference.cpp +++ b/src/shogun/machine/gp/LaplaceInference.cpp @@ -34,7 +34,6 @@ #include -#include #include using namespace shogun; diff --git a/src/shogun/machine/gp/LogitLikelihood.cpp b/src/shogun/machine/gp/LogitLikelihood.cpp index fb4924c0691..94884392df2 100644 --- a/src/shogun/machine/gp/LogitLikelihood.cpp +++ b/src/shogun/machine/gp/LogitLikelihood.cpp @@ -32,7 +32,9 @@ #include +#ifdef USE_GPL_SHOGUN #include +#endif //USE_GPL_SHOGUN #include #include @@ -388,8 +390,12 @@ SGVector CLogitLikelihood::get_log_zeroth_moments( g->set_a(y[i]); // evaluate integral on (-inf, inf) +#ifdef USE_GPL_SHOGUN r[i]=CIntegration::integrate_quadgk(h, -CMath::INFTY, mu[i])+ CIntegration::integrate_quadgk(h, mu[i], CMath::INFTY); +#else + SG_GPL_ONLY +#endif //USE_GPL_SHOGUN } SG_UNREF(h); @@ -430,15 +436,18 @@ float64_t CLogitLikelihood::get_first_moment(SGVector mu, // create an object of k(x)=x*N(x|mu,sigma^2)*sigmoid(x) CProductFunction* k=new CProductFunction(l, h); SG_REF(k); - + float64_t Ex=0; +#ifdef USE_GPL_SHOGUN // compute Z = \int N(x|mu,sigma)*sigmoid(a*x) dx float64_t Z=CIntegration::integrate_quadgk(h, -CMath::INFTY, mu[i])+ CIntegration::integrate_quadgk(h, mu[i], CMath::INFTY); // compute 1st moment: E[x] = Z^-1 * \int x*N(x|mu,sigma)*sigmoid(a*x)dx - float64_t Ex=(CIntegration::integrate_quadgk(k, -CMath::INFTY, mu[i])+ + Ex=(CIntegration::integrate_quadgk(k, -CMath::INFTY, mu[i])+ CIntegration::integrate_quadgk(k, mu[i], CMath::INFTY))/Z; - +#else + SG_GPL_ONLY +#endif //USE_GPL_SHOGUN SG_UNREF(k); return Ex; @@ -482,18 +491,23 @@ float64_t CLogitLikelihood::get_second_moment(SGVector mu, CProductFunction* p=new CProductFunction(q, h); SG_REF(p); + float64_t Ex=0; + float64_t Ex2=0; +#ifdef USE_GPL_SHOGUN // compute Z = \int N(x|mu,sigma)*sigmoid(a*x) dx float64_t Z=CIntegration::integrate_quadgk(h, -CMath::INFTY, mu[i])+ CIntegration::integrate_quadgk(h, mu[i], CMath::INFTY); // compute 1st moment: E[x] = Z^-1 * \int x*N(x|mu,sigma)*sigmoid(a*x)dx - float64_t Ex=(CIntegration::integrate_quadgk(k, -CMath::INFTY, mu[i])+ + Ex=(CIntegration::integrate_quadgk(k, -CMath::INFTY, mu[i])+ CIntegration::integrate_quadgk(k, mu[i], CMath::INFTY))/Z; // compute E[x^2] = Z^-1 * \int x^2*N(x|mu,sigma)*sigmoid(a*x)dx - float64_t Ex2=(CIntegration::integrate_quadgk(p, -CMath::INFTY, mu[i])+ + Ex2=(CIntegration::integrate_quadgk(p, -CMath::INFTY, mu[i])+ CIntegration::integrate_quadgk(p, mu[i], CMath::INFTY))/Z; - +#else + SG_GPL_ONLY +#endif //USE_GPL_SHOGUN SG_UNREF(k); SG_UNREF(p); diff --git a/src/shogun/machine/gp/LogitLikelihood.h b/src/shogun/machine/gp/LogitLikelihood.h index b611a16be18..57251547a3d 100644 --- a/src/shogun/machine/gp/LogitLikelihood.h +++ b/src/shogun/machine/gp/LogitLikelihood.h @@ -179,7 +179,6 @@ class CLogitLikelihood : public CLikelihoodModel */ virtual float64_t get_second_moment(SGVector mu, SGVector s2, const CLabels* lab, index_t i) const; - /** return whether logit likelihood function supports binary classification * * @return true diff --git a/src/shogun/machine/gp/MultiLaplaceInferenceMethod.cpp b/src/shogun/machine/gp/MultiLaplaceInferenceMethod.cpp index 0373305dbcb..2e27678d2b9 100644 --- a/src/shogun/machine/gp/MultiLaplaceInferenceMethod.cpp +++ b/src/shogun/machine/gp/MultiLaplaceInferenceMethod.cpp @@ -44,7 +44,9 @@ #include #include #include +#ifdef USE_GPL_SHOGUN #include +#endif //USE_GPL_SHOGUN using namespace shogun; using namespace Eigen; @@ -53,7 +55,7 @@ namespace shogun { #ifndef DOXYGEN_SHOULD_SKIP_THIS - +#ifdef USE_GPL_SHOGUN /** Wrapper class used for the Brent minimizer */ class CMultiPsiLine : public func_base { @@ -95,6 +97,7 @@ class CMultiPsiLine : public func_base return result; } }; +#endif //USE_GPL_SHOGUN #endif /* DOXYGEN_SHOULD_SKIP_THIS */ @@ -351,7 +354,7 @@ void CMultiLaplaceInferenceMethod::update_alpha() VectorXd &eigen_dalpha=eigen_b; eigen_dalpha+=eigen_E.transpose()*(eigen_M.triangularView().solve(tmp4))-eigen_c-eigen_alpha; - +#ifdef USE_GPL_SHOGUN // perform Brent's optimization CMultiPsiLine func; @@ -368,6 +371,9 @@ void CMultiLaplaceInferenceMethod::update_alpha() float64_t x; Psi_New=local_min(0, m_opt_max, m_opt_tolerance, func, x); +#else + SG_GPL_ONLY +#endif //USE_GPL_SHOGUN m_nlz+=Psi_New; } diff --git a/src/shogun/machine/gp/NumericalVGLikelihood.cpp b/src/shogun/machine/gp/NumericalVGLikelihood.cpp index 51bf1cf9818..7cfbeec7bba 100644 --- a/src/shogun/machine/gp/NumericalVGLikelihood.cpp +++ b/src/shogun/machine/gp/NumericalVGLikelihood.cpp @@ -44,7 +44,9 @@ #include #include #include +#ifdef USE_GPL_SHOGUN #include +#endif //USE_GPL_SHOGUN using namespace Eigen; @@ -254,7 +256,11 @@ bool CNumericalVGLikelihood::set_variational_distribution(SGVector mu { m_xgh=SGVector(m_GHQ_N); m_wgh=SGVector(m_GHQ_N); +#ifdef USE_GPL_SHOGUN CIntegration::generate_gauher(m_xgh, m_wgh); +#else + SG_GPL_ONLY +#endif //USE_GPL_SHOGUN m_is_init_GHQ=true; } diff --git a/src/shogun/machine/gp/SingleFITCLaplaceInferenceMethod.cpp b/src/shogun/machine/gp/SingleFITCLaplaceInferenceMethod.cpp index d0a19ee6ff9..8979f2ab3c9 100644 --- a/src/shogun/machine/gp/SingleFITCLaplaceInferenceMethod.cpp +++ b/src/shogun/machine/gp/SingleFITCLaplaceInferenceMethod.cpp @@ -33,7 +33,9 @@ #include #include +#ifdef USE_GPL_SHOGUN #include +#endif //USE_GPL_SHOGUN #include #include #include @@ -45,7 +47,7 @@ namespace shogun { #ifndef DOXYGEN_SHOULD_SKIP_THIS - +#ifdef USE_GPL_SHOGUN /** Wrapper class used for the Brent minimizer */ class CFITCPsiLine : public func_base { @@ -89,6 +91,7 @@ class CFITCPsiLine : public func_base return result; } }; +#endif //USE_GPL_SHOGUN class SingleFITCLaplaceInferenceMethodCostFunction: public FirstOrderCostFunction { @@ -271,7 +274,7 @@ float64_t CSingleFITCLaplaceNewtonOptimizer::minimize() MatrixXd eigen_RV=eigen_tmp2*eigen_V; //dalpha = dd.*b - (W.*dd).*(RV'*(RV*(dd.*b))) - alpha; % Newt dir + line search VectorXd dalpha=dd.cwiseProduct(b)-eigen_t.cwiseProduct(eigen_RV.transpose()*(eigen_RV*(dd.cwiseProduct(b))))-eigen_al; - +#ifdef USE_GPL_SHOGUN //perform Brent's optimization CFITCPsiLine func; @@ -289,6 +292,9 @@ float64_t CSingleFITCLaplaceNewtonOptimizer::minimize() float64_t x; Psi_New=local_min(0, m_opt_max, m_opt_tolerance, func, x); +#else + SG_GPL_ONLY +#endif //USE_GPL_SHOGUN } if (Psi_Old-Psi_New>m_tolerance && iter>=m_iter) @@ -563,10 +569,8 @@ void CSingleFITCLaplaceInferenceMethod::update_alpha() if (opt) { opt->set_target(this); -#ifdef USE_REFERENCE_COUNTING if(this->ref_count()>1) cleanup=true; -#endif opt->minimize(); opt->unset_target(cleanup); } @@ -577,10 +581,8 @@ void CSingleFITCLaplaceInferenceMethod::update_alpha() SingleFITCLaplaceInferenceMethodCostFunction *cost_fun=new SingleFITCLaplaceInferenceMethodCostFunction(); cost_fun->set_target(this); -#ifdef USE_REFERENCE_COUNTING if(this->ref_count()>1) cleanup=true; -#endif minimizer->set_cost_function(cost_fun); minimizer->minimize(); minimizer->unset_cost_function(false); diff --git a/src/shogun/machine/gp/SingleLaplaceInferenceMethod.cpp b/src/shogun/machine/gp/SingleLaplaceInferenceMethod.cpp index bb9417906df..a17e1470b20 100644 --- a/src/shogun/machine/gp/SingleLaplaceInferenceMethod.cpp +++ b/src/shogun/machine/gp/SingleLaplaceInferenceMethod.cpp @@ -19,7 +19,9 @@ #include #include +#ifdef USE_GPL_SHOGUN #include +#endif //USE_GPL_SHOGUN #include #include @@ -29,6 +31,7 @@ using namespace Eigen; namespace shogun { +#ifdef USE_GPL_SHOGUN #ifndef DOXYGEN_SHOULD_SKIP_THIS /** Wrapper class used for the Brent minimizer */ class PsiLine : public func_base @@ -68,6 +71,7 @@ class PsiLine : public func_base return result; } }; +#endif //USE_GPL_SHOGUN class SingleLaplaceInferenceMethodCostFunction: public FirstOrderCostFunction { @@ -236,6 +240,7 @@ float64_t CSingleLaplaceNewtonOptimizer::minimize() VectorXd dalpha=b-eigen_sW.cwiseProduct( L.solve(eigen_sW.cwiseProduct(eigen_ktrtr*b*CMath::exp((m_obj->m_log_scale)*2.0))))-eigen_alpha; +#ifdef USE_GPL_SHOGUN // perform Brent's optimization PsiLine func; @@ -253,6 +258,9 @@ float64_t CSingleLaplaceNewtonOptimizer::minimize() float64_t x; Psi_New=local_min(0, m_opt_max, m_opt_tolerance, func, x); +#else + SG_GPL_ONLY +#endif //USE_GPL_SHOGUN } if (Psi_Old-Psi_New>m_tolerance && iter>=m_iter) @@ -506,10 +514,8 @@ void CSingleLaplaceInferenceMethod::update_alpha() if (opt) { opt->set_target(this); -#ifdef USE_REFERENCE_COUNTING if(this->ref_count()>1) cleanup=true; -#endif opt->minimize(); opt->unset_target(cleanup); } @@ -517,18 +523,19 @@ void CSingleLaplaceInferenceMethod::update_alpha() { FirstOrderMinimizer* minimizer= dynamic_cast(m_minimizer); REQUIRE(minimizer, "The provided minimizer is not supported\n"); - +#ifdef USE_GPL_SHOGUN SingleLaplaceInferenceMethodCostFunction *cost_fun=new SingleLaplaceInferenceMethodCostFunction(); cost_fun->set_target(this); -#ifdef USE_REFERENCE_COUNTING if(this->ref_count()>1) cleanup=true; -#endif minimizer->set_cost_function(cost_fun); minimizer->minimize(); minimizer->unset_cost_function(false); cost_fun->unset_target(cleanup); SG_UNREF(cost_fun); +#else + SG_GPL_ONLY +#endif //USE_GPL_SHOGUN } // get mean vector and create eigen representation of it Map eigen_mean(m_mean_f.vector, m_mean_f.vlen); diff --git a/src/shogun/machine/gp/SingleSparseInference.cpp b/src/shogun/machine/gp/SingleSparseInference.cpp index f5fd2c4f3bf..2d50137f906 100644 --- a/src/shogun/machine/gp/SingleSparseInference.cpp +++ b/src/shogun/machine/gp/SingleSparseInference.cpp @@ -363,10 +363,8 @@ void CSingleSparseInference::optimize_inducing_features() SingleSparseInferenceCostFunction *cost_fun=new SingleSparseInferenceCostFunction(); cost_fun->set_target(this); bool cleanup=false; -#ifdef USE_REFERENCE_COUNTING if(this->ref_count()>1) cleanup=true; -#endif #ifdef USE_GPL_SHOGUN #ifdef HAVE_NLOPT diff --git a/src/shogun/machine/gp/StudentsTLikelihood.cpp b/src/shogun/machine/gp/StudentsTLikelihood.cpp index 4a9c2c56294..d031747350c 100644 --- a/src/shogun/machine/gp/StudentsTLikelihood.cpp +++ b/src/shogun/machine/gp/StudentsTLikelihood.cpp @@ -35,7 +35,9 @@ #include +#ifdef USE_GPL_SHOGUN #include +#endif //USE_GPL_SHOGUN #include #include #include @@ -617,9 +619,13 @@ SGVector CStudentsTLikelihood::get_log_zeroth_moments( // set Stundent's-t pdf parameters g->set_mu(y[i]); +#ifdef USE_GPL_SHOGUN // evaluate integral on (-inf, inf) r[i]=CIntegration::integrate_quadgk(h, -CMath::INFTY, mu[i])+ CIntegration::integrate_quadgk(h, mu[i], CMath::INFTY); +#else + SG_ERROR("StudentsT likelihood moments only supported under GPL.\n") +#endif //USE_GPL_SHOGUN } SG_UNREF(h); @@ -658,15 +664,19 @@ float64_t CStudentsTLikelihood::get_first_moment(SGVector mu, CProductFunction* k=new CProductFunction(new CLinearFunction(), h); SG_REF(k); + float64_t Ex=0; +#ifdef USE_GPL_SHOGUN // compute Z = \int N(x|mu,sigma)*t(x|mu,sigma,nu) dx float64_t Z=CIntegration::integrate_quadgk(h, -CMath::INFTY, mu[i])+ CIntegration::integrate_quadgk(h, mu[i], CMath::INFTY); // compute 1st moment: // E[x] = Z^-1 * \int x*N(x|mu,sigma)*t(x|mu,sigma,nu)dx - float64_t Ex=(CIntegration::integrate_quadgk(k, -CMath::INFTY, mu[i])+ + Ex=(CIntegration::integrate_quadgk(k, -CMath::INFTY, mu[i])+ CIntegration::integrate_quadgk(k, mu[i], CMath::INFTY))/Z; - +#else + SG_ERROR("StudentsT likelihood moments only supported under GPL.\n") +#endif //USE_GPL_SHOGUN SG_UNREF(k); return Ex; @@ -704,19 +714,24 @@ float64_t CStudentsTLikelihood::get_second_moment(SGVector mu, CProductFunction* p=new CProductFunction(new CQuadraticFunction(), h); SG_REF(p); + float64_t Ex=0; + float64_t Ex2=0; +#ifdef USE_GPL_SHOGUN // compute Z = \int N(x|mu,sigma)*t(x|mu,sigma,nu) dx float64_t Z=CIntegration::integrate_quadgk(h, -CMath::INFTY, mu[i])+ CIntegration::integrate_quadgk(h, mu[i], CMath::INFTY); // compute 1st moment: // E[x] = Z^-1 * \int x*N(x|mu,sigma)*t(x|mu,sigma,nu)dx - float64_t Ex=(CIntegration::integrate_quadgk(k, -CMath::INFTY, mu[i])+ + Ex=(CIntegration::integrate_quadgk(k, -CMath::INFTY, mu[i])+ CIntegration::integrate_quadgk(k, mu[i], CMath::INFTY))/Z; // compute E[x^2] = Z^-1 * \int x^2*N(x|mu,sigma)*t(x|mu,sigma,nu)dx - float64_t Ex2=(CIntegration::integrate_quadgk(p, -CMath::INFTY, mu[i])+ + Ex2=(CIntegration::integrate_quadgk(p, -CMath::INFTY, mu[i])+ CIntegration::integrate_quadgk(p, mu[i], CMath::INFTY))/Z; - +#else + SG_GPL_ONLY +#endif //USE_GPL_SHOGUN SG_UNREF(k); SG_UNREF(p); diff --git a/src/shogun/mathematics/Integration.cpp b/src/shogun/mathematics/Integration.cpp deleted file mode 100644 index 8633a623c87..00000000000 --- a/src/shogun/mathematics/Integration.cpp +++ /dev/null @@ -1,910 +0,0 @@ -/* - * This program is free software; you can redistribute it and/or modify - * it under the terms of the GNU General Public License as published by - * the Free Software Foundation; either version 3 of the License, or - * (at your option) any later version. - * - * Written (w) 2014 Wu Lin - * Written (W) 2013 Roman Votyakov - * - * The abscissae and weights for Gauss-Kronrod rules are taken form - * QUADPACK, which is in public domain. - * http://www.netlib.org/quadpack/ - * - * See header file for which functions are adapted from GNU Octave, - * file quadgk.m: Copyright (C) 2008-2012 David Bateman under GPLv3 - * http://www.gnu.org/software/octave/ - * - * See header file for which functions are adapted from - * Gaussian Process Machine Learning Toolbox, file util/gauher.m, - * http://www.gaussianprocess.org/gpml/code/matlab/doc/ - */ - -#include - - -#include -#include - -using namespace shogun; -using namespace Eigen; - -namespace shogun -{ - -#ifndef DOXYGEN_SHOULD_SKIP_THIS - -/** @brief Class of the function, which is used for standard infinite - * to finite integral transformation - * - * \f[ - * \int_{-\infty}^{\infty}f(x)dx = \int_{-1}^{1}f(g(t))g'(t)dt - * \f] - * - * where \f$g(t)=\frac{t}{1-t^2}\f$ and - * \f$g'(t)=\frac{1+t^2}{(1-t^2)^2}\f$. - */ -class CITransformFunction : public CFunction -{ -public: - /** constructor - * - * @param f function \f$f(x)\f$ - */ - CITransformFunction(CFunction* f) - { - SG_REF(f); - m_f=f; - } - - virtual ~CITransformFunction() { SG_UNREF(m_f); } - - /** return the real value of the function at given point - * - * @param x argument - * - * @return \f$f(g(x))*g'(x)\f$, where \f$g(x)=\frac{x}{1-x^2}\f$ - * and \f$g'(t)=\frac{1+t^2}{(1-t^2)^2}\f$ - */ - virtual float64_t operator() (float64_t x) - { - float64_t hx=1.0/(1.0-CMath::sq(x)); - float64_t gx=x*hx; - float64_t dgx=(1.0+CMath::sq(x))*CMath::sq(hx); - - return (*m_f)(gx)*dgx; - } - -private: - /** function \f$f(x)\f$ */ - CFunction* m_f; -}; - -/** @brief Class of the function, which is used for singularity - * weakening transform on \f$(-\infty, b]\f$ - * - * \f[ - * \int_{-\infty}^{b} f(x)dx=-\int_{-\infty}^{0} f(b-t^2)2tdt - * \f] - * - * and the finite interval transform - * - * \f[ - * \int_{-\infty}^{0} f(b-t^2)2tdt = \int_{-1}^{0} f(b-g(s)^2)2g(s)g'(s)ds - * \f] - * - * where \f$g(s)=\frac{s}{1+s}\f$ and \f$g'(s)=\frac{1}{(1+s)^2}\f$. - */ -class CILTransformFunction : public CFunction -{ -public: - /** constructor - * - * @param f function \f$f(x)\f$ - * @param b upper bound - */ - CILTransformFunction(CFunction* f, float64_t b) - { - SG_REF(f); - m_f=f; - m_b=b; - } - - virtual ~CILTransformFunction() { SG_UNREF(m_f); } - - /** return the real value of the function at given point - * - * @param x argument of a function - * - * @return \f$f(b-g(x)^2)2g(x)g'(x)dx\f$, where - * \f$g(x)=\frac{x}{1+x}\f$ and \f$g'(x)=\frac{1}{(1+x)^2}\f$ - */ - virtual float64_t operator() (float64_t x) - { - float64_t hx=1.0/(1.0+x); - float64_t gx=x*hx; - float64_t dgx=CMath::sq(hx); - - return -(*m_f)(m_b-CMath::sq(gx))*2*gx*dgx; - } - -private: - /** function \f$f(x)\f$ */ - CFunction* m_f; - - /** upper bound */ - float64_t m_b; -}; - -/** @brief Class of the function, which is used for singularity - * weakening transform on \f$[a, \infty)\f$ - * - * \f[ - * \int_{a}^{\infty} f(x)dx=\int_{0}^{\infty} f(a+t^2)2tdt - * \f] - * - * and the finite interval transform - * - * \f[ - * \int_{0}^{\infty} f(a+t^2)2tdt = \int_{0}^{1} f(a+g(s)^2)2g(s)g'(s)ds - * \f] - * - * where \f$g(s)=\frac{s}{1-s}\f$ and \f$g'(s)=\frac{1}{(1-s)^2}\f$. - */ -class CIUTransformFunction : public CFunction -{ -public: - /** constructor - * - * @param f function \f$f(x)\f$ - * @param a lower bound - */ - CIUTransformFunction(CFunction* f, float64_t a) - { - SG_REF(f); - m_f=f; - m_a=a; - } - - virtual ~CIUTransformFunction() { SG_UNREF(m_f); } - - /** return the real value of the function at given point - * - * @param x argument of a function - * - * @return \f$f(a+g(x)^2)2g(x)g'(x)\f$, where - * \f$g(x)=\frac{x}{1-x}\f$ and \f$g'(x)=\frac{1}{(1-x)^2}\f$ - */ - virtual float64_t operator() (float64_t x) - { - float64_t hx=1.0/(1.0-x); - float64_t gx=x*hx; - float64_t dgx=CMath::sq(hx); - - return (*m_f)(m_a+CMath::sq(gx))*2*gx*dgx; - } - -private: - /** function \f$f(x)\f$ */ - CFunction* m_f; - - /** lower bound */ - float64_t m_a; -}; - -/** @brief Class of a function, which is used for finite integral - * transformation - * - * \f[ - * \int_{a}^{b}f(x)dx = \int_{-1}^{1} f(g(t))g'(t)dt - * \f] - * - * where \f$g(t)=\frac{b-a}{2}(\frac{t}{2}(3-t^2))+\frac{b+a}{2}\f$ - * and \f$g'(t)=\frac{b-a}{4}(3-3t^2)\f$. - */ -class CTransformFunction : public CFunction -{ -public: - /** constructor - * - * @param f function \f$f(x)\f$ - * @param a lower bound - * @param b upper bound - */ - CTransformFunction(CFunction* f, float64_t a, float64_t b) - { - SG_REF(f); - m_f=f; - m_a=a; - m_b=b; - } - - virtual ~CTransformFunction() { SG_UNREF(m_f); } - - /** return the real value of the function at given point - * - * @param x argument of a function - * - * @return \f$f(g(x))g'(x)\f$, where - * \f$g(t)=\frac{b-a}{2}(\frac{t}{2}(3-t^2))+\frac{b+a}{2}\f$ and - * \f$g'(t)=\frac{b-a}{4}(3-3t^2)\f$ - */ - virtual float64_t operator() (float64_t x) - { - float64_t qw=(m_b-m_a)/4.0; - float64_t gx=qw*(x*(3.0-CMath::sq(x)))+(m_b+m_a)/2.0; - float64_t dgx=qw*3.0*(1.0-CMath::sq(x)); - - return (*m_f)(gx)*dgx; - } - -private: - /** function \f$f(x)\f$ */ - CFunction* m_f; - - /** lower bound */ - float64_t m_a; - - /** upper bound */ - float64_t m_b; -}; - -#endif /* DOXYGEN_SHOULD_SKIP_THIS */ - -float64_t CIntegration::integrate_quadgk(CFunction* f, float64_t a, - float64_t b, float64_t abs_tol, float64_t rel_tol, uint32_t max_iter, - index_t sn) -{ - // check the parameters - REQUIRE(f, "Integrable function should not be NULL\n") - REQUIRE(abs_tol>0.0, "Absolute tolerance must be positive, but is %f\n", - abs_tol) - REQUIRE(rel_tol>0.0, "Relative tolerance must be positive, but is %f\n", - rel_tol) - REQUIRE(max_iter>0, "Maximum number of iterations must be greater than 0, " - "but is %d\n", max_iter) - REQUIRE(sn>0, "Initial number of subintervals must be greater than 0, " - "but is %d\n", sn) - - // integral evaluation function - typedef void TQuadGKEvaluationFunction(CFunction* f, - CDynamicArray* subs, CDynamicArray* q, - CDynamicArray* err); - - TQuadGKEvaluationFunction* evaluate_quadgk; - - CFunction* tf; - float64_t ta; - float64_t tb; - float64_t q_sign; - - // negate integral value and swap a and b, if a>b - if (a>b) - { - ta=b; - tb=a; - q_sign=-1.0; - } - else - { - ta=a; - tb=b; - q_sign=1.0; - } - - // transform integrable function and domain of integration - if (a==-CMath::INFTY && b==CMath::INFTY) - { - tf=new CITransformFunction(f); - evaluate_quadgk=&evaluate_quadgk15; - ta=-1.0; - tb=1.0; - } - else if (a==-CMath::INFTY) - { - tf=new CILTransformFunction(f, b); - evaluate_quadgk=&evaluate_quadgk15; - ta=-1.0; - tb=0.0; - } - else if (b==CMath::INFTY) - { - tf=new CIUTransformFunction(f, a); - evaluate_quadgk=&evaluate_quadgk15; - ta=0.0; - tb=1.0; - } - else - { - tf=new CTransformFunction(f, a, b); - evaluate_quadgk=&evaluate_quadgk21; - ta=-1.0; - tb=1.0; - } - - // compute initial subintervals, by dividing domain [a, b] into sn - // parts - CDynamicArray* subs=new CDynamicArray(); - - // width of each subinterval - float64_t sw=(tb-ta)/sn; - - for (index_t i=0; ipush_back(ta+i*sw); - subs->push_back(ta+(i+1)*sw); - } - - // evaluate integrals on initial subintervals - CDynamicArray* q_subs=new CDynamicArray(); - CDynamicArray* err_subs=new CDynamicArray(); - - evaluate_quadgk(tf, subs, q_subs, err_subs); - - // compute value of integral and error on [a, b] - float64_t q=0.0; - float64_t err=0.0; - - for (index_t i=0; iget_num_elements(); i++) - q+=(*q_subs)[i]; - - for (index_t i=0; iget_num_elements(); i++) - err+=(*err_subs)[i]; - - // evaluate tolerance - float64_t tol=CMath::max(abs_tol, rel_tol*CMath::abs(q)); - - // number of iterations - uint32_t iter=1; - - CDynamicArray* new_subs=new CDynamicArray(); - - while (err>tol && iterget_num_elements()/2; i++) - { - if (CMath::abs((*err_subs)[i])>=tol*CMath::abs((*subs)[2*i+1]- - (*subs)[2*i])/(tb-ta)) - { - // bisect subinterval - float64_t mid=((*subs)[2*i]+(*subs)[2*i+1])/2.0; - - new_subs->push_back((*subs)[2*i]); - new_subs->push_back(mid); - new_subs->push_back(mid); - new_subs->push_back((*subs)[2*i+1]); - - // subtract value of the integral and error on this - // subinterval from total value and error - q-=(*q_subs)[i]; - err-=(*err_subs)[i]; - } - } - - subs->set_array(new_subs->get_array(), new_subs->get_num_elements(), - new_subs->get_num_elements()); - - new_subs->reset_array(); - - // break if no new subintervals - if (!subs->get_num_elements()) - break; - - // evaluate integrals on selected subintervals - evaluate_quadgk(tf, subs, q_subs, err_subs); - - for (index_t i=0; iget_num_elements(); i++) - q+=(*q_subs)[i]; - - for (index_t i=0; iget_num_elements(); i++) - err+=(*err_subs)[i]; - - // evaluate tolerance - tol=CMath::max(abs_tol, rel_tol*CMath::abs(q)); - - iter++; - } - - SG_UNREF(new_subs); - - if (err>tol) - { - SG_SWARNING("Error tolerance not met. Estimated error is equal to %g " - "after %d iterations\n", err, iter) - } - - // clean up - SG_UNREF(subs); - SG_UNREF(q_subs); - SG_UNREF(err_subs); - SG_UNREF(tf); - - return q_sign*q; -} - -float64_t CIntegration::integrate_quadgh(CFunction* f) -{ - SG_REF(f); - - // evaluate integral using Gauss-Hermite 64-point rule - float64_t q=evaluate_quadgh64(f); - - SG_UNREF(f); - - return q; -} - -float64_t CIntegration::integrate_quadgh_customized(CFunction* f, - SGVector xgh, SGVector wgh) -{ - REQUIRE(xgh.vlen == wgh.vlen, - "The length of node array (%d) and weight array (%d) should be the same\n", - xgh.vlen, wgh.vlen); - - SG_REF(f); - - float64_t q=evaluate_quadgh(f, xgh.vlen, xgh.vector, wgh.vector); - - SG_UNREF(f); - - return q; -} - -void CIntegration::evaluate_quadgk(CFunction* f, CDynamicArray* subs, - CDynamicArray* q, CDynamicArray* err, index_t n, - float64_t* xgk, float64_t* wg, float64_t* wgk) -{ - // check the parameters - REQUIRE(f, "Integrable function should not be NULL\n") - REQUIRE(subs, "Array of subintervals should not be NULL\n") - REQUIRE(!(subs->get_array_size()%2), "Size of the array of subintervals " - "should be even\n") - REQUIRE(q, "Array of values of integrals should not be NULL\n") - REQUIRE(err, "Array of errors should not be NULL\n") - REQUIRE(n%2, "Order of Gauss-Kronrod should be odd\n") - REQUIRE(xgk, "Gauss-Kronrod nodes should not be NULL\n") - REQUIRE(wgk, "Gauss-Kronrod weights should not be NULL\n") - REQUIRE(wg, "Gauss weights should not be NULL\n") - - // create eigen representation of subs, xgk, wg, wgk - Map eigen_subs(subs->get_array(), 2, subs->get_num_elements()/2); - Map eigen_xgk(xgk, n); - Map eigen_wg(wg, n/2); - Map eigen_wgk(wgk, n); - - // compute half width and centers of each subinterval - VectorXd eigen_hw=(eigen_subs.row(1)-eigen_subs.row(0))/2.0; - VectorXd eigen_center=eigen_subs.colwise().sum()/2.0; - - // compute Gauss-Kronrod nodes x for each subinterval: x=hw*xgk+center - MatrixXd x=eigen_hw*eigen_xgk.adjoint()+eigen_center* - (VectorXd::Ones(n)).adjoint(); - - // compute ygk=f(x) - MatrixXd ygk(x.rows(), x.cols()); - - for (index_t i=0; iset_array(eigen_q.data(), eigen_q.size()); - - // choose function values for Gauss nodes - MatrixXd yg(ygk.rows(), ygk.cols()/2); - - for (index_t i=1, j=0; iset_array(eigen_err.data(), eigen_err.size()); -} - -void CIntegration::generate_gauher(SGVector xgh, SGVector wgh) -{ - REQUIRE(xgh.vlen == wgh.vlen, - "The length of node array (%d) and weight array (%d) should be the same\n", - xgh.vlen, wgh.vlen); - - index_t n = xgh.vlen; - - if (n == 20) - { - generate_gauher20(xgh, wgh); - } - else - { - Map eigen_xgh(xgh.vector, xgh.vlen); - Map eigen_wgh(wgh.vector, wgh.vlen); - - eigen_xgh = MatrixXd::Zero(n,1); - eigen_wgh = MatrixXd::Ones(n,1); - - if (n > 1) - { - MatrixXd v = MatrixXd::Zero(n,n); - - //b = sqrt( (1:N-1)/2 )'; - //[V,D] = eig( diag(b,1) + diag(b,-1) ); - v.block(0, 1, n-1, n-1).diagonal() = (0.5*ArrayXd::LinSpaced(n-1,1,n-1)).sqrt(); - v.block(1, 0, n-1, n-1).diagonal() = v.block(0, 1, n-1, n-1).diagonal(); - EigenSolver eig(v); - - //w = V(1,:)'.^2 - eigen_wgh = eig.eigenvectors().row(0).transpose().real().array().pow(2); - - //x = sqrt(2)*diag(D) - eigen_xgh = eig.eigenvalues().real()*sqrt(2.0); - } - } -} - -void CIntegration::generate_gauher20(SGVector xgh, SGVector wgh) -{ - REQUIRE(xgh.vlen == wgh.vlen, - "The length of node array (%d) and weight array (%d) should be the same\n", - xgh.vlen, wgh.vlen); - REQUIRE(xgh.vlen == 20, "The length of xgh and wgh should be 20\n"); - - static const index_t n = 20; - static float64_t wgh_pre[n]= - { - 0.0000000000001257800672437920121938444754, - 0.0000000002482062362315158465220083577413, - 0.0000000612749025998290679114578012251502, - 0.0000044021210902308611768963750310312832, - 0.0001288262799619289543807260089991473251, - 0.0018301031310804880686271545187082665507, - 0.0139978374471010288959682554832397727296, - 0.0615063720639768204967445797137770568952, - 0.1617393339840000332507941038784338161349, - 0.2607930634495548849471902030927594751120, - 0.2607930634495547739248877405771054327488, - 0.1617393339840003108065502601675689220428, - 0.0615063720639767788633811562704067910090, - 0.0139978374471010080792865437615546397865, - 0.0018301031310804856833823750505985117343, - 0.0001288262799619298488475183095403053812, - 0.0000044021210902308865878847926600414553, - 0.0000000612749025998294252534824241331057, - 0.0000000002482062362315177593771748866178, - 0.0000000000001257800672437921636551382778 - }; - - static float64_t xgh_pre[n]= - { - -7.6190485416797573137159815814811736345291, - -6.5105901570136559541879250900819897651672, - -5.5787388058932032564030123467091470956802, - -4.7345813340460569662582201999612152576447, - -3.9439673506573176275935566081898286938667, - -3.1890148165533904744961546384729444980621, - -2.4586636111723669806394809711491689085960, - -1.7452473208141270344384565760265104472637, - -1.0429453488027506935509336472023278474808, - -0.3469641570813560282893206476728664711118, - 0.3469641570813561393116231101885205134749, - 1.0429453488027513596847484222962521016598, - 1.7452473208141265903492467259638942778111, - 2.4586636111723669806394809711491689085960, - 3.1890148165533904744961546384729444980621, - 3.9439673506573162953259270580019801855087, - 4.7345813340460569662582201999612152576447, - 5.5787388058932014800461729464586824178696, - 6.5105901570136532896526659897062927484512, - 7.6190485416797573137159815814811736345291 - - }; - - for (index_t idx = 0; idx < n; idx++) - { - wgh[idx] = wgh_pre[idx]; - xgh[idx] = xgh_pre[idx]; - } - -} - -void CIntegration::evaluate_quadgk15(CFunction* f, CDynamicArray* subs, - CDynamicArray* q, CDynamicArray* err) -{ - static const index_t n=15; - - // Gauss-Kronrod nodes - static float64_t xgk[n]= - { - -0.991455371120812639206854697526329, - -0.949107912342758524526189684047851, - -0.864864423359769072789712788640926, - -0.741531185599394439863864773280788, - -0.586087235467691130294144838258730, - -0.405845151377397166906606412076961, - -0.207784955007898467600689403773245, - 0.000000000000000000000000000000000, - 0.207784955007898467600689403773245, - 0.405845151377397166906606412076961, - 0.586087235467691130294144838258730, - 0.741531185599394439863864773280788, - 0.864864423359769072789712788640926, - 0.949107912342758524526189684047851, - 0.991455371120812639206854697526329 - }; - - // Gauss weights - static float64_t wg[n/2]= - { - 0.129484966168869693270611432679082, - 0.279705391489276667901467771423780, - 0.381830050505118944950369775488975, - 0.417959183673469387755102040816327, - 0.381830050505118944950369775488975, - 0.279705391489276667901467771423780, - 0.129484966168869693270611432679082 - }; - - // Gauss-Kronrod weights - static float64_t wgk[n]= - { - 0.022935322010529224963732008058970, - 0.063092092629978553290700663189204, - 0.104790010322250183839876322541518, - 0.140653259715525918745189590510238, - 0.169004726639267902826583426598550, - 0.190350578064785409913256402421014, - 0.204432940075298892414161999234649, - 0.209482141084727828012999174891714, - 0.204432940075298892414161999234649, - 0.190350578064785409913256402421014, - 0.169004726639267902826583426598550, - 0.140653259715525918745189590510238, - 0.104790010322250183839876322541518, - 0.063092092629978553290700663189204, - 0.022935322010529224963732008058970 - }; - - // evaluate definite integral on each subinterval using Gauss-Kronrod rule - evaluate_quadgk(f, subs, q, err, n, xgk, wg, wgk); -} - -void CIntegration::evaluate_quadgk21(CFunction* f, CDynamicArray* subs, - CDynamicArray* q, CDynamicArray* err) -{ - static const index_t n=21; - - // Gauss-Kronrod nodes - static float64_t xgk[n]= - { - -0.995657163025808080735527280689003, - -0.973906528517171720077964012084452, - -0.930157491355708226001207180059508, - -0.865063366688984510732096688423493, - -0.780817726586416897063717578345042, - -0.679409568299024406234327365114874, - -0.562757134668604683339000099272694, - -0.433395394129247190799265943165784, - -0.294392862701460198131126603103866, - -0.148874338981631210884826001129720, - 0.000000000000000000000000000000000, - 0.148874338981631210884826001129720, - 0.294392862701460198131126603103866, - 0.433395394129247190799265943165784, - 0.562757134668604683339000099272694, - 0.679409568299024406234327365114874, - 0.780817726586416897063717578345042, - 0.865063366688984510732096688423493, - 0.930157491355708226001207180059508, - 0.973906528517171720077964012084452, - 0.995657163025808080735527280689003 - }; - - // Gauss weights - static float64_t wg[n/2]= - { - 0.066671344308688137593568809893332, - 0.149451349150580593145776339657697, - 0.219086362515982043995534934228163, - 0.269266719309996355091226921569469, - 0.295524224714752870173892994651338, - 0.295524224714752870173892994651338, - 0.269266719309996355091226921569469, - 0.219086362515982043995534934228163, - 0.149451349150580593145776339657697, - 0.066671344308688137593568809893332 - }; - - // Gauss-Kronrod weights - static float64_t wgk[n]= - { - 0.011694638867371874278064396062192, - 0.032558162307964727478818972459390, - 0.054755896574351996031381300244580, - 0.075039674810919952767043140916190, - 0.093125454583697605535065465083366, - 0.109387158802297641899210590325805, - 0.123491976262065851077958109831074, - 0.134709217311473325928054001771707, - 0.142775938577060080797094273138717, - 0.147739104901338491374841515972068, - 0.149445554002916905664936468389821, - 0.147739104901338491374841515972068, - 0.142775938577060080797094273138717, - 0.134709217311473325928054001771707, - 0.123491976262065851077958109831074, - 0.109387158802297641899210590325805, - 0.093125454583697605535065465083366, - 0.075039674810919952767043140916190, - 0.054755896574351996031381300244580, - 0.032558162307964727478818972459390, - 0.011694638867371874278064396062192 - }; - - evaluate_quadgk(f, subs, q, err, n, xgk, wg, wgk); -} - -float64_t CIntegration::evaluate_quadgh(CFunction* f, index_t n, float64_t* xgh, - float64_t* wgh) -{ - // check the parameters - REQUIRE(f, "Integrable function should not be NULL\n"); - REQUIRE(xgh, "Gauss-Hermite nodes should not be NULL\n"); - REQUIRE(wgh, "Gauss-Hermite weights should not be NULL\n"); - - float64_t q=0.0; - - for (index_t i=0; i - - -#include -#include -#include -#include - -namespace shogun -{ -template class SGVector; - -/** @brief Class that contains certain methods related to numerical - * integration - */ -class CIntegration : public CSGObject -{ -public: - /** numerically evaluate definite integral \f$\int_a^b f(x) dx\f$, - * where \f$f(x)\f$ - function of one variable, using adaptive - * Gauss-Kronrod quadrature formula - * - * \f[ - * \int_a^b f(x)\dx \approx \sum_{i=1}^n w_i f(x_i) - * \f] - * - * where x_i and w_i - Gauss-Kronrod nodes and weights - * respectively. - * - * This function applies the Gauss-Kronrod 21-point integration - * rule for finite bounds \f$[a, b]\f$ and 15-point rule for - * infinite ones. - * - * Based on ideas form GNU Octave (file quadgk.m) under GPLv3. - * - * @param f integrable function of one variable - * @param a lower bound of the domain of integration - * @param b upper bound of the domain of integration - * @param abs_tol absolute tolerance of the quadrature - * @param rel_tol relative tolerance of the quadrature - * @param max_iter maximum number of iterations of the method - * @param sn initial number of subintervals - * - * @return approximate value of definite integral of the function - * on given domain - */ - static float64_t integrate_quadgk(CFunction* f, float64_t a, - float64_t b, float64_t abs_tol=1e-10, float64_t rel_tol=1e-5, - uint32_t max_iter=1000, index_t sn=10); - - /** numerically evaluate integral of the following kind - * - * \f[ - * \int_{-\infty}^{\infty}e^{-x^2}f(x)dx - * \f] - * - * using 64-point Gauss-Hermite rule - * - * \f[ - * \int_{-\infty}^{\infty}e^{-x^2}f(x)dx \approx - * \sum_{i=1}^{64} w_if(x_i) - * \f] - * - * where x_i and w_i - ith node and weight for the 64-point - * Gauss-Hermite formula respectively. - * - * @param f integrable function of one variable - * - * @return approximate value of the - * integral \f$\int_{-\infty}^{\infty}e^{-x^2}f(x)dx\f$ - */ - static float64_t integrate_quadgh(CFunction* f); - - /** numerically evaluate integral of the following kind - * - * \f[ - * \int_{-\infty}^{\infty}e^{-x^2}f(x)dx - * \f] - * - * using provided Gauss-Hermite points - * - * \f[ - * \int_{-\infty}^{\infty}e^{-x^2}f(x)dx \approx - * \sum_{i=1}^{64} w_if(x_i) - * \f] - * - * where x_i and w_i - ith node and weight for the provided - * Gauss-Hermite formula respectively. - * - * @param f integrable function of one variable - * @param xgh the provided array of nodes - * @param wgh the provided array of weights - * - * @return approximate value of the - * integral \f$\int_{-\infty}^{\infty}e^{-x^2}f(x)dx\f$ - */ - - static float64_t integrate_quadgh_customized(CFunction* f, - SGVector xgh, SGVector wgh); - - - /** generate Gauss-Hermite nodes - * - * Adapted form Gaussian Process Machine Learning Toolbox - * (file util/gauher.m) - * - * @param xgh nodes are saved in this pre-allocated array - * @param wgh weights are saved in this pre-allocated array - * - */ - static void generate_gauher(SGVector xgh, SGVector wgh); - - - /** generate 20 Gauss-Hermite nodes using precomputed result - * - * Adapted form Gaussian Process Machine Learning Toolbox - * (file util/gauher.m) - * - * @param xgh nodes are saved in this pre-allocated array - * @param wgh weights are saved in this pre-allocated array - * - */ - static void generate_gauher20(SGVector xgh, SGVector wgh); - - /** get object name - * - * @return name Integration - */ - virtual const char* get_name() const { return "Integration"; } - -private: - /** evaluate definite integral of a function and error on each - * subinterval using Gauss-Kronrod quadrature formula of order n - * - * Adapted form GNU Octave (file quadgk.m) under GPLv3. - * - * @param f integrable function of one variable - * @param subs subintervals of integration - * @param q approximate value of definite integral of the function - * on each subinterval - * @param err error on each subinterval - * @param n order of the Gauss-Kronrod rule - * @param xgk Gauss-Kronrod nodes - * @param wg Gauss weights - * @param wgk Gauss-Kronrod weights - */ - static void evaluate_quadgk(CFunction* f, CDynamicArray* subs, - CDynamicArray* q, CDynamicArray* err, index_t n, - float64_t* xgk, float64_t* wg, float64_t* wgk); - - /** evaluate definite integral of a function and error on each - * subinterval using Gauss-Kronrod quadrature formula of order 15. - * - * Gauss-Kronrod nodes, Gauss weights and Gauss-Kronrod weights - * are precomputed. - * - * The abscissae and weights for 15-point rule are taken from from - * QUADPACK (file dqk15.f). - * - * @param f integrable function of one variable - * @param subs subintervals of integration - * @param q approximate value of definite integral of the function - * on each subinterval - * @param err error on each subinterval - */ - static void evaluate_quadgk15(CFunction* f, CDynamicArray* subs, - CDynamicArray* q, CDynamicArray* err); - - /** evaluate definite integral of a function and error on each - * subinterval using Gauss-Kronrod quadrature formula of order 21. - * - * Gauss-Kronrod nodes, Gauss weights and Gauss-Kronrod weights - * are precomputed. - * - * The abscissae and weights for 21-point rule are taken from - * QUADPACK (file dqk21.f). - * - * @param f integrable function of one variable - * @param subs subintervals of integration - * @param q approximate value of definite integral of the function - * on each subinterval - * @param err error on each subinterval - */ - static void evaluate_quadgk21(CFunction* f, CDynamicArray* subs, - CDynamicArray* q, CDynamicArray* err); - - /** evaluate integral \f$\int_{-\infty}^{\infty}e^{-x^2}f(x)dx\f$ - * using Gauss-Hermite quadrature formula of order n - * - * @param f integrable function of one variable - * @param n order of the Gauss-Hermite rule - * @param xh Gauss-Hermite nodes - * @param wh Gauss-Hermite weights - * - * @return approximate value of the integral - * \f$\int_{-\infty}^{\infty}e^{-x^2}f(x)dx\f$ - */ - static float64_t evaluate_quadgh(CFunction* f, index_t n, float64_t* xh, - float64_t* wh); - - /** evaluate integral \f$\int_{-\infty}^{\infty}e^{-x^2}f(x)dx\f$ - * using Gauss-Hermite quadrature formula of order 64. - * - * Gauss-Hermite nodes \f$x_i\f$ and weights \f$w_i\f$ are - * precomputed: \f$x_i\f$ - the i-th zero of \f$H_n(x)\f$, - * \f$w_i=\frac{2^{n-1}n!\sqrt{\pi}}{n^2[H_{n-1}(x_i)]^2}\f$, - * where \f$H_n(x)\f$ is physicists' Hermite polynomials. - * - * @param f integrable function of one variable - * - * @return approximate value of the integral - * \f$\int_{-\infty}^{\infty}e^{-x^2}f(x)dx\f$ - */ - static float64_t evaluate_quadgh64(CFunction* f); -}; -} -#endif /* _INTEGRATION_H_ */ diff --git a/src/shogun/mathematics/JacobiEllipticFunctions.cpp b/src/shogun/mathematics/JacobiEllipticFunctions.cpp deleted file mode 100644 index 8f3cec3e8ef..00000000000 --- a/src/shogun/mathematics/JacobiEllipticFunctions.cpp +++ /dev/null @@ -1,136 +0,0 @@ -/* - * This program is free software; you can redistribute it and/or modify - * it under the terms of the GNU General Public License as published by - * the Free Software Foundation; either version 3 of the License, or - * (at your option) any later version. - * - * Written (W) 2013 Soumyajit De - * - * KRYLSTAT Copyright 2011 by Erlend Aune under GPL2+ - * (few parts rewritten and adjusted for shogun) - */ - -#include -#include -#include - -using namespace shogun; - -void CJacobiEllipticFunctions::ellipKKp(Real L, Real &K, Real &Kp) -{ - REQUIRE(L>=0.0, - "CJacobiEllipticFunctions::ellipKKp(): \ - Parameter L should be non-negative\n"); -#if defined HAVE_ARPREC && defined USE_GPL_SHOGUN - const Real eps=Real(std::numeric_limits::epsilon()); - const Real pi=mp_real::_pi; -#else - const Real eps=std::numeric_limits::epsilon(); - const Real pi=M_PI; -#endif //(HAVE_ARPREC && USE_GPL_SHOGUN) - if (L>10.0) - { - K=pi*0.5; - Kp=pi*L+log(4.0); - } - else - { - Real m=exp(-2.0*pi*L); - Real mp=1.0-m; - if (m::max()); - } - else if (mp::max()); - Kp=compute_quarter_period(sqrt(m)); - } - else - { - K=compute_quarter_period(sqrt(mp)); - Kp=compute_quarter_period(sqrt(m)); - } - } -} - -void CJacobiEllipticFunctions - ::ellipJC(Complex u, Real m, Complex &sn, Complex &cn, Complex &dn) -{ - REQUIRE(m>=0.0 && m<=1.0, - "CJacobiEllipticFunctions::ellipJC(): \ - Parameter m should be >=0 and <=1\n"); - -#if defined HAVE_ARPREC && defined USE_GPL_SHOGUN - const Real eps=sqrt(mp_real::_eps); -#else - const Real eps=sqrt(std::numeric_limits::epsilon()); -#endif //(HAVE_ARPREC && USE_GPL_SHOGUN) - if (m>=(1.0-eps)) - { -#if defined HAVE_ARPREC && defined USE_GPL_SHOGUN - complex128_t _u(dble(u.real),dble(u.imag)); - complex128_t t=CMath::tanh(_u); - complex128_t b=CMath::cosh(_u); - complex128_t twon=b*CMath::sinh(_u); - complex128_t ai=0.25*(1.0-dble(m)); - complex128_t _sn=t+ai*(twon-_u)/(b*b); - complex128_t phi=1.0/b; - complex128_t _cn=phi-ai*(twon-_u); - complex128_t _dn=phi+ai*(twon+_u); - sn=mp_complex(_sn.real(),_sn.imag()); - cn=mp_complex(_cn.real(),_cn.imag()); - dn=mp_complex(_dn.real(),_dn.imag()); -#else - Complex t=CMath::tanh(u); - Complex b=CMath::cosh(u); - Complex ai=0.25*(1.0-m); - Complex twon=b*CMath::sinh(u); - sn=t+ai*(twon-u)/(b*b); - Complex phi=Real(1.0)/b; - ai*=t*phi; - cn=phi-ai*(twon-u); - dn=phi+ai*(twon+u); -#endif //(HAVE_ARPREC && USE_GPL_SHOGUN) - } - else - { - const Real prec=4.0*eps; - const index_t MAX_ITER=128; - index_t i=0; - Real kappa[MAX_ITER]; - - while (iprec) - { - Real k; - if (m>0.001) - { - Real mp=sqrt(1.0-m); - k=(1.0-mp)/(1.0+mp); - } - else - k=poly_six(m/4.0); - u/=(1.0+k); - m=k*k; - kappa[i++]=k; - } - Complex sin_u=sin(u); - Complex cos_u=cos(u); - Complex t=Real(0.25*m)*(u-sin_u*cos_u); - sn=sin_u-t*cos_u; - cn=cos_u+t*sin_u; - dn=Real(1.0)+Real(0.5*m)*(cos_u*cos_u); - - i--; - while (i>=0) - { - Real k=kappa[i--]; - Complex ksn2=k*(sn*sn); - Complex d=Real(1.0)+ksn2; - sn*=(1.0+k)/d; - cn*=dn/d; - dn=(Real(1.0)-ksn2)/d; - } - } -} diff --git a/src/shogun/mathematics/JacobiEllipticFunctions.h b/src/shogun/mathematics/JacobiEllipticFunctions.h deleted file mode 100644 index 5a469c7d879..00000000000 --- a/src/shogun/mathematics/JacobiEllipticFunctions.h +++ /dev/null @@ -1,163 +0,0 @@ -/* - * This program is free software; you can redistribute it and/or modify - * it under the terms of the GNU General Public License as published by - * the Free Software Foundation; either version 3 of the License, or - * (at your option) any later version. - * - * Written (W) 2013 Soumyajit De - * - * KRYLSTAT Copyright 2011 by Erlend Aune under GPL2+ - * (few parts rewritten and adjusted for shogun) - * - * NOTE: For higher precision, the methods in this class rely on an external - * library, ARPREC (http://crd-legacy.lbl.gov/~dhbailey/mpdist/), in absense of - * which they fallback to shogun datatypes. To use it with shogun, configure - * ARPREC with `CXX="c++ -fPIC" ./configure' in order to link. - */ - -#ifndef JACOBI_ELLIPTIC_FUNCTIONS_H_ -#define JACOBI_ELLIPTIC_FUNCTIONS_H_ - -#include -#include -#include -#include - -#if defined HAVE_ARPREC && defined USE_GPL_SHOGUN -#include -#include -#endif //(HAVE_ARPREC && USE_GPL_SHOGUN) - -namespace shogun -{ - -/** @brief Class that contains methods for computing Jacobi elliptic functions - * related to complex analysis. These functions are inverse of the elliptic - * integral of first kind, i.e. - * \f[ - * u(k,m)=\int_{0}^{k}\frac{dt}{\sqrt{(1-t^{2})(1-m^{2}t^{2})}} - * =\int_{0}^{\varphi}\frac{d\theta}{\sqrt{(1-m^{2}sin^{2}\theta)}} - * \f] - * where \f$k=sin\varphi\f$, \f$t=sin\theta\f$ and parameter \f$m, 0\le m - * \le 1\f$ is called modulus. Three main Jacobi elliptic functions are defined - * as \f$sn(u,m)=k=sin\theta\f$, \f$cn(u,m)=cos\theta=\sqrt{1-sn(u,m)^{2}}\f$ - * and \f$dn(u,m)=\sqrt{1-m^{2}sn(u,m)^{2}}\f$. - * For \f$k=1\f$, i.e. \f$\varphi=\frac{\pi}{2}\f$, \f$u(1,m)=K(m)\f$ is known - * as the complete elliptic integral of first kind. Similarly, \f$u(1,m'))= - * K'(m')\f$, \f$m'=\sqrt{1-m^{2}}\f$ is called the complementary complete - * elliptic integral of first kind. Jacobi functions are double periodic with - * quardratic periods \f$K\f$ and \f$K'\f$. - * - * This class provides two sets of methods for computing \f$K,K'\f$, and - * \f$sn,cn,dn\f$. Useful for computing rational approximation of matrix - * functions given by Cauchy's integral formula, etc. - */ -class CJacobiEllipticFunctions: public CSGObject -{ -#if defined HAVE_ARPREC && defined USE_GPL_SHOGUN - typedef mp_real Real; - typedef mp_complex Complex; -#else - typedef float64_t Real; - typedef complex128_t Complex; -#endif //(HAVE_ARPREC && USE_GPL_SHOGUN) -private: - static inline Real compute_quarter_period(Real b) - { -#if defined HAVE_ARPREC && defined USE_GPL_SHOGUN - const Real eps=mp_real::_eps; - const Real pi=mp_real::_pi; -#else - const Real eps=std::numeric_limits::epsilon(); - const Real pi=M_PI; -#endif //(HAVE_ARPREC && USE_GPL_SHOGUN) - Real a=1.0; - Real mm=1.0; - - int64_t p=2; - do - { - Real a_new=(a+b)*0.5; - Real b_new=sqrt(a*b); - Real c=(a-b)*0.5; - mm=Real(p)*c*c; - p<<=1; - a=a_new; - b=b_new; - } while (mm>eps); - return pi*0.5/a; - } - - static inline Real poly_six(Real x) - { - return (132*pow(x,6)+42*pow(x,5)+14*pow(x,4)+5*pow(x,3)+2*pow(x,2)+x); - } - -public: - /** Computes the quarter periods (K and K') of Jacobian elliptic functions - * (see class description). - * @param L - * @param K the quarter period (to be computed) on the Real axis - * @param Kp the quarter period (to be computed) on the Imaginary axis - * computed - */ - static void ellipKKp(Real L, Real &K, Real &Kp); - - /** Computes three main Jacobi elliptic functions, \f$sn(u,m)\f$, - * \f$cn(u,m)\f$ and \f$dn(u,m)\f$ (see class description). - * @param u the elliptic integral of the first kind \f$u(k,m)\f$ - * @param m the modulus parameter, \f$0\le m \le 1\f$ - * @param sn Jacobi elliptic function sn(u,m) - * @param cn Jacobi elliptic function cn(u,m) - * @param dn Jacobi elliptic function dn(u,m) - */ - static void ellipJC(Complex u, Real m, Complex &sn, Complex &cn, - Complex &dn); - -#if defined HAVE_ARPREC && defined USE_GPL_SHOGUN - /** Wrapper method for ellipKKp if ARPREC is present (for high precision) - * @param L - * @param K the quarter period (to be computed) on the Real axis - * @param Kp the quarter period (to be computed) on the Imaginary axis - * computed - */ - static void ellipKKp(float64_t L, float64_t &K, float64_t &Kp) - { - mp::mp_init(100, NULL, true); - mp_real _K, _Kp; - ellipKKp(mp_real(L), _K, _Kp); - K=dble(_K); - Kp=dble(_Kp); - mp::mp_finalize(); - } - - /** Wrapper method for ellipJC if ARPREC is present (for high precision) - * @param u the elliptic integral of the first kind \f$u(k,m)\f$ - * @param m the modulus parameter, \f$0\le m \le 1\f$ - * @param sn Jacobi elliptic function sn(u,m) - * @param cn Jacobi elliptic function cn(u,m) - * @param dn Jacobi elliptic function dn(u,m) - */ - static void ellipJC(complex128_t u, float64_t m, - complex128_t &sn, complex128_t &cn, complex128_t &dn) - { - mp::mp_init(100, NULL, true); - mp_complex _sn, _cn, _dn; - ellipJC(mp_complex(u.real(),u.imag()), mp_real(m), _sn, _cn, _dn); - sn=complex128_t(dble(_sn.real),dble(_sn.imag)); - cn=complex128_t(dble(_cn.real),dble(_cn.imag)); - dn=complex128_t(dble(_dn.real),dble(_dn.imag)); - mp::mp_finalize(); - } -#endif //(HAVE_ARPREC && USE_GPL_SHOGUN) - - /** @return object name */ - virtual const char* get_name() const - { - return "JacobiEllipticFunctions"; - } -}; - -} - -#endif /* JACOBI_ELLIPTIC_FUNCTIONS_H_ */ diff --git a/src/shogun/mathematics/Math.cpp b/src/shogun/mathematics/Math.cpp index adf58828ab0..604387f8ea2 100644 --- a/src/shogun/mathematics/Math.cpp +++ b/src/shogun/mathematics/Math.cpp @@ -81,24 +81,6 @@ CMath::~CMath() #endif } -float64_t CMath::dot(const float64_t* v1, const float64_t* v2, int32_t n) -{ - float64_t r=0; - Eigen::Map ev1(v1,n); - Eigen::Map ev2(v2,n); - r = ev1.dot(ev2); - return r; -} - -float32_t CMath::dot(const float32_t* v1, const float32_t* v2, int32_t n) -{ - float32_t r=0; - Eigen::Map ev1(v1,n); - Eigen::Map ev2(v2,n); - r = ev1.dot(ev2); - return r; -} - #ifdef USE_LOGCACHE int32_t CMath::determine_logrange() { diff --git a/src/shogun/mathematics/Math.h b/src/shogun/mathematics/Math.h index 021b4bb214a..401e7b7d326 100644 --- a/src/shogun/mathematics/Math.h +++ b/src/shogun/mathematics/Math.h @@ -140,6 +140,7 @@ class CMath : public CSGObject virtual ~CMath(); //@} +#ifndef SWIG // SWIG should skip this part /**@name min/max/abs functions. */ //@{ @@ -149,10 +150,10 @@ class CMath : public CSGObject * @param b second value * @return minimum value amongst a and b */ - template + template ::value>::type> static inline T min(T a, T b) { - return (a<=b) ? a : b; + return std::min(a, b); } /** Returns the greatest element amongst two input values @@ -160,11 +161,12 @@ class CMath : public CSGObject * @param b second value * @return maximum value amongst a and b */ - template + template ::value>::type> static inline T max(T a, T b) { - return (a>=b) ? a : b; + return std::max(a, b); } +#endif /** Returns the absolute value of a number, that is * if a>0, output is a; if a<0 ,output is -a @@ -205,12 +207,7 @@ class CMath : public CSGObject static T min(T* vec, int32_t len) { ASSERT(len>0) - T minv=vec[0]; - - for (int32_t i=1; i0) - T maxv=vec[0]; - - for (int32_t i=1; i + template ::value>::type> static inline T clamp(T value, T lb, T ub) { if (value<=lb) @@ -254,7 +247,7 @@ class CMath : public CSGObject * @param maxv_ptr pointer to store the maximum value * @return index of the maximum value */ - template + template ::value>::type> static int32_t arg_max(T * vec, int32_t inc, int32_t len, T * maxv_ptr = NULL) { ASSERT(len > 0 || inc > 0) @@ -281,7 +274,7 @@ class CMath : public CSGObject * @param minv_ptr pointer to store the minimum value * @return index of the minimum value */ - template + template ::value>::type> static int32_t arg_min(T * vec, int32_t inc, int32_t len, T * minv_ptr = NULL) { ASSERT(len > 0 || inc > 0) @@ -310,7 +303,7 @@ class CMath : public CSGObject * @param eps threshold for values to be equal/different * @return true if values are equal within eps accuracy, false if not. */ - template + template ::value>::type> static inline bool fequals_abs(const T& a, const T& b, const float64_t eps) { @@ -327,14 +320,13 @@ class CMath : public CSGObject * @param tolerant allows linient check on float equality (within accuracy) * @return true if values are equal within eps accuracy, false if not. */ - template + template ::value>::type> static inline bool fequals(const T& a, const T& b, const float64_t eps, bool tolerant=false) { const T absA = CMath::abs(a); const T absB = CMath::abs(b); const T diff = CMath::abs((a-b)); - T comp; // Handle this separately since NAN is unordered if (CMath::is_nan((float64_t)a) && CMath::is_nan((float64_t)b)) @@ -345,11 +337,7 @@ class CMath : public CSGObject return CMath::fequals_abs(a, b, eps); // handles float32_t and float64_t separately - if (sizeof(T) == 4) - comp = CMath::F_MIN_NORM_VAL32; - - else - comp = CMath::F_MIN_NORM_VAL64; + T comp = (std::is_same::value) ? CMath::F_MIN_NORM_VAL32 : CMath::F_MIN_NORM_VAL64; if (a == b) return true; @@ -365,6 +353,7 @@ class CMath : public CSGObject return (check < eps); } } +#endif /* Get the corresponding absolute tolerance for unit test given a relative tolerance * @@ -618,140 +607,6 @@ class CMath : public CSGObject return ::exp((double) x); } - /// Compute dot product between v1 and v2 (blas optimized) - static inline float64_t dot(const bool* v1, const bool* v2, int32_t n) - { - float64_t r=0; - for (int32_t i=0; i + template ::value>::type> static float64_t* linspace(T start, T end, int32_t n) { float64_t* output = SG_MALLOC(float64_t, n); @@ -1241,6 +1097,7 @@ class CMath : public CSGObject return output; } +#endif /** Returns a vector with n linearly spaced elements between start and end. * @param start beginning of the interval to divide @@ -1325,12 +1182,13 @@ class CMath : public CSGObject */ static void sort(float64_t *a, int32_t*idx, int32_t N); +#ifndef SWIG // SWIG should skip this part /** Performs a quicksort on an array output of length size * it is sorted from in ascending (for type T) * @param output array to be sorted * @param size size of array */ - template + template ::value>::type> static void qsort(T* output, int32_t size) { if (size<=1) @@ -1375,7 +1233,7 @@ class CMath : public CSGObject * @param output array to be sorted * @param size size of array */ - template + template ::value>::type> static void insertion_sort(T* output, int32_t size) { for (int32_t i=0; i + template ::value>::type> inline static void radix_sort(T* array, int32_t size) { radix_sort_helper(array,size,0); } +#endif /** Extract the byte at position p (from left) * of a 64 bit integer. The function is somewhat identical to @@ -1531,13 +1390,14 @@ class CMath : public CSGObject SG_SERROR("CMath::radix_sort_helper():: Not supported for complex128_t\n"); } +#ifndef SWIG // SWIG should skip this part /** Performs a quicksort on an array of pointers. * It is sorted from in ascending (for type T) * Every element is dereferenced once before being compared * @param vector array of pointers to sort * @param length length of array */ - template + template ::value>::type> static void qsort(T** vector, index_t length) { if (length<=1) @@ -1576,20 +1436,15 @@ class CMath : public CSGObject qsort(&vector[left],length-left); } - /// qsort not implemented for complex128_t - static void qsort(complex128_t** vector, index_t length) - { - SG_SERROR("CMath::qsort():: Not supported for complex128_t\n"); - } - /** Quicksort the vector in ascending order (for type T) * @param vector vector to be sorted */ - template + template ::value>::type> static void qsort(SGVector vector) { qsort(vector, vector.size()); } +#endif /** Helper functor for the function argsort */ template @@ -1609,6 +1464,7 @@ class CMath : public CSGObject const T* data; }; +#ifndef SWIG // SWIG should skip this part /** Get sorted index. * * idx = v.argsort() is similar to Matlab [~, idx] = sort(v) @@ -1616,7 +1472,7 @@ class CMath : public CSGObject * @param vector vector to be sorted * @return sorted index for this vector */ - template + template::value>::type> static SGVector argsort(SGVector vector) { IndexSorter cmp(&vector); @@ -1634,7 +1490,7 @@ class CMath : public CSGObject * @param vector input vector * @return true if vector is sorted, false otherwise */ - template + template ::value>::type> static bool is_sorted(SGVector vector) { if (vector.size() < 2) @@ -1648,6 +1504,7 @@ class CMath : public CSGObject return true; } +#endif /** Display bits (useful for debugging) * @param word input to be displayed as bits @@ -2338,23 +2195,6 @@ void CMath::min(float64_t* output, T* index, int32_t size) swap(index[0], index[min_index]); } -/// linspace not implemented for complex128_t, returns null instead -template <> -inline float64_t* CMath::linspace(complex128_t start, complex128_t end, int32_t n) -{ - SG_SERROR("SGVector::linspace():: Not supported for complex128_t\n"); - return NULL; -} - -#define COMPLEX128_ERROR_ONEVECARG_RETURNS_T(function, return_type, return_statement) \ -template <> \ -inline return_type CMath::function(SGVector vector) \ -{ \ - SG_SERROR("CMath::%s():: Not supported for complex128_t\n", \ - #function); \ - return_statement; \ -} - #define COMPLEX128_ERROR_ONEARG_T(function) \ template <> \ inline complex128_t CMath::function(complex128_t a) \ @@ -2364,83 +2204,11 @@ inline complex128_t CMath::function(complex128_t a) \ return complex128_t(0.0, 0.0); \ } -#define COMPLEX128_ERROR_TWOARGS_T(function) \ -template <> \ -inline complex128_t CMath::function(complex128_t a, complex128_t b) \ -{ \ - SG_SERROR("CMath::%s():: Not supported for complex128_t\n",\ - #function);\ - return complex128_t(0.0, 0.0); \ -} - -#define COMPLEX128_ERROR_THREEARGS_T(function) \ -template <> \ -inline complex128_t CMath::function(complex128_t a, complex128_t b, complex128_t c) \ -{ \ - SG_SERROR("CMath::%s():: Not supported for complex128_t\n",\ - #function);\ - return complex128_t(0.0, 0.0); \ -} - -#define COMPLEX128_ERROR_SORT_T(function) \ -template <> \ -inline void CMath::function(complex128_t* output, int32_t b) \ -{ \ - SG_SERROR("CMath::%s():: Not supported for complex128_t\n",\ - #function);\ -} - -#define COMPLEX128_ERROR_ARG_MAX_MIN(function) \ -template <> \ -inline int32_t CMath::function(complex128_t * a, int32_t b, int32_t c, complex128_t * d) \ -{ \ - int32_t maxIdx=0; \ - SG_SERROR("CMath::%s():: Not supported for complex128_t\n",\ - #function);\ - return maxIdx; \ -} - -/// qsort not implemented for complex128_t, returns void instead -COMPLEX128_ERROR_ONEVECARG_RETURNS_T(qsort, void, return;) - -/// argsort not implemented for complex128_t, returns a vector -COMPLEX128_ERROR_ONEVECARG_RETURNS_T(argsort, SGVector, SGVector idx(vector.size());return idx;) - -/// is_sorted not implemented for complex128_t, returns false -COMPLEX128_ERROR_ONEVECARG_RETURNS_T(is_sorted, bool, return false;) - -/// min not implemented for complex128_t, returns (0.0)+i(0.0) instead -COMPLEX128_ERROR_TWOARGS_T(min) - -/// max not implemented for complex128_t, returns (0.0)+i(0.0) instead -COMPLEX128_ERROR_TWOARGS_T(max) - -/// clamp not implemented for complex128_t, returns (0.0)+i(0.0) instead -COMPLEX128_ERROR_THREEARGS_T(clamp) - /// signum not implemented for complex128_t, returns (0.0)+i(0.0) instead // COMPLEX128_ERROR_ONEARG_T(sign) -/// qsort not implemented for complex128_t -COMPLEX128_ERROR_SORT_T(qsort) - -/// insertion_sort not implemented for complex128_t -COMPLEX128_ERROR_SORT_T(insertion_sort) - -/// radix_sort not implemented for complex128_t -COMPLEX128_ERROR_SORT_T(radix_sort) - -/// arg_max not implemented for complex128_t -COMPLEX128_ERROR_ARG_MAX_MIN(arg_max) - -/// arg_min not implemented for complex128_t -COMPLEX128_ERROR_ARG_MAX_MIN(arg_min) - } #undef COMPLEX128_ERROR_ONEARG #undef COMPLEX128_ERROR_ONEARG_T -#undef COMPLEX128_ERROR_TWOARGS_T -#undef COMPLEX128_ERROR_THREEARGS_T #undef COMPLEX128_STDMATH -#undef COMPLEX128_ERROR_SORT_T #endif /** __MATHEMATICS_H_ */ diff --git a/src/shogun/mathematics/SparseInverseCovariance.cpp b/src/shogun/mathematics/SparseInverseCovariance.cpp deleted file mode 100644 index 9e8a60601b3..00000000000 --- a/src/shogun/mathematics/SparseInverseCovariance.cpp +++ /dev/null @@ -1,60 +0,0 @@ -/* - * This program is free software; you can redistribute it and/or modify - * it under the terms of the GNU General Public License as published by - * the Free Software Foundation; either version 3 of the License, or - * (at your option) any later version. - * - * Copyright (C) 2009-2011 Jun Liu, Jieping Ye - * Copyright (C) 2012 Sergey Lisitsyn - */ - - -#include -#ifdef USE_GPL_SHOGUN -#include -#include - -using namespace shogun; - -CSparseInverseCovariance::CSparseInverseCovariance() : - CSGObject(), m_lasso_max_iter(1000), - m_max_iter(1000), m_f_gap(1e-6), m_x_gap(1e-4), - m_xtol(1e-4) -{ - register_parameters(); -} - -CSparseInverseCovariance::~CSparseInverseCovariance() -{ -} - -void CSparseInverseCovariance::register_parameters() -{ - SG_ADD(&m_lasso_max_iter,"lasso_max_iter", - "maximum iteration of LASSO step",MS_NOT_AVAILABLE); - SG_ADD(&m_max_iter,"max_iter","maximum total iteration", - MS_NOT_AVAILABLE); - SG_ADD(&m_f_gap,"f_gap","f gap",MS_NOT_AVAILABLE); - SG_ADD(&m_x_gap,"x_gap","x gap",MS_NOT_AVAILABLE); - SG_ADD(&m_xtol,"xtol","xtol",MS_NOT_AVAILABLE); -} - -SGMatrix CSparseInverseCovariance::estimate(SGMatrix S, float64_t lambda_c) -{ - ASSERT(S.num_cols==S.num_rows) - - int32_t n = S.num_cols; - float64_t sum_S = 0.0; - for (int32_t i=0; i(Theta,n,n); -} -#endif //USE_GPL_SHOGUN diff --git a/src/shogun/mathematics/SparseInverseCovariance.h b/src/shogun/mathematics/SparseInverseCovariance.h deleted file mode 100644 index feda723cc92..00000000000 --- a/src/shogun/mathematics/SparseInverseCovariance.h +++ /dev/null @@ -1,129 +0,0 @@ -/* - * This program is free software; you can redistribute it and/or modify - * it under the terms of the GNU General Public License as published by - * the Free Software Foundation; either version 3 of the License, or - * (at your option) any later version. - * - * Copyright (C) 2012 Sergey Lisitsyn - */ - -#ifndef SPINVCOV_H_ -#define SPINVCOV_H_ - -#include -#ifdef USE_GPL_SHOGUN - -#include -#include - -namespace shogun -{ - -/** @brief used to estimate inverse covariance matrix using graphical lasso - * - * implementation is based on SLEP library's code - */ -class CSparseInverseCovariance : public CSGObject -{ -public: - - /** constructor */ - CSparseInverseCovariance(); - - /** destructor */ - virtual ~CSparseInverseCovariance(); - - /** estimate inverse covariance matrix - * - * @param S empirical covariance matrix - * @param lambda_c regularization constant - */ - SGMatrix estimate(SGMatrix S, float64_t lambda_c); - - /** get name */ - const char* get_name() const { return "SparseInverseCovariance"; }; - - - /** get lasso max iter - * @return lasso max iter - */ - int32_t get_lasso_max_iter() const { return m_lasso_max_iter; } - /** get max iter - * @return max iter - */ - int32_t get_max_iter() const { return m_max_iter; } - /** get lasso max iter - * @return lasso max iter - */ - float64_t get_f_gap() const { return m_f_gap; } - /** get lasso max iter - * @return lasso max iter - */ - float64_t get_x_gap() const { return m_x_gap; } - /** get lasso max iter - * @return lasso max iter - */ - float64_t get_xtol() const { return m_xtol; } - - /** set lasso max iter - * @param lasso_max_iter lasso max iter - */ - void set_lasso_max_iter(int32_t lasso_max_iter) - { - m_lasso_max_iter = lasso_max_iter; - } - /** set max iter - * @param max_iter max iter - */ - void set_max_iter(int32_t max_iter) - { - m_max_iter = max_iter; - } - /** set f gap - * @param f_gap f gap - */ - void set_f_gap(int32_t f_gap) - { - m_f_gap = f_gap; - } - /** set x gap - * @param x_gap x gap - */ - void set_x_gap(int32_t x_gap) - { - m_x_gap = x_gap; - } - /** set xtol - * @param xtol xtol - */ - void set_xtol(int32_t xtol) - { - m_xtol = xtol; - } - -private: - - /** register parameters */ - void register_parameters(); - -protected: - - /** LASSO max iter */ - int32_t m_lasso_max_iter; - - /** max iter */ - int32_t m_max_iter; - - /** fGap */ - float64_t m_f_gap; - - /** xGap */ - float64_t m_x_gap; - - /** xtol */ - float64_t m_xtol; -}; - -} -#endif //USE_GPL_SHOGUN -#endif diff --git a/src/shogun/mathematics/ajd/UWedge.cpp b/src/shogun/mathematics/ajd/UWedge.cpp index e7911c80ee0..8cec2dd4dd0 100644 --- a/src/shogun/mathematics/ajd/UWedge.cpp +++ b/src/shogun/mathematics/ajd/UWedge.cpp @@ -113,7 +113,12 @@ SGMatrix CUWedge::diagonalize(SGNDArray C, SGMatrix #include - #if EIGEN_VERSION_AT_LEAST(3,0,93) - #include - #else - #define EIGEN_YES_I_KNOW_SPARSE_MODULE_IS_NOT_STABLE_YET - #include - - #ifndef DOXYGEN_SHOULD_SKIP_THIS - // Triplet definition for Eigen3 backword compatibility - namespace Eigen { - template struct Triplet - { - Triplet(index_t colIndex, index_t rowIndex, T valueT) : - ecol(colIndex), erow(rowIndex), evalue(valueT) - { - } - index_t col() const { return ecol; }; - index_t row() const { return erow; }; - T value() const { return evalue; }; - index_t ecol; - index_t erow; - T evalue; - }; - - // SimplicialLLT definition for Eigen3 backword compatibility - template class SimplicialLLT - : public SimplicialCholesky - { - public: - SimplicialLLT() - { - SimplicialCholesky::setMode(SimplicialCholeskyLLt); - } - inline const T matrixL() - { - return SimplicialCholesky::m_matrix; - } - inline const T matrixU() - { - return SimplicialCholesky::m_matrix.transpose(); - } - }; - } - #endif //DOXYGEN_SHOULD_SKIP_THIS - - #endif //EIGEN_VERSION_AT_LEAST(3,0,93) + #include #if ((EIGEN_WORLD_VERSION == 3) && (EIGEN_MAJOR_VERSION == 2) && \ ((EIGEN_MINOR_VERSION == 91) || (EIGEN_MINOR_VERSION == 92))) diff --git a/src/shogun/mathematics/lapack.cpp b/src/shogun/mathematics/lapack.cpp index 44318b3af51..8bbe0b67de0 100644 --- a/src/shogun/mathematics/lapack.cpp +++ b/src/shogun/mathematics/lapack.cpp @@ -20,7 +20,10 @@ #include #include -#include +#if defined(EIGEN_USE_LAPACKE) || defined(EIGEN_USE_LAPACKE_STRICT) +// Eigen's lapacke.h will take care of the declaration of the lapacke functions +#include +#endif using namespace shogun; @@ -310,17 +313,17 @@ void wrap_dsyevr(char jobz, char uplo, int n, double *a, int lda, int il, int iu int m; double vl,vu; double abstol = 0.0; - char I = 'I'; - int* isuppz = SG_MALLOC(int, n); + char range = 'I'; + int* isuppz = SG_MALLOC(int, 2 * (iu - il + 1)); #ifdef HAVE_ACML - DSYEVR(jobz,I,uplo,n,a,lda,vl,vu,il,iu,abstol,m, + DSYEVR(jobz,range,uplo,n,a,lda,vl,vu,il,iu,abstol,m, eigenvalues,eigenvectors,n,isuppz,info); #else int lwork = -1; int liwork = -1; double work1 = 0; int work2 = 0; - DSYEVR(&jobz,&I,&uplo,&n,a,&lda,&vl,&vu,&il,&iu,&abstol, + DSYEVR(&jobz,&range,&uplo,&n,a,&lda,&vl,&vu,&il,&iu,&abstol, &m,eigenvalues,eigenvectors,&n,isuppz, &work1,&lwork,&work2,&liwork,info); ASSERT(*info==0) @@ -328,7 +331,7 @@ void wrap_dsyevr(char jobz, char uplo, int n, double *a, int lda, int il, int iu liwork = work2; double* work = SG_MALLOC(double, lwork); int* iwork = SG_MALLOC(int, liwork); - DSYEVR(&jobz,&I,&uplo,&n,a,&lda,&vl,&vu,&il,&iu,&abstol, + DSYEVR(&jobz,&range,&uplo,&n,a,&lda,&vl,&vu,&il,&iu,&abstol, &m,eigenvalues,eigenvectors,&n,isuppz, work,&lwork,iwork,&liwork,info); ASSERT(*info==0) @@ -346,21 +349,21 @@ void wrap_dsygvx(int itype, char jobz, char uplo, int n, double *a, int lda, dou double abstol = 0.0; double vl,vu; int* ifail = SG_MALLOC(int, n); - char I = 'I'; + char range = 'I'; #ifdef HAVE_ACML - DSYGVX(itype,jobz,I,uplo,n,a,lda,b,ldb,vl,vu, + DSYGVX(itype,jobz,range,uplo,n,a,lda,b,ldb,vl,vu, il,iu,abstol,m,eigenvalues, eigenvectors,n,ifail,info); #else int lwork = -1; double work1 = 0; int* iwork = SG_MALLOC(int, 5*n); - DSYGVX(&itype,&jobz,&I,&uplo,&n,a,&lda,b,&ldb,&vl,&vu, + DSYGVX(&itype,&jobz,&range,&uplo,&n,a,&lda,b,&ldb,&vl,&vu, &il,&iu,&abstol,&m,eigenvalues,eigenvectors, &n,&work1,&lwork,iwork,ifail,info); lwork = (int)work1; double* work = SG_MALLOC(double, lwork); - DSYGVX(&itype,&jobz,&I,&uplo,&n,a,&lda,b,&ldb,&vl,&vu, + DSYGVX(&itype,&jobz,&range,&uplo,&n,a,&lda,b,&ldb,&vl,&vu, &il,&iu,&abstol,&m,eigenvalues,eigenvectors, &n,work,&lwork,iwork,ifail,info); SG_FREE(work); diff --git a/src/shogun/mathematics/lapack.h b/src/shogun/mathematics/lapack.h index ca42a6c28c5..6ba466f21fa 100644 --- a/src/shogun/mathematics/lapack.h +++ b/src/shogun/mathematics/lapack.h @@ -32,8 +32,12 @@ extern "C" { #ifdef HAVE_CXX11 #define __VFORCE_H 1 #endif +#if !(defined(EIGEN_USE_LAPACKE) || defined(EIGEN_USE_LAPACKE_STRICT)) #include #else +#include +#endif +#else #include #endif @@ -44,7 +48,7 @@ extern "C" { #ifdef HAVE_ATLAS #include #else -// ACML and MKL do not provide clapack_* routines +// ACML, MKL and Mac OS vecLib do not provide clapack_* routines // double precision int clapack_dpotrf(const CBLAS_ORDER Order, const CBLAS_UPLO Uplo, const int N, double *A, const int lda); @@ -87,7 +91,7 @@ void wrap_dstemr(char jobz, char range, int n, double* d__, double *e, double vl } // only MKL, ACML and Mac OS vector library provide a header file for the lapack routines -#if !defined(HAVE_ACML) && !defined(HAVE_MKL) && !defined(HAVE_MVEC) +#if !defined(HAVE_ACML) && !defined(HAVE_MKL) && !defined(HAVE_MVEC) && !(defined(EIGEN_USE_LAPACKE) || defined(EIGEN_USE_LAPACKE_STRICT)) // double precision int dsyev_(char*, char*, int*, double*, int*, double*, double*, int*, int*); int dgesvd_(char* jobu, char* jobvt, int* m, int* n, double* a, int* lda, diff --git a/src/shogun/mathematics/linalg/GPUMemoryBase.h b/src/shogun/mathematics/linalg/GPUMemoryBase.h index 61f7a3cecd8..8882e1b9118 100644 --- a/src/shogun/mathematics/linalg/GPUMemoryBase.h +++ b/src/shogun/mathematics/linalg/GPUMemoryBase.h @@ -36,24 +36,23 @@ namespace shogun { -/** @brief Interface for GPU memory libraries */ -template -struct GPUMemoryBase -{ - /** Default constructor */ - GPUMemoryBase() + /** @brief Interface for GPU memory libraries */ + template + struct GPUMemoryBase { - } - - /** Clone GPU memory, i.e. vector or matrix - * - * @param GPUMemoryBase structure pointer - * @return A deep-copy of GPUMemoryBase structure pointer - */ - virtual GPUMemoryBase* clone_vector(GPUMemoryBase* gpu_ptr, - index_t vlen) const = 0; -}; + /** Default constructor */ + GPUMemoryBase() + { + } + /** Clone GPU memory, i.e. vector or matrix + * + * @param gpu_ptr GPUMemoryBase structure pointer + * @return A deep-copy of GPUMemoryBase structure pointer + */ + virtual GPUMemoryBase* + clone_vector(GPUMemoryBase* gpu_ptr, index_t vlen) const = 0; + }; } -#endif //GPU_MEMORY_BASE_H__ +#endif // GPU_MEMORY_BASE_H__ diff --git a/src/shogun/mathematics/linalg/GPUMemoryViennaCL.h b/src/shogun/mathematics/linalg/GPUMemoryViennaCL.h index 470927563f5..1deaa0ed1f3 100644 --- a/src/shogun/mathematics/linalg/GPUMemoryViennaCL.h +++ b/src/shogun/mathematics/linalg/GPUMemoryViennaCL.h @@ -36,118 +36,127 @@ #include #ifdef HAVE_VIENNACL -#include -#include #include +#include +#include namespace shogun { -/** @brief ViennaCL memory structure. - * Saves data to GPU and clone data. - * @see SGVector - */ -template -struct GPUMemoryViennaCL : public GPUMemoryBase -{ - friend class LinalgBackendViennaCL; + /** @brief ViennaCL memory structure. + * Saves data to GPU and clone data. + * @see SGVector + */ + template + struct GPUMemoryViennaCL : public GPUMemoryBase + { + friend class LinalgBackendViennaCL; - typedef viennacl::backend::mem_handle VCLMemoryArray; - typedef viennacl::vector_base VCLVectorBase; + typedef viennacl::backend::mem_handle VCLMemoryArray; + typedef viennacl::vector_base + VCLVectorBase; /** @see */ #if VIENNACL_VERSION >= 10600 - typedef viennacl::matrix_base VCLMatrixBase; + typedef viennacl::matrix_base + VCLMatrixBase; #else - typedef viennacl::matrix_base VCLMatrixBase; + typedef viennacl::matrix_base + VCLMatrixBase; #endif - /** Default constructor */ - GPUMemoryViennaCL() : m_data(new VCLMemoryArray()) - { - init(); - }; - - /** Create a new vector - * - * @param len Number of elements - */ - GPUMemoryViennaCL(index_t len): m_data(new VCLMemoryArray()) - { - init(); - viennacl::backend::memory_create(*m_data, sizeof(T)*len, - viennacl::context()); - } - - /** Wrap a vector around an existing memory segment - * - * @param gpu_ptr GPUMemoryBase pointer - */ - GPUMemoryViennaCL(GPUMemoryBase* gpu_ptr) : m_data(new VCLMemoryArray()) - { - GPUMemoryViennaCL* temp_ptr = static_cast*>(gpu_ptr); - init(); - m_data = temp_ptr->m_data; - m_offset = temp_ptr->m_offset; - }; - - /** Clone GPU vector - * - * @param vector GPUMemoryBase pointer - * @param vlen Length of the vector - */ - GPUMemoryBase* clone_vector(GPUMemoryBase* vector, index_t vlen) const - { - GPUMemoryViennaCL* src_ptr = static_cast*>(vector); - GPUMemoryViennaCL* gpu_ptr = new GPUMemoryViennaCL(); - - viennacl::backend::memory_create(*(gpu_ptr->m_data), sizeof(T)*vlen, - viennacl::context()); - viennacl::backend::memory_copy(*(src_ptr->m_data), *(gpu_ptr->m_data), - 0, 0, vlen*sizeof(T)); - - return gpu_ptr; - } - - /** ViennaCL Vector structure that saves the data - * - * @param len Number of elements - */ - VCLVectorBase data_vector(index_t len) - { - return VCLVectorBase(*m_data, len, m_offset, 1); - } - - /** ViennaCL Vector structure that saves the data - * - * @param nrows Row number of the matrix - * @param ncols Column number of the matrix - */ - VCLMatrixBase data_matrix(index_t nrows, index_t ncols) - { - #if VIENNACL_VERSION >= 10600 - return VCLMatrixBase(*m_data, nrows, m_offset, 1, nrows, ncols, 0, 1, ncols, false); - #else - return VCLMatrixBase(*m_data, nrows, m_offset, 1, nrows, ncols, 0, 1, ncols); - #endif - } - -private: - void init() - { - m_offset = 0; - } + /** Default constructor */ + GPUMemoryViennaCL() : m_data(new VCLMemoryArray()) + { + init(); + }; + + /** Create a new vector + * + * @param len Number of elements + */ + GPUMemoryViennaCL(index_t len) : m_data(new VCLMemoryArray()) + { + init(); + viennacl::backend::memory_create( + *m_data, sizeof(T) * len, viennacl::context()); + } + + /** Wrap a vector around an existing memory segment + * + * @param gpu_ptr GPUMemoryBase pointer + */ + GPUMemoryViennaCL(GPUMemoryBase* gpu_ptr) + : m_data(new VCLMemoryArray()) + { + GPUMemoryViennaCL* temp_ptr = + static_cast*>(gpu_ptr); + init(); + m_data = temp_ptr->m_data; + m_offset = temp_ptr->m_offset; + }; + + /** Clone GPU vector + * + * @param vector GPUMemoryBase pointer + * @param vlen Length of the vector + */ + GPUMemoryBase* + clone_vector(GPUMemoryBase* vector, index_t vlen) const + { + GPUMemoryViennaCL* src_ptr = + static_cast*>(vector); + GPUMemoryViennaCL* gpu_ptr = new GPUMemoryViennaCL(); + + viennacl::backend::memory_create( + *(gpu_ptr->m_data), sizeof(T) * vlen, viennacl::context()); + viennacl::backend::memory_copy( + *(src_ptr->m_data), *(gpu_ptr->m_data), 0, 0, vlen * sizeof(T)); + + return gpu_ptr; + } + + /** ViennaCL Vector structure that saves the data + * + * @param len Number of elements + */ + VCLVectorBase data_vector(index_t len) + { + return VCLVectorBase(*m_data, len, m_offset, 1); + } + + /** ViennaCL Vector structure that saves the data + * + * @param nrows Row number of the matrix + * @param ncols Column number of the matrix + */ + VCLMatrixBase data_matrix(index_t nrows, index_t ncols) + { +#if VIENNACL_VERSION >= 10600 + return VCLMatrixBase( + *m_data, nrows, m_offset, 1, nrows, ncols, 0, 1, ncols, false); +#else + return VCLMatrixBase( + *m_data, nrows, m_offset, 1, nrows, ncols, 0, 1, ncols); +#endif + } - /** Memory segment holding the data for the vector */ - alignas(CPU_CACHE_LINE_SIZE) std::shared_ptr m_data; + private: + void init() + { + m_offset = 0; + } - /** Offset for the memory segment, i.e the data of the vector - * starts at vector+offset - */ - alignas(CPU_CACHE_LINE_SIZE) index_t m_offset; -}; + /** Memory segment holding the data for the vector */ + alignas(CPU_CACHE_LINE_SIZE) std::shared_ptr m_data; + /** Offset for the memory segment, i.e the data of the vector + * starts at vector+offset + */ + alignas(CPU_CACHE_LINE_SIZE) index_t m_offset; + }; } #endif // HAVE_VIENNACL -#endif //GPU_MEMORY_VIENNACL_H__ +#endif // GPU_MEMORY_VIENNACL_H__ diff --git a/src/shogun/mathematics/linalg/LinalgBackendBase.h b/src/shogun/mathematics/linalg/LinalgBackendBase.h index 82b70239ea1..1e6a9471d6f 100644 --- a/src/shogun/mathematics/linalg/LinalgBackendBase.h +++ b/src/shogun/mathematics/linalg/LinalgBackendBase.h @@ -33,412 +33,730 @@ #ifndef LINALG_BACKEND_BASE_H__ #define LINALG_BACKEND_BASE_H__ -#include -#include -#include -#include +#include #include +#include +#include +#include +#include +#include #include +#include #include -#include namespace shogun { -/** @brief Base interface of generic linalg methods - * and generic memory transfer methods. - */ -class LinalgBackendBase -{ -public: - #define DEFINE_FOR_ALL_PTYPE(METHODNAME, Container) \ - METHODNAME(bool, Container); \ - METHODNAME(char, Container); \ - METHODNAME(int8_t, Container); \ - METHODNAME(uint8_t, Container); \ - METHODNAME(int16_t, Container); \ - METHODNAME(uint16_t, Container); \ - METHODNAME(int32_t, Container); \ - METHODNAME(uint32_t, Container); \ - METHODNAME(int64_t, Container); \ - METHODNAME(uint64_t, Container); \ - METHODNAME(float32_t, Container); \ - METHODNAME(float64_t, Container); \ - METHODNAME(floatmax_t, Container); \ - METHODNAME(complex128_t, Container); \ - - #define DEFINE_FOR_REAL_PTYPE(METHODNAME, Container) \ - METHODNAME(bool, Container); \ - METHODNAME(char, Container); \ - METHODNAME(int8_t, Container); \ - METHODNAME(uint8_t, Container); \ - METHODNAME(int16_t, Container); \ - METHODNAME(uint16_t, Container); \ - METHODNAME(int32_t, Container); \ - METHODNAME(uint32_t, Container); \ - METHODNAME(int64_t, Container); \ - METHODNAME(uint64_t, Container); \ - METHODNAME(float32_t, Container); \ - METHODNAME(float64_t, Container); \ + /** @brief Base interface of generic linalg methods + * and generic memory transfer methods. + */ + class LinalgBackendBase + { + public: +#define DEFINE_FOR_ALL_PTYPE(METHODNAME, Container) \ + METHODNAME(bool, Container); \ + METHODNAME(char, Container); \ + METHODNAME(int8_t, Container); \ + METHODNAME(uint8_t, Container); \ + METHODNAME(int16_t, Container); \ + METHODNAME(uint16_t, Container); \ + METHODNAME(int32_t, Container); \ + METHODNAME(uint32_t, Container); \ + METHODNAME(int64_t, Container); \ + METHODNAME(uint64_t, Container); \ + METHODNAME(float32_t, Container); \ + METHODNAME(float64_t, Container); \ + METHODNAME(floatmax_t, Container); \ + METHODNAME(complex128_t, Container); + +#define DEFINE_FOR_REAL_PTYPE(METHODNAME, Container) \ + METHODNAME(bool, Container); \ + METHODNAME(char, Container); \ + METHODNAME(int8_t, Container); \ + METHODNAME(uint8_t, Container); \ + METHODNAME(int16_t, Container); \ + METHODNAME(uint16_t, Container); \ + METHODNAME(int32_t, Container); \ + METHODNAME(uint32_t, Container); \ + METHODNAME(int64_t, Container); \ + METHODNAME(uint64_t, Container); \ + METHODNAME(float32_t, Container); \ + METHODNAME(float64_t, Container); \ METHODNAME(floatmax_t, Container); - #define DEFINE_FOR_NON_INTEGER_PTYPE(METHODNAME, Container) \ - METHODNAME(float32_t, Container); \ - METHODNAME(float64_t, Container); \ - METHODNAME(floatmax_t, Container); \ +#define DEFINE_FOR_NON_INTEGER_PTYPE(METHODNAME, Container) \ + METHODNAME(float32_t, Container); \ + METHODNAME(float64_t, Container); \ + METHODNAME(floatmax_t, Container); \ METHODNAME(complex128_t, Container); - /** - * Wrapper method of add operation the operation result = alpha*a + beta*b. - * - * @see linalg::add - */ - #define BACKEND_GENERIC_IN_PLACE_ADD(Type, Container) \ - virtual void add(Container& a, Container& b, Type alpha, Type beta, Container& result) const \ - { \ - SG_SNOTIMPLEMENTED; \ - } - DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_IN_PLACE_ADD, SGVector) - DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_IN_PLACE_ADD, SGMatrix) - #undef BACKEND_GENERIC_IN_PLACE_ADD - - /** - * Wrapper method of Cholesky decomposition. - * - * @see linalg::cholesky_factor - */ - #define BACKEND_GENERIC_CHOLESKY_FACTOR(Type, Container) \ - virtual Container cholesky_factor(const Container& A, \ - const bool lower) const \ - { \ - SG_SNOTIMPLEMENTED; \ - return 0; \ - } - DEFINE_FOR_NON_INTEGER_PTYPE(BACKEND_GENERIC_CHOLESKY_FACTOR, SGMatrix) - #undef BACKEND_GENERIC_CHOLESKY_FACTOR - - /** - * Wrapper triangular solver with Choleksy decomposition. - * - * @see linalg::cholesky_solver - */ - #define BACKEND_GENERIC_CHOLESKY_SOLVER(Type, Container) \ - virtual SGVector cholesky_solver(const Container& L, \ - const SGVector& b, const bool lower) const \ - { \ - SG_SNOTIMPLEMENTED; \ - return 0; \ - } - DEFINE_FOR_NON_INTEGER_PTYPE(BACKEND_GENERIC_CHOLESKY_SOLVER, SGMatrix) - #undef BACKEND_GENERIC_CHOLESKY_SOLVER - - /** - * Wrapper method of vector dot-product that works with generic vectors. - * - * @see linalg::dot - */ - #define BACKEND_GENERIC_DOT(Type, Container) \ +/** + * Wrapper method of add operation the operation result = alpha*a + beta*b. + * + * @see linalg::add + */ +#define BACKEND_GENERIC_IN_PLACE_ADD(Type, Container) \ + virtual void add( \ + Container& a, Container& b, Type alpha, Type beta, \ + Container& result) const \ + { \ + SG_SNOTIMPLEMENTED; \ + } + DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_IN_PLACE_ADD, SGVector) + DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_IN_PLACE_ADD, SGMatrix) +#undef BACKEND_GENERIC_IN_PLACE_ADD + +/** + * Wrapper method of add column vector result = alpha*A.col(i) + beta*b. + * + * @see linalg::add_col_vec + */ +#define BACKEND_GENERIC_ADD_COL_VEC(Type, Container) \ + virtual void add_col_vec( \ + const SGMatrix& A, index_t i, const SGVector& b, \ + Container& result, Type alpha, Type beta) const \ + { \ + SG_SNOTIMPLEMENTED; \ + return; \ + } + DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_ADD_COL_VEC, SGVector) + DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_ADD_COL_VEC, SGMatrix) +#undef BACKEND_GENERIC_ADD_COL_VEC + +/** + * Wrapper method of add vector to each column of matrix. + * + * @see linalg::add_vector + */ +#define BACKEND_GENERIC_ADD_VECTOR(Type, Container) \ + virtual void add_vector( \ + const SGMatrix& A, const SGVector& b, \ + SGMatrix& result, Type alpha, Type beta) const \ + { \ + SG_SNOTIMPLEMENTED; \ + return; \ + } + DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_ADD_VECTOR, SGMatrix) +#undef BACKEND_GENERIC_ADD_VECTOR + +/** + * Wrapper method of add scalar operation. + * + * @see linalg::add_scalar + */ +#define BACKEND_GENERIC_ADD_SCALAR(Type, Container) \ + virtual void add_scalar(Container& a, Type b) const \ + { \ + SG_SNOTIMPLEMENTED; \ + } + DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_ADD_SCALAR, SGVector) + DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_ADD_SCALAR, SGMatrix) +#undef BACKEND_GENERIC_ADD_SCALAR + +/** + * Wrapper method of center matrix operation. + * + * @see linalg::center_matrix + */ +#define BACKEND_GENERIC_CENTER_MATRIX(Type, Container) \ + virtual void center_matrix(Container& A) const \ + { \ + SG_SNOTIMPLEMENTED; \ + } + DEFINE_FOR_NON_INTEGER_PTYPE(BACKEND_GENERIC_CENTER_MATRIX, SGMatrix) +#undef BACKEND_GENERIC_CENTER_MATRIX + +/** + * Wrapper method of Cholesky decomposition. + * + * @see linalg::cholesky_factor + */ +#define BACKEND_GENERIC_CHOLESKY_FACTOR(Type, Container) \ + virtual Container cholesky_factor( \ + const Container& A, const bool lower) const \ + { \ + SG_SNOTIMPLEMENTED; \ + return 0; \ + } + DEFINE_FOR_NON_INTEGER_PTYPE(BACKEND_GENERIC_CHOLESKY_FACTOR, SGMatrix) +#undef BACKEND_GENERIC_CHOLESKY_FACTOR + +/** + * Wrapper triangular solver with Choleksy decomposition. + * + * @see linalg::cholesky_solver + */ +#define BACKEND_GENERIC_CHOLESKY_SOLVER(Type, Container) \ + virtual SGVector cholesky_solver( \ + const Container& L, const SGVector& b, const bool lower) \ + const \ + { \ + SG_SNOTIMPLEMENTED; \ + return 0; \ + } + DEFINE_FOR_NON_INTEGER_PTYPE(BACKEND_GENERIC_CHOLESKY_SOLVER, SGMatrix) +#undef BACKEND_GENERIC_CHOLESKY_SOLVER + +/** + * Wrapper method of cross entropy. + * + * @see linalg::cross_entropy + */ +#define BACKEND_GENERIC_CROSS_ENTROPY(Type, Container) \ + virtual Type cross_entropy( \ + const Container& P, const Container& Q) const \ + { \ + SG_SNOTIMPLEMENTED; \ + return 0; \ + } + DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_CROSS_ENTROPY, SGMatrix) +#undef BACKEND_GENERIC_CROSS_ENTROPY + +/** + * Wrapper method of vector dot-product that works with generic vectors. + * + * @see linalg::dot + */ +#define BACKEND_GENERIC_DOT(Type, Container) \ virtual Type dot(const Container& a, const Container& b) const \ - { \ - SG_SNOTIMPLEMENTED; \ - return 0; \ + { \ + SG_SNOTIMPLEMENTED; \ + return 0; \ } - DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_DOT, SGVector) - #undef BACKEND_GENERIC_DOT + DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_DOT, SGVector) +#undef BACKEND_GENERIC_DOT - /** - * Wrapper method of in-place matrix elementwise product. - * - * @see linalg::element_prod - */ - #define BACKEND_GENERIC_IN_PLACE_ELEMENT_PROD(Type, Container) \ - virtual void element_prod(Container& a, Container& b,\ - Container& result) const \ - { \ - SG_SNOTIMPLEMENTED; \ - } - DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_IN_PLACE_ELEMENT_PROD, SGMatrix) - #undef BACKEND_GENERIC_IN_PLACE_ELEMENT_PROD - - /** - * Wrapper method of in-place matrix block elementwise product. - * - * @see linalg::element_prod - */ - #define BACKEND_GENERIC_IN_PLACE_BLOCK_ELEMENT_PROD(Type, Container) \ - virtual void element_prod(linalg::Block>& a, \ - linalg::Block>& b, Container& result) const \ - { \ - SG_SNOTIMPLEMENTED; \ - } - DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_IN_PLACE_BLOCK_ELEMENT_PROD, SGMatrix) - #undef BACKEND_GENERIC_IN_PLACE_BLOCK_ELEMENT_PROD - - /** - * Wrapper method of logistic function f(x) = 1/(1+exp(-x)) - * - * @see linalg::logistic - */ - #define BACKEND_GENERIC_LOGISTIC(Type, Container) \ - virtual void logistic(Container& a, Container& result) const \ - { \ - SG_SNOTIMPLEMENTED; \ - } - DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_LOGISTIC, SGMatrix) - #undef BACKEND_GENERIC_LOGISTIC - - /** - * Wrapper method of matrix product method. - * - * @see linalg::matrix_prod - */ - #define BACKEND_GENERIC_IN_PLACE_MATRIX_PROD(Type, Container) \ - virtual void matrix_prod(SGMatrix& a, Container& b,\ - Container& result, bool transpose_A, bool transpose_B) const \ - { \ - SG_SNOTIMPLEMENTED; \ - } - DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_IN_PLACE_MATRIX_PROD, SGVector) - DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_IN_PLACE_MATRIX_PROD, SGMatrix) - #undef BACKEND_GENERIC_IN_PLACE_MATRIX_PROD - - /** - * Wrapper method of max method. Return the largest element in a vector or matrix. - * - * @see linalg::max - */ - #define BACKEND_GENERIC_MAX(Type, Container) \ - virtual Type max(const Container& a) const \ - { \ - SG_SNOTIMPLEMENTED; \ - return 0; \ - } - DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_MAX, SGVector) - DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_MAX, SGMatrix) - #undef BACKEND_GENERIC_MAX - - /** - * Wrapper method that computes mean of SGVectors and SGMatrices - * that are composed of real numbers. - * - * @see linalg::mean - */ - #define BACKEND_GENERIC_REAL_MEAN(Type, Container) \ - virtual float64_t mean(const Container& a) const \ - { \ - SG_SNOTIMPLEMENTED; \ - return 0; \ - } - DEFINE_FOR_REAL_PTYPE(BACKEND_GENERIC_REAL_MEAN, SGVector) - DEFINE_FOR_REAL_PTYPE(BACKEND_GENERIC_REAL_MEAN, SGMatrix) - #undef BACKEND_GENERIC_REAL_MEAN - - /** - * Wrapper method that computes mean of SGVectors and SGMatrices - * that are composed of complex numbers. - * - * @see linalg::mean - */ - #define BACKEND_GENERIC_COMPLEX_MEAN(Container) \ - virtual complex128_t mean(const Container& a) const \ - { \ - SG_SNOTIMPLEMENTED; \ - return 0; \ - } - BACKEND_GENERIC_COMPLEX_MEAN(SGVector) - BACKEND_GENERIC_COMPLEX_MEAN(SGMatrix) - #undef BACKEND_GENERIC_COMPLEX_MEAN - - /** - * Wrapper method that range fills a vector of matrix. - * - * @see linalg::range_fill - */ - #define BACKEND_GENERIC_RANGE_FILL(Type, Container) \ - virtual void range_fill(Container& a, const Type start) const \ - { \ - SG_SNOTIMPLEMENTED; \ - } - DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_RANGE_FILL, SGVector) - DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_RANGE_FILL, SGMatrix) - #undef BACKEND_GENERIC_RANGE_FILL - - /** - * Wrapper method of scale operation the operation result = alpha*A. - * - * @see linalg::scale - */ - #define BACKEND_GENERIC_IN_PLACE_SCALE(Type, Container) \ - virtual void scale(Container& a, Type alpha, Container& result) const \ - { \ - SG_SNOTIMPLEMENTED; \ - } - DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_IN_PLACE_SCALE, SGVector) - DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_IN_PLACE_SCALE, SGMatrix) - #undef BACKEND_GENERIC_IN_PLACE_SCALE - - /** - * Wrapper method that sets const values to vectors or matrices. - * - * @see linalg::set_const - */ - #define BACKEND_GENERIC_SET_CONST(Type, Container) \ - virtual void set_const(Container& a, const Type value) const \ - { \ - SG_SNOTIMPLEMENTED; \ - } - DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_SET_CONST, SGVector) - DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_SET_CONST, SGMatrix) - #undef BACKEND_GENERIC_SET_CONST - - /** - * Wrapper method of sum that works with generic vectors or matrices. - * - * @see linalg::sum - */ - #define BACKEND_GENERIC_SUM(Type, Container) \ - virtual Type sum(const Container& a, bool no_diag) const \ - { \ - SG_SNOTIMPLEMENTED; \ - return 0; \ - } - DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_SUM, SGVector) - DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_SUM, SGMatrix) - #undef BACKEND_GENERIC_SUM - - /** - * Wrapper method of sum that works with matrix blocks. - * - * @see linalg::sum - */ - #define BACKEND_GENERIC_BLOCK_SUM(Type, Container) \ - virtual Type sum(const linalg::Block>& a, bool no_diag) const \ - { \ - SG_SNOTIMPLEMENTED; \ - return 0; \ - } - DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_BLOCK_SUM, SGMatrix) - #undef BACKEND_GENERIC_BLOCK_SUM - - /** - * Wrapper method of sum that works with symmetric matrices. - * - * @see linalg::sum_symmetric - */ - #define BACKEND_GENERIC_SYMMETRIC_SUM(Type, Container) \ - virtual Type sum_symmetric(const Container& a, bool no_diag) const \ - { \ - SG_SNOTIMPLEMENTED; \ - return 0; \ - } - DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_SYMMETRIC_SUM, SGMatrix) - #undef BACKEND_GENERIC_SYMMETRIC_SUM - - /** - * Wrapper method of sum that works with symmetric matrix blocks. - * - * @see linalg::sum - */ - #define BACKEND_GENERIC_SYMMETRIC_BLOCK_SUM(Type, Container) \ - virtual Type sum_symmetric(const linalg::Block>& a, bool no_diag) const \ - { \ - SG_SNOTIMPLEMENTED; \ - return 0; \ - } - DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_SYMMETRIC_BLOCK_SUM, SGMatrix) - #undef BACKEND_GENERIC_SYMMETRIC_BLOCK_SUM - - /** - * Wrapper method of matrix rowwise sum that works with dense matrices. - * - * @see linalg::colwise_sum - */ - #define BACKEND_GENERIC_COLWISE_SUM(Type, Container) \ - virtual SGVector colwise_sum(const Container& a, bool no_diag) const \ - { \ - SG_SNOTIMPLEMENTED; \ - return 0; \ - } - DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_COLWISE_SUM, SGMatrix) - #undef BACKEND_GENERIC_COLWISE_SUM - - /** - * Wrapper method of matrix colwise sum that works with dense matrices. - * - * @see linalg::colwise_sum - */ - #define BACKEND_GENERIC_BLOCK_COLWISE_SUM(Type, Container) \ - virtual SGVector colwise_sum(const linalg::Block>& a, bool no_diag) const \ - { \ - SG_SNOTIMPLEMENTED; \ - return 0; \ - } - DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_BLOCK_COLWISE_SUM, SGMatrix) - #undef BACKEND_GENERIC_BLOCK_COLWISE_SUM - - /** - * Wrapper method of matrix rowwise sum that works with dense matrices. - * - * @see linalg::rowwise_sum - */ - #define BACKEND_GENERIC_ROWWISE_SUM(Type, Container) \ - virtual SGVector rowwise_sum(const Container& a, bool no_diag) const \ - { \ - SG_SNOTIMPLEMENTED; \ - return 0; \ - } - DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_ROWWISE_SUM, SGMatrix) - #undef BACKEND_GENERIC_ROWWISE_SUM - - /** - * Wrapper method of matrix rowwise sum that works with dense matrices. - * - * @see linalg::rowwise_sum - */ - #define BACKEND_GENERIC_BLOCK_ROWWISE_SUM(Type, Container) \ - virtual SGVector rowwise_sum(const linalg::Block>& a, bool no_diag) const \ - { \ - SG_SNOTIMPLEMENTED; \ - return 0; \ - } - DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_BLOCK_ROWWISE_SUM, SGMatrix) - #undef BACKEND_GENERIC_BLOCK_ROWWISE_SUM - - /** - * Wrapper method of Transferring data to GPU memory. - * Does nothing if no GPU backend registered. - * - * @see linalg::to_gpu - */ - #define BACKEND_GENERIC_TO_GPU(Type, Container) \ - virtual GPUMemoryBase* to_gpu(const Container&) const \ - { \ - SG_SNOTIMPLEMENTED; \ - return 0; \ - } - DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_TO_GPU, SGVector) - DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_TO_GPU, SGMatrix) - #undef BACKEND_GENERIC_TO_GPU - - /** - * Wrapper method of fetching data from GPU memory. - * - * @see linalg::from_gpu - */ - #define BACKEND_GENERIC_FROM_GPU(Type, Container) \ - virtual void from_gpu(const Container&, Type* data) const \ - { \ - SG_SNOTIMPLEMENTED; \ +/** + * Wrapper method of eigenvalues and eigenvectors computation. + * + * @see linalg::eigen_solver + */ +#define BACKEND_GENERIC_EIGEN_SOLVER(Type, Container) \ + virtual void eigen_solver( \ + const Container& A, SGVector& eigenvalues, \ + SGMatrix& eigenvectors) const \ + { \ + SG_SNOTIMPLEMENTED; \ + } + DEFINE_FOR_NON_INTEGER_PTYPE(BACKEND_GENERIC_EIGEN_SOLVER, SGMatrix) +#undef BACKEND_GENERIC_EIGEN_SOLVER + +/** + * Wrapper method of eigenvalues and eigenvectors computation + * for symmetric matrices. + * + * @see linalg::eigen_solver_symmetric + */ +#define BACKEND_GENERIC_EIGEN_SOLVER_SYMMETRIC(Type, Container) \ + virtual void eigen_solver_symmetric( \ + const Container& A, SGVector& eigenvalues, \ + SGMatrix& eigenvectors, index_t k) const \ + { \ + SG_SNOTIMPLEMENTED; \ + } + DEFINE_FOR_NON_INTEGER_PTYPE( + BACKEND_GENERIC_EIGEN_SOLVER_SYMMETRIC, SGMatrix) +#undef BACKEND_GENERIC_EIGEN_SOLVER_SYMMETRIC + +/** + * Wrapper method of in-place matrix elementwise product. + * + * @see linalg::element_prod + */ +#define BACKEND_GENERIC_IN_PLACE_ELEMENT_PROD(Type, Container) \ + virtual void element_prod( \ + Container& a, Container& b, Container& result) const \ + { \ + SG_SNOTIMPLEMENTED; \ + } + DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_IN_PLACE_ELEMENT_PROD, SGMatrix) +#undef BACKEND_GENERIC_IN_PLACE_ELEMENT_PROD + +/** + * Wrapper method of in-place matrix block elementwise product. + * + * @see linalg::element_prod + */ +#define BACKEND_GENERIC_IN_PLACE_BLOCK_ELEMENT_PROD(Type, Container) \ + virtual void element_prod( \ + linalg::Block>& a, linalg::Block>& b, \ + Container& result) const \ + { \ + SG_SNOTIMPLEMENTED; \ + } + DEFINE_FOR_ALL_PTYPE( + BACKEND_GENERIC_IN_PLACE_BLOCK_ELEMENT_PROD, SGMatrix) +#undef BACKEND_GENERIC_IN_PLACE_BLOCK_ELEMENT_PROD + +/** + * Wrapper method of in-place exponent method. + * + * @see linalg::exponent + */ +#define BACKEND_GENERIC_EXPONENT(Type, Container) \ + virtual void exponent(const Container& a, Container& result) \ + const \ + { \ + SG_SNOTIMPLEMENTED; \ + } + DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_EXPONENT, SGVector) + DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_EXPONENT, SGMatrix) +#undef BACKEND_GENERIC_EXPONENT + +/** + * Wrapper method of set matrix to identity. + * + * @see linalg::identity + */ +#define BACKEND_GENERIC_IDENTITY(Type, Container) \ + virtual void identity(Container& identity_matrix) const \ + { \ + SG_SNOTIMPLEMENTED; \ + return; \ } - DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_FROM_GPU, SGVector) - DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_FROM_GPU, SGMatrix) - #undef BACKEND_GENERIC_FROM_GPU + DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_IDENTITY, SGMatrix) +#undef BACKEND_GENERIC_IDENTITY + +/** + * Wrapper method of logistic function f(x) = 1/(1+exp(-x)) + * + * @see linalg::logistic + */ +#define BACKEND_GENERIC_LOGISTIC(Type, Container) \ + virtual void logistic(Container& a, Container& result) const \ + { \ + SG_SNOTIMPLEMENTED; \ + } + DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_LOGISTIC, SGMatrix) +#undef BACKEND_GENERIC_LOGISTIC + +/** + * Wrapper method of matrix product method. + * + * @see linalg::matrix_prod + */ +#define BACKEND_GENERIC_IN_PLACE_MATRIX_PROD(Type, Container) \ + virtual void matrix_prod( \ + SGMatrix& a, Container& b, Container& result, \ + bool transpose_A, bool transpose_B) const \ + { \ + SG_SNOTIMPLEMENTED; \ + } + DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_IN_PLACE_MATRIX_PROD, SGVector) + DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_IN_PLACE_MATRIX_PROD, SGMatrix) +#undef BACKEND_GENERIC_IN_PLACE_MATRIX_PROD + +/** + * Wrapper method of max method. Return the largest element in a vector or + * matrix. + * + * @see linalg::max + */ +#define BACKEND_GENERIC_MAX(Type, Container) \ + virtual Type max(const Container& a) const \ + { \ + SG_SNOTIMPLEMENTED; \ + return 0; \ + } + DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_MAX, SGVector) + DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_MAX, SGMatrix) +#undef BACKEND_GENERIC_MAX + +/** +* Wrapper method that computes mean of SGVectors and SGMatrices +* that are composed of real numbers. +* +* @see linalg::mean +*/ +#define BACKEND_GENERIC_REAL_MEAN(Type, Container) \ + virtual float64_t mean(const Container& a) const \ + { \ + SG_SNOTIMPLEMENTED; \ + return 0; \ + } + DEFINE_FOR_REAL_PTYPE(BACKEND_GENERIC_REAL_MEAN, SGVector) + DEFINE_FOR_REAL_PTYPE(BACKEND_GENERIC_REAL_MEAN, SGMatrix) +#undef BACKEND_GENERIC_REAL_MEAN + +/** +* Wrapper method that computes mean of SGVectors and SGMatrices +* that are composed of complex numbers. +* +* @see linalg::mean +*/ +#define BACKEND_GENERIC_COMPLEX_MEAN(Container) \ + virtual complex128_t mean(const Container& a) const \ + { \ + SG_SNOTIMPLEMENTED; \ + return 0; \ + } + BACKEND_GENERIC_COMPLEX_MEAN(SGVector) + BACKEND_GENERIC_COMPLEX_MEAN(SGMatrix) +#undef BACKEND_GENERIC_COMPLEX_MEAN + +/** + * Wrapper method of multiply_by_logistic_derivative function f(x) = + * 1/(1+exp(-x)) + * + * @see linalg::multiply_by_logistic_derivative + */ +#define BACKEND_GENERIC_MULTIPLY_BY_LOGISTIC_DERIV(Type, Container) \ + virtual void multiply_by_logistic_derivative( \ + Container& a, Container& result) const \ + { \ + SG_SNOTIMPLEMENTED; \ + } + DEFINE_FOR_ALL_PTYPE( + BACKEND_GENERIC_MULTIPLY_BY_LOGISTIC_DERIV, SGMatrix) +#undef BACKEND_GENERIC_MULTIPLY_BY_LOGISTIC_DERIV + +/** + * Wrapper method of multiply_by_rectified_linear_derivative + * + * @see linalg::multiply_by_rectified_linear_derivative + */ +#define BACKEND_GENERIC_MULTIPLY_BY_RECTIFIED_LINEAR_DERIV(Type, Container) \ + virtual void multiply_by_rectified_linear_derivative( \ + Container& a, Container& result) const \ + { \ + SG_SNOTIMPLEMENTED; \ + } + DEFINE_FOR_ALL_PTYPE( + BACKEND_GENERIC_MULTIPLY_BY_RECTIFIED_LINEAR_DERIV, SGMatrix) +#undef BACKEND_GENERIC_MULTIPLY_BY_RECTIFIED_LINEAR_DERIV + +/** +* Wrapper method that range fills a vector of matrix. +* +* @see linalg::range_fill +*/ +#define BACKEND_GENERIC_RANGE_FILL(Type, Container) \ + virtual void range_fill(Container& a, const Type start) const \ + { \ + SG_SNOTIMPLEMENTED; \ + } + DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_RANGE_FILL, SGVector) + DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_RANGE_FILL, SGMatrix) +#undef BACKEND_GENERIC_RANGE_FILL + +/** + * Wrapper method of rectified_linear method f(x) = max(0, x) + * + * @see linalg::rectified_linear + */ +#define BACKEND_GENERIC_RECTIFIED_LINEAR(Type, Container) \ + virtual void rectified_linear(Container& a, Container& result) \ + const \ + { \ + SG_SNOTIMPLEMENTED; \ + } + DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_RECTIFIED_LINEAR, SGMatrix) +#undef BACKEND_GENERIC_RECTIFIED_LINEAR + +/** + * Wrapper method that solves a system of linear equations + * using QR decomposition. + * + * @see linalg::qr_solver + */ +#define BACKEND_GENERIC_QR_SOLVER(Type, Container) \ + virtual Container qr_solver( \ + const SGMatrix& A, const Container& b) const \ + { \ + SG_SNOTIMPLEMENTED; \ + return 0; \ + } + DEFINE_FOR_NON_INTEGER_PTYPE(BACKEND_GENERIC_QR_SOLVER, SGVector) + DEFINE_FOR_NON_INTEGER_PTYPE(BACKEND_GENERIC_QR_SOLVER, SGMatrix) +#undef BACKEND_GENERIC_QR_SOLVER + +/** + * Wrapper method of scale operation the operation result = alpha*A. + * + * @see linalg::scale + */ +#define BACKEND_GENERIC_IN_PLACE_SCALE(Type, Container) \ + virtual void scale( \ + Container& a, Type alpha, Container& result) const \ + { \ + SG_SNOTIMPLEMENTED; \ + } + DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_IN_PLACE_SCALE, SGVector) + DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_IN_PLACE_SCALE, SGMatrix) +#undef BACKEND_GENERIC_IN_PLACE_SCALE + +/** + * Wrapper method that sets const values to vectors or matrices. + * + * @see linalg::set_const + */ +#define BACKEND_GENERIC_SET_CONST(Type, Container) \ + virtual void set_const(Container& a, const Type value) const \ + { \ + SG_SNOTIMPLEMENTED; \ + } + DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_SET_CONST, SGVector) + DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_SET_CONST, SGMatrix) +#undef BACKEND_GENERIC_SET_CONST + +/** +* Wrapper method of sum that works with generic vectors or matrices. +* +* @see linalg::sum +*/ +#define BACKEND_GENERIC_SUM(Type, Container) \ + virtual Type sum(const Container& a, bool no_diag) const \ + { \ + SG_SNOTIMPLEMENTED; \ + return 0; \ + } + DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_SUM, SGVector) + DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_SUM, SGMatrix) +#undef BACKEND_GENERIC_SUM + +/** + * Wrapper method of softmax method. + * + * @see linalg::softmax + */ +#define BACKEND_GENERIC_SOFTMAX(Type, Container) \ + virtual void softmax(Container& a) const \ + { \ + SG_SNOTIMPLEMENTED; \ + } + DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_SOFTMAX, SGMatrix) +#undef BACKEND_GENERIC_SOFTMAX + +/** +* Wrapper method of squared error method. +* +* @see linalg::squared_error +*/ +#define BACKEND_GENERIC_SQUARED_ERROR(Type, Container) \ + virtual Type squared_error( \ + const Container& P, const Container& Q) const \ + { \ + SG_SNOTIMPLEMENTED; \ + return 0; \ + } + DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_SQUARED_ERROR, SGMatrix) +#undef BACKEND_GENERIC_SQUARED_ERROR + +/** +* Wrapper method of sum that works with matrix blocks. +* +* @see linalg::sum +*/ +#define BACKEND_GENERIC_BLOCK_SUM(Type, Container) \ + virtual Type sum(const linalg::Block>& a, bool no_diag) \ + const \ + { \ + SG_SNOTIMPLEMENTED; \ + return 0; \ + } + DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_BLOCK_SUM, SGMatrix) +#undef BACKEND_GENERIC_BLOCK_SUM + +/** +* Wrapper method of sum that works with symmetric matrices. +* +* @see linalg::sum_symmetric +*/ +#define BACKEND_GENERIC_SYMMETRIC_SUM(Type, Container) \ + virtual Type sum_symmetric(const Container& a, bool no_diag) const \ + { \ + SG_SNOTIMPLEMENTED; \ + return 0; \ + } + DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_SYMMETRIC_SUM, SGMatrix) +#undef BACKEND_GENERIC_SYMMETRIC_SUM + +/** +* Wrapper method of sum that works with symmetric matrix blocks. +* +* @see linalg::sum +*/ +#define BACKEND_GENERIC_SYMMETRIC_BLOCK_SUM(Type, Container) \ + virtual Type sum_symmetric( \ + const linalg::Block>& a, bool no_diag) const \ + { \ + SG_SNOTIMPLEMENTED; \ + return 0; \ + } + DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_SYMMETRIC_BLOCK_SUM, SGMatrix) +#undef BACKEND_GENERIC_SYMMETRIC_BLOCK_SUM + +/** + * Wrapper method of matrix rowwise sum that works with dense matrices. + * + * @see linalg::colwise_sum + */ +#define BACKEND_GENERIC_COLWISE_SUM(Type, Container) \ + virtual SGVector colwise_sum(const Container& a, bool no_diag) \ + const \ + { \ + SG_SNOTIMPLEMENTED; \ + return 0; \ + } + DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_COLWISE_SUM, SGMatrix) +#undef BACKEND_GENERIC_COLWISE_SUM + +/** +* Wrapper method of matrix colwise sum that works with dense matrices. +* +* @see linalg::colwise_sum +*/ +#define BACKEND_GENERIC_BLOCK_COLWISE_SUM(Type, Container) \ + virtual SGVector colwise_sum( \ + const linalg::Block>& a, bool no_diag) const \ + { \ + SG_SNOTIMPLEMENTED; \ + return 0; \ + } + DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_BLOCK_COLWISE_SUM, SGMatrix) +#undef BACKEND_GENERIC_BLOCK_COLWISE_SUM + +/** + * Wrapper method of matrix rowwise sum that works with dense matrices. + * + * @see linalg::rowwise_sum + */ +#define BACKEND_GENERIC_ROWWISE_SUM(Type, Container) \ + virtual SGVector rowwise_sum(const Container& a, bool no_diag) \ + const \ + { \ + SG_SNOTIMPLEMENTED; \ + return 0; \ + } + DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_ROWWISE_SUM, SGMatrix) +#undef BACKEND_GENERIC_ROWWISE_SUM + +/** +* Wrapper method of matrix rowwise sum that works with dense matrices. +* +* @see linalg::rowwise_sum +*/ +#define BACKEND_GENERIC_BLOCK_ROWWISE_SUM(Type, Container) \ + virtual SGVector rowwise_sum( \ + const linalg::Block>& a, bool no_diag) const \ + { \ + SG_SNOTIMPLEMENTED; \ + return 0; \ + } + DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_BLOCK_ROWWISE_SUM, SGMatrix) +#undef BACKEND_GENERIC_BLOCK_ROWWISE_SUM + +/** + * Wrapper method of svd computation. + * + * @see linalg::svd + */ +#define BACKEND_GENERIC_SVD(Type, Container) \ + virtual void svd( \ + const Container& A, SGVector s, SGMatrix U, \ + bool thin_U, linalg::SVDAlgorithm alg) const \ + { \ + SG_SNOTIMPLEMENTED; \ + } + DEFINE_FOR_NON_INTEGER_PTYPE(BACKEND_GENERIC_SVD, SGMatrix) +#undef BACKEND_GENERIC_SVD + +/** + * Wrapper method of trace computation. + * + * @see linalg::trace + */ +#define BACKEND_GENERIC_TRACE(Type, Container) \ + virtual Type trace(const Container& A) const \ + { \ + SG_SNOTIMPLEMENTED; \ + return 0; \ + } + DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_TRACE, SGMatrix) +#undef BACKEND_GENERIC_TRACE + +/** + * Wrapper method of trace computation. + * + * @see linalg::transpose_matrix + */ +#define BACKEND_GENERIC_TRANSPOSE_MATRIX(Type, Container) \ + virtual Container transpose_matrix(const Container& A) const \ + { \ + SG_SNOTIMPLEMENTED; \ + return 0; \ + } + DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_TRANSPOSE_MATRIX, SGMatrix) +#undef BACKEND_GENERIC_TRANSPOSE_MATRIX + +/** + * Wrapper method of triangular solver. + * + * @see linalg::triangular_solver + */ +#define BACKEND_GENERIC_TRIANGULAR_SOLVER(Type, Container) \ + virtual Container triangular_solver( \ + const SGMatrix& L, const Container& b, \ + const bool lower = true) const \ + { \ + SG_SNOTIMPLEMENTED; \ + return 0; \ + } + DEFINE_FOR_NON_INTEGER_PTYPE( + BACKEND_GENERIC_TRIANGULAR_SOLVER, SGVector) + DEFINE_FOR_NON_INTEGER_PTYPE( + BACKEND_GENERIC_TRIANGULAR_SOLVER, SGMatrix) +#undef BACKEND_GENERIC_TRIANGULAR_SOLVER + +/** + * Wrapper method of set vector or matrix to zero. + * + * @see linalg::zero + */ +#define BACKEND_GENERIC_ZERO(Type, Container) \ + virtual void zero(Container& a) const \ + { \ + SG_SNOTIMPLEMENTED; \ + return; \ + } + DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_ZERO, SGVector) + DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_ZERO, SGMatrix) +#undef BACKEND_GENERIC_ZERO + +/** + * Wrapper method of Transferring data to GPU memory. + * Does nothing if no GPU backend registered. + * + * @see linalg::to_gpu + */ +#define BACKEND_GENERIC_TO_GPU(Type, Container) \ + virtual GPUMemoryBase* to_gpu(const Container&) const \ + { \ + SG_SNOTIMPLEMENTED; \ + return 0; \ + } + DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_TO_GPU, SGVector) + DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_TO_GPU, SGMatrix) +#undef BACKEND_GENERIC_TO_GPU + +/** + * Wrapper method of fetching data from GPU memory. + * + * @see linalg::from_gpu + */ +#define BACKEND_GENERIC_FROM_GPU(Type, Container) \ + virtual void from_gpu(const Container&, Type* data) const \ + { \ + SG_SNOTIMPLEMENTED; \ + } + DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_FROM_GPU, SGVector) + DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_FROM_GPU, SGMatrix) +#undef BACKEND_GENERIC_FROM_GPU #undef DEFINE_FOR_ALL_PTYPE #undef DEFINE_FOR_REAL_PTYPE #undef DEFINE_FOR_NON_INTEGER_PTYPE -}; - + }; } -#endif //LINALG_BACKEND_BASE_H__ +#endif // LINALG_BACKEND_BASE_H__ diff --git a/src/shogun/mathematics/linalg/LinalgBackendEigen.h b/src/shogun/mathematics/linalg/LinalgBackendEigen.h index 99c410b862a..96b80e0da9c 100644 --- a/src/shogun/mathematics/linalg/LinalgBackendEigen.h +++ b/src/shogun/mathematics/linalg/LinalgBackendEigen.h @@ -33,684 +33,625 @@ #ifndef LINALG_BACKEND_EIGEN_H__ #define LINALG_BACKEND_EIGEN_H__ -#include +#include +#include #include #include -#include +#include +#include namespace shogun { -/** @brief Linalg methods with Eigen3 backend */ -class LinalgBackendEigen : public LinalgBackendBase -{ -public: - #define DEFINE_FOR_ALL_PTYPE(METHODNAME, Container) \ - METHODNAME(bool, Container); \ - METHODNAME(char, Container); \ - METHODNAME(int8_t, Container); \ - METHODNAME(uint8_t, Container); \ - METHODNAME(int16_t, Container); \ - METHODNAME(uint16_t, Container); \ - METHODNAME(int32_t, Container); \ - METHODNAME(uint32_t, Container); \ - METHODNAME(int64_t, Container); \ - METHODNAME(uint64_t, Container); \ - METHODNAME(float32_t, Container); \ - METHODNAME(float64_t, Container); \ - METHODNAME(floatmax_t, Container); \ - METHODNAME(complex128_t, Container); \ - - #define DEFINE_FOR_REAL_PTYPE(METHODNAME, Container) \ - METHODNAME(bool, Container); \ - METHODNAME(char, Container); \ - METHODNAME(int8_t, Container); \ - METHODNAME(uint8_t, Container); \ - METHODNAME(int16_t, Container); \ - METHODNAME(uint16_t, Container); \ - METHODNAME(int32_t, Container); \ - METHODNAME(uint32_t, Container); \ - METHODNAME(int64_t, Container); \ - METHODNAME(uint64_t, Container); \ - METHODNAME(float32_t, Container); \ - METHODNAME(float64_t, Container); \ - METHODNAME(floatmax_t, Container); - - #define DEFINE_FOR_NON_INTEGER_PTYPE(METHODNAME, Container) \ - METHODNAME(float32_t, Container); \ - METHODNAME(float64_t, Container); \ - METHODNAME(floatmax_t, Container); \ - METHODNAME(complex128_t, Container); - - #define DEFINE_FOR_NUMERIC_PTYPE(METHODNAME, Container) \ - METHODNAME(char, Container); \ - METHODNAME(int8_t, Container); \ - METHODNAME(uint8_t, Container); \ - METHODNAME(int16_t, Container); \ - METHODNAME(uint16_t, Container); \ - METHODNAME(int32_t, Container); \ - METHODNAME(uint32_t, Container); \ - METHODNAME(int64_t, Container); \ - METHODNAME(uint64_t, Container); \ - METHODNAME(float32_t, Container); \ - METHODNAME(float64_t, Container); \ - METHODNAME(floatmax_t, Container); - - /** Implementation of @see LinalgBackendBase::add */ - #define BACKEND_GENERIC_IN_PLACE_ADD(Type, Container) \ - virtual void add(Container& a, Container& b, Type alpha, \ - Type beta, Container& result) const \ - { \ - add_impl(a, b, alpha, beta, result); \ - } - DEFINE_FOR_NUMERIC_PTYPE(BACKEND_GENERIC_IN_PLACE_ADD, SGVector) - DEFINE_FOR_NUMERIC_PTYPE(BACKEND_GENERIC_IN_PLACE_ADD, SGMatrix) - #undef BACKEND_GENERIC_IN_PLACE_ADD - - /** Implementation of @see LinalgBackendBase::cholesky_factor */ - #define BACKEND_GENERIC_CHOLESKY_FACTOR(Type, Container) \ - virtual Container cholesky_factor(const Container& A, \ - const bool lower) const \ - { \ - return cholesky_factor_impl(A, lower); \ - } - DEFINE_FOR_NON_INTEGER_PTYPE(BACKEND_GENERIC_CHOLESKY_FACTOR, SGMatrix) - #undef BACKEND_GENERIC_CHOLESKY_FACTOR - - /** Implementation of @see LinalgBackendBase::cholesky_solver */ - #define BACKEND_GENERIC_CHOLESKY_SOLVER(Type, Container) \ - virtual SGVector cholesky_solver(const Container& L, \ - const SGVector& b, const bool lower) const \ - { \ - return cholesky_solver_impl(L, b, lower); \ - } - DEFINE_FOR_NON_INTEGER_PTYPE(BACKEND_GENERIC_CHOLESKY_SOLVER, SGMatrix) - #undef BACKEND_GENERIC_CHOLESKY_SOLVER - - /** Implementation of @see LinalgBackendBase::dot */ - #define BACKEND_GENERIC_DOT(Type, Container) \ - virtual Type dot(const Container& a, const Container& b) const \ - { \ - return dot_impl(a, b); \ - } - DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_DOT, SGVector) - #undef BACKEND_GENERIC_DOT - - /** Implementation of @see LinalgBackendBase::element_prod */ - #define BACKEND_GENERIC_IN_PLACE_ELEMENT_PROD(Type, Container) \ - virtual void element_prod(Container& a, Container& b,\ - Container& result) const \ - { \ - element_prod_impl(a, b, result); \ - } - DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_IN_PLACE_ELEMENT_PROD, SGMatrix) - #undef BACKEND_GENERIC_IN_PLACE_ELEMENT_PROD - - /** Implementation of @see LinalgBackendBase::element_prod */ - #define BACKEND_GENERIC_IN_PLACE_BLOCK_ELEMENT_PROD(Type, Container) \ - virtual void element_prod(linalg::Block>& a, \ - linalg::Block>& b, Container& result) const \ - { \ - element_prod_impl(a, b, result); \ - } - DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_IN_PLACE_BLOCK_ELEMENT_PROD, SGMatrix) - #undef BACKEND_GENERIC_IN_PLACE_BLOCK_ELEMENT_PROD - - /** Implementation of @see LinalgBackendBase::logistic */ - #define BACKEND_GENERIC_LOGISTIC(Type, Container) \ - virtual void logistic(Container& a, Container& result) const \ - { \ - logistic_impl(a, result); \ - } - DEFINE_FOR_NUMERIC_PTYPE(BACKEND_GENERIC_LOGISTIC, SGMatrix) - #undef BACKEND_GENERIC_LOGISTIC - - /** Implementation of @see LinalgBackendBase::matrix_prod */ - #define BACKEND_GENERIC_IN_PLACE_MATRIX_PROD(Type, Container) \ - virtual void matrix_prod(SGMatrix& a, Container& b,\ - Container& result, bool transpose_A, bool transpose_B) const \ - { \ - matrix_prod_impl(a, b, result, transpose_A, transpose_B); \ - } - DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_IN_PLACE_MATRIX_PROD, SGVector) - DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_IN_PLACE_MATRIX_PROD, SGMatrix) - #undef BACKEND_GENERIC_IN_PLACE_MATRIX_PROD - - /** Implementation of @see LinalgBackendBase::max */ - #define BACKEND_GENERIC_MAX(Type, Container) \ - virtual Type max(const Container& a) const \ - { \ - return max_impl(a); \ - } - DEFINE_FOR_REAL_PTYPE(BACKEND_GENERIC_MAX, SGVector) - DEFINE_FOR_REAL_PTYPE(BACKEND_GENERIC_MAX, SGMatrix) - #undef BACKEND_GENERIC_MAX - - /** Implementation of @see LinalgBackendBase::mean */ - #define BACKEND_GENERIC_REAL_MEAN(Type, Container) \ - virtual float64_t mean(const Container& a) const \ - { \ - return mean_impl(a); \ - } - DEFINE_FOR_REAL_PTYPE(BACKEND_GENERIC_REAL_MEAN, SGVector) - DEFINE_FOR_REAL_PTYPE(BACKEND_GENERIC_REAL_MEAN, SGMatrix) - #undef BACKEND_GENERIC_REAL_MEAN - - /** Implementation of @see LinalgBackendBase::mean */ - #define BACKEND_GENERIC_COMPLEX_MEAN(Container) \ - virtual complex128_t mean(const Container& a) const \ - { \ - return mean_impl(a); \ - } - BACKEND_GENERIC_COMPLEX_MEAN(SGVector) - BACKEND_GENERIC_COMPLEX_MEAN(SGMatrix) - #undef BACKEND_GENERIC_COMPLEX_MEAN - - /** Implementation of @see LinalgBackendBase::range_fill */ - #define BACKEND_GENERIC_RANGE_FILL(Type, Container) \ - virtual void range_fill(Container& a, const Type start) const \ - { \ - range_fill_impl(a, start); \ - } - DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_RANGE_FILL, SGVector) - DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_RANGE_FILL, SGMatrix) - #undef BACKEND_GENERIC_RANGE_FILL - - /** Implementation of @see linalg::scale */ - #define BACKEND_GENERIC_IN_PLACE_SCALE(Type, Container) \ - virtual void scale(Container& a, Type alpha, Container& result) const \ - { \ - scale_impl(a, alpha, result); \ - } - DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_IN_PLACE_SCALE, SGVector) - DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_IN_PLACE_SCALE, SGMatrix) - #undef BACKEND_GENERIC_IN_PLACE_SCALE - - /** Implementation of @see LinalgBackendBase::set_const */ - #define BACKEND_GENERIC_SET_CONST(Type, Container) \ - virtual void set_const(Container& a, const Type value) const \ - { \ - set_const_impl(a, value); \ - } - DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_SET_CONST, SGVector) - DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_SET_CONST, SGMatrix) - #undef BACKEND_GENERIC_SET_CONST - - /** Implementation of @see LinalgBackendBase::sum */ - #define BACKEND_GENERIC_SUM(Type, Container) \ - virtual Type sum(const Container& a, bool no_diag) const \ - { \ - return sum_impl(a, no_diag); \ - } - DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_SUM, SGVector) - DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_SUM, SGMatrix) - #undef BACKEND_GENERIC_SUM - - /** Implementation of @see LinalgBackendBase::sum */ - #define BACKEND_GENERIC_BLOCK_SUM(Type, Container) \ - virtual Type sum(const linalg::Block>& a, bool no_diag) const \ - { \ - return sum_impl(a, no_diag); \ - } - DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_BLOCK_SUM, SGMatrix) - #undef BACKEND_GENERIC_BLOCK_SUM - - /** Implementation of @see LinalgBackendBase::sum_symmetric */ - #define BACKEND_GENERIC_SYMMETRIC_SUM(Type, Container) \ - virtual Type sum_symmetric(const Container& a, bool no_diag) const \ - { \ - return sum_symmetric_impl(a, no_diag); \ - } - DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_SYMMETRIC_SUM, SGMatrix) - #undef BACKEND_GENERIC_SYMMETRIC_SUM - - /** Implementation of @see LinalgBackendBase::sum_symmetric */ - #define BACKEND_GENERIC_SYMMETRIC_BLOCK_SUM(Type, Container) \ - virtual Type sum_symmetric(const linalg::Block>& a, bool no_diag) const \ - { \ - return sum_symmetric_impl(a, no_diag); \ - } - DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_SYMMETRIC_BLOCK_SUM, SGMatrix) - #undef BACKEND_GENERIC_SYMMETRIC_BLOCK_SUM - - /** Implementation of @see LinalgBackendBase::colwise_sum */ - #define BACKEND_GENERIC_COLWISE_SUM(Type, Container) \ - virtual SGVector colwise_sum(const Container& a, bool no_diag) const \ - { \ - return colwise_sum_impl(a, no_diag); \ - } - DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_COLWISE_SUM, SGMatrix) - #undef BACKEND_GENERIC_COLWISE_SUM - - /** Implementation of @see LinalgBackendBase::colwise_sum */ - #define BACKEND_GENERIC_BLOCK_COLWISE_SUM(Type, Container) \ - virtual SGVector colwise_sum(const linalg::Block>& a, bool no_diag) const \ - { \ - return colwise_sum_impl(a, no_diag); \ - } - DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_BLOCK_COLWISE_SUM, SGMatrix) - #undef BACKEND_GENERIC_BLOCK_COLWISE_SUM - - /** Implementation of @see LinalgBackendBase::rowwise_sum */ - #define BACKEND_GENERIC_ROWWISE_SUM(Type, Container) \ - virtual SGVector rowwise_sum(const Container& a, bool no_diag) const \ - { \ - return rowwise_sum_impl(a, no_diag); \ - } - DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_ROWWISE_SUM, SGMatrix) - #undef BACKEND_GENERIC_ROWWISE_SUM - - /** Implementation of @see LinalgBackendBase::rowwise_sum */ - #define BACKEND_GENERIC_BLOCK_ROWWISE_SUM(Type, Container) \ - virtual SGVector rowwise_sum(const linalg::Block>& a, bool no_diag) const \ - { \ - return rowwise_sum_impl(a, no_diag); \ - } - DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_BLOCK_ROWWISE_SUM, SGMatrix) - #undef BACKEND_GENERIC_BLOCK_ROWWISE_SUM - - #undef DEFINE_FOR_ALL_PTYPE - -private: - /** Eigen3 vector result = alpha*A + beta*B method */ - template - void add_impl(SGVector& a, SGVector& b, T alpha, T beta, SGVector& result) const - { - typename SGVector::EigenVectorXtMap a_eig = a; - typename SGVector::EigenVectorXtMap b_eig = b; - typename SGVector::EigenVectorXtMap result_eig = result; - - result_eig = alpha * a_eig + beta * b_eig; - } - - /** Eigen3 matrix result = alpha*A + beta*B method */ - template - void add_impl(SGMatrix& a, SGMatrix& b, T alpha, T beta, SGMatrix& result) const + /** @brief Linalg methods with Eigen3 backend */ + class LinalgBackendEigen : public LinalgBackendBase { - typename SGMatrix::EigenMatrixXtMap a_eig = a; - typename SGMatrix::EigenMatrixXtMap b_eig = b; - typename SGMatrix::EigenMatrixXtMap result_eig = result; + public: +/** Implementation of @see LinalgBackendBase::add */ +#define BACKEND_GENERIC_IN_PLACE_ADD(Type, Container) \ + virtual void add( \ + Container& a, Container& b, Type alpha, Type beta, \ + Container& result) const; + DEFINE_FOR_NUMERIC_PTYPE(BACKEND_GENERIC_IN_PLACE_ADD, SGVector) + DEFINE_FOR_NUMERIC_PTYPE(BACKEND_GENERIC_IN_PLACE_ADD, SGMatrix) +#undef BACKEND_GENERIC_IN_PLACE_ADD + +/** Implementation of @see LinalgBackendBase::add_col_vec */ +#define BACKEND_GENERIC_ADD_COL_VEC(Type, Container) \ + virtual void add_col_vec( \ + const SGMatrix& A, index_t i, const SGVector& b, \ + Container& result, Type alpha, Type beta) const; + DEFINE_FOR_NUMERIC_PTYPE(BACKEND_GENERIC_ADD_COL_VEC, SGVector) + DEFINE_FOR_NUMERIC_PTYPE(BACKEND_GENERIC_ADD_COL_VEC, SGMatrix) +#undef BACKEND_GENERIC_ADD_COL_VEC + +/** Implementation of @see LinalgBackendBase::add_vector */ +#define BACKEND_GENERIC_ADD(Type, Container) \ + virtual void add_vector( \ + const SGMatrix& A, const SGVector& b, \ + SGMatrix& result, Type alpha, Type beta) const; + DEFINE_FOR_NUMERIC_PTYPE(BACKEND_GENERIC_ADD, SGMatrix) +#undef BACKEND_GENERIC_ADD + +/** Implementation of @see LinalgBackendBase::add_scalar */ +#define BACKEND_GENERIC_ADD_SCALAR(Type, Container) \ + virtual void add_scalar(Container& a, Type b) const; + DEFINE_FOR_NUMERIC_PTYPE(BACKEND_GENERIC_ADD_SCALAR, SGVector) + DEFINE_FOR_NUMERIC_PTYPE(BACKEND_GENERIC_ADD_SCALAR, SGMatrix) +#undef BACKEND_GENERIC_ADD_SCALAR + +/** Implementation of @see LinalgBackendBase::center_matrix */ +#define BACKEND_GENERIC_CENTER_MATRIX(Type, Container) \ + virtual void center_matrix(Container& A) const; + DEFINE_FOR_NON_INTEGER_PTYPE(BACKEND_GENERIC_CENTER_MATRIX, SGMatrix) +#undef BACKEND_GENERIC_CENTER_MATRIX + +/** Implementation of @see LinalgBackendBase::cholesky_factor */ +#define BACKEND_GENERIC_CHOLESKY_FACTOR(Type, Container) \ + virtual Container cholesky_factor( \ + const Container& A, const bool lower) const; + DEFINE_FOR_NON_INTEGER_PTYPE(BACKEND_GENERIC_CHOLESKY_FACTOR, SGMatrix) +#undef BACKEND_GENERIC_CHOLESKY_FACTOR + +/** Implementation of @see LinalgBackendBase::cholesky_solver */ +#define BACKEND_GENERIC_CHOLESKY_SOLVER(Type, Container) \ + virtual SGVector cholesky_solver( \ + const Container& L, const SGVector& b, const bool lower) \ + const; + DEFINE_FOR_NON_INTEGER_PTYPE(BACKEND_GENERIC_CHOLESKY_SOLVER, SGMatrix) +#undef BACKEND_GENERIC_CHOLESKY_SOLVER + +/** Implementation of @see linalg::cross_entropy */ +#define BACKEND_GENERIC_CROSS_ENTROPY(Type, Container) \ + virtual Type cross_entropy( \ + const Container& P, const Container& Q) const; + DEFINE_FOR_NON_INTEGER_REAL_PTYPE( + BACKEND_GENERIC_CROSS_ENTROPY, SGMatrix) +#undef BACKEND_GENERIC_CROSS_ENTROPY + +/** Implementation of @see LinalgBackendBase::dot */ +#define BACKEND_GENERIC_DOT(Type, Container) \ + virtual Type dot(const Container& a, const Container& b) const; + DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_DOT, SGVector) +#undef BACKEND_GENERIC_DOT + +/** Implementation of @see LinalgBackendBase::eigen_solver */ +#define BACKEND_GENERIC_EIGEN_SOLVER(Type, Container) \ + virtual void eigen_solver( \ + const Container& A, SGVector& eigenvalues, \ + SGMatrix& eigenvectors) const; + DEFINE_FOR_NON_INTEGER_PTYPE(BACKEND_GENERIC_EIGEN_SOLVER, SGMatrix) +#undef BACKEND_GENERIC_EIGEN_SOLVER + +/** Implementation of @see LinalgBackendBase::eigen_solver_symmetric */ +#define BACKEND_GENERIC_EIGEN_SOLVER_SYMMETRIC(Type, Container) \ + virtual void eigen_solver_symmetric( \ + const Container& A, SGVector& eigenvalues, \ + SGMatrix& eigenvectors, index_t k) const; + DEFINE_FOR_NON_INTEGER_PTYPE( + BACKEND_GENERIC_EIGEN_SOLVER_SYMMETRIC, SGMatrix) +#undef BACKEND_GENERIC_EIGEN_SOLVER_SYMMETRIC + +/** Implementation of @see LinalgBackendBase::element_prod */ +#define BACKEND_GENERIC_IN_PLACE_ELEMENT_PROD(Type, Container) \ + virtual void element_prod( \ + Container& a, Container& b, Container& result) \ + const; + DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_IN_PLACE_ELEMENT_PROD, SGMatrix) +#undef BACKEND_GENERIC_IN_PLACE_ELEMENT_PROD + +/** Implementation of @see LinalgBackendBase::element_prod */ +#define BACKEND_GENERIC_IN_PLACE_BLOCK_ELEMENT_PROD(Type, Container) \ + virtual void element_prod( \ + linalg::Block>& a, linalg::Block>& b, \ + Container& result) const; + DEFINE_FOR_ALL_PTYPE( + BACKEND_GENERIC_IN_PLACE_BLOCK_ELEMENT_PROD, SGMatrix) +#undef BACKEND_GENERIC_IN_PLACE_BLOCK_ELEMENT_PROD + +/** Implementation of @see linalg::exponent */ +#define BACKEND_GENERIC_EXPONENT(Type, Container) \ + virtual void exponent(const Container& a, Container& result) \ + const + DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_EXPONENT, SGVector) + DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_EXPONENT, SGMatrix) +#undef BACKEND_GENERIC_EXPONENT + +/** Implementation of @see LinalgBackendBase::identity */ +#define BACKEND_GENERIC_IDENTITY(Type, Container) \ + virtual void identity(Container& identity_matrix) const; + DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_IDENTITY, SGMatrix) +#undef BACKEND_GENERIC_IDENTITY + +/** Implementation of @see LinalgBackendBase::logistic */ +#define BACKEND_GENERIC_LOGISTIC(Type, Container) \ + virtual void logistic(Container& a, Container& result) const; + DEFINE_FOR_NUMERIC_PTYPE(BACKEND_GENERIC_LOGISTIC, SGMatrix) +#undef BACKEND_GENERIC_LOGISTIC + +/** Implementation of @see LinalgBackendBase::matrix_prod */ +#define BACKEND_GENERIC_IN_PLACE_MATRIX_PROD(Type, Container) \ + virtual void matrix_prod( \ + SGMatrix& a, Container& b, Container& result, \ + bool transpose_A, bool transpose_B) const; + DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_IN_PLACE_MATRIX_PROD, SGVector) + DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_IN_PLACE_MATRIX_PROD, SGMatrix) +#undef BACKEND_GENERIC_IN_PLACE_MATRIX_PROD + +/** Implementation of @see LinalgBackendBase::max */ +#define BACKEND_GENERIC_MAX(Type, Container) \ + virtual Type max(const Container& a) const; + DEFINE_FOR_REAL_PTYPE(BACKEND_GENERIC_MAX, SGVector) + DEFINE_FOR_REAL_PTYPE(BACKEND_GENERIC_MAX, SGMatrix) +#undef BACKEND_GENERIC_MAX + +/** Implementation of @see LinalgBackendBase::mean */ +#define BACKEND_GENERIC_REAL_MEAN(Type, Container) \ + virtual float64_t mean(const Container& a) const; + DEFINE_FOR_REAL_PTYPE(BACKEND_GENERIC_REAL_MEAN, SGVector) + DEFINE_FOR_REAL_PTYPE(BACKEND_GENERIC_REAL_MEAN, SGMatrix) +#undef BACKEND_GENERIC_REAL_MEAN + +/** Implementation of @see LinalgBackendBase::mean */ +#define BACKEND_GENERIC_COMPLEX_MEAN(Container) \ + virtual complex128_t mean(const Container& a) const; + BACKEND_GENERIC_COMPLEX_MEAN(SGVector) + BACKEND_GENERIC_COMPLEX_MEAN(SGMatrix) +#undef BACKEND_GENERIC_COMPLEX_MEAN + +/** Implementation of @see linalg::multiply_by_logistic_derivative */ +#define BACKEND_GENERIC_MULTIPLY_BY_LOGISTIC_DERIV(Type, Container) \ + virtual void multiply_by_logistic_derivative( \ + Container& a, Container& result) const; + DEFINE_FOR_NUMERIC_PTYPE( + BACKEND_GENERIC_MULTIPLY_BY_LOGISTIC_DERIV, SGMatrix) +#undef BACKEND_GENERIC_MULTIPLY_BY_LOGISTIC_DERIV + +/** Implementation of @see linalg::multiply_by_rectified_linear_derivative */ +#define BACKEND_GENERIC_MULTIPLY_BY_RECTIFIED_LINEAR_DERIV(Type, Container) \ + virtual void multiply_by_rectified_linear_derivative( \ + Container& a, Container& result) const; + DEFINE_FOR_NON_INTEGER_REAL_PTYPE( + BACKEND_GENERIC_MULTIPLY_BY_RECTIFIED_LINEAR_DERIV, SGMatrix) +#undef BACKEND_GENERIC_MULTIPLY_BY_RECTIFIED_LINEAR_DERIV + +/** Implementation of @see LinalgBackendBase::qr_solver */ +#define BACKEND_GENERIC_QR_SOLVER(Type, Container) \ + virtual Container qr_solver( \ + const SGMatrix& A, const Container& b) const; + DEFINE_FOR_NON_INTEGER_PTYPE(BACKEND_GENERIC_QR_SOLVER, SGVector) + DEFINE_FOR_NON_INTEGER_PTYPE(BACKEND_GENERIC_QR_SOLVER, SGMatrix) +#undef BACKEND_GENERIC_QR_SOLVER + +/** Implementation of @see LinalgBackendBase::range_fill */ +#define BACKEND_GENERIC_RANGE_FILL(Type, Container) \ + virtual void range_fill(Container& a, const Type start) const; + DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_RANGE_FILL, SGVector) + DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_RANGE_FILL, SGMatrix) +#undef BACKEND_GENERIC_RANGE_FILL + +/** Implementation of @see linalg::rectified_linear */ +#define BACKEND_GENERIC_RECTIFIED_LINEAR(Type, Container) \ + virtual void rectified_linear(Container& a, Container& result) \ + const; + DEFINE_FOR_REAL_PTYPE(BACKEND_GENERIC_RECTIFIED_LINEAR, SGMatrix) +#undef BACKEND_GENERIC_RECTIFIED_LINEAR + +/** Implementation of @see linalg::scale */ +#define BACKEND_GENERIC_IN_PLACE_SCALE(Type, Container) \ + virtual void scale( \ + Container& a, Type alpha, Container& result) const; + DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_IN_PLACE_SCALE, SGVector) + DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_IN_PLACE_SCALE, SGMatrix) +#undef BACKEND_GENERIC_IN_PLACE_SCALE + +/** Implementation of @see LinalgBackendBase::set_const */ +#define BACKEND_GENERIC_SET_CONST(Type, Container) \ + virtual void set_const(Container& a, const Type value) const; + DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_SET_CONST, SGVector) + DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_SET_CONST, SGMatrix) +#undef BACKEND_GENERIC_SET_CONST + +/** Implementation of @see linalg::softmax */ +#define BACKEND_GENERIC_SOFTMAX(Type, Container) \ + virtual void softmax(Container& a) const; + DEFINE_FOR_NON_INTEGER_REAL_PTYPE(BACKEND_GENERIC_SOFTMAX, SGMatrix) +#undef BACKEND_GENERIC_SOFTMAX + +/** Implementation of @see linalg::squared_error */ +#define BACKEND_GENERIC_SQUARED_ERROR(Type, Container) \ + virtual Type squared_error( \ + const Container& P, const Container& Q) const; + DEFINE_FOR_NON_INTEGER_REAL_PTYPE( + BACKEND_GENERIC_SQUARED_ERROR, SGMatrix) +#undef BACKEND_GENERIC_SQUARED_ERROR + +/** Implementation of @see LinalgBackendBase::sum */ +#define BACKEND_GENERIC_SUM(Type, Container) \ + virtual Type sum(const Container& a, bool no_diag) const; + DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_SUM, SGVector) + DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_SUM, SGMatrix) +#undef BACKEND_GENERIC_SUM + +/** Implementation of @see LinalgBackendBase::sum */ +#define BACKEND_GENERIC_BLOCK_SUM(Type, Container) \ + virtual Type sum(const linalg::Block>& a, bool no_diag) \ + const; + DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_BLOCK_SUM, SGMatrix) +#undef BACKEND_GENERIC_BLOCK_SUM + +/** Implementation of @see LinalgBackendBase::sum_symmetric */ +#define BACKEND_GENERIC_SYMMETRIC_SUM(Type, Container) \ + virtual Type sum_symmetric(const Container& a, bool no_diag) const; + DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_SYMMETRIC_SUM, SGMatrix) +#undef BACKEND_GENERIC_SYMMETRIC_SUM + +/** Implementation of @see LinalgBackendBase::sum_symmetric */ +#define BACKEND_GENERIC_SYMMETRIC_BLOCK_SUM(Type, Container) \ + virtual Type sum_symmetric( \ + const linalg::Block>& a, bool no_diag) const; + DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_SYMMETRIC_BLOCK_SUM, SGMatrix) +#undef BACKEND_GENERIC_SYMMETRIC_BLOCK_SUM + +/** Implementation of @see LinalgBackendBase::colwise_sum */ +#define BACKEND_GENERIC_COLWISE_SUM(Type, Container) \ + virtual SGVector colwise_sum(const Container& a, bool no_diag) \ + const; + DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_COLWISE_SUM, SGMatrix) +#undef BACKEND_GENERIC_COLWISE_SUM + +/** Implementation of @see LinalgBackendBase::colwise_sum */ +#define BACKEND_GENERIC_BLOCK_COLWISE_SUM(Type, Container) \ + virtual SGVector colwise_sum( \ + const linalg::Block>& a, bool no_diag) const; + DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_BLOCK_COLWISE_SUM, SGMatrix) +#undef BACKEND_GENERIC_BLOCK_COLWISE_SUM + +/** Implementation of @see LinalgBackendBase::rowwise_sum */ +#define BACKEND_GENERIC_ROWWISE_SUM(Type, Container) \ + virtual SGVector rowwise_sum(const Container& a, bool no_diag) \ + const; + DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_ROWWISE_SUM, SGMatrix) +#undef BACKEND_GENERIC_ROWWISE_SUM + +/** Implementation of @see LinalgBackendBase::rowwise_sum */ +#define BACKEND_GENERIC_BLOCK_ROWWISE_SUM(Type, Container) \ + virtual SGVector rowwise_sum( \ + const linalg::Block>& a, bool no_diag) const; + DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_BLOCK_ROWWISE_SUM, SGMatrix) +#undef BACKEND_GENERIC_BLOCK_ROWWISE_SUM + +/** Implementation of @see LinalgBackendBase::svd */ +#define BACKEND_GENERIC_SVD(Type, Container) \ + virtual void svd( \ + const Container& A, SGVector s, Container U, \ + bool thin_U, linalg::SVDAlgorithm alg) const; + DEFINE_FOR_NON_INTEGER_PTYPE(BACKEND_GENERIC_SVD, SGMatrix) +#undef BACKEND_GENERIC_SVD + +/** Implementation of @see LinalgBackendBase::trace */ +#define BACKEND_GENERIC_TRACE(Type, Container) \ + virtual Type trace(const Container& A) const; + DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_TRACE, SGMatrix) +#undef BACKEND_GENERIC_TRACE + +/** Implementation of @see LinalgBackendBase::transpose_matrix */ +#define BACKEND_GENERIC_TRANSPOSE_MATRIX(Type, Container) \ + virtual Container transpose_matrix(const Container& A) const; + DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_TRANSPOSE_MATRIX, SGMatrix) +#undef BACKEND_GENERIC_TRANSPOSE_MATRIX + +/** Implementation of @see LinalgBackendBase::triangular_solver */ +#define BACKEND_GENERIC_TRIANGULAR_SOLVER(Type, Container) \ + virtual Container triangular_solver( \ + const SGMatrix& L, const Container& b, const bool lower) \ + const; + DEFINE_FOR_NON_INTEGER_PTYPE( + BACKEND_GENERIC_TRIANGULAR_SOLVER, SGVector) + DEFINE_FOR_NON_INTEGER_PTYPE( + BACKEND_GENERIC_TRIANGULAR_SOLVER, SGMatrix) +#undef BACKEND_GENERIC_TRIANGULAR_SOLVER + +/** Implementation of @see LinalgBackendBase::zero */ +#define BACKEND_GENERIC_ZERO(Type, Container) \ + virtual void zero(Container& a) const; + DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_ZERO, SGVector) + DEFINE_FOR_ALL_PTYPE(BACKEND_GENERIC_ZERO, SGMatrix) +#undef BACKEND_GENERIC_ZERO - result_eig = alpha * a_eig + beta * b_eig; - } +#undef DEFINE_FOR_ALL_PTYPE +#undef DEFINE_FOR_REAL_PTYPE +#undef DEFINE_FOR_NON_INTEGER_PTYPE +#undef DEFINE_FOR_NUMERIC_PTYPE - /** Eigen3 Cholesky decomposition */ - template - SGMatrix cholesky_factor_impl(const SGMatrix& A, const bool lower) const - { - SGMatrix c(A.num_rows, A.num_cols); - set_const_impl(c, 0); - typename SGMatrix::EigenMatrixXtMap A_eig = A; - typename SGMatrix::EigenMatrixXtMap c_eig = c; - - Eigen::LLT > llt(A_eig); - - //compute matrix L or U - if(lower==false) - c_eig = llt.matrixU(); - else - c_eig = llt.matrixL(); - - /* - * checking for success - * - * 0: Eigen::Success. Decomposition was successful - * 1: Eigen::NumericalIssue. The provided data did not satisfy the prerequisites. + private: + /** Eigen3 vector result = alpha*A + beta*B method */ + template + void add_impl( + SGVector& a, SGVector& b, T alpha, T beta, + SGVector& result) const; + + /** Eigen3 matrix result = alpha*A + beta*B method */ + template + void add_impl( + SGMatrix& a, SGMatrix& b, T alpha, T beta, + SGMatrix& result) const; + + /** Eigen3 add column vector method */ + template + void add_col_vec_impl( + const SGMatrix& A, index_t i, const SGVector& b, + SGMatrix& result, T alpha, T beta) const; + + /** Eigen3 add column vector method */ + template + void add_col_vec_impl( + const SGMatrix& A, index_t i, const SGVector& b, + SGVector& result, T alpha, T beta) const; + + /** Eigen3 add vector to each column of matrix method */ + template + void add_vector_impl( + const SGMatrix& A, const SGVector& b, SGMatrix& result, + T alpha, T beta) const; + + /** Eigen3 vector add scalar method */ + template + void add_scalar_impl(SGVector& a, T b) const; + + /** Eigen3 matrix add scalar method */ + template + void add_scalar_impl(SGMatrix& a, T b) const; + + /** Eigen3 center matrix method */ + template + void center_matrix_impl(SGMatrix& A) const; + + /** Eigen3 Cholesky decomposition */ + template + SGMatrix + cholesky_factor_impl(const SGMatrix& A, const bool lower) const; + + /** Eigen3 Cholesky solver */ + template + SGVector cholesky_solver_impl( + const SGMatrix& L, const SGVector& b, const bool lower) const; + + /** Eigen3 cross_entropy method + * The cross entropy is defined as \f$ H(P,Q) = - \sum_{ij} + * P[i,j]log(Q[i,j]) \f$ */ - REQUIRE(llt.info()!=Eigen::NumericalIssue, "Matrix is not Hermitian positive definite!\n"); + template + T cross_entropy_impl(const SGMatrix& p, const SGMatrix& q) const; - return c; - } - - /** Eigen3 Cholesky solver */ - template - SGVector cholesky_solver_impl(const SGMatrix& L, const SGVector& b, - const bool lower) const - { - SGVector x(b.size()); - set_const_impl(x, 0); - typename SGMatrix::EigenMatrixXtMap L_eig = L; - typename SGVector::EigenVectorXtMap b_eig = b; - typename SGVector::EigenVectorXtMap x_eig = x; - - if (lower == false) - { - Eigen::TriangularView::EigenMatrixXt, - 0, Eigen::Stride<0,0> >, Eigen::Upper> tlv(L_eig); - - x_eig = (tlv.transpose()).solve(tlv.solve(b_eig)); - } - else - { - Eigen::TriangularView::EigenMatrixXt, - 0, Eigen::Stride<0,0> >, Eigen::Lower> tlv(L_eig); - x_eig = (tlv.transpose()).solve(tlv.solve(b_eig)); - } - - return x; - } - - /** Eigen3 vector dot-product method */ - template - T dot_impl(const SGVector& a, const SGVector& b) const - { - return (typename SGVector::EigenVectorXtMap(a)).dot(typename SGVector::EigenVectorXtMap(b)); - } - - /** Eigen3 matrix in-place elementwise product method */ - template - void element_prod_impl(SGMatrix& a, SGMatrix& b, SGMatrix& result) const - { - typename SGMatrix::EigenMatrixXtMap a_eig = a; - typename SGMatrix::EigenMatrixXtMap b_eig = b; - typename SGMatrix::EigenMatrixXtMap result_eig = result; - - result_eig = a_eig.array() * b_eig.array(); - } - - /** Eigen3 logistic method. Calculates f(x) = 1/(1+exp(-x)) */ - template - void logistic_impl(SGMatrix& a, SGMatrix& result) const - { - typename SGMatrix::EigenMatrixXtMap a_eig = a; - typename SGMatrix::EigenMatrixXtMap result_eig = result; - - result_eig = (T)1 / (1 + ((-1 * a_eig).array()).exp()); - } - - /** Eigen3 matrix block in-place elementwise product method */ - template - void element_prod_impl(linalg::Block>& a, - linalg::Block>& b, SGMatrix& result) const - { - typename SGMatrix::EigenMatrixXtMap a_eig = a.m_matrix; - typename SGMatrix::EigenMatrixXtMap b_eig = b.m_matrix; - typename SGMatrix::EigenMatrixXtMap result_eig = result; - - Eigen::Block::EigenMatrixXtMap> a_block = - a_eig.block(a.m_row_begin, a.m_col_begin, a.m_row_size, a.m_col_size); - Eigen::Block::EigenMatrixXtMap> b_block = - b_eig.block(b.m_row_begin, b.m_col_begin, b.m_row_size, b.m_col_size); - - result_eig = a_block.array() * b_block.array(); - } - - /** Eigen3 matrix * vector in-place product method */ - template - void matrix_prod_impl(SGMatrix& a, SGVector& b, SGVector& result, - bool transpose, bool transpose_B=false) const - { - typename SGMatrix::EigenMatrixXtMap a_eig = a; - typename SGVector::EigenVectorXtMap b_eig = b; - typename SGVector::EigenVectorXtMap result_eig = result; - - if (transpose) - result_eig = a_eig.transpose() * b_eig; - else - result_eig = a_eig * b_eig; - } - - /** Eigen3 matrix in-place product method */ - template - void matrix_prod_impl(SGMatrix& a, SGMatrix& b, SGMatrix& result, - bool transpose_A, bool transpose_B) const - { - typename SGMatrix::EigenMatrixXtMap a_eig = a; - typename SGMatrix::EigenMatrixXtMap b_eig = b; - typename SGMatrix::EigenMatrixXtMap result_eig = result; - - if (transpose_A && transpose_B) - result_eig = a_eig.transpose() * b_eig.transpose(); - - else if (transpose_A) - result_eig = a_eig.transpose() * b_eig; - - else if (transpose_B) - result_eig = a_eig * b_eig.transpose(); - - else - result_eig = a_eig * b_eig; - } - - /** Return the largest element in the vector with Eigen3 library */ - template - T max_impl(const SGVector& vec) const - { - return (typename SGVector::EigenVectorXtMap(vec)).maxCoeff(); - } - - /** Return the largest element in the matrix with Eigen3 library */ - template - T max_impl(const SGMatrix& mat) const - { - return (typename SGMatrix::EigenMatrixXtMap(mat)).maxCoeff(); - } - - /** Real eigen3 vector and matrix mean method */ - template class Container> - typename std::enable_if::value, float64_t>::type - mean_impl(const Container& a) const - { - return sum_impl(a)/(float64_t(a.size())); - } - - /** Complex eigen3 vector and matrix mean method */ - template