# shogun-toolbox/shogun

The Shogun Machine Learning Toolbox (Source Code)
C++ Python CMake C Matlab R Other
Latest commit 112dd79 Apr 25, 2016 Merge pull request #3181 from yorkerlin/fix1
fix an issue related to enum class for the precise build 2
 Failed to load latest commit information. applications Sep 12, 2015 benchmarks Mar 23, 2016 cmake Apr 9, 2016 configs Apr 7, 2016 data @ c964e86 Mar 30, 2016 doc Apr 22, 2016 examples Apr 16, 2016 scripts Apr 30, 2015 src Apr 25, 2016 tests Apr 25, 2016 .clang_complete Aug 14, 2013 .gitignore Mar 4, 2016 .gitmodules Feb 10, 2016 .travis.yml Mar 12, 2016 CMakeLists.txt Apr 9, 2016 COPYING Dec 22, 2014 CTestConfig.cmake Aug 9, 2013 NEWS Apr 6, 2016 README.md Mar 3, 2016

# The SHOGUN machine learning toolbox

Develop branch build status:

Other links that may be useful:

• See INSTALL for first steps on installation and running SHOGUN.
• See README.developer for the developer documentation.
• See README.cmake for setting particular build options with SHOGUN and cmake.

## Introduction

The machine learning toolbox's focus is on large scale kernel methods and especially on Support Vector Machines (SVM) [1]. It provides a generic SVM object interfacing to several different SVM implementations, among them the state of the art LibSVM [2] and SVMlight [3]. Each of the SVMs can be combined with a variety of kernels. The toolbox not only provides efficient implementations of the most common kernels, like the Linear, Polynomial, Gaussian and Sigmoid Kernel but also comes with a number of recent string kernels as e.g. the Locality Improved [4], Fischer [5], TOP [6], Spectrum [7], Weighted Degree Kernel (with shifts) [8, 9, 10]. For the latter the efficient LINADD [10] optimizations are implemented. Also SHOGUN offers the freedom of working with custom pre-computed kernels. One of its key features is the combined kernel which can be constructed by a weighted linear combination of a number of sub-kernels, each of which not necessarily working on the same domain. An optimal sub-kernel weighting can be learned using Multiple Kernel Learning [11, 12, 16]. Currently SVM 2-class classification and regression problems can be dealt with. However SHOGUN also implements a number of linear methods like Linear Discriminant Analysis (LDA), Linear Programming Machine (LPM), (Kernel) Perceptrons and features algorithms to train hidden markov models. The input feature-objects can be dense, sparse or strings, and of types int/short/double/char. In addition, they can be converted into different feature types. Chains of preprocessors (e.g. substracting the mean) can be attached to each feature object allowing for on-the-fly pre-processing.

Shogun got initiated by Soeren Sonnenburg and Gunnar Raetsch (thats where the name ShoGun originates from). It is now developed by a much larger Team cf. AUTHORS and would not have been possible without the patches and bug reports by various people and by the various authors of other machine learning packages that we utilize. See CONTRIBUTIONS for a detailled list.

## Interfaces

SHOGUN is implemented in C++ and interfaces to Matlab(tm), R, Octave, Java, C#, Ruby, Lua and Python.

The following table depicts the status of each interface available in SHOGUN:

Interface Status
python_modular mature (no known problems)
octave_modular mature (no known problems)
java_modular stable (no known problems; not all examples are ported)
ruby_modular stable (no known problems; only few examples ported)
csharp_modular stable (no known problems; not all examples are ported)
lua_modular alpha (some examples work, string typemaps are unstable
perl_modular pre-alpha (work in progress quality)
r_modular pre-alpha (SWIG does not properly handle reference counting and thus only for the brave,
--disable-reference-counting to get it to work, but beware that it will leak memory; disabled by default)
octave_static mature (no known problems)
matlab_static mature (no known problems)
python_static mature (no known problems)
r_static mature (no known problems)
libshogun_static mature (no known problems)
cmdline_static stable (but some data types incomplete)
elwms_static this is the eierlegendewollmilchsau interface, a chimera that in one file interfaces with python, octave, r, matlab and provides the run_python command to run code in python using the in octave,r,matlab available variables, etc)

Visit http://www.shogun-toolbox.org/doc/en/current for further information.

## Platforms

Debian GNU/Linux, Mac OSX and WIN32/CYGWIN are supported platforms (see the INSTALL file for generic and platform specific installation instructions).

## Contents

The following directories are found in the source distribution.

• src - source code.
• data - data sets (required for some examples / applications - these need to be downloaded separately via the download site or git submodule update --init from the root of the git checkout.
• doc - documentation (to be built using doxygen), ipython notebooks, and the PDF tutorial.
• examples - example files for all interfaces.
• applications - applications of SHOGUN.
• benchmarks - speed benchmarks.
• tests - unit and integration tests.
• cmake - cmake build scripts

## Applications

We have successfully used this toolbox to tackle the following sequence analysis problems: Protein Super Family classification[6], Splice Site Prediction [8, 13, 14], Interpreting the SVM Classifier [11, 12], Splice Form Prediction [8], Alternative Splicing [9] and Promotor Prediction [15]. Some of them come with no less than 10 million training examples, others with 7 billion test examples.

Except for the files classifier/svm/Optimizer.{cpp,h}, classifier/svm/SVM_light.{cpp,h}, regression/svr/SVR_light.{cpp,h} and the kernel caching functions in kernel/Kernel.{cpp,h} which are (C) Torsten Joachims and follow a different licensing scheme (cf. LICENSE_SVMlight) SHOGUN is generally licensed under the GPL version 3 or any later version (cf. LICENSE) with code borrowed from various GPL compatible libraries from various places (cf. CONTRIBUTIONS). See also LICENSE_msufsort and LICENSE_tapkee.

## References

[1] C. Cortes and V.N. Vapnik. Support-vector networks. Machine Learning, 20(3):273--297, 1995.

[2] J. Liu, S. Ji, and J. Ye. SLEP: Sparse Learning with Efficient Projections. Arizona State University, 2009. http://www.public.asu.edu/~jye02/Software/SLEP.

[3] C.C. Chang and C.J. Lin. Libsvm: Introduction and benchmarks. Technical report, Department of Computer Science and Information Engineering, National Taiwan University, Taipei, 2000.

[4] T. Joachims. Making large-scale SVM learning practical. In B.~Schoelkopf, C.J.C. Burges, and A.J. Smola, editors, Advances in Kernel Methods - Support Vector Learning, pages 169--184, Cambridge, MA, 1999. MIT Press.

[5] A.Zien, G.Raetsch, S.Mika, B.Schoelkopf, T.Lengauer, and K.-R. Mueller. Engineering Support Vector Machine Kernels That Recognize Translation Initiation Sites. Bioinformatics, 16(9):799-807, September 2000.

[6] T.S. Jaakkola and D.Haussler.Exploiting generative models in discriminative classifiers. In M.S. Kearns, S.A. Solla, and D.A. Cohn, editors, Advances in Neural Information Processing Systems, volume 11, pages 487-493, 1999.

[7] K.Tsuda, M.Kawanabe, G.Raetsch, S.Sonnenburg, and K.R. Mueller. A new discriminative kernel from probabilistic models. Neural Computation, 14:2397--2414, 2002.

[8] C.Leslie, E.Eskin, and W.S. Noble. The spectrum kernel: A string kernel for SVM protein classification. In R.B. Altman, A.K. Dunker, L.Hunter, K.Lauderdale, and T.E. Klein, editors, Proceedings of the Pacific Symposium on Biocomputing, pages 564-575, Kaua'i, Hawaii, 2002.

[9] G.Raetsch and S.Sonnenburg. Accurate Splice Site Prediction for Caenorhabditis Elegans, pages 277-298. MIT Press series on Computational Molecular Biology. MIT Press, 2004.

[10] G.Raetsch, S.Sonnenburg, and B.Schoelkopf. RASE: recognition of alternatively spliced exons in c. elegans. Bioinformatics, 21:i369--i377, June 2005.

[11] S.Sonnenburg, G.Raetsch, and B.Schoelkopf. Large scale genomic sequence SVM classifiers. In Proceedings of the 22nd International Machine Learning Conference. ACM Press, 2005.

[12] S.Sonnenburg, G.Raetsch, and C.Schaefer. Learning interpretable SVMs for biological sequence classification. In RECOMB 2005, LNBI 3500, pages 389-407. Springer-Verlag Berlin Heidelberg, 2005.

[13] G.Raetsch, S.Sonnenburg, and C.Schaefer. Learning Interpretable SVMs for Biological Sequence Classification. BMC Bioinformatics, Special Issue from NIPS workshop on New Problems and Methods in Computational Biology Whistler, Canada, 18 December 2004, 7:(Suppl. 1):S9, March 2006.

[14] S. Sonnenburg. New methods for splice site recognition. Master's thesis, Humboldt University, 2002. Supervised by K.R. Mueller H.D. Burkhard and G. Raetsch.

[15] S. Sonnenburg, G. Raetsch, A. Jagota, and K.R. Mueller. New methods for splice-site recognition. In Proceedings of the International Conference on Artifical Neural Networks, 2002. Copyright by Springer.

[16] S. Sonnenburg, A. Zien, and G. Raetsch. ARTS: Accurate Recognition of Transcription Starts in Human. 2006.

[17] S. Sonnenburg, G. Raetsch, C.Schaefer, and B.Schoelkopf, Large Scale Multiple Kernel Learning, Journal of Machine Learning Research, 2006, K.Bennett and E.P. Hernandez Editors.