release 2015-04 follow-up

[stonebig]

WinPython 2015-04 (in 2015-05-12: IPython-3.1+, Qt5 minimal , mingwpy, Theano, Flask, ...)

features:

• improve WPPM so it doesn't show 'failed to install' packages (Spyder in WinPython-64bit-3.4.3.2 missing jedi and sphinx #74)
• Sphinx is now included
• Numba 0.18.2 + llvmlite 0.4.0
• Theano
• workaround Vitables.bat missing Spyder option Vitables doesn't work #75
• werkzeug + Flask
• redis + pymongo interfaces
• new mingw toolchain (mingwpy 2015-04-17 09h00)
• brings back openblas 0.2.14 (0.2.15 in time?), as theano lasagne uses it
• standard test package (nose)
• workaround pip not refreshing it's package list (once loaded)
• Shift WPPM to PyQt5 compatible code (in theory)
• fixed historic mingwpy configuration miss uncovered by Theano testing
• add missing adodbapi package.
• in winpython_checker: remove old titles, reduce db.py, add seaborn and theano
• fix build parameter bug that was avoiding pip 6.1.1 upgrade
• http://xkcd.com/1513/
• jedi
• a fix for pip refresh pip.get_installed_distributions() doesn't check if a package has been uninstalled (or added) pypa/pip#2695 (comment)
• remove un-released lasagne
• include a seaborn demo from Jakevdp
• pyqtgraph (a replacement to un-maintained PyQwt/GuiQwt, that should survive Qt5)
• pandas 0.16.1
• update website front page
• upload package and source used for release

Infrastructure:

• improve Build process so that 'pip wheel' install failures don't happen unnoticed (Spyder in WinPython-64bit-3.4.3.2 missing jedi and sphinx #74)
• flavors are not popular: reduce effort to notebooks 'how-to install this'
• baby-step1 on AppVeyor, after Olivier Grisel Pycon2015 talk (https://www.youtube.com/watch?v=d-p6lJJObLU) (applied on sqlite_bro package)
• python 3.5 status: we may need a plan B for mingwpy
• looking for a spare wheel for PyQt5: there is none yet
• rodeo (or equivalent) Stand-alone application yhat/rodeo#30 (comment)
• considering theanets, keras, blocks, pylearn2, sklearn-theano, scikit-neuralnetwork (looking for a pypi page, then a py3 wheel, then )

Other consideration:

• WEB Gui helpers: Vispy 0.4 and Qt 5.5 are hoped for next version,
• Deep Learning: theano is included (but not the big graphic card),
• Qt4 -> Qt5 move: PyQtGraph is added as preparatory step to remove un-maintained PyQwt.

MD5	SHA-1	Binary
aa82fa67756bd1880ee7c20df1aecb66	249320fb396023ad4182fd89bf49792af0cc3965	winpython-32bit-3.4.3.3.exe
21837dda642c6c1ae3d011bf13b383f4	da6222a79e58bd9b5759b7be926b80776d423b3e	winpython-64bit-3.4.3.3.exe
04849a7f9209fb6bdb05f9cf2f2ba50a	24c3a5d80e698a013d0c2e2792885373df4caf8b	winpython-32bit-2.7.9.5.exe
cf4c4d064ddafeec898ab6100203beb4	834d7659ea5495528b2209d70be900740252f23c	winpython-64bit-2.7.9.5.exe
48cbc498565492a1c21d53c573187818	17eaf6da91f812294401a1f9dc16c0eb22ac2265	winpython-32bit-3.3.5.8.exe
01c8c48709b65a74682a710cfb1908f0	4b0bbdf70ece3ddbdd74b02c68d2078be6e14dff	winpython-64bit-3.3.5.8.exe

History of changes for WinPython 3.4.3.3

The following changes were made to WinPython distribution since version 3.4.3.2.

Python packages

New packages:

• Babel 1.3 (Internationalization utilities)
• Flask 0.10.1 (A microframework based on Werkzeug, Jinja2 and good intentions)
• Theano 0.7.0 (Optimizing compiler for evaluating mathematical expressions on CPUs and GPUs.)
• Werkzeug 0.10.4 (The Swiss Army knife of Python web development)
• adodbapi 2.6.0.7 (A pure Python package implementing PEP 249 DB-API using Microsoft ADO.)
• alabaster 0.7.3 (A configurable sidebar-enabled Sphinx theme)
• click 4.0 (A simple wrapper around optparse for powerful command line utilities.)
• docopt 0.6.2 (Pythonic argument parser, that will make you smile)
• itsdangerous 0.24 (Various helpers to pass trusted data to untrusted environments and back.)
• jedi 0.8.1 (An autocompletion tool for Python that can be used for text editors)
• pkginfo 1.2.1 (Query metadatdata from sdists / bdists / installed packages.)
• pymongo 3.0.1 (Python driver for MongoDB http://www.mongodb.org)
• pyqtgraph 0.9.10 (Scientific Graphics and GUI Library for Python)
• redis 2.10.3 (Python client for Redis key-value store)
• snowballstemmer 1.2.0 (This package provides 16 stemmer algorithms (15 + Poerter English stemmer) generated from Snowball algorithms.)
• sphinx_rtd_theme 0.1.8 (ReadTheDocs.org theme for Sphinx, 2013 version.)
• twine 1.5.0 (Collection of utilities for interacting with PyPI)

Upgraded packages:

• Pillow 2.7.0 → 2.8.1 (Python Imaging Library (fork))
• PuLP 1.5.6 → 1.5.9 (PuLP is an LP modeler written in python. PuLP can generate MPS or LP files and call GLPK, COIN CLP/CBC, CPLEX, and GUROBI to solve linear problems)
• SQLAlchemy 0.9.9 → 1.0.4 (SQL Toolkit and Object Relational Mapper)
• XlsxWriter 0.7.1 → 0.7.2 (A Python module for creating Excel XLSX files.)
• certifi 14.5.14 → 2015.4.28 (Python package for providing Mozilla's CA Bundle.)
• h5py 2.4.0 → 2.5.0 (General-purpose Python interface to HDF5 files (unlike PyTables, h5py provides direct access to the full HDF5 C library))
• husl 4.0.1 → 4.0.2 (Human-friendly HSL (Hue-Saturation-Lightness))
• ipython 3.0.0 → 3.1.0 (Enhanced Python shell)
• llvmlite 0.2.2 → 0.4.0 (lightweight wrapper around basic LLVM functionality)
• lxml 3.4.2 → 3.4.4 (Powerful and Pythonic XML processing library combining libxml2/libxslt with the ElementTree API.)
• nose 1.3.4 → 1.3.6 (nose is a discovery-based unittest extension (e.g. NumPy test module is using nose))
• numba 0.17.0 → 0.18.2 (compiling Python code using LLVM)
• numexpr 2.4 → 2.4.3 (Fast evaluation of array expressions elementwise by using a vector-based virtual machine)
• pandas 0.16.0 → 0.16.1 (Powerful data structures for data analysis, time series and statistics)
• pg8000 1.10.1 → 1.10.2 (PostgreSQL interface library)
• pip 6.0.8 → 6.1.1 (A tool for installing and managing Python packages)
• pycparser 2.10 → 2.12 (C parser in Python)
• pyodbc 3.0.7 → 3.0.9 (DB API Module for ODBC)
• python_dateutil 2.4.0 → 2.4.2 (Powerful extensions to the standard datetime module)
• pyzmq 14.5.0 → 14.6.0 (Lightweight and super-fast messaging based on ZeroMQ library (required for IPython Qt console))
• requests 2.6.0 → 2.7.0 (Requests is an Apache2 Licensed HTTP library, written in Python, for human beings.)
• scikit_image 0.11.2 → 0.11.3 (Image processing toolbox for SciPy)
• scikit_learn 0.16.0 → 0.16.1 (A set of Python modules for machine learning and data mining)
• setuptools 14.3.1 → 15.2 (Download, build, install, upgrade, and uninstall Python packages - easily)
• sqlite_bro 0.8.7.4 → 0.8.8 (a graphic SQLite Client in 1 Python file)
• sqlparse 0.1.14 → 0.1.15 (Non-validating SQL parser)
• tables 3.1.1 → 3.2.0 (Package based on HDF5 library for managing hierarchical datasets (extremely large amounts of data))

---

[stonebig]

build1 (2015-04-19: IPython3.1, WPPM rework, mingwpy Compiler Toolchain)

feedback whish: does Winpython Package Manager still work well for non-latin character windows PC ? (because of this pip workaround fix 850c257)

MD5	SHA-1	Binary
4570294a17cb9a15da2a7bbc5b07314e	de488ee968377521e20aa6094e20f43c220fe39b	winpython-64bit-3.4.3.3_build1.exe

History of changes for WinPython 3.4.3.3

The following changes were made to WinPython distribution since version 3.4.3.2.

Python packages

New packages:

• Babel 1.3 (Internationalization utilities)
• Flask 0.10.1 (A microframework based on Werkzeug, Jinja2 and good intentions)
• Theano 0.7.0 (Optimizing compiler for evaluating mathematical expressions on CPUs and GPUs.)
• Werkzeug 0.10.4 (The Swiss Army knife of Python web development)
• alabaster 0.7.3 (A configurable sidebar-enabled Sphinx theme)
• click 4.0 (A simple wrapper around optparse for powerful command line utilities.)
• itsdangerous 0.24 (Various helpers to pass trusted data to untrusted environments and back.)
• pkginfo 1.2.1 (Query metadatdata from sdists / bdists / installed packages.)
• pymongo 3.0 ()
• redis 2.10.3 (Python client for Redis key-value store)
• snowballstemmer 1.2.0 (This package provides 16 stemmer algorithms (15 + Poerter English stemmer) generated from Snowball algorithms.)
• sphinx_rtd_theme 0.1.7 (ReadTheDocs.org theme for Sphinx, 2013 version.)
• twine 1.5.0 (Collection of utilities for interacting with PyPI)

Upgraded packages:

• Pillow 2.7.0 → 2.8.1 (Python Imaging Library (fork))
• PuLP 1.5.6 → 1.5.8 (PuLP is an LP modeler written in python. PuLP can generate MPS or LP files and call GLPK, COIN CLP/CBC, CPLEX, and GUROBI to solve linear problems)
• SQLAlchemy 0.9.9 → 1.0.0 (SQL Toolkit and Object Relational Mapper)
• XlsxWriter 0.7.1 → 0.7.2 (A Python module for creating Excel XLSX files.)
• h5py 2.4.0 → 2.5.0 (General-purpose Python interface to HDF5 files (unlike PyTables, h5py provides direct access to the full HDF5 C library))
• ipython 3.0.0 → 3.1.0 (Enhanced Python shell)
• llvmlite 0.2.2 → 0.4.0 (lightweight wrapper around basic LLVM functionality)
• lxml 3.4.2 → 3.4.3 (Powerful and Pythonic XML processing library combining libxml2/libxslt with the ElementTree API.)
• nose 1.3.4 → 1.3.6 (nose is a discovery-based unittest extension (e.g. NumPy test module is using nose))
• numba 0.17.0 → 0.18.2 (compiling Python code using LLVM)
• numexpr 2.4 → 2.4.1 (Fast evaluation of array expressions elementwise by using a vector-based virtual machine)
• python_dateutil 2.4.0 → 2.4.2 (Powerful extensions to the standard datetime module)
• scikit_image 0.11.2 → 0.11.3 (Image processing toolbox for SciPy)
• scikit_learn 0.16.0 → 0.16.1 (A set of Python modules for machine learning and data mining)
• setuptools 14.3.1 → 15.1 (Download, build, install, upgrade, and uninstall Python packages - easily)
• sqlite_bro 0.8.7.4 → 0.8.8 (a graphic SQLite Client in 1 Python file)

---

[hiccup7]

When you built OpenBLAS v0.2.14, did you use MAX_STACK_ALLOC=2048? See JuliaLang/julia#10780

[stonebig]

Hi @hiccup7,

Julia uses a different OpenBLAS dll, because of a numpy compatibility issue, so Python OpenBLAS won't speed-up your Julia code unless you shift that part back to Python.

Carl's release pdf ( https://bitbucket.org/carlkl/mingw-w64-for-python/downloads/mingwpy-2015-04-readme.pdf) points to this commit of OpenBLAS: OpenMathLib/OpenBLAS@fb02cb0

This is 5 commits after the one you are waiting for: OpenMathLib/OpenBLAS@6c3a0b5

So this build of OpenBLAS may have MAX_STACK_ALLOC=2048

(but the commit after could mean troubles: OpenMathLib/OpenBLAS@a4c96ec , see OpenMathLib/OpenBLAS#543)

[hiccup7]

The Julia team is creating the libopenblas.dll I need for Julia. So my needs are well taken care of.

I brought up the issue of MAX_STACK_ALLOC=2048 because you are upgrading OpenBLAS to v0.2.14, and perhaps neither you nor @carlkl were aware of the need for this build option with that specific release version. I am trying to prevent a bug in theano lasagne in the next WinPython release.

[carlkl]

The latest openblas build for mingwpy:

https://bitbucket.org/carlkl/mingw-w64-for-python/downloads/openblas-fb02cb0_amd64.7z
https://bitbucket.org/carlkl/mingw-w64-for-python/downloads/openblas-fb02cb0-win32.7z

have been build with commit 37b9033 (wrong name of the archive files) and MAX_STACK_ALLOC=2048

the DLLs have been renamed to libopenblaspy.dll to avoid a name clash with julialang's openblas build

[stonebig]

So I just need to replace the previous ones and I'm good ?

mingw32\bin\libopenblaspy.dll
mingw32\x86_64-w64-mingw32\lib\libopenblaspy.dll.a
mingw32\x86_64-w64-mingw32\lib\libopenblaspy.a
mingw32\x86_64-w64-mingw32\include\openblas\f77blas.h (and friends)

[carlkl]

that's correct.

[hiccup7]

@carlkl , I glad to know that you already knew about the MAX_STACK_ALLOC=2048 detail. Thanks for your big contribution to the Python community!

[stonebig]

build2 (2015-04-23: mingwpy + theano ok , lasagne)

feedback whish:

• does Winpython Package Manager still work well for non-latin character windows PC ? (because of this pip workaround fix 850c257)
• is result better than http://www.deeplearning.net/software/theano/install_windows.html#install-windows ?

MD5	SHA-1	Binary
0587704f33923f8df10b76eca6cd922c	e80703d0b711551ad6b02b6009be8edf677dda20	winpython-64bit-3.4.3.3_build2.exe

History of changes for WinPython 3.4.3.3

The following changes were made to WinPython distribution since version 3.4.3.2.

Python packages

New packages:

• Babel 1.3 (Internationalization utilities)
• Flask 0.10.1 (A microframework based on Werkzeug, Jinja2 and good intentions)
• Lasagne 0.0.1 ()
• Theano 0.7.0 (Optimizing compiler for evaluating mathematical expressions on CPUs and GPUs.)
• Werkzeug 0.10.4 (The Swiss Army knife of Python web development)
• adodbapi 2.6.0.7 (A pure Python package implementing PEP 249 DB-API using Microsoft ADO.)
• alabaster 0.7.3 (A configurable sidebar-enabled Sphinx theme)
• click 4.0 (A simple wrapper around optparse for powerful command line utilities.)
• itsdangerous 0.24 (Various helpers to pass trusted data to untrusted environments and back.)
• pkginfo 1.2.1 (Query metadatdata from sdists / bdists / installed packages.)
• pymongo 3.0.1 (Python driver for MongoDB http://www.mongodb.org)
• redis 2.10.3 (Python client for Redis key-value store)
• snowballstemmer 1.2.0 (This package provides 16 stemmer algorithms (15 + Poerter English stemmer) generated from Snowball algorithms.)
• sphinx_rtd_theme 0.1.7 (ReadTheDocs.org theme for Sphinx, 2013 version.)
• twine 1.5.0 (Collection of utilities for interacting with PyPI)

Upgraded packages:

• Pillow 2.7.0 → 2.8.1 (Python Imaging Library (fork))
• PuLP 1.5.6 → 1.5.9 (PuLP is an LP modeler written in python. PuLP can generate MPS or LP files and call GLPK, COIN CLP/CBC, CPLEX, and GUROBI to solve linear problems)
• SQLAlchemy 0.9.9 → 1.0.0 (SQL Toolkit and Object Relational Mapper)
• XlsxWriter 0.7.1 → 0.7.2 (A Python module for creating Excel XLSX files.)
• h5py 2.4.0 → 2.5.0 (General-purpose Python interface to HDF5 files (unlike PyTables, h5py provides direct access to the full HDF5 C library))
• husl 4.0.1 → 4.0.2 (Human-friendly HSL (Hue-Saturation-Lightness))
• ipython 3.0.0 → 3.1.0 (Enhanced Python shell)
• llvmlite 0.2.2 → 0.4.0 (lightweight wrapper around basic LLVM functionality)
• lxml 3.4.2 → 3.4.3 (Powerful and Pythonic XML processing library combining libxml2/libxslt with the ElementTree API.)
• nose 1.3.4 → 1.3.6 (nose is a discovery-based unittest extension (e.g. NumPy test module is using nose))
• numba 0.17.0 → 0.18.2 (compiling Python code using LLVM)
• numexpr 2.4 → 2.4.1 (Fast evaluation of array expressions elementwise by using a vector-based virtual machine)
• pg8000 1.10.1 → 1.10.2 (PostgreSQL interface library)
• pip 6.0.8 → 6.1.1 (A tool for installing and managing Python packages)
• python_dateutil 2.4.0 → 2.4.2 (Powerful extensions to the standard datetime module)
• scikit_image 0.11.2 → 0.11.3 (Image processing toolbox for SciPy)
• scikit_learn 0.16.0 → 0.16.1 (A set of Python modules for machine learning and data mining)
• setuptools 14.3.1 → 15.1 (Download, build, install, upgrade, and uninstall Python packages - easily)
• sqlite_bro 0.8.7.4 → 0.8.8 (a graphic SQLite Client in 1 Python file)
• sqlparse 0.1.14 → 0.1.15 (Non-validating SQL parser)

---

[carlkl]

@stonebig, build2: in settings there is no .theanorc file. You should include it with the following input:

[blas]
ldflags = -lopenblaspy

With that settings theano links blas/lapack dependant binary extensions automagically to <..>\tools\mingw32\bin\libopenblaspy.dll

You can check it with

python check_blas.py   # run this script located in <...>\Lib\site-packages\theano\misc

[stonebig]

ok, I'll do.

[stonebig]

on the MNIST dataset lasagne test, it doesn't seem to make a difference with previous NOparameters #82 (comment) , on my i3-350m limited and old cpu.

Starting training...
Epoch 1 of 500 took 21.039s
  training loss:        1.382395
  validation loss:      0.479808
  validation accuracy:      87.24 %%
Epoch 2 of 500 took 20.673s
  training loss:        0.595778
  validation loss:      0.335485
  validation accuracy:      90.45 %%
Epoch 3 of 500 took 20.156s
  training loss:        0.469481
  validation loss:      0.282895
  validation accuracy:      91.88 %%

[stonebig]

python ..\Lib\site-packages\theano\misc\check_blas.py

        Some results that you can compare against. They were 10 executions
        of gemm in float64 with matrices of shape 2000x2000 (M=N=K=2000).
        All memory layout was in C order.

        CPU tested: Xeon E5345(2.33Ghz, 8M L2 cache, 1333Mhz FSB),
                    Xeon E5430(2.66Ghz, 12M L2 cache, 1333Mhz FSB),
                    Xeon E5450(3Ghz, 12M L2 cache, 1333Mhz FSB),
                    Xeon X5560(2.8Ghz, 12M L2 cache, hyper-threads?)
                    Core 2 E8500, Core i7 930(2.8Ghz, hyper-threads enabled),
                    Core i7 950(3.07GHz, hyper-threads enabled)
                    Xeon X5550(2.67GHz, 8M l2 cache?, hyper-threads enabled)


        Libraries tested:
            * numpy with ATLAS from distribution (FC9) package (1 thread)
            * manually compiled numpy and ATLAS with 2 threads
            * goto 1.26 with 1, 2, 4 and 8 threads
            * goto2 1.13 compiled with multiple threads enabled

                          Xeon   Xeon   Xeon  Core2 i7    i7     Xeon   Xeon
        lib/nb threads    E5345  E5430  E5450 E8500 930   950    X5560  X5550

        numpy 1.3.0 blas                                                775.92s
        numpy_FC9_atlas/1 39.2s  35.0s  30.7s 29.6s 21.5s 19.60s
        goto/1            18.7s  16.1s  14.2s 13.7s 16.1s 14.67s
        numpy_MAN_atlas/2 12.0s  11.6s  10.2s  9.2s  9.0s
        goto/2             9.5s   8.1s   7.1s  7.3s  8.1s  7.4s
        goto/4             4.9s   4.4s   3.7s  -     4.1s  3.8s
        goto/8             2.7s   2.4s   2.0s  -     4.1s  3.8s
        openblas/1                                        14.04s
        openblas/2                                         7.16s
        openblas/4                                         3.71s
        openblas/8                                         3.70s
        mkl 11.0.083/1            7.97s
        mkl 10.2.2.025/1                                         13.7s
        mkl 10.2.2.025/2                                          7.6s
        mkl 10.2.2.025/4                                          4.0s
        mkl 10.2.2.025/8                                          2.0s
        goto2 1.13/1                                                     14.37s
        goto2 1.13/2                                                      7.26s
        goto2 1.13/4                                                      3.70s
        goto2 1.13/8                                                      1.94s
        goto2 1.13/16                                                     3.16s

        Test time in float32

        cuda version      6.5    6.0    5.5    5.0    4.2    4.1    4.0    3.2
  3.0   # note
        gpu
        K6000/NOECC       0.06s         0.06s
        K40                             0.07s
        K20m/ECC          0.08s 0.08s          0.07s
        K20/NOECC                              0.07s
        M2090                           0.19s
        C2075                                         0.25s
        M2075                                  0.25s
        M2070                                  0.25s         0.27s         0.32s

        M2070-Q                                0.48s         0.27s         0.32s

        M2050(Amazon)                          0.25s
        C1060                                                              0.46s

        K600                            1.04s

        GTX Titan Black                 0.05s
        GTX Titan(D15U-50)              0.06s  0.06s  don't work
        GTX 780                         0.06s
        GTX 980           0.06s
        GTX 970           0.08s
        GTX 680                         0.11s  0.12s  0.154s               0.218
s
        GRID K520         0.14s
        GTX 580                         0.16s  0.16s  0.164s               0.203
s
        GTX 480                         0.19s  0.19s  0.192s               0.237
s 0.27s
        GTX 750 Ti        0.20s
        GTX 470                         0.23s  0.23s  0.238s               0.297
s 0.34s
        GTX 660                         0.18s  0.20s  0.23s
        GTX 560                                       0.30s
        GTX 650 Ti                             0.27s
        GTX 765M                 0.27s
        GTX 460                                0.37s                0.45s
        GTX 285                         0.42s         0.452s        0.452s
  0.40s # cuda 3.0 seems faster? driver version?
        750M                                   0.49s
        GT 610            2.38s
        GTX 550 Ti                                                  0.57s
        GT 520                                        2.68s                3.06s

        520M                                   2.44s                       3.19s
        # with bumblebee on Ubuntu 12.04
        GT 220                                                             3.80s

        GT 210                                                      6.35s
        8500 GT
  10.68s

Some Theano flags:
    blas.ldflags= -lopenblaspy
    compiledir= C:\Users\famille\AppData\Local\Theano\compiledir_Windows-7-6.1.7
601-SP1-Intel64_Family_6_Model_37_Stepping_2_GenuineIntel-3.4.3-64
    floatX= float64
    device= cpu
Some OS information:
    sys.platform= win32
    sys.version= 3.4.3 (v3.4.3:9b73f1c3e601, Feb 24 2015, 22:44:40) [MSC v.1600
64 bit (AMD64)]
    sys.prefix= D:\WinPython\basedir34\build\winpython-3.4.3.amd64\python-3.4.3.
amd64
Some environment variables:
    MKL_NUM_THREADS= None
    OMP_NUM_THREADS= None
    GOTO_NUM_THREADS= None

Numpy config: (used when the Theano flag "blas.ldflags" is empty)
blas_mkl_info:
    libraries = ['mkl_lapack95_lp64', 'mkl_blas95_lp64', 'mkl_intel_lp64', 'mkl_
intel_thread', 'mkl_core', 'libiomp5md', 'libifportmd']
    define_macros = [('SCIPY_MKL_H', None)]
    library_dirs = ['C:/Program Files (x86)/Intel/Composer XE/mkl/lib/intel64']
    include_dirs = ['C:/Program Files (x86)/Intel/Composer XE/mkl/include']
blas_opt_info:
    libraries = ['mkl_lapack95_lp64', 'mkl_blas95_lp64', 'mkl_intel_lp64', 'mkl_
intel_thread', 'mkl_core', 'libiomp5md', 'libifportmd']
    define_macros = [('SCIPY_MKL_H', None)]
    library_dirs = ['C:/Program Files (x86)/Intel/Composer XE/mkl/lib/intel64']
    include_dirs = ['C:/Program Files (x86)/Intel/Composer XE/mkl/include']
lapack_opt_info:
    libraries = ['mkl_lapack95_lp64', 'mkl_blas95_lp64', 'mkl_intel_lp64', 'mkl_
intel_thread', 'mkl_core', 'libiomp5md', 'libifportmd', 'mkl_lapack95_lp64', 'mk
l_blas95_lp64', 'mkl_intel_lp64', 'mkl_intel_thread', 'mkl_core', 'libiomp5md',
'libifportmd']
    define_macros = [('SCIPY_MKL_H', None)]
    library_dirs = ['C:/Program Files (x86)/Intel/Composer XE/mkl/lib/intel64']
    include_dirs = ['C:/Program Files (x86)/Intel/Composer XE/mkl/include']
lapack_mkl_info:
    libraries = ['mkl_lapack95_lp64', 'mkl_blas95_lp64', 'mkl_intel_lp64', 'mkl_
intel_thread', 'mkl_core', 'libiomp5md', 'libifportmd', 'mkl_lapack95_lp64', 'mk
l_blas95_lp64', 'mkl_intel_lp64', 'mkl_intel_thread', 'mkl_core', 'libiomp5md',
'libifportmd']
    define_macros = [('SCIPY_MKL_H', None)]
    library_dirs = ['C:/Program Files (x86)/Intel/Composer XE/mkl/lib/intel64']
    include_dirs = ['C:/Program Files (x86)/Intel/Composer XE/mkl/include']
openblas_lapack_info:
  NOT AVAILABLE
mkl_info:
    libraries = ['mkl_lapack95_lp64', 'mkl_blas95_lp64', 'mkl_intel_lp64', 'mkl_
intel_thread', 'mkl_core', 'libiomp5md', 'libifportmd']
    define_macros = [('SCIPY_MKL_H', None)]
    library_dirs = ['C:/Program Files (x86)/Intel/Composer XE/mkl/lib/intel64']
    include_dirs = ['C:/Program Files (x86)/Intel/Composer XE/mkl/include']
Numpy dot module: numpy.core._dotblas
Numpy location: D:\WinPython\basedir34\build\winpython-3.4.3.amd64\python-3.4.3.
amd64\lib\site-packages\numpy\__init__.py
Numpy version: 1.9.2

We executed 10 calls to gemm with a and b matrices of shapes (2000, 2000) and (2
000, 2000).

Total execution time: 12.95s on CPU (with direct Theano binding to blas).

Try to run this script a few times. Experience shows that the first time is not
as fast as followings calls. The difference is not big, but consistent.

[stonebig]

without .theanorc:

python ..\Lib\site-packages\theano\misc\check_blas.py

        Some results that you can compare against. They were 10 executions
        of gemm in float64 with matrices of shape 2000x2000 (M=N=K=2000).
        All memory layout was in C order.

        CPU tested: Xeon E5345(2.33Ghz, 8M L2 cache, 1333Mhz FSB),
                    Xeon E5430(2.66Ghz, 12M L2 cache, 1333Mhz FSB),
                    Xeon E5450(3Ghz, 12M L2 cache, 1333Mhz FSB),
                    Xeon X5560(2.8Ghz, 12M L2 cache, hyper-threads?)
                    Core 2 E8500, Core i7 930(2.8Ghz, hyper-threads enabled),
                    Core i7 950(3.07GHz, hyper-threads enabled)
                    Xeon X5550(2.67GHz, 8M l2 cache?, hyper-threads enabled)


        Libraries tested:
            * numpy with ATLAS from distribution (FC9) package (1 thread)
            * manually compiled numpy and ATLAS with 2 threads
            * goto 1.26 with 1, 2, 4 and 8 threads
            * goto2 1.13 compiled with multiple threads enabled

                          Xeon   Xeon   Xeon  Core2 i7    i7     Xeon   Xeon
        lib/nb threads    E5345  E5430  E5450 E8500 930   950    X5560  X5550

        numpy 1.3.0 blas                                                775.92s
        numpy_FC9_atlas/1 39.2s  35.0s  30.7s 29.6s 21.5s 19.60s
        goto/1            18.7s  16.1s  14.2s 13.7s 16.1s 14.67s
        numpy_MAN_atlas/2 12.0s  11.6s  10.2s  9.2s  9.0s
        goto/2             9.5s   8.1s   7.1s  7.3s  8.1s  7.4s
        goto/4             4.9s   4.4s   3.7s  -     4.1s  3.8s
        goto/8             2.7s   2.4s   2.0s  -     4.1s  3.8s
        openblas/1                                        14.04s
        openblas/2                                         7.16s
        openblas/4                                         3.71s
        openblas/8                                         3.70s
        mkl 11.0.083/1            7.97s
        mkl 10.2.2.025/1                                         13.7s
        mkl 10.2.2.025/2                                          7.6s
        mkl 10.2.2.025/4                                          4.0s
        mkl 10.2.2.025/8                                          2.0s
        goto2 1.13/1                                                     14.37s
        goto2 1.13/2                                                      7.26s
        goto2 1.13/4                                                      3.70s
        goto2 1.13/8                                                      1.94s
        goto2 1.13/16                                                     3.16s

        Test time in float32

        cuda version      6.5    6.0    5.5    5.0    4.2    4.1    4.0    3.2
  3.0   # note
        gpu
        K6000/NOECC       0.06s         0.06s
        K40                             0.07s
        K20m/ECC          0.08s 0.08s          0.07s
        K20/NOECC                              0.07s
        M2090                           0.19s
        C2075                                         0.25s
        M2075                                  0.25s
        M2070                                  0.25s         0.27s         0.32s

        M2070-Q                                0.48s         0.27s         0.32s

        M2050(Amazon)                          0.25s
        C1060                                                              0.46s

        K600                            1.04s

        GTX Titan Black                 0.05s
        GTX Titan(D15U-50)              0.06s  0.06s  don't work
        GTX 780                         0.06s
        GTX 980           0.06s
        GTX 970           0.08s
        GTX 680                         0.11s  0.12s  0.154s               0.218
s
        GRID K520         0.14s
        GTX 580                         0.16s  0.16s  0.164s               0.203
s
        GTX 480                         0.19s  0.19s  0.192s               0.237
s 0.27s
        GTX 750 Ti        0.20s
        GTX 470                         0.23s  0.23s  0.238s               0.297
s 0.34s
        GTX 660                         0.18s  0.20s  0.23s
        GTX 560                                       0.30s
        GTX 650 Ti                             0.27s
        GTX 765M                 0.27s
        GTX 460                                0.37s                0.45s
        GTX 285                         0.42s         0.452s        0.452s
  0.40s # cuda 3.0 seems faster? driver version?
        750M                                   0.49s
        GT 610            2.38s
        GTX 550 Ti                                                  0.57s
        GT 520                                        2.68s                3.06s

        520M                                   2.44s                       3.19s
        # with bumblebee on Ubuntu 12.04
        GT 220                                                             3.80s

        GT 210                                                      6.35s
        8500 GT
  10.68s

Some Theano flags:
    blas.ldflags=
    compiledir= C:\Users\famille\AppData\Local\Theano\compiledir_Windows-7-6.1.7
601-SP1-Intel64_Family_6_Model_37_Stepping_2_GenuineIntel-3.4.3-64
    floatX= float64
    device= cpu
Some OS information:
    sys.platform= win32
    sys.version= 3.4.3 (v3.4.3:9b73f1c3e601, Feb 24 2015, 22:44:40) [MSC v.1600
64 bit (AMD64)]
    sys.prefix= D:\WinPython\basedir34\build\winpython-3.4.3.amd64\python-3.4.3.
amd64
Some environment variables:
    MKL_NUM_THREADS= None
    OMP_NUM_THREADS= None
    GOTO_NUM_THREADS= None

Numpy config: (used when the Theano flag "blas.ldflags" is empty)
lapack_mkl_info:
    library_dirs = ['C:/Program Files (x86)/Intel/Composer XE/mkl/lib/intel64']
    define_macros = [('SCIPY_MKL_H', None)]
    include_dirs = ['C:/Program Files (x86)/Intel/Composer XE/mkl/include']
    libraries = ['mkl_lapack95_lp64', 'mkl_blas95_lp64', 'mkl_intel_lp64', 'mkl_
intel_thread', 'mkl_core', 'libiomp5md', 'libifportmd', 'mkl_lapack95_lp64', 'mk
l_blas95_lp64', 'mkl_intel_lp64', 'mkl_intel_thread', 'mkl_core', 'libiomp5md',
'libifportmd']
openblas_lapack_info:
  NOT AVAILABLE
blas_opt_info:
    library_dirs = ['C:/Program Files (x86)/Intel/Composer XE/mkl/lib/intel64']
    define_macros = [('SCIPY_MKL_H', None)]
    include_dirs = ['C:/Program Files (x86)/Intel/Composer XE/mkl/include']
    libraries = ['mkl_lapack95_lp64', 'mkl_blas95_lp64', 'mkl_intel_lp64', 'mkl_
intel_thread', 'mkl_core', 'libiomp5md', 'libifportmd']
lapack_opt_info:
    library_dirs = ['C:/Program Files (x86)/Intel/Composer XE/mkl/lib/intel64']
    define_macros = [('SCIPY_MKL_H', None)]
    include_dirs = ['C:/Program Files (x86)/Intel/Composer XE/mkl/include']
    libraries = ['mkl_lapack95_lp64', 'mkl_blas95_lp64', 'mkl_intel_lp64', 'mkl_
intel_thread', 'mkl_core', 'libiomp5md', 'libifportmd', 'mkl_lapack95_lp64', 'mk
l_blas95_lp64', 'mkl_intel_lp64', 'mkl_intel_thread', 'mkl_core', 'libiomp5md',
'libifportmd']
mkl_info:
    library_dirs = ['C:/Program Files (x86)/Intel/Composer XE/mkl/lib/intel64']
    define_macros = [('SCIPY_MKL_H', None)]
    include_dirs = ['C:/Program Files (x86)/Intel/Composer XE/mkl/include']
    libraries = ['mkl_lapack95_lp64', 'mkl_blas95_lp64', 'mkl_intel_lp64', 'mkl_
intel_thread', 'mkl_core', 'libiomp5md', 'libifportmd']
blas_mkl_info:
    library_dirs = ['C:/Program Files (x86)/Intel/Composer XE/mkl/lib/intel64']
    define_macros = [('SCIPY_MKL_H', None)]
    include_dirs = ['C:/Program Files (x86)/Intel/Composer XE/mkl/include']
    libraries = ['mkl_lapack95_lp64', 'mkl_blas95_lp64', 'mkl_intel_lp64', 'mkl_
intel_thread', 'mkl_core', 'libiomp5md', 'libifportmd']
Numpy dot module: numpy.core._dotblas
Numpy location: D:\WinPython\basedir34\build\winpython-3.4.3.amd64\python-3.4.3.
amd64\lib\site-packages\numpy\__init__.py
Numpy version: 1.9.2

We executed 10 calls to gemm with a and b matrices of shapes (2000, 2000) and (2
000, 2000).

Total execution time: 12.03s on CPU (without direct Theano binding to blas but w
ith numpy/scipy binding to blas).

Try to run this script a few times. Experience shows that the first time is not
as fast as followings calls. The difference is not big, but consistent.````

[carlkl]

Theano uses scipy's gemv if no blas library is found. Linking to an external Blas library seems to be necessary only if scipy isn't installed. In the context of WinPython and Theano you may exclude OpenBLAS again, but to be sure ask the theano ML.

[stonebig]

ok. I'll remove the .theanorc file from next build

[stonebig]

build3 (2015-05-07: 32bit with latest updates, + jedi + seaborn demo - lasagne)

feedback whish:

• issues (if you find some),
• jedi in spyder.

MD5	SHA-1	Binary
4dca5dabec277f7fcf18075d2275b2bb	4c5c1ef9b07e3c2f169190780a313f2e3a21afe9	winpython-32bit-3.4.3.3_build3.exe

History of changes for WinPython 3.4.3.3

The following changes were made to WinPython distribution since version 3.4.3.2.

Python packages

New packages:

• Babel 1.3 (Internationalization utilities)
• Flask 0.10.1 (A microframework based on Werkzeug, Jinja2 and good intentions)
• Theano 0.7.0 (Optimizing compiler for evaluating mathematical expressions on CPUs and GPUs.)
• Werkzeug 0.10.4 (The Swiss Army knife of Python web development)
• adodbapi 2.6.0.7 (A pure Python package implementing PEP 249 DB-API using Microsoft ADO.)
• alabaster 0.7.3 (A configurable sidebar-enabled Sphinx theme)
• click 4.0 (A simple wrapper around optparse for powerful command line utilities.)
• docopt 0.6.2 (Pythonic argument parser, that will make you smile)
• itsdangerous 0.24 (Various helpers to pass trusted data to untrusted environments and back.)
• jedi 0.8.1 (An autocompletion tool for Python that can be used for text editors)
• pkginfo 1.2.1 (Query metadatdata from sdists / bdists / installed packages.)
• pymongo 3.0.1 (Python driver for MongoDB http://www.mongodb.org)
• redis 2.10.3 (Python client for Redis key-value store)
• snowballstemmer 1.2.0 (This package provides 16 stemmer algorithms (15 + Poerter English stemmer) generated from Snowball algorithms.)
• sphinx_rtd_theme 0.1.8 (ReadTheDocs.org theme for Sphinx, 2013 version.)
• twine 1.5.0 (Collection of utilities for interacting with PyPI)

Upgraded packages:

• Pillow 2.7.0 → 2.8.1 (Python Imaging Library (fork))
• PuLP 1.5.6 → 1.5.9 (PuLP is an LP modeler written in python. PuLP can generate MPS or LP files and call GLPK, COIN CLP/CBC, CPLEX, and GUROBI to solve linear problems)
• SQLAlchemy 0.9.9 → 1.0.3 (SQL Toolkit and Object Relational Mapper)
• XlsxWriter 0.7.1 → 0.7.2 (A Python module for creating Excel XLSX files.)
• certifi 14.5.14 → 2015.4.28 (Python package for providing Mozilla's CA Bundle.)
• h5py 2.4.0 → 2.5.0 (General-purpose Python interface to HDF5 files (unlike PyTables, h5py provides direct access to the full HDF5 C library))
• husl 4.0.1 → 4.0.2 (Human-friendly HSL (Hue-Saturation-Lightness))
• ipython 3.0.0 → 3.1.0 (Enhanced Python shell)
• llvmlite 0.2.2 → 0.4.0 (lightweight wrapper around basic LLVM functionality)
• lxml 3.4.2 → 3.4.4 (Powerful and Pythonic XML processing library combining libxml2/libxslt with the ElementTree API.)
• nose 1.3.4 → 1.3.6 (nose is a discovery-based unittest extension (e.g. NumPy test module is using nose))
• numba 0.17.0 → 0.18.2 (compiling Python code using LLVM)
• numexpr 2.4 → 2.4.3 (Fast evaluation of array expressions elementwise by using a vector-based virtual machine)
• pg8000 1.10.1 → 1.10.2 (PostgreSQL interface library)
• pip 6.0.8 → 6.1.1 (A tool for installing and managing Python packages)
• pycparser 2.10 → 2.12 (C parser in Python)
• pyodbc 3.0.7 → 3.0.9 (DB API Module for ODBC)
• python_dateutil 2.4.0 → 2.4.2 (Powerful extensions to the standard datetime module)
• pyzmq 14.5.0 → 14.6.0 (Lightweight and super-fast messaging based on ZeroMQ library (required for IPython Qt console))
• requests 2.6.0 → 2.7.0 (Requests is an Apache2 Licensed HTTP library, written in Python, for human beings.)
• scikit_image 0.11.2 → 0.11.3 (Image processing toolbox for SciPy)
• scikit_learn 0.16.0 → 0.16.1 (A set of Python modules for machine learning and data mining)
• setuptools 14.3.1 → 15.2 (Download, build, install, upgrade, and uninstall Python packages - easily)
• sqlite_bro 0.8.7.4 → 0.8.8 (a graphic SQLite Client in 1 Python file)
• sqlparse 0.1.14 → 0.1.15 (Non-validating SQL parser)
• tables 3.1.1 → 3.2.0 (Package based on HDF5 library for managing hierarchical datasets (extremely large amounts of data))

---