Python wrappers for libsdca

[121onto]

I wrote some wrappers that make it easy to use the libsdca library with Python 3.5. I tried to stick with your coding conventions. One exception is that I use distutils and Cython to build things (not too familiar with cmake).

The wrappers could use a user-friendly interface, to help with trouble-shooting potential bugs. Wouldn't be too difficult to write a nice python class for this purpose.

Would love to hear any feedback or thoughts on this.

[mlapin]

Hey @121onto, thanks a lot for the contribution!

I cannot really test it with python 3.5 right now, but I decided to go ahead with the merge anyway - we can fix things after the holidays.

With Python 2.7.3, there is a compilation issue:

Error compiling Cython file:
------------------------------------------------------------
...

    cdef DATASET* dat = new DATASET()
    dat.num_dimensions = features.shape[0]
    dat.num_examples = features.shape[1]
    dat.num_classes = num_classes
    dat.labels.assign(&labels[0], &labels[0] + features.shape[1])
                     ^
------------------------------------------------------------

libsdca.pyx:84:22: Cannot assign type 'int *' to 'size_t'

[121onto]

Interesting. I downloaded the latest anaconda release for python 2.7 and was able to compile things. Here is my version info:

$ python -c 'import sys, cython; print sys.version, "\n", cython.__version__'
2.7.11...
0.23.4

Are you willing to install the latest Anaconda build and try again? Its super easy to do. Everything gets placed in ~/anaconda and so uninstalling it is as easy as deleting that directory. Find install instructions here.

If it is a version issue, we should make that explicit in setup.py. Here is an updated setup.py that accomplishes this:

Edited on Wed Dec 23 10:36:42 PST 2015
Edited on Wed Dec 23 22:30:43 PST 2015

import sys, os                                                                                                          
from Cython.Build import cythonize                                                                                      
from distutils.version import LooseVersion                                                                              

import numpy as np                                                                                                      

try:                                                                                                                    
    from setuptools import setup, Extension                                                                             
    _have_setuptools = True                                                                                             
except ImportError:                                                                                                     
    # no setuptools installed                                                                                           
    from distutils.core import setup, Extension                                                                         
    _have_setuptools = False                                                                                            

setuptools_kwargs = {}                                                                                                  
min_cython_ver = '0.23.4'                                                                                               
try:                                                                                                                    
    import Cython                                                                                                       
except ImportError:                                                                                                     
    sys.exit('You are missing Cython.  Install with:'                                                                   
             '\n$ pip install cython')                                                                                  


try:                                                                                                                    
    cython_ver = Cython.__version__                                                                                     
    assert(cython_ver >= LooseVersion(min_cython_ver))                                                                  
except AssertionError:                                                                                                  
    sys.exit('Your version of Cython is not supported.  '                                                               
             'Upgrade  with: '                                                                                          
             '\n$ pip install cython --upgrade')                                                                                

# create build dir if it doesn't already exist                                                                          
try:                                                                                                                    
    os.makedirs('libsdca')                                                                                              
except OSError:                                                                                                         
    if not os.path.isdir('libsdca'):                                                                                    
        raise                                                                                                           


# Modifiy this if BLAS and LAPACK libraries are not in /usr/lib.                                                        
BLAS_LIB_DIR = '/usr/lib'                                                                                               
BLAS_LIB = ['blas']                                                                                                     
LAPACK_LIB = ['lapack']                                                                                                 

BLAS_COMPILE_ARGS = ['-DBLAS_DEFAULT_LOCAL_HEADER']                                                                     
BASE_COMPILE_ARGS = ['-std=c++11', '-lstdc++']                                                                          
BASE_LINK_ARGS = ['-std=c++11', '-lstdc++']                                                                             

INCLUDE_DIRS = [np.get_include(), '../']                                                                                

extension = Extension(name = 'libsdca',                                                                                 
                 language = 'c++',                                                                                      
                 libraries = LAPACK_LIB + BLAS_LIB,                                                                     
                 include_dirs = INCLUDE_DIRS,                                                                           
                 library_dirs = [ BLAS_LIB_DIR ],                                                                       
                 extra_compile_args = BASE_COMPILE_ARGS + BLAS_COMPILE_ARGS,                                            
                 extra_link_args = BASE_LINK_ARGS,                                                                      
                 sources = ['libsdca.pyx'])                                                                             

setup(name = 'libsdca',                                                                                                 
      description = 'Python wrappers for libsdca.',                                                                     
      version = '0.0.1',                                                                                                
      long_description = '''                                                                                            
      Python wrappers for libsdca.                                                                                      
      libsdca is a library for multiclass classification                                                                
      based on stochastic dual coordinate ascent (SDCA).                                                                
      ''',                                                                                                              
      author = '121onto',                                                                                               
      author_email = '121onto@gmail.com',                                                                               
      ext_package = 'libsdca',                                                                                          
      ext_modules = cythonize(extension),                                                                               
      packages = 'libsdca',                                                                                             
      classifiers = [                                                                                                   
          'Development Status :: 2 - Pre-Alpha',                                                                        
          'Intended Audience :: Science/Research',                                                                      
          'Operating System :: OS Independent',                                                                         
          'Programming Language :: C++',                                                                                
          'Programming Language :: Python',                                                                             
          'Programming Language :: Python :: 3',                                                                        
          'Topic :: Scientific/Engineering',                                                                            
      ],                                                                                                                
      **setuptools_kwargs                                                                                               
)