Installing libsvm-3.0 for Python on OSX 10.6

This was very frustrating to solve and I eventually had to get my friend Kieran (thanks Kieran!) to help me.  Basically the vanilla installation instructions that come with are insufficient (at least for me) to get a working module.  Here are the steps that were required to get everything to work.

1. Run 'make' in the libsvm-3.0 directory

2. Run 'make' in the libsvm-3.0/python directory

In the libsvm-3.0 directory there now should be a .so2 file

3. Create a new directory in your site-packages directory (your pythonpath) called libsvm

4. Copy the .so2 file from libsvm-3.0 and the svm.py, svm.pyc, svmutil.py files from libsvm-3.0/python to site-packages/libsvm

There are a couple things that are missing so now we need to make them.

5. In site-packages/libsvm create a file called __init__.py.  This is an empty file, but it is necessary to get the directory recognized as a python module.

6. Edit svm.py and add the following two lines after the other import statements at the top of the file:

          import os.path
    _PATH = os.path.join( *os.path.split(__file__)[:-1] )


7.  At around line 7 you will see this statement


          # For unix the prefix 'lib' is not considered.

    if find_library('svm'):
       libsvm = CDLL(find_library('svm'))
    elif find_library('libsvm'):
       libsvm = CDLL(find_library('libsvm'))
    else:
       if sys.platform == 'win32':
          libsvm = CDLL('../windows/libsvm.dll')
       else:
          libsvm = CDLL('../libsvm.so.2')

8.  Change this to look like this:


    # For unix the prefix 'lib' is not considered.
    if find_library('svm'):
  libsvm = CDLL(find_library('svm'))
    elif find_library('libsvm'):
  libsvm = CDLL(find_library('libsvm'))
    else:
  if sys.platform == 'win32':
  libsvm = CDLL(os.path.join(_PATH,'windows','libsvm.dll'))
  else:
  libsvm = CDLL(os.path.join(_PATH,'libsvm.so.2'))

9. Once you save svm.py, you should be able to fire up a python interpreter and do 'from libsvm import svm'.  If that works, and a dir(svm) shows you a ton of functions, then you are good to go.