Implement libSVM adaptation in sklearn

[ThomasPel1]

Hi Everyone,

For my Masters thesis I am comparing a number of ordinal regression techniques. Here I found an adaptation of the libSVM library to include SVORIM, a support vector machine for ordinal regression with implicit constraints. The code for this adaptation can be downloaded here. Since I only have programming experience in python I am not sure how I could/should run this code. The sklearn svm module calls libSVM and enables the use of these models within python. Can I download the sklearn svm code and change parts of it to incorporate SVORIM and also be able to use it? If so, how should I approach this?

If it is not possible to incorporate this into python, how can I then run this code in another way?

I hope to hear from you and thank you in advance!

[jjerphan]

Hi @ThomasPel1,

Thanks for your pointers.

Can I download the sklearn svm code and change parts of it to incorporate SVORIM and also be able to use it? If so, how should I approach this?

Yes, you can. libSVM is bound to the svm module using Cython.

You should be able to modify scikit-learn to use this adaptation by:

• changing the vendored libsvm C/C++ code base.
• adapting the Cython headers (for any parameters that you want to expose from the C/C++ interface)
• adapting the Cython sources (here and potentially there) (for any parameters that you want to expose)
• adapting the Python internal calls to use any new parameters that you want to expose from the Cython interface. This likely sits in svm.BaseLibSVM and svm.BaseSVC.
• adapting the Python API for your personal use

Those are general guidelines, but when modifying one would surely need more inspection (that I haven't had time to go through yet).

Here are some relevant sections from the Cython's docs:

• Embedding Cython modules in C/C++ applications
• Interfacing with External C Code
• Using C++ in Cython