Know any good c++ support vector machine (SVM) libraries ?

Question

Hey everyone,

Do you know of any good c++ svm libraries out there ?

I tried libsvm (http://www.csie.ntu.edu.tw/~cjlin/libsvm/) but so far I'm not flabbergasted (no documentation, or close to none).

I have also heard of SVMLight and TinySVM. Have you tried them ? Any new players ?

Thanks !

JC

Answer 1

• SVMTorch (support vector machines for large-scale regression problems) implemented in the torch machine learning library.
• mySVM - based on the optimization algorithm of SVM-Light.

A comprehensive list of SVM libraries can be found here.

Answer 2

Here's another monster list of SVM packages, libraries and SVM applications.

Answer 3

I've used SVMLight before and found it to be very stable and fast. I had a good experience using it and would recommend it.

However, I think there is probably less documentation on SVMLight than libSVM; just the papers by Thorsten Joachims and the comments in the source code. I didn't find the source too hard to follow in general, but you need to read the papers beforehand to understand the background. It's also written in pure C, not C++, if that matters to you.

As for 'new players', the new research is mostly into making the SVM optimisation algorithms more efficient. For example, using stochastic gradient descent as in svmsgd and pegasos. I haven't looked at the implementations of these algorithms, but it's research code so I wouldn't expect that they are particularly easy to follow, if that's your primary concern.