Multi-Class Classifier

[dbl001]

Hi,

Can support vector machines be used as multi-class classifiers?

Thanks in Advance

[yassin.ezbakhe]

LIBSVM supports multi-class classification. If you're using Python, you can use scikit-learn, which under the hood uses LIBSVM; for an example, go here: http://scikit-learn.sourceforge.net/...classification

[hsolo]

Not a direct followup to this question but I was wondering how Multinomial Softmax (generalization of logistic for multiple classes) compares with SVMs for multi-class problem, in practice.

One difference is the softmax directly gives the probability whereas SVM probabilities are 'indirect'

I was wondering how does this comparison pan out in practice?

[rimidan]

That is an interesting question. Maybe yassin is still around and has an answer to that.

[htlin]

Originally Posted by hsolo

Not a direct followup to this question but I was wondering how Multinomial Softmax (generalization of logistic for multiple classes) compares with SVMs for multi-class problem, in practice.

One difference is the softmax directly gives the probability whereas SVM probabilities are 'indirect'

I was wondering how does this comparison pan out in practice?

In my experience one-versus-all linear SVM and multinomial logistic regression are somewhat comparable in practice.