SVM implementation supporting weighted samples

Hi,

I'm looking for an SVM (or other kernelizable classifier) implementation that support both multiclass classification and instance weighting. (Actually, instance weighting is much more important, but it would be nice to have out-of-the-box multiclass as well).

Any ideas?

asked Jun 18 '13 at 17:14

yoavg
741●12●23●31

One Answer:

oldest newest most voted

In scikit-learn you can use the Nystrom kernel approximation + SGDClassifier with hinge loss and l2 penalty to approximate a kernelized SVM while still benefiting from the weighted instance features of the linear model. To workaround the learning rate issue you can replace SGDClassifier with PassiveAggressiveClassifier.

SGDClassifier comes with One vs the Rest multiclass by default.

answered Jun 18 '13 at 17:40

ogrisel
4989●9●55●91

Subscription:

Once you sign in you will be able to subscribe for any updates here

Tags:

svm ×134
implementation ×39
weighting ×4

Asked: Jun 18 '13 at 17:14

Seen: 737 times

Last updated: Jun 18 '13 at 17:40

SVM implementation supporting weighted samples

Subscription:

Related questions