search

Xtra-Computing / thundersvm

ThunderSVM: A Fast SVM Library on GPUs and CPUs
Apache License 2.0
1.56k stars 217 forks source link

Scikitlearn wrapper kernel parameters #85

Closed ikiskin closed 6 years ago

ikiskin commented 6 years ago

New users may be confused about kernel selection, due to the syntax difference with libsvm's scikitlearn: SVC(kernel='rbf') and thundersvm's: SVC(kernel=2) Running thundersvm with kernel='rbf' does not return an error, and the optimization fails (objective -inf). I would suggest either updating the interface to match libsvm's scikitlearn syntax and/or alerting the user when an invalid kernel (and possibly others) argument is passed.

QinbinLi commented 6 years ago

Thanks for your feedback. We have improved the interface and the usage of kernel parameter is the same with libsvm now.

ikiskin commented 6 years ago

Great, just a few spots left in the documentation to fix, under the Usage heading on https://github.com/zeyiwen/thundersvm/tree/master/python subheadings: "SVM classification", "One-class SVMs", and "SVM regression". class SVC(kernel = 'rbf', ... etc

QinbinLi commented 6 years ago

Thank you very much. The doc has been fixed.