Orange Forum • View topic - Sparse Data SVM Question

Sparse Data SVM Question

A place to ask questions about methods in Orange and how they are used and other general support.

Sparse Data SVM Question

Postby jacoblyles » Tue Mar 17, 2009 10:49

I am interested in using the sparse SVM learner on a sparse dataset. My understanding is that this can be done by storing the data in a .basket file. If so, how do I denote which attribute is the class attribute?

If not, how do I use the sparse SVM?

Thanks,
Jacob

Postby Ales » Tue Mar 17, 2009 13:10

Store the data in a tab file with a class attribute, and list sparse values using the "basket attribute" (see this for more info: http://www.ailab.si/orange/doc/reference/tabdelimited.htm#basket

Postby jacoblyles » Tue Mar 17, 2009 21:57

How does time and memory performance compare to, say, libsvm?

I like orange's clean layout and flexibility, but I am getting the impression that the dense data features are the most mature.

Postby Blaz » Tue Mar 17, 2009 22:26

Jacob, Orange uses libsvm (provides a Python and Orange interface to libsvm rutines), and incorporates libsvm's C code, so performance should match.

Libsvm new version

Postby aonlazio » Sun Apr 05, 2009 21:49

Hi,
Does Orange have plan to update the version of libsvm in its package also? I mean, the version libsvm Orange currently uses is 8.1 now it's 8.9 .

Postby Ales » Mon Apr 06, 2009 11:54

I will try to update orange with the new version of svmlib as soon as possible. Check back in about a week for an update.

Postby Ales » Mon Apr 20, 2009 9:08

Just posting to let you know that orange (daily snapshot) now uses libsvm 2.89


Return to Questions & Support