Run time SVM learners

Dear all!

I hope you can give me some answers on why my model is running for multiple hours and doesn't give any output.

I want to apply a SVM learner, I tried the normal SVM learner under Mining, the LIBSVM, the WEKA libSVM, SMO and the SMOreg. I tried all the SVM learner nodes I could find.  The SMOreg node didn't work and the others run for a very long time.

Only the WEKA libSVM learner worked, however the WEKA predictor gave the error that there where no nominal or string variables. But i changed my predictor variable in a string so this was a realy strange error.

 

So I have 3 questions:

1. Are there more people that have the problem that it takes a long time, if so, how did you solve this?

2. What can I do so that the WEKA predictor doesn't give me this error anymore?

3. Is there a maximum in the usage of variables and/or cases for the SVM learners?

I really hope that you can help me!

Chantine

 

Hi Chantine,

1. The LibSVM documentation mentions it is better used with normalized values (2.2 Scaling). In my experience that makes running SVM algorithms much faster.

2. You might need to node description of the string variable.

3. I do not think there is a hard limit (probably they should fit in memory).

Cheers, gabor

Hi Gabor!

Thanks for the quick response! I am now going to test your suggestion. I will come back to you with my findings!!

Thankss!!!

Chantine