Orange Forum • View topic - LinearSVM performance differences

LinearSVM performance differences

General discussions about Orange and with Orange connected things (data mining, machine learning, bioinformatics...).

LinearSVM performance differences

Postby re832003 » Wed Mar 06, 2013 22:39


I used LinearSVM - which is a wrapper around LIBLINEAR - and noticed big differences between the results of the wrapper and the pure implementation? The difference is up to 10% higher for LinearSVM.

I'm kind of confused about the reason. I tried LIBLINEAR with the same parameters set in the documentation of LinearSVM but I still get this big difference.

The LinearSVM doesn't mention how the normalization is done. Is normalization one reason for this performance difference?

Finally, if I end up using LinearSVM from Orange, is there a way to save the trained model to use it for future on new data?


Return to General Discussions