Orange Forum • View topic - SVM classifier with a Multi-class label

SVM classifier with a Multi-class label

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

SVM classifier with a Multi-class label

Postby bonjovi2012 » Sun Jul 01, 2012 15:07

Hi

I am a newbee to Orange. In the past, I used other tools such as RapidMiner. RapidMiner cannot support a SVM learner on a dataset with (i) a multi-class or (ii) categorical input variables. During the data preprocessing, I had to
1. Convert the categorical variables to continuous variables
2. Build multiple SVM models (i.e. one for each class)

Today, I started to use Orange Canvas for a college project and I was pleasantly surprised that the SVM widget can support both a multi-class label and categorical input variables. This is great! As it saves me time.

But I would like to know how the Orange SVM widget can support multi-class label and categorical input variables? Does it automnatically convert categorical variables to numeric? etc

thanks very much
Bonjovi

Re: SVM classifier with a Multi-class label

Postby Ales » Mon Jul 02, 2012 9:41

Categorical input variables are converted to dummy indicator variables (one for each value).

And for K multi-class problem multiple binary classifiers are build automatically (using 1 vs 1 scheme). This is a feature of the underlying LibSVM library which is used in Orange.

Re: SVM classifier with a Multi-class label

Postby bonjovi2012 » Mon Jul 02, 2012 14:17

Thanks Ales


Return to Questions & Support



cron