Orange Forum • View topic - More than one class feature

More than one class feature

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

More than one class feature

Postby arauzo » Tue May 31, 2005 19:11

Have you considered including support for problems with more than one class (i.e. predicting/target) feature?

There are many of these datasets in information retrieval. For example:

Term1 Term2 Term3 Category1 Category2 ...
0 1 1 1 1 ...
1 1 0 0 1 ...

i.e. each document could belong to more than one category.

Neural Networks are able to handle this, and other classifiers too. Do you think it would be much work to support this? I mean for me to do it.

Regards,

Postby Janez » Fri Jun 03, 2005 21:46

Orange does not support multiple class attributes, that is neither can the ExampleTable (or Example) store such data, nor can a classifier return such predictions. If you've had only two class attributes you could cheat your way by Cartesian products, but if you have multiple, than I guess the best way to treat them in Orange would be to construct one classifier for each class and then make them predict probabilities of belonging to a certain class.

Besides Orange's inability to predict all the classes at once, don't you think that is better anyway to have a separate model for each class? (But I never did any real text mining, so I wouldn't know.)


Return to Questions & Support