Orange Forum • View topic - Incremental Updates to Classifiers

Incremental Updates to Classifiers

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

Incremental Updates to Classifiers

Postby Cerin » Fri Jul 11, 2008 0:50

Is it possible to train classifiers incrementally, instead of requiring the entire training corpus at once? The pre-built classifiers are fantastic, but as my corpus changes, I'd like to feed the changes into the classifier instead of having to rebuild the entire corpus and create an entirely new classifier from scratch.

Postby Janez » Sat Jul 12, 2008 19:49

No, more or less.

Orange has no incremental learning algorithms. One could update a trained classifier - this would be easy for naive bayesian classifier or kNN, less so for trees, for instance - but I don't think anybody tried it yet.

I'm sorry, but that's one thing we never needed ourselves and thus never implemented it.

Postby Cerin » Sat Jul 12, 2008 21:14

Thanks, I suspected that wasn't the normal use-case.

However, I've found another tool that does what I need. It doesn't have all the features of Orange's Bayesian classifier, but Divmod's Reverend module supports incremental updates by default.

Return to Questions & Support