Orange Forum • View topic - Weight data for boosting

Weight data for boosting

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

Weight data for boosting

Postby wkim » Mon Oct 09, 2006 17:50

Hi,

I would like to try boosting on Naive Bayes.

Is there any way to use arbitrary weighting scheme to data instead of frequency for e.g. naive bayes ?

Postby Blaz » Sat Oct 14, 2006 17:31

Boosting is a wrapper enterly written in Python. The learner class, BoostedLearnerClass (orngEnsemble.py module in Orange's root), is about 30 lines long and for what you are proposing the simplest think is to take its code and modify it.

I must admit, though, that I did not entirely understand your question. orngEnsamble starts with all examples having equal weight, which is then adjusted using procedure from AdaBoost.M1 (that is, the weights are not based on frequencies but instead on (in)correct classification of particular data instance).


Return to Questions & Support



cron