PDA

View Full Version : Weka Wrapper



snassimr
02-20-2007, 04:37 PM
I try to use WrapperSubsetEval to evaluate subset "merit" but I need the classifier that was built for my further purpose . getClassifier method bring me classifier with option but without tree to able me to classifiy.

Need I to use somthing else different ?

Mark
03-01-2007, 05:04 PM
Hi,

WrapperSubsetEval performs internal cross-validation on the training data in order to determine the merit of a given subset of attributes. Therefore, in your case, there are multiple trees constructed in the process of evaluating each subset. This is done in order to stabilize the estimate of the merit (error rate in this case). So, it doesn't make much sense to obtain the classifier from WrapperSubsetEval as it will just be the last tree built from the last run of cross-validation (and will have used only 4/5ths of the training data to boot). The best thing to do is to obtain the final selected attribute set, reduce the full training data to contain only these attributes, and then construct a decision tree using this.

Cheers,
Mark.