Re: percentage split versus re-evaluated complete data set

> So, in order to produce a model for production from a single data set, one has to split the dataset into the training and the test set beforehand (e.g. with a filter) instead of using the percentage split in the explorer.
> Am I right that this behaviour (final saved model is based on learning on the complete dataset) is also used during usage of the CLI for learning?

Whatever is defined as training set via -t on the command-line or
loaded as main dataset in the Explorer will get used for building the
final model.

> That behaviour also explains, why I was never able to reproduce the good results of AutoWeka in the Explorer. The AutoWeka stats are apparently based on learning and predicting on the full dataset but the stats printed in the Explorer are based on learning on the selected subset (e.g percentage split or cross validation) and predicting on the rest.

I've never used AutoWeka, so I can't comment on it, unfortunately.
However, if it is using the training set, then the statistics will be
quite optimistic.

Cheers, Peter
-- 
Peter Reutemann
Dept. of Computer Science
University of Waikato, NZ
+64 (7) 577-5304
http://www.cms.waikato.ac.nz/~fracpete/
http://www.data-mining.co.nz/
_______________________________________________
Wekalist mailing list -- [email protected]
Send posts to [email protected]
To unsubscribe send an email to [email protected]
To subscribe, unsubscribe, etc., visit https://list.waikato.ac.nz/postorius/lists/wekalist.list.waikato.ac.nz
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html