[pymvpa] Class weighting in unbalanced EEG data

[Ben McCartney]

 I tried the Balancer with a range of classifiers, and it seemed to improve some and impair others. My worry with this is that we are taking our already small dataset and making it smaller, 100 trials seems quite small to me to classify reliably, although I could just be wrong on that. I did also try implementing oversampling myself wheich did seem to help a little bit, however this seems like a more awkward solution to the problem than if I could give the classes the correct weights during training as in my first post.

 Is changing the dataset (eg undersampling) a more widely practiced solution to this problem with my kind of data?

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.alioth.debian.org/pipermail/pkg-exppsy-pymvpa/attachments/20160506/6cbd10c8/attachment.html>