Hi all, Thanks Jo and Taku for the responses. I will try using either the sub-sampling or over-sampling methods instead then. I thought that weighting the classes would do functionally the same thing -- why isn't it common practice?
Taku, or anyone else, do you have any references for the bootstrapping method you mentioned? Neither I nor my collaborators have heard of that method for fMRI, but it would be nice to be able to use most or all of our data from the larger class. Jo, we are changing the cross-validation partition to leave-one-subject-out, but it still ends up pretty imbalanced. So we're still going to have to use a method to correct the imbalance... Thanks, William
_______________________________________________ Pkg-ExpPsy-PyMVPA mailing list [email protected] http://lists.alioth.debian.org/cgi-bin/mailman/listinfo/pkg-exppsy-pymvpa
