[pymvpa] No samples of a class in a chunk
Jan Derrfuss
j.derrfuss at donders.ru.nl
Fri Aug 2 11:49:01 UTC 2013
Hello,
I would like to run an exploratory analysis where the presence of
samples in the chunks was not under experimental control. As a result,
there are chunks where only one of the two target classes I'm decoding
is present in the chunk the classifier is tested on (there is also a
single chunk in one subject where neither of the two classes is
present). I'm running a searchlight analysis with 6-fold
cross-validation, use a linear SVM, and compute the mean true positive rate.
Is there a preferred way to deal with such a situation?
Jan
PS. The temperature here in the Lower Rhine region is currently 33 °C
(91 °F). Wherever you are, I hope it's colder there! :-)
More information about the Pkg-ExpPsy-PyMVPA
mailing list