[pymvpa] Test on a sample attribute other than 'targets' in searchlight?

[Dillon Plunkett]

Hi all,

I'm looking to perform a searchlight analysis with normal
leave-one-chunk-out cross validation, but to train the classifier on one
set of labels and test on another (in other words, train on 'targets', but
have the error be calculated by comparing predicted values for the held out
samples to the values of a different attribute of the samples in the test
set, rather than the true target values).

However, I don't see an obvious way to construct a CrossValidation object
that accomplishes this. Is there a straightforward way to do this that I've
missed so far or has anyone done something similar in the past?

Thanks very much for any guidance!!


Dillon Plunkett
Greene Lab
Harvard University
dillonplunkett.com