[pymvpa] hyperalignment back to nifti

[Kiefer Katovich]

Hello,

I would like to be able to hyperalign data, run classification on the
hyperaligned data, then export the classification maps back into brain
space so that they can be viewed.

Currently I am using a FixedNElementTailSelector to select out 3000
features prior to hyperalignment. I apply the same feature selector to
the data that will be hyperaligned.

I believe doing this feature selection is preventing me from saving
the data back into brain space. If I attempt to use map2nifti on the
hyperaligned or feature selected data (or classification results) then
I get a mapping error. The specific error is:

mvpa2/featsel/base.pyc in _reverse_data(self, data)
    142                       refcheck=False)
    143         mapped.fill(self.filler)
--> 144         mapped[:, self._slicearg] = data
    145         return mapped
    146

ValueError: array is not broadcastable to correct shape


This makes sense, I think. The feature selection cuts out most of the
voxels and those voxels have no way of being mapped back into their
original position? Does this mean that hyperalignment using feature
selection prevents a mapping of the classification back into brain
space?

I'm wondering if something can be used in place of feature selection
to preform hyperalignment so that the results can be visualized
overlayed on the brain. Hyperalignment seems to be capable of running
just feeding in the entire brain, but that is very strenuous on the
computer.

On another note, I am still having trouble with SVD convergence, but
for now am just using a subset of the brains that the hyperalignment
is able to converge with.

Thanks,
Kiefer