[pymvpa] Dissimilarity vs classification

[Alyson Saenz]

Dear PyMVPA community,

I'm struggling to get my head around this. I'm running a searchlight with
20 participants, trying to classify two categories. I run a one sample t
test (50% chance) and I find that region X classifies accurately the two
categories, but no other region does. Then I created a dissimilarity map
for each participant comparing the two same categories. Then I run a one
sample t test on the dissimilarity maps (chance calculated by permutating
the labels) I find that region Y has the biggest difference between the two
categories.

The way I always understood the classification was that if the two
categories were consistently different, then the classifier would be able
to predict which is which. So when I test for dissimilarity, two patterns
sufficiently different should be accurately distinguished, right?

Can anyone point me towards a paper or someplace where I can understand why
I find these different results?

Best regards,

Alyson
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://alioth-lists.debian.net/pipermail/pkg-exppsy-pymvpa/attachments/20181206/47f32106/attachment.html>