[pymvpa] Using ridge regression in a searchlight analysis

[VINCE TASCHEREAU-DUMOUCHEL]

Hi everyone,

I am trying to run a searchlight with a linear outcome ranging from 0 to 6
(I recoded the Targets of my dataset accordingly). I have 1 average pattern
for each of these target values for each of my 6 chunks (I should probably
include more trials instead of using averages, but this is what I am
playing with so far to have an even dataset) .

I thought of using ridge regression and here is the code I have so far:

    clfsd = RidgeReg(lm=None, enable_ca=['stats'])

    cv = CrossValidation(clfsd,
                     NFoldPartitioner(),
                     errorfx=lambda x,y: pearsonr(x,y)[0],
                     enable_ca=['training_stats','stats'])

    sl = sphere_searchlight(cv, radius=3, postproc=mean_sample())
    sl_results = sl(ds_alll)

I am having a hard time figuring out what the pearson r means in this
context. Is it the correlation between Y  and Y-Hat? I end up with some
negative values. Any chance it could be interpreted has a negative relation
between the voxels within the searchlight and the continuous target values?

Also, since I am fairly new to pyMVPA, anyone has any suggestion on how to
figure that out by myself in the code? I tried running in debug mode but I
seem to run into an error that I don't have if I run the code normally.

Thanks!

V
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.alioth.debian.org/pipermail/pkg-exppsy-pymvpa/attachments/20170117/2aeba6c8/attachment.html>