[pymvpa] Group level searchlight

[Pegah Kassraian Fard]

Many thanks. A last question: I might go for the classical approach of
t-testing, and keeping the results per voxel which are significant against
chance level - say however that this is given for a certain voxel, how is
its accuracy then determined based on the subjects accuracy? Would that be
an average?

On Sat, Oct 21, 2017 at 7:27 AM, Nick Oosterhof <
n.n.oosterhof at googlemail.com> wrote:

> On 20 October 2017 at 21:28, Pegah Kassraian Fard <pegahkf at gmail.com>
> wrote:
>
>> Many thanks, very useful indeed. A related question: So the input to the
>> groupwise calculations are the permutated searchlight maps from single
>> subjects - and computation-wise this seems to me to be quite a bottleneck,
>> given that one whole brain searchlight already takes several hours. Is
>> there any way to speed up this part?
>>
>
> You could use a faster classifier combined with the searchlight, in
> particular a gaussian naive bayesian (GNB) classifier; see
> http://www.pymvpa.org/generated/mvpa2.measures.
> gnbsearchlight.GNBSearchlight.html
>
> For typical fMRI data GNB tends to be less accurate than linear SVM, in
> other words there is a speed - accuracy trade-off.
>
> _______________________________________________
> Pkg-ExpPsy-PyMVPA mailing list
> Pkg-ExpPsy-PyMVPA at lists.alioth.debian.org
> http://lists.alioth.debian.org/cgi-bin/mailman/listinfo/pkg-exppsy-pymvpa
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.alioth.debian.org/pipermail/pkg-exppsy-pymvpa/attachments/20171021/40015f5a/attachment-0001.html>