[pymvpa] Q about faster GNB searchlight script: quick way to compute sphere voxels? Plus some Matlab code, in case it helps
Francisco Pereira
fpereira at princeton.edu
Thu Apr 29 15:37:49 UTC 2010
Rajeev,
You don't need to reimplement this, as it all started with my talking
with Yarik about the fast GNB in MATLAB that we already had, and Yarik
producing a python version in 2 days flat :)
On Thu, Apr 29, 2010 at 10:53 AM, Rajeev Raizada
<rajeev.raizada at dartmouth.edu> wrote:
>
> The script gains its speed by exploiting a nice trick
> that I hadn't thought of:
> because GNB doesn't use any info about particular
> combinations of voxels, you can calculate the means and s.d.s
> of all the voxels at once at the beginning of each cross-validation rep,
> and then reuse those precomputed values for any searchlight sphere.
It does that...
> But I was wondering if you guys have a quick trick
> for computing the voxel coords of searchlight spheres which
> are of larger radius. Once you get past radius 1 or 2,
> the lookup-table approach starts to get pretty impractical.
... and uses a pretty economic lookup table, so I think you might be
able to go up to 3.
We're beta testing the MATLAB one for general use (it's been in use
internally since last year, so the core functionality is fine). If
you'd like to be a guinea pig let me know and I can get you in touch
with the code ;)
Francisco
On Thu, Apr 29, 2010 at 11:02 AM, Francisco Pereira
<francisco.pereira at gmail.com> wrote:
> Rajeev,
>
> You don't need to reimplement this, as it all started with my talking
> with Yarik about the fast GNB in MATLAB that we already had, and Yarik
> producing a python version in 2 days flat :)
>
> On Thu, Apr 29, 2010 at 10:53 AM, Rajeev Raizada
> <rajeev.raizada at dartmouth.edu> wrote:
>>
>> The script gains its speed by exploiting a nice trick
>> that I hadn't thought of:
>> because GNB doesn't use any info about particular
>> combinations of voxels, you can calculate the means and s.d.s
>> of all the voxels at once at the beginning of each cross-validation rep,
>> and then reuse those precomputed values for any searchlight sphere.
>
> It does that...
>
>> But I was wondering if you guys have a quick trick
>> for computing the voxel coords of searchlight spheres which
>> are of larger radius. Once you get past radius 1 or 2,
>> the lookup-table approach starts to get pretty impractical.
>
> ... and uses a pretty economic lookup table, so I think you might be
> able to go up to 3.
>
> We're beta testing the MATLAB one for general use (it's been in use
> internally since last year, so the core functionality is fine). If
> you'd like to be a guinea pig let me know and I can get you in touch
> with the code ;)
>
> Francisco
>
More information about the Pkg-ExpPsy-PyMVPA
mailing list