[pymvpa] Searchlight question

[Per B. Sederberg]

On Thu, Jun 10, 2010 at 10:07 AM, Michael Hanke <michael.hanke at gmail.com> wrote:
> Hey,
>
> On Thu, Jun 10, 2010 at 09:48:44AM -0400, Per B. Sederberg wrote:
>> Hi Folks:
>>
>> I'm gearing up to do a searchlight analysis (with the bleeding edge
>> development code) and, when reading through the searchlight.py example
>> (and making some minor corrections to the text), I learned how
>> center_ids work.  This is an awesome feature, but I wonder if it
>> should be attached to the searchlight run call instead of to the init
>> call.
>>
>> Specifically, we currently set the center_ids like this:
>>
>> sl = sphere_searchlight(cv, radius=radius, space='voxel_indices',
>>                              center_ids=center_ids,
>>                              postproc=mean_sample())
>>
>> But might it be better to keep the searchlight instance divorced from
>> the dataset and pass it in at runtime:
>>
>> sl_map = sl(ds, center_ids=center_ids)
>>
>>
>> I have not looked closely at the underlying code, so I don't know if
>> there is a specific reason it works as it does.  Any thoughts?
>
> We cannot attach the center_ids to the __call__(). All objects in PyMVPA 0.5
> (when it will be finally released) will take a dataset on call() and do
> something useful. A dataset alone has to be sufficient, since they also
> will return a dataset when called and that way can be combined into
> arbitrary processing chains.
>
> Hope that explains,
>

Ahhh, so it's less that the SearchLight instance is distinct from the
dataset and more that the __call__ methods work only with datasets so
you can set up cool pipelines.

Sounds good to me :)

Best,
P

>
> Michael
>
> --
> GPG key:  1024D/3144BE0F Michael Hanke
> http://mih.voxindeserto.de
>
> _______________________________________________
> Pkg-ExpPsy-PyMVPA mailing list
> Pkg-ExpPsy-PyMVPA at lists.alioth.debian.org
> http://lists.alioth.debian.org/mailman/listinfo/pkg-exppsy-pymvpa
>