[Pkg-exppsy-pymvpa-dev] Question about new fmri_dataset

[Valentin Haenel]

* Yaroslav Halchenko <debian at onerussian.com> [100105]:
> 
> On Tue, 05 Jan 2010, Valentin Haenel wrote:
> > > On Tue, 29 Dec 2009, Valentin Haenel wrote:
> > > > > So, they look very close but not an exact match... how close your results
> > > > > matched before? may be I am taking some other results for matlab? (see path
> > > > > above)
> > > > Looks like the right file
> > > but do you remember what differences you saw?
> > not sure what you mean by that.
> Differences in the results between matlab and pymvpa implementations
I still have the files here, the look really similar in mricron, but i couldn't
find any kind of systematic pattern. I attributed the differences to numerical
instabilities.

I also remember having used a so called "large" searchlight. This means the 
comparison, to figure out if a voxel is in the current searchlight is performed
with a less than or equal to test. But as far as i can tell this should be the
case here. (at least from what i see in neighborhood.py in current mh/master on
alioth)

Still, 86% correlation isn't really an awefull lot..

V-