[pymvpa] Paper on searchlight radius?

[Etzel, Jo]

Yaroslav beat me to the response, thanks! Another potentially useful 
paper is https://arxiv.org/abs/1210.6317v1 (On the geometric structure 
of fMRI searchlight-based information maps. Shivakumar Viswanathan, 
Matthew Cieslak, Scott T. Grafton).

I personally haven't been working on interpreting searchlight analyses 
lately; my general interpretation is that results (particularly group 
level) are more believable if they appear over a sensible range of 
searchlight radii (e.g., with both 2 and 3-voxel). Effects that only 
appear with a particular searchlight radius would need to be accompanied 
by some convincing evidence for why that specificity is meaningful 
(instead of some quirk).

Jo



On 10/8/2018 7:39 AM, Yaroslav Halchenko wrote:
> I would say it all depends on many factors, such as problem at hand, 
> classifier, etc. So not sure if anything systematic and actually 
> generically useful could be done/stated. But have a look may be at Jo's 
> paper.
> 
> Etzel JA, Zacks JM, Braver TS. Searchlight analysis: Promise, pitfalls, 
> and potential. Neuroimage. 2013. PMCID: PMC3988828
> 
> On October 8, 2018 7:55:43 AM EDT, Alyson Saenz <neuro.alyson at gmail.com> 
> wrote:
> 
>     Dear PyMVPA community,
> 
>     Does anyone know a paper in which someone has explored the effect of
>     different searchlight radius?
> 
>     I know that a bigger radius increases performance but also
>     processing time, and that at some point you will get bad
>     performances, but, has anyone systematically explored this?
> 
>     Thanks for your help
> 
>     Alyson
> 
> 
> 
> 
> 
> -- 
> Yaroslav O. Halchenko (mobile version)
> Center for Open Neuroscience http://centerforopenneuroscience.org
> Dartmouth College, NH, USA
> 
> 
-- 
Joset A. Etzel, Ph.D.
Staff Scientist
Cognitive Control & Psychopathology Lab
Washington University in St. Louis
mvpa.blogspot.com