[pymvpa] How you run permutation test for search-light analysis?
debian at onerussian.com
Tue Mar 27 00:19:55 UTC 2012
Do you believe in GNB? ;) or -- does it perform on your data?
I am asking because gnb searchligting is fast so you can easily
accumulate enough permutations in feasible time.
NB. Hopefully tomorrow/day after we would release bugfix release
addressing rare misdemeanor of CachedQueryEngine which should be
used while performing permutation testing.
another underexplored approach which wasn't yet mentioned could be to
assume the family of the chance distribution (e.g. binomial which for
binary would be well approx by normal) and then collect a
reasonable but small number of sample (e.g. 30-50) which would be used
to estimate the parameters of the distribution per each voxel.
In PyMVPA MCNullDist can take arbitrary distribution as 2nd parameter
(by default is 'Nonparametric') which would be fit based on the
permutted estimates, e.g. what I believe we do in our unittests:
On Sun, 25 Mar 2012, Vadim Axel wrote:
> When I run a whole-brain search-light analysis with non-overlaping lights
> (each voxel participates only in one light) I get about 3000 lights (half
> a hour classification). Running 1000 times such permutation analysis would
> obviously take weeks.� And this is a limited non-overlapping lights case.
> Is there any more realistic approach? I found here that people ran
> permutation analysis only 10 times while they stack the predictions of all
> voxels together to have their chance distribution:
> (see "Searchlight analysis" section second paragraph). Does it make sense
> for you?
> Visible links
> 1. http://cercor.oxfordjournals.org/content/early/2011/12/20/cercor.bhr357.full
> Pkg-ExpPsy-PyMVPA mailing list
> Pkg-ExpPsy-PyMVPA at lists.alioth.debian.org
Keep in touch www.onerussian.com
Yaroslav Halchenko www.ohloh.net/accounts/yarikoptic
More information about the Pkg-ExpPsy-PyMVPA