[pymvpa] Help needed untangling searchlight accuracy bias

[Mike E. Klein]

So the zscoring/averaging order mixup is absolutely what caused these
strange distributions… swapping the order or removing zscoring returns them
to normal-ish (with the zscored results having a much narrower
distribution). That being said, I can't get my mind around why inverting the
zscore/averaging would lead to this particular strange result: not only were
accuracies increased high the board, but my observed searchlight accuracy
peaks were right around the hypothesized region of interest. In take 2 of
the analysis, at least initially, zscoring (at the correct time) seems to
obliterate any effect.. although it appears to return a bit if I leave out
the zscoring completely. I'm also wondering if there's relatively simple way
to compromise between raw 4D files and standard scores by just going for
effect sizes.

Thanks again,
Mike

On Thu, Oct 20, 2011 at 4:10 PM, Yaroslav Halchenko
<debian at onerussian.com>wrote:

>
> On Thu, 20 Oct 2011, Mike E. Klein wrote:
> > I wasn't getting that error… at the time of zscoring I hadn't yet
> > removed category #3, so it was seeing -exactly- 3 samples per chunk.
>
> aha... indeed... may be we should warn if number of samples per chunk <
> some reasonable number, e.g. at least 5
>
> --
> =------------------------------------------------------------------=
> Keep in touch                                     www.onerussian.com
> Yaroslav Halchenko                 www.ohloh.net/accounts/yarikoptic
>
> _______________________________________________
> Pkg-ExpPsy-PyMVPA mailing list
> Pkg-ExpPsy-PyMVPA at lists.alioth.debian.org
> http://lists.alioth.debian.org/cgi-bin/mailman/listinfo/pkg-exppsy-pymvpa
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.alioth.debian.org/pipermail/pkg-exppsy-pymvpa/attachments/20111024/e8bdd72d/attachment.html>