[pymvpa] How to set the chunks with the independent samples?

Michael Hanke michael.hanke at gmail.com
Thu May 26 12:03:14 UTC 2011

On Thu, May 26, 2011 at 07:35:37PM +0800, Zhen Zonglei wrote:
> Hi Michael
> Thanks very much. If we set he chunks to a unique value per subject. Then
> all chunk-wise preprocessing of the raw data (e.g. zscore) will be improper.
> We thus need to preprocess the data out of the whole ChainMapper. Am I
> right?

Not exactly. If you are using the 0.6 series, all mappers can be told to
use arbitrary attributes to do their work, e.g. you can zscore based on
groups defines by a 'usethisoneinstead' attribute. Just look up the
name of the argument to the mappers you are using. Moreover, in the 0.6
series the name 'chunks' should now be mere convention, not necessity. Many
helper functions will create such an attribute for you, but your are
totally free to ignore or modify or rename it, as you please.

And lastly, if you don't need all subjects to be processed individually
(e.g. in a cross-validation procedure). You can broaden your chunks to
include multiple subjects each -- afterall they are all mututally independent
and so arbitrary grouping should have no effect.


Michael Hanke

More information about the Pkg-ExpPsy-PyMVPA mailing list