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

[Michael Hanke]

On Thu, May 26, 2011 at 02:31:21PM +0800, Zhen Zonglei wrote:
>     We want to train a classifier to decode the state of subject (Patient
> vs. Normal) from a neuroimaging parameter map and evaluate the accuracy. The
> samples are independent and no clear group of the subjects exists. How
> should we set the chunk in pymvpa?
> 
> 
> 
> Another similar question: If we try to predict a mental score from the
> neuroimaging parameter map for each subject with regression analysis, again,
> how should we set the chunks?

Independent of what you want to do with the data, the chunks indicate
mutually independent portions of data. If you suspect that there are no
dependencies between you subjects, you should set the chunks to a unique
value per subject. I.e.

  chunks = np.arange(len(dataset))


Michael

-- 
Michael Hanke
http://mih.voxindeserto.de