[pymvpa] sensitivity analysis as feature selection in searchlight analysis

Michael Hanke mih at debian.org
Mon May 6 11:52:19 UTC 2013


On Wed, Apr 24, 2013 at 11:11:55AM +0200, Danilo Bzdok wrote:
> Hi there,
> I am wondering whether the PyMVPA API allows not only ANOVA-based
> feature selection as a preprocessing step during a searchlight
> analysis but also a sensitivity analysis based on LinearNuSVMC for
> example? In a way, this would probably be equivalent to a
> single-iteration recursive feature selection before actual
> classification in each searchlight. I would be happy for any advice...

Take a look at sensanas.py in the example folder. It shows you a bunch
of sensitivity measures all used with the ANOVA API. You can plug them
in, where you have ANOVA right now, and you would achieve the desired


Michael Hanke

More information about the Pkg-ExpPsy-PyMVPA mailing list