[pymvpa] kNN and sensitivity map

[Yaroslav Halchenko]

On Tue, 31 May 2016, marco tettamanti wrote:

> Dear all,
> since kNN performs best on a particular dataset, I am trying to obtain a
> sensitivity map based on the code that I have used for other classifiers.
> However, I receive an error message, and I am not able to proceed further.
> I have been struggling with the documentation, but I still cannot find any
> solutions for this problem.
> Can somebody please help?

kNN (depending on your k) could be a highly non-linear classifier and it
wouldn't be as easy to estimate its sensitivity as in the case of linear
classifiers.  With relatively large k, decision surface could be
approximately linear and then some kind of 'trivial sensitivity' could
be assessed (I remember Francisco Pereira talking about that in one of
his paper) but we don't have it implemented.  

The only quick way to get a sensitivity for you would be to try

NoisePerturbationSensitivity

but it might take awhile depending on the size of your dataset ;)

Myself, haven't used that functionality for ages, so although
(unit)tested -- mileage might vary

-- 
Yaroslav O. Halchenko
Center for Open Neuroscience     http://centerforopenneuroscience.org
Dartmouth College, 419 Moore Hall, Hinman Box 6207, Hanover, NH 03755
Phone: +1 (603) 646-9834                       Fax: +1 (603) 646-1419
WWW:   http://www.linkedin.com/in/yarik