[pymvpa] PCA transformation prior to SVM classification

[Yaroslav Halchenko]

On Mon, 29 Nov 2010, Jakob Scherer wrote:
> > or did I misunderstand entirely?
> Actually i wanted to ask: is it possible to get a higher performance
> by feature selection?

yes! (given you have irrelevant features as well, or some times just
redundant noisy ones)
for a simple example just run doc/examples/clfs_examples.py and see
effects of feature selections, or just look at Table 1 of our elderly
paper
http://dx.doi.org/10.1007/s12021-008-9041-y

moreover, with improper (e.g. double-dipping) feature-selection you
might get unrealistically good "performance" (once with incorrectly
approached RFE I got 100% generalization across subjects without any
alignment ;))

-- 
=------------------------------------------------------------------=
Keep in touch                                     www.onerussian.com
Yaroslav Halchenko                 www.ohloh.net/accounts/yarikoptic