[pymvpa] Recovering original image dimensions after remove_invariant_features

[Yaroslav Halchenko]

On Thu, 23 Oct 2014, Nick Oosterhof wrote:


> On 23 Oct 2014, at 16:27, Yaroslav Halchenko <debian at onerussian.com> wrote:


> > On Tue, 21 Oct 2014, Shane Hoversten wrote:

> >>   Hi Nick -
> >>   Thanks for the reply.
> >>   An update: for the time being I just went back to not using
> >>   remove_invariant_features and ignoring those warning messages.A  However,
> >>   processing one of my subject's data (so far only just one hangs, out of 7
> >>   that finish) hangs PyMVPA s.t. it never resolves.A  My understanding is
> >>   that these warnings I'm getting:
> >>   WARNING: Obtained degenerate data with zero norm for training of
> >>   <LinearCSVMC>.A  Scaling of C cannot be done.

> > sounds more like a degenerate sample somewhere

> Yes, that’s my impression too.

> If we were to add a mapper (StaticFeatureSelection to be precise) to the dataset when removing invariant features, then this would get rid of the degenerate samples *and* cause no issues when mapping the data back to NIFTI.

> The question (for developers): would it be better to add a new function for this, or rewrite the current remove_invariant_features?

imho it shouldn't be necessary since Dataset itself should have done
it when you select features, see mvpa2/datasets/base.py  __getitem__

the question would be what was the original .a.mapper (in addition to
.summary() since the norm issue is probably related ;) )

sorry if I am missing some points

-- 
Yaroslav O. Halchenko, Ph.D.
http://neuro.debian.net http://www.pymvpa.org http://www.fail2ban.org
Research Scientist,            Psychological and Brain Sciences Dept.
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