[pymvpa] sensitivity analyzer for regression models?

[Yaroslav Halchenko]

Hi Daniel,

it is indeed bug... probably in our swig wrapping for libsvm - we will
look into it

meanwhile, if you have shogun python-modular installed you could
experiment with C-SVRs with following:

        from mvpa.clfs import sg
		cc = sg.SVM(svm_impl='libsvr', kernel_type='linear', regression=True)


> I'd be happy to switch to the development version if it's reasonably
> stable.

well -- development version is probably "as stable" now but still
more fluid in terms of API, so, although you might prefer to switch to
it to get advantage from new features, for this given problem, there
would be no advantage I guess :-/ sorry

On Mon, 15 Mar 2010, Daniel Kimberg wrote:

> Hi.  I wanted to get a sensitivity map for a regression problem, but I keep getting segfaults.  I've reduced it to a simple case that's repeatable on my machines:

>  from mvpa.suite import *
>  import numpy as n
>  myds=Dataset(samples=n.random.normal(size=(10,5)),labels=N.random.normal(size=10))
>  cc=SVM(svm_impl='NU_SVR',kernel_type='linear',regression=True)
>  ss=cc.getSensitivityAnalyzer()
>  res=ss(myds)

> Am I misunderstanding how this is supposed to work, or otherwise doing something wrong?  (Segfaulting seems like bad behavior in any case.)  If I change it to a classification problem, I do get sensible results.  I'm using 0.4.4 on a 64-bit Fedora 12 system.  I'd be happy to switch to the development version if it's reasonably stable.  Thanks for any help,

> dan

> _______________________________________________
> Pkg-ExpPsy-PyMVPA mailing list
> Pkg-ExpPsy-PyMVPA at lists.alioth.debian.org
> http://lists.alioth.debian.org/mailman/listinfo/pkg-exppsy-pymvpa


-- 
                                  .-.
=------------------------------   /v\  ----------------------------=
Keep in touch                    // \\     (yoh@|www.)onerussian.com
Yaroslav Halchenko              /(   )\               ICQ#: 60653192
                   Linux User    ^^-^^    [175555]