[pymvpa] Seg on OSX when using libsvm

[Michael Hanke]

On Thu, Mar 11, 2010 at 09:04:42AM -0500, Yaroslav Halchenko wrote:
> 
> On Thu, 11 Mar 2010, Fabian Pedregosa wrote:
> > It is true that I achieved better performance, but at the long run
> > I'm a bit concerned on maintainability, i.e. what will happen when
> > libsvm makes a new release.
> So far, we haven't observed any major API changes in libsvm -- they
> seems to simply maintain it -- not much of new development goes into it

Their main problem (IMHO) is that they don't see it as an actual library,
but just a bunch of code. Doing the MILK way, turning the whole thing
into an extension solves some performance problems, but you also inherit
the fun of maintaining it forever. So far nobody was brave enough to do it.

> > Every binding I've seen has this problem, so in the long run I'm
> > pretty sure that either libsvm will have to change their API, either
> > a new svm library with a better API will take up its place.
> do you mean shogun may be? ;)

Here is where our hopes lie. Those guys take care of a large amount of
SVM implementations (and not just SMVs). They know what they are doing
and they keep doing it. With its many language bindings, they make many
people use the same code from different worlds -- hopefully catching
bugs earlier than what would happen with just C or just Python users.
However, if you look at PyMVPA's interface to shogun (all Yarik's work),
you'll see that is sometimes doesn'`t blend nicely with the rest.

For scikits-learn shogun is probably a bit heavy.

Michael

-- 
GPG key:  1024D/3144BE0F Michael Hanke
http://mih.voxindeserto.de