[pymvpa] RPy and ENET

[Michael Hanke]

On Mon, Jan 11, 2010 at 09:39:31AM -0500, Per B. Sederberg wrote:
> On Mon, Jan 11, 2010 at 9:31 AM, Michael Hanke <michael.hanke at gmail.com> wrote:
> > The problem was RPy1 -- they didn't play nicely together. Once I removed
> > RPy1, things were fine.
> >
> 
> Strange.  I think it's something about 64-bit, too.  Tests for rpy and
> rpy2 dependent classifiers all run without a problem on my laptop,
> which is 32-bit and has both rpy and rpy2, but I just tested on my
> 64-bit desktop and got the same error you did.

I did that on my laptop, which is 32bit as well...
>
> Given that all of the code to test for rpy and rpy2 run from ipython,
> I'm really thinking this is an error in the exists method.  There's a
> part in there that I don't understand about rpy dependency checking,
> so I couldn't do the same for rpy2.  Did you write that?

Nope, Yarik did.

> > That raises the question, what we should do with the RPy1 dependent
> > code? Per, could you port ithe rest to RPy2 as well (possibly removing ENET, if
> > it is fully replaced by GLMNET now)?
> >
> 
> GLMNET actually replaces both ENET and LARS and does it faster than
> both, so I think we should just end up removing both the others.  I'll
> try and write an example to go into the docs that does all three with
> GLMNET.
> 
> If someone can voice a reason to keep LARS and ENET, then I'm happy to
> port them, too.

I vote for removal unless a valid reason pops up.

> > Sidenote: If refactored the glmnet tests a bit and included them into
> > the main battery. That revealed a few problems that I am addressing
> > right now -- will push what I have now nevertheless and push the fixes
> > later on.
> 
> Sounds good.  Let me know when you've pushed the other changes and
> I'll give it a whirl.

Right now test_datameasure and test_clf are failing. I work on the
latter, Yarik on the former.


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