[Pkg-exppsy-maintainers] Was: ridge... now VALGRIND time

Yaroslav Halchenko debian at onerussian.com
Tue Jan 15 04:10:06 UTC 2008

what numpy version is that?

I wonder what would happen if you valgrind it? I added suppression file
where I enabled all recommended suppressions for non-patched (to do so)

unfortunately on our cluster (amd64 etch) I run into lots of msgs
similar to reported on
but I will backport numpy and install it -- may be it will fix some of

on i386 sid box it is not that bad but also puked a bit. I adjusted
suppressions (they are under tools), so now just use THE TOOL, in a way:
tools/valgrind_python blah.py
where blah.py is your code. Running it on almost (it is still running
since it is slow but so far it is salient) all the tests is salient
(verify that first since I adjusted suppressions only for somewhat
uptodate Debian sid i686 with python2.4).

so if it pukes out smth on your data (but not on tests) -- see
where that happens -- it might provide you with a hint.
Indeed I bet it is not your code, although everything could happen -- I
can write a single line with quite a few subtle bugs ( ;-) )

> I'm actually having a big error with it right now, however, which is
> segfaulting python when trying to free some memory.  Both the unit
> test and the pylab example (modified to use RidgeReg) run without a
> problem, but when I try and use it to train on my real data it
> segfaults upon returning from the self._train call:

> > /usr/lib/python2.4/site-packages/mvpa/clfs/classifier.py(164)train()
> -> result = self._train(dataset)
> (Pdb) n
> *** glibc detected *** python: free(): invalid pointer: 0x914e5008 ***

> I don't see anything wrong in my code, so my current guess is that
> it's an error in numpy.
Yaroslav Halchenko
Research Assistant, Psychology Department, Rutgers-Newark
Student  Ph.D. @ CS Dept. NJIT
Office: (973) 353-5440x263 | FWD: 82823 | Fax: (973) 353-1171
        101 Warren Str, Smith Hall, Rm 4-105, Newark NJ 07102
WWW:     http://www.linkedin.com/in/yarik        

More information about the Pkg-exppsy-maintainers mailing list