[pymvpa] Never-ending searchlight

[Yaroslav Halchenko]

On Fri, 30 Jan 2015, Christopher J Markiewicz wrote:

> Naturally, the single-threaded attempt also succeeded, so I have no
> useful debugging information.

> My guess would have to be that there's a pathological case lurking in
> the balancing, where, for some subset of balanced training sets, on a
> given searchlight (probably high voxel count), the SVM fails to
> converge. 

aha -- so there is a random factor in all this kitchen (Balancer) so
reproducibility is not trivial.. gotcha

then if you still could reproduce parallel one, I would recommend to do
something like  running that script with

  MVPA_DEBUG_METRICS=pid,asctime MVPA_DEBUG=SLC.*,SVM.*,CLF.*

while dumping all output to some file

  > /tmp/debug.log 2>&1

when it gets stuck -- note the PID of the stuck process, so we could grep
for it and see if anything useful for us gets reported.

then -- additionally... if you are using Debian-based system (with python DEBUG
builds e.g. python-numpy-dbg),  hen we could easily get right at the
heart of the stall and still be able to see what is going on... just let me
know what you are using first

> Since we did two balanced partitions per attempt, what would
> be required was for both partitions fail to fall in the pathological
> case for that searchlight.

> (This is not to say that the balancer is failing but that the balancer
> is why it sometimes fails and sometimes succeeds.)
-- 
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