Skip to content
Prev 6317 / 20628 Next

lme4a crash

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1
On 06/29/2011 03:24 AM, Christopher Philipson wrote:
Thanks.

  The core dump is not so useful (although perhaps one of the other lme4
authors will find it so), but bandwidth is cheap.

  Just for general information: we have had trouble, for some time now,
managing to set up the lme4a code (which has lots of shiny Rcpp stuff
under the surface) so that errors emanating from inside C++ code are
caught correctly and passed back to the R "surface" rather than crashing
R.  As in your case, the results seem to be platform-dependent.

   We have some examples to work on (and just lack time, and in my case
knowledge, to spend tracking down and fixing the problems), but another
example would still be welcome.

  Of course, even if the program "only" stops with an error rather than
crashing you will still have a problem (albeit a less annoying one), in
that the optimization heads to a point that produces NaNs.  In the
spirit of working around that problem, here are a couple of thoughts:

  * can you specify "better" starting values, e.g. by starting with the
fixed-effect analogue of the model above?  (Perhaps you are already
doing this.)
  * your model is very complicated.  I hope you have a big data set ...
does it work if you eliminate one or more of the random effects?  Can
you use that approach to sneak up on better starting values?
  * there doesn't seem to be anything in your model that wouldn't work
with nlme (i.e. no crossed random effects -- although if your data set
is big nlmer will probably be faster than nlme).  Have you tried it in
nlme, perhaps with a subset of the data to start?
_______________________________________________
R-sig-mixed-models at r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.10 (GNU/Linux)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org/

iEYEARECAAYFAk4LcxQACgkQc5UpGjwzenNzagCfbye2bUzQdEzJKKZ0xnz9Aj2m
qKUAn2rNSZr6u/oDD34J2o26HbVNydm5
=2+va
-----END PGP SIGNATURE-----