[R-sig-ME] previous posts about error message in glmmADMB

Ben Bolker bbolker at gmail.com
Mon May 20 20:45:19 CEST 2013


Julia Sommerfeld <Julia.Sommerfeld at ...> writes:

> 
> Dear list,
> 
> I'm trying to fit a glmmadmb, but I'm always getting the following error
> message:
> 
> *mod <- glmmadmb(anzahl_round~wea1*seg + offset(log(flaeche)) + (1|day_ID)
> + (1|seg), zeroInflation=TRUE, family="poisson", data=dat1)  *
> 
> *Parameters were estimated, but not standard errors were not: the most
> likely problem is that the curvature at MLE was zero or negative
> Error in glmmadmb(anzahl_round ~ wea1 * seg + offset(log(flaeche)) + (1 |
> :
>   The function maximizer failed (couldn't find STD file) Troubleshooting
> steps include (1) run with 'save.dir' set and inspect output files; (2)
> change run parameters: see '?admbControl'

 [snip]

> Previous posts suggested to run "..., save.dir="tmp",
> admb.opts=admbControl(run=FALSE), ...".  However, when I run this example
> provided in an older post (see old post below), I get another error
> message 'run=FALSE'
> specified, STD file not found: stopping
> 

  Let me try to explain in a little more detail.

 debug=TRUE is just a generic debugging option; you don't
need it if you don't want to see explicitly where glmmADMB
is looking for the binary files, putting the results, etc.

 setting 'save.dir' *and* setting 'run=FALSE' is intended
for (1) the case where you want to look at the model setup
files yourself, or send them to an AD Model Builder guru,
because glmmADMB is running out of memory or otherwise doing
something funny, or (2) the case where you have already run ADMB 
(and the output files exist in your save-directory) and just
want to read the output files into ADMB.

  It's a bit hard to diagnose precisely what is happening when
the curvature of the MLE is zero/negative (technically, the
second-derivative matrix of the likelihood surface is 
non-positive definite).  Very often the problem is overfitting
(you are trying to fit a model that is more complex than
your data can accommodate), but sometimes there is something
else wrong with the model, and sometimes it's not your fault
at all -- the numerics are just a little too sensitive.  If you
send me the 'dat1' data set I can try to see if there is
something that can be tweaked to make the model succeed
with these data.

  For future reference, it can help to give
summary(dat1) so we can see what your variables are, how
many levels the grouping variables day_ID and seg
have, etc ...



More information about the R-sig-mixed-models mailing list