R help - Oct 2006 - Execution halting of lmer on UNIX when no problem on windows

Dear R-users, 

I have a frustrating problem that I am hoping has a simple fix.  I am running a
series of lmer models from the lme4 package of the general form:

model<-lmer(y~x1 + x2 ..... + xn +
(1|site),data=dataframe,family=poisson,method="Laplace",control=list(usePQL=FALSE,msVerbose=TRUE))

where the same model is executed multiple times on a bootstrapped dataframe. 
For each bootstrapped model run the resulting model object is used to return the
AIC (and models are then compared using a bootstrapped weight - frequency of
runs a given model had the lowest AIC).

This works just fine when running interactively on my windows machine (so there
is nothing the matter with the code, hence I have not bored you with the details
here).... however when I submit the job as a batch to a UNIX system it usually
(but not always) fails and the execution is halted after an error message is
produced:

Error in objective (.par,...): Leading minor of order 1 in downdated X'X is
not positive definite
In addition: There were 12 warnings (use warnings() to see them)
Error in logLik(model) : no applicable method for "logLik"
Execution halted

On windows when the execution of a single model run fails to estimate the logLik
(an unstable model...) R just continues past the error and still runs to the end
of the script (i.e. runs through all the bootstraps), and I can then inspect the
output and any errors at the end.

My question, then is when using UNIX on batch mode, how can I get the job to NOT
halt it's execution on the production of an unstable model (not positive
definite) and continue running? If the models are not bootstrapped then the
script is executed without any problem in UNIX (so the script is successfuly
submitted as a batch job), so it seems that some of the bootstrap runs are
producing unstable models in rare instances, but just one is sufficient to halt
the execution of the script.

I am running R.2.3.1 on both Windows and UNIX, 

Many thanks in advance, 

Toby Gardner 

School of Environmental Sciences
University of East Anglia
Norwich, NR4 7TJ
United Kingdom
Email: t.gardner@uea.ac.uk
Website: www.uea.ac.uk/~e387495

	[[alternative HTML version deleted]]

See

?try
?tryCatch

FAQ 7.32

This is not a Windows/UNIX difference as you suppose, but an 
interactive/batch difference.  You could also look into using 
options(error=).

On Sun, 15 Oct 2006, Toby Gardner wrote:
> Dear R-users,
>
> I have a frustrating problem that I am hoping has a simple fix.  I am 
> running a series of lmer models from the lme4 package of the general 
> form:
>
> model<-lmer(y~x1 + x2 ..... + xn +
(1|site),data=dataframe,family=poisson,method="Laplace",control=list(usePQL=FALSE,msVerbose=TRUE))
>
> where the same model is executed multiple times on a bootstrapped 
> dataframe.  For each bootstrapped model run the resulting model object 
> is used to return the AIC (and models are then compared using a 
> bootstrapped weight - frequency of runs a given model had the lowest 
> AIC).
>
> This works just fine when running interactively on my windows machine 
> (so there is nothing the matter with the code, hence I have not bored 
> you with the details here).... however when I submit the job as a batch 
> to a UNIX system it usually (but not always) fails and the execution is 
> halted after an error message is produced:
>
> Error in objective (.par,...): Leading minor of order 1 in downdated
X'X is not positive definite
> In addition: There were 12 warnings (use warnings() to see them)
> Error in logLik(model) : no applicable method for "logLik"
> Execution halted
>
> On windows when the execution of a single model run fails to estimate 
> the logLik (an unstable model...) R just continues past the error and 
> still runs to the end of the script (i.e. runs through all the 
> bootstraps), and I can then inspect the output and any errors at the 
> end.
>
> My question, then is when using UNIX on batch mode, how can I get the 
> job to NOT halt it's execution on the production of an unstable model 
> (not positive definite) and continue running? If the models are not 
> bootstrapped then the script is executed without any problem in UNIX (so 
> the script is successfuly submitted as a batch job), so it seems that 
> some of the bootstrap runs are producing unstable models in rare 
> instances, but just one is sufficient to halt the execution of the 
> script.
>
> I am running R.2.3.1 on both Windows and UNIX,
>
> Many thanks in advance,
>
> Toby Gardner
>
> School of Environmental Sciences
> University of East Anglia
> Norwich, NR4 7TJ
> United Kingdom
> Email: t.gardner at uea.ac.uk
> Website: www.uea.ac.uk/~e387495
>
> 	[[alternative HTML version deleted]]
>
> ______________________________________________
> R-help at stat.math.ethz.ch mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide
http://www.R-project.org/posting-guide.html
> and provide commented, minimal, self-contained, reproducible code.
>
-- 
Brian D. Ripley,                  ripley at stats.ox.ac.uk
Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
University of Oxford,             Tel:  +44 1865 272861 (self)
1 South Parks Road,                     +44 1865 272866 (PA)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595