[ADMB Users] ADMB NBMM for R: running ADMB in parallel?

[James R. Bence]

Following up on Weihai's response:

I think the problem you are encountering is caused, as you suspect, by 
conflicts in files that are being created as the multiple copies of admb 
program run.  Each time you call glmm.admb it is running the same admb 
executable program in the same directory and thus you have multiple calls 
modifying the same output files such as admb.cov.  When we run admb 
programs in parallel on the HPCC (Michigan State's High Performance Compute 
Center) servers, we either set up multiple copies of our compiled programs 
in different directories or we have our shell script make the copies.

Ideally we would just pass calls to our program an arg that would tell it 
where to save the output files different from where the program starts 
(does anyone know a way to do this?) and then we would not need to set up 
multiple copies of the executable in different places.

You have the added challenge that even if there is any easy way to change 
the location for output files, you need to be able to pass that info 
through your R call to glmm.admb.  I recall that recently there was 
expression that it would be nice to add the capability of passing arbitrary 
args to glmm.admb, so I suspect that capability does not yet exist.

Jim Bence

So At 02:38 PM 7/29/2011, Tara Crewe wrote:
>Content-type: multipart/alternative;
>  boundary="Boundary_(ID_oBtCKwtWoaJicJtof93Qog)"
>Content-language: en
>
>Good afternoon,
>
>I'm running glmmADMB (0.5-2) with negative binomial and zinegbin 
>distribution on count data in R (version 2.12.2).  Currently I'm using a 
>cluster (www.sharcnet.ca) to submit multiple datasets to be analyzed by 
>the same R program in parallel (i.e., serial farming).
>
>Other R packages work fine in parallel this way (e.g., glmer), and the 
>code for ADMB works fine on a single dataset.  However, as soon as I 
>submit a second dataset to be analyzed in parallel, ADMB fails with the 
>following error messages:
>
>  Error in dvar_matrix write
>Error in glmm.admb(ObservationCount ~ c.year + poly(c.day, 2) + 
>poly(c.hour,  :
>   The function maximizer failed
>Calls: glmm.zinb.trnd.admb.yr -> glmm.admb
>In addition: Warning messages:
>1: In system(cmd2, intern = !verbose) :
>   line 1 may be truncated in call to system(, intern = TRUE)
>2: In system(cmd2, intern = !verbose) :
>   line 1 may be truncated in call to system(, intern = TRUE)
>3: In system(cmd2, intern = !verbose) :
>   line 1 may be truncated in call to system(, intern = TRUE)
>4: In system(cmd2, intern = !verbose) :
>   line 1 may be truncated in call to system(, intern = TRUE)
>5: running command './nbmm -maxfn 500 ' had status 1
>Execution halted
>
>I searched these error messages online, and haven't found any 
>documentation.  Question is, is it possible to run ADMB in parallel?  I 
>suspect (with my very limited knowledge) that the error may have to do 
>with how temporary files are written/stored by ADMB??
>
>I haven't attached sample data, because for now I'd simply like to know if 
>parallel processing is possible with ADMB? If so, is there documentation 
>that could help me out?  I am also happy to supply data and R code if it 
>will help.
>
>Thanks so much,
>
>Tara
>
>--
>Tara L. Crewe, PhD Candidate
>
>Department of Biology
>University of Western Ontario
>1151 Richmond Street
>London, Ontario, Canada,
>N6A 3K7
>
>Tel: 519-661-2111 x 84646
>email: tcrewe2 at uwo.ca
>_______________________________________________
>Users mailing list
>Users at admb-project.org
>http://lists.admb-project.org/mailman/listinfo/users

Jim Bence
Dept. of Fisheries and Wildlife
Michigan State University
http://www.msu.edu/user/bence/ 
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.admb-project.org/pipermail/users/attachments/20110729/f2f0aac3/attachment.html>