Re: [ADMB Users] glmmADMB error message

[Evans, Tracie M.]

Hi,

 

I am trying to run a model using glmmADMB but am getting an error message which I am struggling to understand, can anyone please shed any light?

 

gnb3 <- glmmadmb(Seeds ~ Treatment+(1|Square)+(1|Plant),data=seed.set,zeroInflation=F, family="nbinom")

Parameters were estimated, but standard errors were not: the most likely problem is that the curvature at MLE was zero or negative

Error in glmmadmb(Seeds ~ Treatment + (1 | Square) + (1 | Plant), data = seed.set,  :

  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'

In addition: Warning message:

running command 'C:\Windows\system32\cmd.exe /c "C:/Users/traeva/Documents/R/win-library/3.0/glmmADMB/bin/windows64/glmmadmb.exe" -maxfn 500 -maxph 5 -noinit -shess' had status 1

 

The experiment was a block design with 4 spatially separated blocks, each with 4 populations containing 3 plants. The data is seed counts from two treatments (2 populations per treatment). I am fitting a mixed effects model to account for the variation between blocks and plants and I am using negative binomial error distribution as this has a better fit then poisson. I have checked for zero inflation but models where zero inflation is false have a better fit. The model above does however seem to work when zeroinflation=T.

 

I have tried:

extra.args= "-ndi 60000"

and

 

save.dir="tmp"

and

 

shess=FALSE

 

and have also increased my memory in r but I am still getting the same error message.

 

Thanks for the help,

Best,

Tracie Evans

PhD Candidate,

The university of Edinburgh

s124...@sms.ed.ac.uk

 

  ________________________________  

This message (and any attachments) is for the recipient only. NERC is subject to the Freedom of Information Act 2000 and the contents of this email and any reply you make may be disclosed by NERC unless it is exempt from release under the Act. Any material supplied to NERC may be stored in an electronic records management system.

  ________________________________  

[fromdust...@gmail.com]

I've never used glmmADMB, but if you are running the model with -nohess, it won't estimate standard errors and won't generate the .std file. Try removing the '-nohess' flag.

[Fowler, Mark]

I don’t use glmmadmb either but I use both ADMB and R, so can see glmmadmb is just intercepting ADMB error trapping. I waited on answering, hoping somebody familiar with glmmadmb would respond. I’m stuck with guessing here. Normally I would diagnose the problem via familiar ADMB outputs, track the fitting progress, and output key estimates (but have to modify the tpl for that). In your case it looks the job was submitted with default settings, and the hessian was attempted, but I have no experience with either zero-inflated models or sparse estimation of the hessian (the –shess option). The error message is just the usual consequence of a failure to achieve convergence. When the model fails it (usually) won’t produce the file of standard deviations of estimated parameters. Not a syntax issue or anything wrong with handshaking between R and ADMB, but maybe not the most intuitive error message. My first guess is an inappropriate model, either type or particular specifications. A second guess is too little data per design partitions. I.e. just how many plants per block/treatment, and what sort of variance in seed counts do we see within these partitions. Sorry, best I can do.

--
You received this message because you are subscribed to the Google Groups "ADMB Users" group.
For more options, visit https://groups.google.com/a/admb-project.org/d/optout.

To unsubscribe from this group and stop receiving emails from it, send an email to users+un...@admb-project.org.

[Richard Schuster]

Hi Tracy,

There is a couple things you could try:

Ben Bolker suggests downgrading the glmmADMB version in one case, but that might be outdated:
http://lists.admb-project.org/pipermail/users/2012-January/001667.html

What I usually do is this:

• Make sure my y-variable is not messed up. I have often seen people use a beta-distribution that forgot to scale % values from 0-100 to 0-1, plus due to the logit link neither 0's nor 1's are allowed. But as you have a nbinom you should not have that problem, just wanted to mention it.
• Have you tried poisson, nbinom1 or nbinom2 yet? If so, do you get the same error?
• Now if its nothing with your data or the error distribution, I would go about trying different options:
• admb.opts=admbControl(shess=FALSE,noinit=FALSE)
• admb.opts=admbControl(shess=FALSE,noinit=FALSE, impSamp=200,maxfn=1000,imaxfn=500,maxph=5)
• the above blus starting values:
  start=list(RE_sd=rep(1,2),RE_cor=rep(0,2))
  You could also try with leaving sd out first. Essentially you just need to give one value for each random term in your model
• If everything else fails:
  xtra.args="-ams 500000000 -simplex"
  but be warned, this will take a very long time to run and in many cases does not fit either.

This might have helpful text:
http://glmmadmb.r-forge.r-project.org/glmmADMB.pdf

And if everything else fails, try to get in touch with Ben Bolker (http://ms.mcmaster.ca/~bolker/)

Cheers,
Richard

[xdavef...@gmail.com]

  If you use the save.dir option and send me the zipped up contents of the directory,
  I will take a look and let you know what is going on.

          Dave

[Oscar C]

Hi,

I know this is an old post but I'm wondering if this was ever solved? I have a very similar model structure for seed counts that I'm using in my Masters project and getting the same error. Any help much appreciated.

Cheers,

Oscar

[Ben Bolker]

I have no desire to poach users away from ADMB, but development of
glmmADMB has stopped. Have you tried this with glmmTMB, which is a
nearly drop-in replacement ... ?

gnb3 <- glmmTMB(Seeds ~
Treatment+(1|Square)+(1|Plant),data=seed.set, family = "nbinom2")

A *singular* fit (in this case, most likely an estimate of zero
variance among squares because you don't have very many levels - 4, if
"blocks" are equivalent to "squares" in your description below) is quite
likely to give you a bit of trouble in any case -- e.g. look for the
GLMM FAQ online and read about singular fits there ...

cheers
Ben Bolker

> _  ________________________________ _

> This message (and any attachments) is for the recipient only. NERC
> is subject to the Freedom of Information Act 2000 and the contents
> of this email and any reply you make may be disclosed by NERC unless
> it is exempt from release under the Act. Any material supplied to
> NERC may be stored in an electronic records management system.

> _  ________________________________ _
>
> --

> To unsubscribe from this group and stop receiving emails from it, send
> an email to users+un...@admb-project.org

> <mailto:users+un...@admb-project.org>.

--
Dr. Benjamin Bolker
Professor, Mathematics & Statistics and Biology, McMaster University
Director, School of Computational Science and Engineering
(Acting) Graduate chair, Mathematics & Statistics