[ADMB Users] function maximizer failed (couldn't find STD file)??

[Ben Bolker]

On 12-01-25 05:05 PM, Shawn McCracken wrote:
> Dear glmmADMB users,
> 
> Yesterday I was running my full & reduced models with several different
> distribution families and had no problems except for the warning message
> "Estimated covariance matrix may not be positive definite" followed by a
> string of numbers for a few of the distribution famlies. But when I ran the
> summary() command on the model everything seemed fine. I decided I would
> update to the latest version of glmmADMB to see what happens. Ran the same
> models and the warning was gone. Great! I was on Mac OSX 10.6.8, running R
> v.2.14.0 64-bit, and package glmmADMB v.0.6.5 and updated to glmmADMB
> v.0.7.2.5. I ran the following distributions: poisson, zero-inflated
> poisson, negative binomial, zero-inflated negative binomial, and negative
> binomial type 1. Today I come back to continue working on reduced models
> and I get the following error:
> Error in glmmadmb(count ~ treat + bromleaf + (1 | tree), data = ecpad,  :
>   The function maximizer failed (couldn't find STD file)
> In addition: Warning message:
> running command './glmmadmb -maxfn 500 -maxph 5 -noinit -shess' had status
> 1
> 
> I have now gone through restarting everything, downgrading glmmADMB to
> v.0.7.2.4 then back to v.0.7.2.5, and upgraded R to v.2.14.1 with no luck.
> I looked for glmmADMB version 0.6.5 but couldn't find it. Just some
> preliminary testing with models that ran fine yesterday it appears
> that negative binomial (family="nbinom") and negative binomial type 1
> (family="nbinom1") are the two that now give the above error. It is strange
> that zero-inflated negative binomial works fine. I have tried to follow the
> previous few posts about dealing with this problem but with no luch and its
> a little unclear to me if they were on a Mac or not for some of those
> remedies.
> 
> Really loving all the new functionality of glmmADMB so big thanks to the
> developers. Just bummed its not working for me now, particularly negative
> binomial type 1.
> 
> Where can I get the glmmADMB v. 0.6.5 that was working?

  A couple of thoughts:

 * with a little bit of work I can retrieve the relevant version from
the source code repository, build it, and put it up somewhere.  In the
meantime you could go to
http://www.math.mcmaster.ca/~bolker/R/src/contrib/ to download version
0.7 (then put it somewhere, setwd() to the appropriate directory, and
install.packages("glmmADMB_0.7.tar.gz",repos=NULL,type="source"))
 * a lot of the recent changes in ADMB have to do with tweaks to the TPL
(model definition) file and to ADMB itself that are _supposed_ to make
things more stable in general, but *may* tip a marginally stable model
over the line to unstable -- think of it as a perturbation in a
generally good direction but with some random component ...
 * try running with admb.opts=admbControl(shess=FALSE,noinit=FALSE) and
see if that helps at all ...

 Ben Bolker