[ADMB Users] help with glmmADMB 0.6.4 - function maximizer failed

dave fournier davef at otter-rsch.com
Fri Sep 23 09:53:44 PDT 2011

I ran your problem form the command line.  That is always interesting as
you can see exactly where the program fails if it does.  It was hard to make
it fail.  I think there are a few problems with the default values set 
out by
the R script.   The value imaxfn=10 is way too small. 100 or 500 would be
better.  every once and a while it hits a bad spot and needs more evals
to get the gradient small enough so that the newton-raphson step
converges nicely.  If it is wasting too much time in the inner 
optimization one can
tune the convergence criterion  with the -icrit option.   Also the 
-noinit option
should be used. This uses the previously found maximum for the random
effects.  It tends to be faster and more stable (but not always) So 
something like


with  optional  command line arguments

    -noinit -icrit .01 -shess

and it should run fine (well my Linux version does)
If you still have trouble you can find the glmmadmb.exe (windows) and 
the glmmadmb.dat and
glmmadmb.pin files and put them all in one folder for simplicity and type

       glmmadmb.exe -shess -noinit -icrit .01 -imax 500

If that fails save all the results to a file with

   glmmadmb.exe -shess -noinit -icrit .01 -imax 500 > output

(How to get std error as well on windows -- I forget)  and send me output.

More information about the Users mailing list