NaN values for random effects.

[michael.v.clawson]

Hi all I am using ADMB 12.0. I am running a model with a single annual vector of random effects vector. 

Does anyone know or have a theory as to why the random effects values themselves would be "nan"? 

Thanks,

Mike

[Johnoel]

A nan is generated when the result of the computation is undefined.  For example, a number divided by

zero.  To fix it, you will need to check the tpl file for possible coding errors.  Also, ADMB-12.3 has just 

been released which has many bug fixes and improvements (see download link below).

http://www.admb-project.org/downloads/admb-12.3/

Let me know if you need any help with tracking down the nans.

Johnoel

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

[Michael Clawson]

I declare :

random_effects_vector tt_f(0,nyears_f-1,2);

the values of the tt_f vector are nan and the optimization fails. 

I ended up changing the optimization phases and removing a different parameter (which was poorly estimated) and it worked. I am not totally clear on why that may have solved the problem, except that the poorly estimated parameter was hindering the optimizer maybe? Do RE's need to be in the final optimization phase? 

--

Dr. Michael V. Clawson

Boone and Crockett Fellow, Post-doctoral Researcher

College of Forestry & Conservation, University of Montana.