[ADMB Users] non-convergence and "inner maxg = 0"
mmaunder at iattc.org
Sat Mar 20 08:09:30 PDT 2010
I think Dave probably means his trick where you assume that the random
effect is N(0,1) and when you use it you multiply it by the sd
From: users-bounces at admb-project.org
[mailto:users-bounces at admb-project.org] On Behalf Of Ian Fiske
Sent: Saturday, March 20, 2010 7:53 AM
To: users at admb-project.org
Subject: Re: [ADMB Users] non-convergence and "inner maxg = 0"
but anyway I think the sign is wrong. I think it should be
g = nroute*log_sigma_mu + 0.5*norm2(mu_r)/mfexp(2*log_sigma_mu);
g += nobs*log_sigma_y+0.5*norm2(y - Ey)/mfexp(2*log_sigma_y);
Thanks, Dave! How embarrassing, given the emphasis on "ADMB does
minimization" throughout the docs.
Also it is better to handle the std dev parameterization differently.
Is it better to put std dev inside the norm like the following?
0.5*norm2((y - Ey)/exp(log_sigma_y))
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the Users