[ADMB Users] profiling glitch: starting from MLE gives bad profile output
davef at otter-rsch.com
Fri Dec 7 11:32:41 PST 2012
On 12-12-07 11:28 AM, Ben Bolker wrote:
> On 12-12-07 02:21 PM, dave fournier wrote:
It is one of those trivial cases. If you add say -crit 1.e-7 to the command
line the problem will go away. More serious is why it is there. The
gradient at the first function evaluation is already smaller than the
convergence value so it never gets into the minimization routine so the
variable that hold the best log-likelihood value never gets set. That
should be fixed.
> Does that mean "yes, I reproduced the problem on my system" ... ?
> Looking back at my e-mail, I see that there were two explicit questions:
> * Does anyone have any insight into the following problem?
> * Is there anything I can do to fix it [return status "1" when profiling] ?
> Maybe you were answering one of those questions instead?
More information about the Users