Meaning of -noest

[Bill Clark]

When I run a model with the -noest option, the parameter values that are written in the REPORT_SECTION are slightly different from the values in the .pin file. I'd like to get a report of all the values calculated in the PROCEDURE_SECTION using the exact initial parameter values from the .pin file. Is that possible?

[Ian Taylor - NOAA Federal]

I use
-maxfn 0 -phase 10 -nohess
Which I learned from Jim Ianelli. I wish I knew how and why this differs from -nohess.

On Tuesday, February 25, 2020, Bill Clark <old.bil...@gmail.com> wrote:
> When I run a model with the -noest option, the parameter values that are written in the REPORT_SECTION are slightly different from the values in the .pin file. I'd like to get a report of all the values calculated in the PROCEDURE_SECTION using the exact initial parameter values from the .pin file. Is that possible?
>

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

[Richard Methot - NOAA Federal]

-noest bypasses the PROCEDURE section, so some quantities needed for reporting are bypassed.  Our team had a substantial deliberation on this topic in issue #70048

--

Richard D. Methot Jr. Ph.D.
NOAA Fisheries Senior Scientist for Stock Assessments

Mobile: 301-787-0241

[Bill Clark]

Thanks to Ian, Rick and Jim for helping out. I am using Ian's fix happily although I'm a bit puzzled that the function gets evaluated with zero function evaluations. As to the small differences between the REPORTED initial parameter values and the values in the .pin file, they arose because -noest bypasses the REPORT_SECTION too, so I was actually looking at .rep files that had been written when I was fitting the model using true initial parameter values. I.e., the .rep files were not written by the -noest runs. The slight differences (and slight lack of fit) resulted entirely from logging and antilogging the (true) initial values.

[Jim Ianelli - NOAA Federal]

I think the subtlety is -maxfn 0 refers to how many calls to the function to do from the optimizer, but still "initializes" the model by calling the function once. so -noest is an odd argument since it omits everything in the procedure section. I suppose it could be useful for some simulation studies where the functions are called outside of the procedure section...it's description from -help could be improved!

On Wed, Feb 26, 2020 at 4:29 PM Bill Clark <old.bil...@gmail.com> wrote:

Thanks to Ian, Rick and Jim for helping out. I am using Ian's fix happily although I'm a bit puzzled that the function gets evaluated with zero function evaluations. As to the small differences between the REPORTED initial parameter values and the values in the .pin file, they arose because -noest bypasses the REPORT_SECTION too, so I was actually looking at .rep files that had been written when I was fitting the model using true initial parameter values. I.e., the .rep files were not written by the -noest runs. The slight differences (and slight lack of fit) resulted entirely from logging and antilogging the (true) initial values.

--

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

--

James Ianelli
NMFS/NOAA Building 4
7600 Sand Pt Way NE
Seattle WA 98115

 206 526 6510

Visit the ADMB project http://admb-project.org/

Also see Alaska groundfish stock assessments

[Dave Fournier]

Actually the idea was that the function got called one more time after all the calls to the function minimizer.

That way one could ensure that the parameters corresponding to the best fit got used.

Presumably one would like that to happen even for the -noest option but that part of the code

could be getting bypassed with that option.  Another case where two features combine to produce

a "bug".