search

admb-project / admb

AD Model Builder
http://admb-project.org
Other
64 stars 19 forks source link

"-output" switch does not appear to allow the old ADMB runtime output #288

Closed wilbergmj closed 1 year ago

wilbergmj commented 1 year ago

I just updated to ADMB 13.1, and I would like to be able to run models with the old output that shows the gradient component information for all of the parameters. I find this information useful when developing models. Although it's not documented using the help "-?" switch, it looks like the old ADMB output should be produced if a model is run with the "-output 2" switch. However, the model output does not change back to the old default with that switch. Hopefully this description of the issue is clear enough.

Cole-Monnahan-NOAA commented 1 year ago

The argument is "-display 2". It was renamed at some point from output. The help file is correct though:

 -display N       N sets display output with '0' for quiet display, '1' for concise display, or '2' for legacy display (default is 1).

If that doesn't work please reopen this

iantaylor-NOAA commented 1 year ago

Thanks @Cole-Monnahan-NOAA, both -display 0 and -display 2 worked for me just now. Linking original discussion #219 and PR #220.

I'm closing this since that appears to be what you intended.

Cole-Monnahan-NOAA commented 1 year ago

@iantaylor-NOAA yes thanks, I don't actually have repo permissions so I can't close issues

wilbergmj commented 1 year ago

Thanks! That works for me now.

Mike

On Fri, Jun 2, 2023 at 11:57 AM Cole Monnahan @.***> wrote:

@iantaylor-NOAA https://github.com/iantaylor-NOAA yes thanks, I don't actually have repo permissions so I can't close issues

— Reply to this email directly, view it on GitHub https://github.com/admb-project/admb/issues/288#issuecomment-1573962291, or unsubscribe https://github.com/notifications/unsubscribe-auth/AL637ZVTL3V52WPJZKK2PGTXJIEN5ANCNFSM6AAAAAAYYH7V3U . You are receiving this because you authored the thread.Message ID: @.***>

-- Michael Wilberg Professor Chesapeake Biological Laboratory University of Maryland Center for Environmental Science P.O. Box 38 Solomons, Maryland 20688, USA @.*** Office: 410-326-7273 Lab website: wilberglab.cbl.umces.edu Pronouns: he/him/his

Cole-Monnahan-NOAA commented 1 year ago

Out of curiosity what do you find useful about the old output? If you have any suggestions for the new concise output please let us know

wilbergmj commented 1 year ago

I find having the values for all of the gradient components useful when I'm trying to debug a model that is having trouble finding a solution. The main thing I look at is when the likelihood becomes undefined (nan) which parameters appear to be the cause.

Having the magnitude of the gradient components for each variable lets me see which variables are having issues during minimization. It also allows for a quick check if the scaling* of the likelihood function is similar across variables. I know I can also get at this with other parts of the ADMB output, but having it right on the screen is easy.

The other part is just habit - I've been using ADMB for over 20 years now and have gotten used to the OG output.

Thanks for the changes, though. The automatic warning about variables near bounds is really useful. Also, for models that are working well, the new default output is really slick.

Thanks, Mike

*by scaling I mean that the likelihood surface is very responsive to small changes in parameter values for some parameters, but not for others.

On Fri, Jun 2, 2023 at 12:25 PM Cole Monnahan @.***> wrote:

Out of curiosity what do you find useful about the old output? If you have any suggestions for the new concise output please let us know

— Reply to this email directly, view it on GitHub https://github.com/admb-project/admb/issues/288#issuecomment-1574007960, or unsubscribe https://github.com/notifications/unsubscribe-auth/AL637ZWWVWHXIYC6NGNP2ITXJIHXPANCNFSM6AAAAAAYYH7V3U . You are receiving this because you authored the thread.Message ID: @.***>

-- Michael Wilberg Professor Chesapeake Biological Laboratory University of Maryland Center for Environmental Science P.O. Box 38 Solomons, Maryland 20688, USA @.*** Office: 410-326-7273 Lab website: wilberglab.cbl.umces.edu Pronouns: he/him/his

Cole-Monnahan-NOAA commented 1 year ago

Hi Mike yeah that totally makes sense for debugging and why we left it is but not as the default. Glad to hear you find the new stuff useful. Thanks -Cole