search

nmfs-ost / ss3-source-code

The source code for Stock Synthesis (SS3).
https://nmfs-ost.github.io/ss3-website/
Creative Commons Zero v1.0 Universal
37 stars 17 forks source link

add option to adjust recdev settings automatically for retrospectives? #35

Open k-doering-NOAA opened 4 years ago

k-doering-NOAA commented 4 years ago

Imported from redmine, Issue #55783 Opened by @iantaylor-NOAA on 2018-10-02 Status when imported: Under Review

Chantel pointed out in an r4ss issue (https://github.com/r4ss/r4ss/issues/245) that when the age and length comps from teh final years of the model are removed by the retrospective input in the starter file, the best practice would be to shift the "last year of main recr_devs" input back as well, as this break is often dependent on the final year of comp data with information about recruitment. Likewise, the "last_yr_fullbias_adj_in_MPD" and "first_recent_yr_nobias_adj_in_MPD" inputs should likely be adjusted in parallel with the break between main and late recdevs when data are removed from the end of the model. These changes are easy to make manually, but typically aren't done, especially when using the r4ss function SS_doReto which iteratively modifies the starter file and runs the model in a successive set of folders.

It would be possible to implemented this suite of linked changes in r4ss but would likely be more straightforward in SS itself. It might be good to give the user the option to control whether this happens or not, rather than forcing it on them.

k-doering-NOAA commented 4 years ago

comment from @kellijohnson-NOAA on 2020-01-24: The current behaviour of SS with respect to "last year of main recr_devs" is not consistent. Some of which is noted above, but not all. I am running a model where "last year of main recr_devs" is set to be "last year of data - 1". So, for a retro of 1 the "last year of main recr_devs" is not changed, but beyond a retro of 1 the control.ss_new file has the value changed to equal the last year of data for that retro. This led to non-convergence for this particular model. I don't necessarily think that there is a good best practices to implement for SS with respect to automatically changing the year when it is in the future relative to the last year of data, but potentially we could have SS put a warning in the warning.sso file when this happens to alert users what is going on.