posteriors.sso different if MCMC output detail 1 or 2

[kellijohnson-NOAA]

I cannot figure out why posteriors.sso would be different for FORECASTS if MCMC output detail is 1 or 2 in starter.ss. When MCMC output detail is 1, the results match those from a previous version of SS3. When I change MCMC output detail to 2 the forecasts are different for some rows but not all. I fixed catches at zero in the forecast file and the problem did not go away. I will say that having zero catches made the SSB values for the second year of the forecast match regardless of what was in starter.ss where they did not before. I am guessing this is a timing issue and I am happy to help in any way that I can but on my own I am not quick enough to diagnose what is going on in the time frame that we need. In the mean time, do you think for Pacific Hake we should run all of our forecasts with MCMC output detail to 1? I have attached two derived_posteriors.sso files where the integer is the integer used for MCMC output detail in the starter file.
Related to #218
derived_posteriors_1.txt
derived_posteriors_2.txt

[Rick-Methot-NOAA]

working on it; mostly this afternoon

[Rick-Methot-NOAA]

think I found it. It is in ss_proced and where get_forecast is called.
Current logic in PROCEDURE_SECTION is:
if(sd_phase() || mceval_phase())
{ get_time_series, Get_Benchmarks }
if(dyn_Bzero) {get_time_series, Get_Forecast} // with F off
get_time_series, Get_Forecast // restore time series and forecast with F on
Process_STDquant
get_posteriors
if(mcmc_output_detail>=2)
{
get_time_series // why???
write_bigoutput
set no_report=1 // , so will not repeat reporting in final_section, but this does not depend on mcmc_output_detail
}
else
{finish procedure and go to final section in globals}

FINAL_SECTION
if(no_report==0) // so is not run if doing mceval
{
if(dyn_Bzero) {get_time_series, Get_Forecast} // with F off
get_time_series
if(did_MSY==0) get_benchmark
Get_Forecast // restore time series and forecast with F on
Process_STDquant
get_posteriors
write_bigoutput
}

[Rick-Methot-NOAA]

@kellijohnson-NOAA
I got identical posteriors.sso files when only changing this line from 1 to 2
1 # MCMC output detail: integer part (0=default; 1=adds obj func components); and decimal part (added to SR_LN(R0) on first call to mcmc)

Please send me the files you used to create the difference

[Rick-Methot-NOAA]

results identical when using the current ss.
posteriors_1.sso.txt
posteriors_2.sso.txt

I will try to figure out what commit after yours could have caused the fix.

[Rick-Methot-NOAA]

latest tweaks to this code are in: fe2c8ac from Jan 7

[Rick-Methot-NOAA]

The hake assessment is using version from Feb 10 a0f982d for which there was a tweak in write_report to resolve report file name problem. There are no changes to ss_proced at that time.

I checked out a0f982d and re-compiled. Re-ran the mceval using option 1 and option 2. Seeing no differences based on character-by-character compare of the files.

I am out of options to explore the difference you are seeing.
I am seeing value in getting the commit code into the versioninfo such that it can be part of the model output.

[kellijohnson-NOAA]

@Rick-Methot-NOAA I just noticed that you are comparing posteriors.sso, where the differences for us were only in derived_posteriors.sso.

[Rick-Methot-NOAA]

problem related to updating of fecundity

with mceval reporting option 2, something is changing about what gets updated in the forecast, so after the first iteration it does not updated bodywts in first year of forecast, so wtatage carried forward from 2021 gets used for 2022.

will fix after dinner.
Got it.
Problem is due to an interaction between the mceval save each report and the new code for streamlining wtatage storage.
When the saving for report is happening, the save_for_report flag gets turned on.
with save_for_report flag turned on, biology processes every year and code in biofxn (lines 1461) was saving the current value of fec() back into Wt_Age_t(t,-2,g)= fec(g).
That must not happen! when empirical wtatage is used. Wt_Age_t is created from the input file back in prelim and needs to be static.
problem rectified by making that assignment in biofxn conditional on WTage_rd==0
examination of code for fishselex verifies that this is not a problem for the storage of wtatage for fleets.

Because the problem only occurs when save_for_report comes on, this issue does not affect the wtatage used during the fitting of the model. save_for_report does is not on during mcmc; only during mceval.

[kellijohnson-NOAA]

Thank you @Rick-Methot-NOAA, I will try it out first thing in the morning.