sdmTMB/RV; comparison: difference in parameters/R pkg version/data set

[landrylys]

https://gccode.ssc-spc.gc.ca/dfo-gulf-science/fishframsci-group/jolene/sdmtmb.git All comparison will be done with the data_test file/mesh_coarse_test and the best models of each species.

• 0 vs1 (no intercept vs intercept)
• year vs fyear,
• poly(var, 2) vs var + I(var^2)
• depth.scaled vs depth

[landrylys]

• creation of a script in the RV_2.X part for each comparison, only best models tested in RV_3.X are included
• LookModel fonction have to be modified to include the possibility to not print in a .pdf the results from it
• fyear variable (as.factor(year)) have been included in the dat and dat.test files.

[landrylys]

Result for intercept:

• no difference between both: intercept vs no intercept see Results of 2023-07-19 (add details where to find the result here)

[landrylys]

Result for factor year:

• It isn't possible to test time = 'fyear', spatiotemporal = "ar1",
• I test instead time = 'fyear', spatiotemporal = "iid" vs time = 'year', spatiotemporal = "iid" and there is no difference between both model see Results of 2023-07-19 for Cod only

[landrylys]

Result for poly vs I (also resolve the question about depth.scaled vs depth, to discuss)

• The results were expected but we have to choose what is the best for us. I propose to follow Isabella Ghement steps. https://stats.stackexchange.com/questions/421988/why-prefer-poly-to-i-are-they-different see Results of this in:
  • log_2023-07-19_cod_models_diag_intercept.txt
  • RV_02.4_test_models_cov_dist.R Idea in message that I sent to Jolene: "The associated figures are also really interesting...I think my diagnostic figures aren't really good/properly selected to diagnose all problems..."

[landrylys]

Jolene want to compare first selected model result for cod and redfish with the result we obtain now and find out why it has a differences in model results between the first and second R scripts run.

• verify again if the same dataset were used.

[landrylys]

compare result estimate of hurdle vs tweedie. in biomass. / comparer directement les estimé des modèles avec la réalité pour connaître la vrai issue du modèle.

[landrylys]

RV_02.9_test_diff_btw_dataset.R answer the question about the difference between datasets. No difference observed between cod and redfish but more observation in the second one.

[landrylys]

Details on poly() vs I() : It stays a difference between models but we can see it only in the .txt files, the plots are similars. We have to find why the plots are similar.

[landrylys]

See the RV_02 scripts and also the folder C:\Users\landryl\Documents\gitlab_gccode\sdmtmb\00_Temporary_Comparisions

• Librairies version/ R and R studio version and interaction between all is important, and create problem over time, it's important to report with which version of the libraries, R and R studio we are working to produce the analysis/models: session_Info()
• To import the RV survey data, it's important to load the lastest version of the Gulf package and the exact same script, but for some reason sometimes, it will remains some difference between two downloaded files from two different computers. So we have download the whole RV survey dataset on 2023-07-06 and will use it until the end of the project. See C:\Users\landryl\Documents\gitlab_gccode\sdmtmb\RV_Input_Files\ReadME_Seriously.txt