mdscheuerell / aukecoho Goto Github PK

Okay @DavidTallmon, here's what you'll need to do to build your forked copy of salmonIPM. In the R project for your version, do the following:

pkgbuild::compile_dll()
roxygen2::roxygenize()
devtools::install(quick = FALSE, upgrade = FALSE)

More info here, but that ought to do it. When you make a change and are ready to rebuild, you will probably find that you need to close and restart any R project that currently has your package loaded. I'm not sure why just detaching, and even removing the package isn't enough, but that's been my experience.

It would be good to have a detailed description of our workflow, which could ultimately be included as supplemental material with a peer-reviewed submission.

Hey guys. Apologies for being AWOL. I'm back to working on this with some updated data and covariates. The data are complete through 2017 but lack age structure information from scale aging for 2018-2019. Not sure if such data are beneficial. The covariates are ones shown to be somewhat informative from recent LME work. These covariates overlap with the last set we examined but have some differences.
covariates_1980-2019.xlsx

auke_coho_data_1980-2019.xlsx

I'm not sure of the best way to set priors. Recall that we have some parameters that are going to be near 0, such as the observation errors for M and S and so will need some help on the prior end of things. Is there a best practice for setting these near 0 in R or need I enter the land of STAN to do this? If so, will this require a re-compile? Please advise.

Hey @DavidTallmon, just a heads-up that I realized there was an issue in the IPM_SMaS_np Stan code in salmonIPM that affects our inference re environmental covariates at the end of yesterday's code-jam session. As part of my efforts to diagnose the parallel-chains error, I had intentionally commented out the lines where the regression predictors enter into smolt production and SAR, and forgot to restore them before we fit the fit_BH_env model. I've done that now, and pushed the "real" model to salmonIPM@ICchinook-models, so the next time you work on the analysis you should unleash these lines to reinstall the package:

#unleash below to update master branch version of IPM
detach(package:salmonIPM, unload = TRUE)
devtools::install_github("ebuhle/salmonIPM@ICchinook-models")
library("salmonIPM")

I'll try to make sure the package repo is fully functional (for everyone but me) from now on, although I can't guarantee that in the immediate future.

FWIW, there's still nothing going on with stream temp or hatchery releases, but there is a 96% probability that PDO has a positive effect on SAR.

As @ebuhle demonstrated on 10/26, shinystan() is a convenient way to examine some of the diagnostics and summaries of model fits. However, we will like want some additional R scripts for customizing our output. Eric and I have several options from other projects that we can draw upon.