107 estimation of bayes factors for models including random effects requires brmssave pars

[GidonFrischkorn]

Summary

• add function save_pars_bmm2brms to the package.
• add test to evaluate that the function works as intended

I wanted to avoid that there is a conflict if users have loaded both brms and bmm and therefore named the function differently. Generally, we might want to clarify that we recommend that users load both brms and bmm so that they can access the post-processing of the fit-objects that is supported by brms.

Tests

[x] Confirm that all tests passed
[x] Confirm that devtools::check() produces no errors

Release notes

[venpopov]

Thanks for doing this. I see a couple of problems around the current implementation. If people didn't use save_pars= argument and attempt bridge sampling they will get the error:

stop2(
  "Bridgesampling failed. Perhaps you did not set ",
  "'save_pars = save_pars(all = TRUE)' when fitting your model? ",)

which then doesn't math the function name in bmm, so they will still have to load brms. We could add internal checks and rewrite this, but why? Also, if save_pars changes arguments in the future, we'll have to change our function too, which we might not knot until someone reports an error.

A much simpler way is like this:

#' @inherit brms::save_pars title params return
#' @description Thin wrapper around [brms::save_pars()]. When calling
#'   [fit_model] with additional information to save parameters you can use this
#'   function to pass information about saving parameter draws to `brms` without
#'   having to load `brms`. Alternatively, you can also load `brms` and call
#'   `save_pars`. For details see ?brms::save_pars.
#' @export
save_pars <- brms::save_pars

I implemented this and pushed the changes

[GidonFrischkorn]

Yes, this looks way nicer. I tried to import the save_pars function, but this only works for internal development not when loading the package regularly. Thanks for adapting this.