kfold(): Arguments (from ...?) not passed to brm()

[fweber144]

This is similar to #1373, but for kfold.brmsfit() instead of update.brmsfit(). So here it seems like kfold.brmsfit() does not pass all arguments from the original brm() call to the K fold-specific brm() calls (I guess arguments from the ellipsis ... are missing). For example: argument adapt_delta in case of the cmdstanr backend:

library(brms)
options(mc.cores = parallel::detectCores(logical = FALSE))
fit_cmdstanr <- brm(rate ~ conc + state,
                    data = Puromycin,
                    refresh = 0,
                    seed = 11403,
                    backend = "cmdstanr",
                    adapt_delta = 0.9)
unique(sapply(fit_cmdstanr$fit@stan_args, "[[", "delta"))
## Gives: 0.9
kfold_cmdstanr <- kfold(fit_cmdstanr, K = 2, save_fits = TRUE, seed = 789)
fit_cmdstanr_fold1 <- kfold_cmdstanr$fits[1, "fit"]$fit
unique(sapply(fit_cmdstanr_fold1$fit@stan_args, "[[", "delta"))
## Gives: 0.8

Specifying adapt_delta in the kfold() call helps (even though the ... documentation at ?kfold.brmsfit only mentions "post-processing functions"):

kfold_cmdstanr_add <- kfold(fit_cmdstanr, K = 2, save_fits = TRUE, seed = 789,
                            adapt_delta = 0.9)
fit_cmdstanr_fold1_add <- kfold_cmdstanr_add$fits[1, "fit"]$fit
unique(sapply(fit_cmdstanr_fold1_add$fit@stan_args, "[[", "delta"))
## Gives: 0.9

Tested with brms v2.17.0.

[paul-buerkner]

Fixed.

[fweber144]

Thanks! I can confirm that the example is working now.