`get_refmodel()` for GAMMs: `object not found` error

[fweber144]

On branch develop (commit 96d8e11), this reprex:

options(mc.cores = parallel::detectCores(logical = FALSE))
nobsv <- 40L
set.seed(1235)
df_gamm <- mgcv::gamSim(eg = 6, n = nobsv, dist = "normal", scale = 0)
df_gamm$wobs_col <- sample(1:4, nobsv, replace = TRUE)
df_gamm$y <- rbinom(nobsv, df_gamm$wobs_col, binomial()$linkinv(df_gamm$y))
fit_gamm_binom <- rstanarm::stan_gamm4(
  cbind(y, wobs_col - y) ~ s(f0) + s(f1) + s(f2),
  random = ~ (f0 | fac),
  family = binomial(),
  data = df_gamm,
  iter = 500,
  seed = 943205
)

library(projpred)
refmod <- get_refmodel(fit_gamm_binom)

throws the following error:

Error in eval(formula[[2]], data, environment(formula)) : 
  object 'wobs_col' not found

[AlejandroCatalina]

Apparently, the internal data is transformed inside of rstanarm::stan_gamm4 and replaces the target and weights column with an unexpected form. This can be fixed if you explicitly pass data = df_gamm to get_refmodel, but it's not ideal. I'll see why the data is changed inside of rstanarm and maybe adapt the column names again in get_refmodel.