posterior_predict() fails with newdata for models estimated with family = mgcv::betar

Summary:

Similar to #324, I can run posterior_predict and posterior_linpred for betareg models, but as soon as I specify newdata = XX I get an error.

Description:

Running posterior_linpred with newdata (even if the newdata argument is the same as the data used to estimate the model) results in the following error:

Error in (function (optimizer = "nloptwrap", restart_edge = TRUE, boundary.tol = 1e-05, : unused arguments (tolPwrss = 1e-07, compDev = TRUE, nAGQ0initStep = TRUE, checkControl = list("ignore", "ignore", "ignore", "stop", "ignore", "message+drop.cols", "warning", "stop", "ignore"), checkConv = list(list("ignore", 0.001, NULL), list("ignore", 1e-04), list("ignore", 1e-06)))

The workaround mentioned in #324 (adding class) does not help.

Reproducible Steps:

library(rstanarm)
test <- data.frame(y = c(0.1, 0.3), x = c(TRUE, FALSE))

SEED=101
res <- stan_glmer(y ~ (1|x), family=mgcv::betar(link="logit"), data=test, seed = SEED, iter = 200)

class(res) <- c(class(res), "betareg") #Workaround from issue 324
P <- posterior_linpred(res, newdata=test)

RStanARM Version:

2.19.2

R Version:

3.6.0

Operating System:

Windows 10

Start with the posterior_linpred and posterior_predict entry points and reproduce the reported mgcv::betar case using the supplied data and model. Trace why passing newdata produces the unused-arguments error; done means both prediction calls accept newdata for this model without requiring the class workaround.