Most recent commit breaks CmdStanR models with 1 chain

[jgabry]

@paul-buerkner I think this commit 79d4521 is causing problems for CmdStanR models with 1 chain. The reason is that with that commit a list of a single data frame now gets passed to posterior::: as_array_matrix_list because is_matrix_list_like is TRUE. This results in an error because of these lines in as_array_matrix_list:

posterior/R/as_draws_array.R

Lines 232 to 234 in 79d4521

	x <- x[[1]]
	dim(x) <- c(dim(x), 1)
	dimnames(x) <- tmp

If the list has two data frames there's no error, which is why it only affects runs of a single chain.

This can be demonstrated without CmdStanR with a toy example. Here I first check the CRAN version (no error), then the latest posterior on GitHub (error), then install from posterior on GitHub from the commit before the latest commit (no error):

remotes::install_cran("posterior")
test <- list(
  data.frame(x = rnorm(100), y = rnorm(100))
)
posterior::as_draws_array(test) # no error

# need to restart R
remotes::install_github("stan-dev/posterior") # includes latest commit
test <- list(
  data.frame(x = rnorm(100), y = rnorm(100))
)
posterior::as_draws_array(test) # error

# need to restart R
remotes::install_github("stan-dev/posterior", ref = "d0f8b76") # this commit is from the PR before your latest commit
test <- list(
  data.frame(x = rnorm(100), y = rnorm(100))
)
posterior::as_draws_array(test) # no error 

When a CmdStanR model has a single chain this error happens when as_draws_array() is called on the list of a single data frame that is created when reading in the single CSV file.

[jgabry]

I believe this is the source of the error in stan-dev/cmdstanr#1050. It took me a while (longer than it should have!) to notice that the example from @tillahoffmann was using the latest commit of posterior.

[jgabry]

@tillahoffmann also confirmed in stan-dev/cmdstanr#1050 (comment) that he can replicate this posterior behavior, so I'm pretty sure this is the issue causing the errors in CmdStanR.

[jgabry]

Another instance of a user running into this: https://discourse.mc-stan.org/t/error-in-dim-x-c-dim-x-1/37661

[paul-buerkner]

Thank you for finding this error. I am sorry I didn't respond earlier. I didn't get email notifications from github. Do you have a suggestion for how to fix this without having to revert to the commit? I would like to keep this feature if we can make it work safely with cmdStanR

[jgabry]

For now @paul-buerkner and I decided to revert the commit, and we will add this feature back in without breaking CmdStanR. In order to do that it might be sufficient to just avoid applying this to lists of data frames, but will have to double check if that's enough.

[jgabry]

Closing now that #387 is merged