error with already compiled stored models

[bnicenboim]

Summary:

I'm getting an error quite often when the rds of a model is saved.

Description:

This:

N <- 10000
mu <- .002
sigma <- .0004
rt <- 1/rnorm(N, mu, sigma) 
fit_rec <- stan(normal_recrt.stan, data = list(N = N, recRT = 1/rt))

If the model was already ran (and stored), sometimes produces this error.

Error in if (!file.exists(file.rds) || (mtime.rds <- file.info(file.rds)$mtime) <  : 
  missing value where TRUE/FALSE needed

Here is the model and the stored model in rds: normal_recrt.zip

Reproducible Steps:

I cannot really reproduce it, it just happens and then I remove the rds file and continues working. I'm mostly running the models as part of a bookdown, and I get that error a lot!. But once I get it, the stan() command will give me errors until I delete the rds file.

RStan Version:

‘2.26.3’

R Version:

R version 4.1.1 (2021-08-10)

Operating System:

Platform: x86_64-pc-linux-gnu (64-bit) Running under: Ubuntu 20.04.3 LTS

[ShaunMcDonald1021]

I'm getting the same issue on Windows 10 with rstan 2.26.4.

[martinmodrak]

I am getting the same issue on Windows 10, rstan 2.26.6.

The underlying problem seems to be that at https://github.com/stan-dev/rstan/blob/da2fc9c079534a82d3d26adda51ad17bf22f5e2b/rstan/rstan/R/rstan.R#L91 the call to packageDescription("rstan")$Date returns NULL, which then makes the subexpression

(mtime.rds <- file.info(file.rds)$mtime) <
       as.POSIXct(packageDescription("rstan")$Date)

evaluate to logical(0). Since the file exists and is valid, the next subexpression (!is(obj <- readRDS(file.rds), "stanmodel")) evaluates to FALSE and logical(0) || FALSE evaluates to NA.

[martinmodrak]

The problem was probably introduced with https://github.com/stan-dev/rstan/commit/b6360c07af56f2cffa65aed40a2c7de0b3e4d24c - whether or not CRAN sets the Date field on release, the version at https://mc-stan.org/r-packages/ doesn't have it. Indeed building a version of the package with Date added to DESCRIPTION resolves the issue.

[martinmodrak]

So there is a couple possible solutions roughly in order of complexity:

1. Put Date back in DESCRIPTION and keep it manually up-to-date
2. When date is not found in DESCRIPTION, try to use a different source of package date, e.g. file.info(system.file("DESCRIPTION", package = "rstan"))$mtime
3. Modify the deployment process to https://mc-stan.org/r-packages/ so that it automatically fills the Date field in.

Are there opinions on preferred one?

[CLRafaelR]

@martinmodrak

Thank you for pointing the source of the problem! I solved the same issue I had in my local environment, by following your (first? and) second approach: I ran file.info(system.file("DESCRIPTION", package = "rstan"))$mtime and put that output into the Date field of DESCRIPTION file of the rstan package installed on my computer.

However, I think that restoring the Date field (by, for example, reverting https://github.com/stan-dev/rstan/commit/b6360c07af56f2cffa65aed40a2c7de0b3e4d24c ?) would be the best. The reason is that it might be a long time before CRAN accepts the latest version of rstan (due to the difficulty stated here) and automatically sets the release date in the Date field of DESCRIPTION file.

[tomshafer]

Note: I had to do something similar to @CLRafaelR, but I needed to modify Meta/package.rds instead of DESCRIPTION.