R fatal error when calling pairs

[dana-and-monsters]

Summary:

Calling pairs() on two different models has resulted in 2 R fatal errors/R sessions aborted.

Description:

I created the models with brm() in the BRMS package. I received warnings that there were many (~200) divergent transitions with the first model that crashed and just 1 divergent transition the second time it crashed. I have successfully run the pairs() function once but there were no red dots in it which I expected based on my reading of what the pairs() function does. The other time I managed to get a plot, the plot was broken.

Reproducible Steps:

Here is the data I used: SampleData.xlsx

And the code I used for the second time it crashed:

prior<-get_prior(formula = bf(value ~ 1 + allowed_vote + (1|Participant) + (1|variable))+
    lf(disc ~ 0 + allowed_vote, cmc = FALSE),
  data = data,
  family = acat("probit"))

fit_G3_withPriors<-brm(
  formula = bf(value ~ 1 + allowed_vote + (1|Participant) + (1|variable))+
    lf(disc ~ 0 + allowed_vote, cmc = FALSE),
  data = data,
  family = cumulative("probit"),
  save_all_pars = TRUE,
  prior = prior
)

pairs(fit_G3_withPriors)

Current Output:

The message when crashing with a cumulative model with unequal variances:

The plot I got that was clearly broken (from running adjacent category model with category-specific effects):

The plot I got when pairs() actually ran with a cumulative model with equal variances:

Expected Output:

The pairs plot is supposed to show divergent transitions with red points but even when pairs ran, I don't see any red points in the pairs() plot (though I believe in that case there was only one divergent transition).

Versions:

rstan_2.21.2 brms_2.13.5 R version 4.0.2

[dana-and-monsters]

And the call that crashed before, when re-run after restarting R provided this plot:

[andrjohns]

Hi Dana, sorry about the delay in getting to this. I'm able to run your code above on Windows with no crashes. The main difference I see that I'm using the github version of brms. Can you try installing that and trying your code again? Installed via:

if (!requireNamespace("remotes")) {
  install.packages("remotes")
}
remotes::install_github("paul-buerkner/brms")

If you still have crashes, then I'd recommend installing cmdstanR: https://mc-stan.org/cmdstanr/articles/cmdstanr.html, and then running your brms models with the option backend="cmdstanr", since that tends to be more stable.