I’m fitting this bernoulli logit model with two varying-effect terms and fairly weak data, drawing 1,000,000 posterior samples in order to eliminate Monte Carlo SE from elpd_loo.
But when trying to calculate the loo object using moment_match = T and reloo = T, I get the error message in the title. What’s going on?
There’s also the following warning:
In addition: Warning message: In parallel::mclapply(mc.cores = cores, X = seq_len(dim(x)[3]), : scheduled cores 17, 18 did not deliver results, all values of the jobs will be affected
Although I don’t know if that’s actually a problem.
Hi, looks like you are using R, first can you tell whether you are using CmdStanR, RStan, or brms, and post the output from sessionInfo()? Can you make reproducible example which you can share?
Thanks for the response. I was ultimately able to calculate the loo object simply by calling loo(mymod), given that there were turned out to be no problematic observations. Perhaps something about “preparing for” moment_matching with such a gigantic posterior matrix was causing the problem.
Since the problem is no longer urgent for me (and messing with such huge model objects is time and memory intensive), maybe we’ll leave the thread open in case the problem reappears at a later time.
1 Like