Is it possible to parallelize the chains when I run
loo(m, reloo = TRUE)
? I typically only have one bad observation, and my model takes a couple of days to fit with 4 chains on 4 cores. I never waited for loo to finish, but the CPU usage suggests only one core is active during the refitting for reloo. Any hints are appreciated.
P.S.: moment_match doesn’t solve my problem, it tells me to use reloo
Ping @paul.buerkner
Use argument reloo_args = list(cores = 4) in loo() for example.
Thank you for your response!
Am I using the syntax incorrectly here? Using
loo(my_brmsfit, reloo = TRUE, reloo_args = list(cores = 4))
still fits the chains sequentially.
However, when I set (as is helpfully suggested when I directly load the loo-package)
options(mc.cores = 4)
in the script, it works. So I found a solution that works for me.
Yeah, there was a problem but it should be fixed now in the brms github version.