Stanfit object size too big for memory

General
1

Hello everyone,

I have a model that has 3 individual-level parameters for 60,000 observations (~200,000 parameters in total for the full model). The resulting stanfit object ends up being 7gb (warmup = 1000, iter =1500, chains = 4). I then save the object with saveRDS(). When I reload it, however, I get a memory allocation error and R crashes. Is there any way I can reduce the size of the object so I don’t run into this problem? I’ve already prevented all of the parameters that aren’t of interest from being saved in the object (include = FALSE, pars = c(…)), but I’m out of ideas otherwise.

Thanks!

2

Just this morning I shared some code on github to strip out extra parameters from a stanfit object.

Maybe you can use that to split it up into two or three stanfit objects?

3

Thanks for this Aaron.

I’ve already stripped on the extra parameters using the arguments include = FALSE and pars = c(…) when calling sampling(). I wonder if there’s a way to remove the warmup iterations?

4

save_warmup = FALSE

5

Wow. That’s easy. When I looked through the help file for sampling(), it didn’t show a save_warmup argument.

I see now that it’s included in the R documentation for stan(). I’ve been using the stan_model() to sampling() route. Thanks a lot Ben. This is perfect.