Quarto Rendering library(cmdstanr)

Interfaces CmdStan
1

Hi,

I’m new to using cmdstanr. I’m trying to save time when rendering my Quarto file by not having to fit the Stan model every time I render the document. Is there a way to do this?

I get the following error when I use

#| cache: true

or

saveRDS(fit_lin_mod, file = "fit_lin_mod.rds")
fit_lin_mod <- readRDS("fit_lin_mod.rds")

processing file: 4_stan_models.qmd
|… | 57% [unnamed-chunk-6] Error in read_cmdstan_csv():
! Assertion on ‘files’ failed: File does not exist: ‘C:/Users/olivi/AppData/Local/Temp/RtmpoBlpPk/linear_model-202507070840-1-61cef6.csv’.

The #| cache: true works the first time I render the document, but the error appears with the next times I try to render the document. I’m guessing it’s because cache or saveRDS doesn’t save all the data. Is there a way to get around this, I don’t want to have to wait minutes each time I render.

Thanks in advance for your help!

data_list <- list(
  N = nrow(grades),
  x8 = grades$y8,
  x9 = grades$y9,
  y10 = grades$y10
)

lin_mod <- cmdstan_model(stan_file = 'linear_model.stan')

fit_lin_mod <- lin_mod$sample(
  dat = data_list,
  chains = 4,
  parallel_chains = 4,
  iter_warmup = 2000,
  iter_sampling = 2000,
  adapt_delta = 0.99
  )

fit_lin_mod$cmdstan_diagnose()

fit_lin_mod$summary(
  variables = c(
    "mu8", "alpha9", "alpha10",
    "phi8", "phi9", "phi10",
    "beta1", "beta2", "beta3",
    "direct_effect", "indirect_effect", "total_effect")
  ) |>
  kable(digits = 2)
2

Does it work if you specify the output_dir argument when fitting the model? By default cmdstanr writes files to a temporary directory that maybe doesn’t exist the next time you render.

3

I’m not sure what you mean. Could you provide me with an example Quarto chunk based on my code?

Thank you,

Olivier

4

I think Jonah is referring to the argument that you would pass to the cmdstanr::sample function when you call it. By default, the function just places things into a temp directory (this is different than the Quarto cache) that disappears. With the output_dir specified, you can make it persistent, which may fix your issue.

1 Like
5

Yeah what @ssp3nc3r said is what I meant. When you do lin_mod$sample() try adding the argument output_dir and point it to a non-temporary location to store the files written by cmdstanr. Does that solve the problem?

6

Is this what you mean? It still takes a long time to render. Is there an example documented somewhere?

```{r}
#| output: FALSE
#| cache: true

# Create an output directory if it doesn't exist
if (!dir.exists("stan_output")) dir.create("stan_output")

fit_lin_mod <- lin_mod$sample(
  dat = data_list,
  chains = 4,
  parallel_chains = 4,
  iter_warmup = 2000,
  iter_sampling = 2000,
  adapt_delta = 0.99,
  output_dir = "stan_output",         # <- This is the fix
  save_warmup = FALSE                 # optional, to reduce file size
  )
7

Yeah that’s what I meant. So now it still takes a long time but doesn’t error? I haven’t tested this with quarto before, it was just my best guess as to what could help. @ssp3nc3r have you gotten caching to work with quarto?

You mentioned you also got an error using this. I forgot to mention that you should be able to avoid that error by using fit_lin_mod$save_object() instead of calling saveRDS() yourself. This calls saveRDS() for you but only after making sure everything that’s needed to be saved is available in memory:

There’s a section in the vignette that gives more details and suggestions about saving fitted model objects:

And the reasons for all this behavior are explained more here (essentially CmdStanR initially writes posterior draws to CSV and then reads them into memory):

8

Thank you! This is exactly what I was looking for.