save_object() for variational fit loses return code

[ksvanhorn]

Describe the bug
After running a variational estimation, saving the fit with save_object(), and reading it back with readRDS(), return_codes() returns NA instead of the return code from estimation.

To Reproduce

    file <- file.path(cmdstanr::cmdstan_path(), "examples/bernoulli/bernoulli.stan")
    mod <- cmdstanr::cmdstan_model(file)
    stan_data <- list(N = 10, y = c(0,1,0,0,0,0,0,0,0,1))
    fit_vb <- mod$variational(data = stan_data, seed = 123)
    fit_vb$return_codes() # returns 0 as expected
    fit_vb$save_object('~/Tmp/vbexample.rds')
    fit_vb1 <- readRDS('~/Tmp/vbexample.rds')
    fit_vb1$return_codes() # returns NA

Expected behavior
The call fit_vb1$return_codes() should also return 0.

Operating system
Mac OS Ventura 13.4

CmdStanR version number
0.5.3.

Additional context
R version 4.2.2.

[jgabry]

Thanks for reporting this. It would indeed be good to get the right return codes after loading the saved model.

…

On Tue, Jun 27, 2023 at 12:24 PM Kevin S. Van Horn ***@***.***> wrote: *Describe the bug* After running a variational estimation, saving the fit with save_object(), and reading it back with readRDS(), return_codes() returns NA instead of the return code from estimation. *To Reproduce* file <- file.path(cmdstanr::cmdstan_path(), "examples/bernoulli/bernoulli.stan") mod <- cmdstanr::cmdstan_model(file) stan_data <- list(N = 10, y = c(0,1,0,0,0,0,0,0,0,1)) fit_vb <- mod$variational(data = stan_data, seed = 123) fit_vb$return_codes() # returns 0 as expected fit_vb$save_object('/Tmp/vbexample.rds') fit_vb1 <- readRDS('/Tmp/vbexample.rds') fit_vb1$return_codes() # returns NA *Expected behavior* The call fit_vb1$return_codes() should also return 0. *Operating system* Mac OS Ventura 13.4 *CmdStanR version number* 0.5.3. *Additional context* R version 4.2.2. — Reply to this email directly, view it on GitHub <#780>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AB3PQQ4BLZVSTVOCKQOYLGTXNMQMJANCNFSM6AAAAAAZV7IWPU> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[jgabry]

Following up on this, it's an issue not only for variational but also for sampling (and presumably all fitting methods). The problem is that fit$return_codes() calls fit$runset$procs$return_codes(). And that function tries to query the processes:

      for (id in private$proc_ids_) {
        ret <- c(ret, self$get_proc(id)$get_exit_status())
      }

Presumably the solution to this is to store the return codes instead of trying to get the exit status every time the return codes are asked for.

[jgabry]

@ksvanhorn I have a PR open that fixes this: #798

Would you mind trying it out and making sure it resolves the issue you were having?