Store the information of how CmdStan was called

[avehtari]

Sometimes for debugging purposes or in order to repeat the same call from commandline, it would be useful if CmdStanFit object would include the actual options used to call CmdStan

[jgabry]

By 'options used to call CmdStan' do you mean something different than what is returned as part of fit$metadata()?

Edit: here's an example of what is currently there:

fit <- cmdstanr_example()
str(fit$metadata())

List of 40
 $ stan_version_major  : num 2
 $ stan_version_minor  : num 33
 $ stan_version_patch  : num 0
 $ start_datetime      : chr "2024-01-23 18:34:47 UTC"
 $ method              : chr "sample"
 $ save_warmup         : num 0
 $ thin                : num 1
 $ gamma               : num 0.05
 $ kappa               : num 0.75
 $ t0                  : num 10
 $ init_buffer         : num 75
 $ term_buffer         : num 50
 $ window              : num 25
 $ algorithm           : chr "hmc"
 $ engine              : chr "nuts"
 $ metric              : chr "diag_e"
 $ stepsize_jitter     : num 0
 $ num_chains          : num 1
 $ id                  : num [1:4] 1 2 3 4
 $ init                : num [1:4] 2 2 2 2
 $ seed                : num 8.02e+08
 $ refresh             : num 100
 $ sig_figs            : num -1
 $ profile_file        : chr "/var/folders/s0/zfzm55px2nd2v__zlw5xfj2h0000gn/T/RtmpNmr6Mp/logistic-profile-202401231134-1-0be96e.csv"
 $ stanc_version       : chr "stanc3 v2.33.1"
 $ sampler_diagnostics : chr [1:6] "accept_stat__" "stepsize__" "treedepth__" "n_leapfrog__" ...
 $ variables           : chr [1:105] "lp__" "alpha" "beta[1]" "beta[2]" ...
 $ step_size_adaptation: num [1:4] 0.808 0.689 0.743 0.719
 $ model_name          : chr "logistic_model"
 $ adapt_engaged       : num 1
 $ adapt_delta         : num 0.8
 $ max_treedepth       : num 10
 $ step_size           : num [1:4] 1 1 1 1
 $ iter_warmup         : num 1000
 $ iter_sampling       : num 1000
 $ threads_per_chain   : num 1
 $ time                :'data.frame':   4 obs. of  4 variables:
  ..$ chain_id: num [1:4] 1 2 3 4
  ..$ warmup  : num [1:4] 0.023 0.023 0.024 0.021
  ..$ sampling: num [1:4] 0.07 0.075 0.074 0.072
  ..$ total   : num [1:4] 0.093 0.098 0.098 0.093
 $ stan_variable_sizes :List of 4
  ..$ lp__   : num 1
  ..$ alpha  : num 1
  ..$ beta   : num 3
  ..$ log_lik: num 100
 $ stan_variables      : chr [1:4] "lp__" "alpha" "beta" "log_lik"
 $ model_params        : chr [1:105] "lp__" "alpha" "beta[1]" "beta[2]" ...

[avehtari]

Given that metadata, how do I call cmdstan with those options in command line?

[jgabry]

Oh I see, yeah there's no function that would take that info and compose the call to cmdstan.

What about setting options("cmdstanr_verbose"=TRUE), which results in printing the call to cmdstanr when you're running a model. See https://mc-stan.org/cmdstanr/articles/cmdstanr-internals.html#troubleshooting-and-debugging. If you scroll down a bit in that section past the compilation output, the call to cmdstan is printed and you could copy and paste it. Is that sufficient?

Running ./bernoulli 'id=1' random 'seed=90958316' data \
  'file=/var/folders/s0/zfzm55px2nd2v__zlw5xfj2h0000gn/T/RtmpXOnqh8/standata-742874baeb02.json' \
  output \
  'file=/var/folders/s0/zfzm55px2nd2v__zlw5xfj2h0000gn/T/RtmpXOnqh8/bernoulli-202312131009-1-4db120.csv' \
  'profile_file=/var/folders/s0/zfzm55px2nd2v__zlw5xfj2h0000gn/T/RtmpXOnqh8/bernoulli-profile-202312131009-1-1c1f83.csv' \
  'method=sample' 'num_samples=100' 'num_warmup=100' 'save_warmup=0' \
  'algorithm=hmc' 'engine=nuts' adapt 'engaged=1'

[avehtari]

It would be nice to have cmdstanr_verbose=TRUE as documented argument for $sampler(), $pathfinder(), etc

It could be useful to include that call to cmdstan to the returned object, too