Run MCMC for an Opal object

Sampling uses a fresh RTMB objective. It never changes the stored optimum. init = "auto" uses the fitted point when available and the configured active parameters otherwise. Failed sampler attempts are recorded and warned about, preserving any existing posterior. The default sampler skips internal optimisation; an MLE is optional. The default stan metric supports bounds. Random effects are sampled jointly by default, over their full domain, while fixed effects retain the object's bounds. Change bounds with opal_update(). Use laplace = TRUE and metric = "unit" for marginal sampling.

Usage

opal_mcmc(
  x,
  sampler = "snuts",
  init = "auto",
  check = TRUE,
  check_args = list(),
  ...
)

Arguments

x

A configured or fitted opal_obj.

sampler

"snuts" or a function accepting an obj argument and sampler settings.

init

Initial-value policy or explicit sampler initial values.

check

Run posterior diagnostics.

check_args

Arguments passed to opal_check().

...

Named sampler settings, including seed, chains, cores, num_samples, and num_warmup. The obj, globals, lower, and upper arguments are reserved.

Value

An updated opal_obj, including portable samples and attempt history.

Details

Defaults are four chains, four cores, 1,000 warm-up iterations, and 500 retained iterations per chain. Set seed and choose effort appropriate to the assessment. After sampling, inspect x$validation$mcmc and x$mcmc_history; a failed or unchecked attempt cannot replace a previously checked, passing posterior. opal_as_tmbfit() exposes the selected samples to SparseNUTS plotting and diagnostic tools.

See also

Other assessment workflow: opal_attach_fit(), opal_attach_mcmc(), opal_build(), opal_check(), opal_derived(), opal_example_inputs(), opal_fit(), opal_from_fit(), opal_io, opal_obj(), opal_posterior(), opal_project(), opal_report(), opal_update()

Examples

if (FALSE) { # \dontrun{
assessment <- opal_read("assessment.rds")
assessment <- opal_mcmc(assessment, seed = 42, chains = 4, cores = 4,
                        num_warmup = 1000, num_samples = 500)
summary(assessment)
opal_save(assessment, "assessment-sampled.rds")
} # }