Brms Fitter

Fitter implementation for brms models. Extends the abstract Fitter class with brms-specific configuration properties.

Usage

BrmsFitter(
  name = "brms",
  supports_predictions = TRUE,
  supports_log_lik = TRUE,
  supports_loo = TRUE,
  supports_epred = TRUE,
  backend = "cmdstanr",
  chains = 4L,
  iter = 2000L,
  warmup = 1000L,
  thin = 1L,
  refresh = 0L,
  silent = 2L,
  cores = 1L,
  precompile = TRUE,
  allow_default_priors = FALSE,
  stan_args = list()
)

Arguments

name

Character string identifying the fitter (inherited from Fitter)

supports_predictions

Logical indicating if predictions are supported (inherited)

supports_log_lik

Logical indicating if log-likelihood is supported (inherited)

supports_loo

Logical indicating if LOO-CV is supported (inherited)

supports_epred

Logical indicating if posterior expectation predictions are supported (inherited)

backend

Character string for Stan backend ("cmdstanr" or "rstan")

chains

Integer number of MCMC chains

iter

Integer total iterations per chain

warmup

Integer warmup iterations per chain

thin

Integer thinning interval

refresh

Integer refresh rate for progress output

silent

Integer verbosity level (0, 1, or 2)

cores

Integer number of cores for parallel processing

precompile

Logical; if TRUE (default), the model bank compiles each distinct model spec once via brms::brm(chains = 0) and reuses the prefit across tasks via stats::update(recompile = FALSE). Set to FALSE to fall back to a fresh brms::brm() per task. When precompiling, specify priors explicitly: some brms defaults are derived from the template dataset and would otherwise remain embedded in the reused compiled model.

allow_default_priors

Logical, default FALSE. When FALSE, precompiled model banks reject model specs without an explicit prior with a fatal bayesim_config_error(): brms derives data-dependent default priors from the template dataset, and they stay embedded in the compiled model that the bank reuses for every task (the whole study would silently be fit with the template's priors). Set TRUE to permit brms data-derived default priors to be embedded from the template data (rarely what you want; a notice is emitted once per run). Ignored when precompile is FALSE.

stan_args

Named list of Stan/brms arguments passed through to the fit, e.g. list(adapt_delta = 0.95, max_treedepth = 12, init = 0.1, threads = 2). NULL (default) uses brms/Stan defaults.

Value

An S7 BrmsFitter object