Construct a prior-predictive data generator

Draws a parameter vector theta from the model prior (via a sample_prior = "only" brmsfit) and simulates data y ~ p(y | theta) using brms::posterior_predict(). Each task uses a distinct prior draw, deterministically indexed by task_ctx$rep_idx, so Simulation-Based Calibration ranks are well-defined and resume is reproducible.

Usage

prior_predictive_generator(
  prior_fit,
  predictor_generator = NULL,
  vars_of_interest = NULL,
  response = NULL
)

Arguments

prior_fit

A brmsfit compiled with sample_prior = "only" (or a formula + family + prior combination to be compiled; see Details). Must contain prior draws.

predictor_generator

Function (data_spec, task_ctx) -> data.frame producing the design matrix of predictors (everything except the response). Must consume the ambient RNG state. If NULL, the prior_fit's own data is reused at its original size.

vars_of_interest

Character vector naming the prior parameters to report as true_params (defaults to all population-level effects "b_<name>", renamed to <name>).

response

Name of the response column (defaults to the LHS of prior_fit's formula).

Value

A generator function (data_spec, task_ctx) -> data_bundle.

Details

The prior model is compiled once (a sample_prior = "only" brmsfit); reuse it across tasks via the model bank or by passing the same object. Predictor covariates not implied by the prior are supplied by predictor_generator.