Shared simulate-and-drop preamble for sbc() and tarp()

Both diagnostics draw "true" parameters from prior, simulate data for them, and drop trials whose simulation failed. The prior-width check lives here rather than in each caller, so sbc() and tarp() get it in one place: prior exists to be overridden (see their docs), and a prior of the wrong width would otherwise reach sweep()/z-scoring downstream and silently recycle against the fit's width.

Usage

sbc_draws(fit, simulator, prior, n, sim_args, what)

Arguments

fit

An nsbi_npe, nsbi_nle or nsbi_nre fit.

simulator

The simulator used for inference.

prior

The prior to draw true parameters from.

n

Number of trials to draw.

sim_args

Named list of extra simulator arguments.

what

Label for drop_failed_sims()'s warning (e.g. "SBC trials").

Value

A list with theta and x (the surviving trials), n (their count, i.e. nrow(theta) after dropping), and n_dropped.