Drop simulations whose parameters or output are not finite

A row survives only when every value in both theta and x is finite, and the two are subset together so the training set stays paired. Checking theta matters on the pre-computed path, where a user-supplied NA would otherwise reach chol() in the estimator and surface as a linear-algebra error. Returns the surviving pair and the number dropped; warns once, and errors when nothing survives. See the "Failed simulations" section of nsbi_simulator.

Usage

drop_failed_sims(
  theta,
  x,
  what = "simulations",
  x_hint = "Check the simulator on a single prior draw."
)

Arguments

theta

Parameter matrix, or NULL when the caller has no parameters to keep aligned.

x

Simulated data matrix.

what

Noun for the warning ("simulations", "SBC trials", ...).

x_hint

What to check when the non-finite values are in x. The default points at the simulator, which is only useful advice when one was run; the pre-computed (theta, x) path passes advice about x itself.