ref_freeze

Usage

ref_freeze(
  fit,
  criteria = NULL,
  units = NULL,
  missing_policy = "complete-case per outcome; predictors are never imputed"
)

The bundle omits raw training observations. Each fitted model is reduced to what prediction needs (coefficients, their covariance, the smooth constructions, terms, and factor levels): the model frame, fitted values, residuals, weights, and the mgcv family object (rebuilt when predicting) are dropped, so the bundle is a small fraction of the fit’s size. Predictions from the bundle reproduce those of the fit exactly. The bundle keeps what is required for support checks, adaptation, calibration, and provenance. Printing a ref_freeze shows the model card.

Arguments

Argument	Description
fit	A ref_fit.
criteria, units, missing_policy	Optional provenance fields.

Value

An object of class ref_freeze (also a ref_fit).