ref_crossfit

Usage

ref_crossfit(
  spec,
  data,
  outcomes,
  folds = 5,
  strata = NULL,
  cluster = NULL,
  id = NULL,
  uncertainty = c("conditional", "total"),
  n_draw = NULL,
  ...
)

Every reference observation receives a score from a model that did not include that observation. A final deployment model is then refit on all reference rows.

Arguments

Argument	Description
spec	A ref_spec.
data	Reference data.
outcomes	Outcome selection, as in ref_fit().
folds	Number of folds (at least 2).
strata	Optional stratification column. Rows with a missing stratum form a stratum of their own.
cluster	Optional cluster / subject column. Entire clusters stay in one fold.
id	Optional identifier stored on scores.
uncertainty	Passed to predict.ref_fit() when the held-out fold is scored, as in ref_assess(). Use "total" to match the predictive that predict.ref_fit() returns by default, which is what a map fitted on these scores by ref_calibrate() will be applied to.
n_draw	Number of coefficient draws when uncertainty = "total". The value is retained as calibration provenance.
...	Passed to ref_fit().

Value

A ref_scores object with .in_sample = FALSE, .row indexing rows of data, a per-row crps column, and a .fold column, plus a deployment fit attribute.