kfold_cv: K-fold cross-validation for qbrms models (ordinal and standard families)

Description

Performs K-fold cross-validation either from a fitted model or from formula + data. For ordinal (cumulative/ordinal) families, you can choose the re-fit engine used inside CV: TMB (qbrmO) or a robust fallback using MASS::polr that avoids TMB compilation in each fold. Your original fitted model is unchanged.

Usage

kfold_cv(
  object,
  data = NULL,
  family = gaussian(),
  K = 10,
  folds = NULL,
  seed = NULL,
  stratify = TRUE,
  parallel = FALSE,
  workers = NULL,
  keep_fits = FALSE,
  engine = c("auto", "tmb", "polr"),
  verbose = TRUE,
  ...
)

Value

An object of class qbrms_kfold with ELPD, pointwise elpd, SE, etc.

Arguments

object: Either a fitted qbrms/qbrmO object or a formula.
data: Required only if object is a formula. Ignored if object is a fit.
family: Optional family override (used if object is a formula; fits use their own).
K: Number of folds (default 10).
folds: Optional integer vector of length nrow(data) giving fold IDs.
seed: Optional seed for stratified folds.
stratify: Logical; stratify on response if factor/ordered (default TRUE).
parallel: Logical; use future.apply if available (default FALSE).
workers: Optional workers when parallel and no plan is set.
keep_fits: Logical; keep per-fold fits (default FALSE).
engine: Ordinal CV engine: "auto" (default), "tmb", or "polr". Only used for ordinal families during CV refits. "auto" uses getOption("qbrms.kfold.ordinal_engine", "polr").
verbose: Logical; brief progress (default TRUE).
...: Passed to qbrms() when refitting folds (non-ordinal or engine="tmb").