Skip to contents

Ranger random-forest backend for xplus

Source: R/backends.R
xplus_backend_ranger.Rd

Random forests via ranger::ranger(). Binary targets use a probability forest and fractional pseudo-labels a regression forest, so soft final-fit targets are supported natively. The backend performs no internal cross-validation; its hyperparameters are fixed at construction. Out-of-bag predictions supply fit_predict, so the stored predictions and classification cutoff use honest scores rather than in-sample memorization.

Usage

xplus_backend_ranger(
  trees = 500,
  mtry = NULL,
  min_n = 1,
  max.depth = 0,
  num.threads = 1,
  seed = NULL,
  importance.mode = c("impurity", "permutation", "none")
)

Arguments

trees

Number of trees (ranger num.trees); the parsnip-harmonized name is used so a parsnip bridge can map rand_forest(trees = ...) later.

mtry

Number of variables sampled per split; NULL uses the ranger default.

min_n

Minimum node size (ranger min.node.size).

max.depth, num.threads

Engine-specific ranger controls forwarded unchanged.

seed

Optional seed passed to each ranger fit. When NULL, one is drawn from R's RNG per fit so that xplus(seed = ...) remains reproducible.

importance.mode

Variable-importance mode ("impurity", "permutation", or "none").

Value

An "xplus_backend" object.

See also

xplus_backend(), xplus()

Examples

if (FALSE) { # \dontrun{
fit <- xplus(x, y, backend = xplus_backend_ranger(trees = 1000))
} # }