Skip to contents

Concentration-index gain for a prediction partition

Source: R/ci_parsnip_formula.R
ci_gain_vec.Rd

Computes the concentration-index impurity reduction from a root partition to groups defined by terminal-node ids or another supplied partition column. This is the yardstick-compatible counterpart to ci_tree_validation_gain().

Usage

ci_gain_vec(
  truth,
  estimate,
  rank,
  node = NULL,
  case_weights = NULL,
  type = "CIg",
  root_impurity = NULL,
  na_rm = TRUE,
  ...
)

Arguments

truth

Numeric or logical observed outcomes.

estimate

Numeric predictions. Used as the partition only when node is NULL.

rank

Socioeconomic rank values.

node

Optional terminal-node or partition ids. True CI gain requires a partition; using predictions as a fallback is only an approximation.

case_weights

Optional non-negative case weights.

type

One of "CI", "CIg", "CIc", or "L".

root_impurity

Optional pre-computed root impurity.

na_rm

Should incomplete rows be removed?

...

Not used.

Value

A numeric concentration-index gain.