Skip to contents

Batch-compute RATEs for each outcome in a margot_causal_forest result

Source: R/margot_rate.R
margot_rate_batch.Rd

This replaces the legacy internal RATE code and is the only function that actually talks to grf. It flips the CATE vector on the fly when policy is "withhold_best".

Usage

margot_rate_batch(
  model_results,
  policy = c("treat_best", "withhold_best"),
  target = c("AUTOC", "QINI"),
  level = 0.95,
  round_digits = 3,
  model_prefix = "model_",
  verbose = TRUE,
  seed = 12345,
  q = seq(0.1, 1, by = 0.1),
  use_evaluation_subset = TRUE,
  ...
)

Arguments

model_results

List returned by margot_causal_forest(), containing results and full_models.

policy

Character; either "treat_best" (default) or "withhold_best".

target

Character; weighting scheme: "AUTOC" (default) or "QINI".

level

Numeric; Wald confidence level (default 0.95).

round_digits

Integer; decimal places to keep (default 3).

model_prefix

Character; common prefix on model names (default "model_").

verbose

Logical; print progress with cli (default TRUE).

seed

Integer; base seed for reproducible RATE computations (default 12345).

q

Numeric vector of quantiles at which to evaluate. Default is seq(0.1, 1, by = 0.1) which matches the GRF default.

use_evaluation_subset

Logical; if TRUE, use test indices from qini_metadata when available for proper out-of-sample evaluation (default TRUE).

...

Additional arguments passed to compute_rate_on_demand() and ultimately to grf::rank_average_treatment_effect().