Apply frozen quantile calibration to new samples

Maps test-sample feature values to the training distribution using the frozen empirical-CDF grid from fit_frozen_quantile. Each test value is located in the training CDF, and the corresponding training reference value is returned. This aligns the test distribution to the training distribution in a monotone, feature-wise manner.

Usage

apply_frozen_quantile(fit, x_test, missing_features = c("error", "skip"))

Arguments

fit

A frozen_quantile_fit object from fit_frozen_quantile.

x_test

Numeric matrix (samples \(\times\) features). All values must be non-negative.

missing_features

"error" (default, fail-closed when training features are absent) or "skip" (process only shared features and return a reduced-column matrix).

Value

Numeric matrix with columns for the shared features, values on the training quantile scale.