Fit a frozen monotone quantile calibrator from training data

Estimates per-feature empirical quantile grids on a training cohort and freezes them for deployment. At apply time (apply_frozen_quantile), test-sample feature values are mapped to their empirical quantile under the training distribution and then transformed back to the corresponding training-scale value. Tail values beyond the training range are clamped (intentional winsorisation, documented limitation).

Usage

fit_frozen_quantile(
  x_train,
  n_quantiles = 1000L,
  na_action = c("median", "leave_na")
)

Arguments

x_train

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

n_quantiles

Integer — number of quantile grid levels (resolution). Default 1000.

na_action

"median" (default, replaces NA test values with per-feature training median) or "leave_na".

Value

Object of class frozen_quantile_fit with components: quantile_ref (matrix of quantile values), quantile_levels, feature_names, median_per_feat, n_quantiles, na_action.

References

Bolstad BM, et al. (2003) Bioinformatics 19: 185–193.

Examples

if (FALSE) { # \dontrun{
set.seed(42)
x_train <- matrix(abs(rnorm(50 * 20, 100, 30)), nrow = 50,
                  dimnames = list(NULL, paste0("miR-", seq_len(20))))
fit <- fit_frozen_quantile(x_train, n_quantiles = 100L)
} # }