Apply frozen logistic-normal EB shrinkage to new samples

Shrinks CLR-transformed test samples toward the per-feature training prior using the model fitted by fit_logistic_normal_eb.

Missing test features are handled by missing_features: "error" (default, fail-closed) or "fill_with_prior" (pads missing features with NA in the test matrix, then the posterior collapses fully to the prior mean for those features).

Usage

apply_logistic_normal_eb(
  fit,
  x_test,
  missing_features = c("error", "fill_with_prior")
)

Arguments

fit

A logistic_normal_eb_fit object from fit_logistic_normal_eb.

x_test

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

missing_features

"error" (default) or "fill_with_prior".

Value

Numeric matrix (same row/col structure as ordered training features) of posterior-shrunk CLR coordinates.