Score the DA-cVAE single-sample novelty (out-of-distribution) capability

Returns, for each row of X, the mean Euclidean distance in the frozen latent space from its encoded \(\mu\) to its model$novelty_k nearest TRAIN control (reference) latents — a kNN-distance novelty score (design F9: a multi-reference score, not a single-centroid Mahalanobis). Larger = more out-of-distribution relative to the training reference population. This is a SECONDARY capability (it supports the cross-platform / OOD use case) and is NOT the disease score. Floored / degenerate specimens return 0. Pure base-R, row-equivariant, exactly scale-invariant.

Usage

score_dacvae_novelty(model, X, meta = NULL)

Arguments

model

A dacvae_model from fit_dacvae.

X

Numeric matrix (samples x features) of non-negative abundances; named cols.

meta

Optional per-sample metadata; accepted for interface uniformity, ignored.

Value

Finite numeric vector of length nrow(X); higher = more novel.

See also

fit_dacvae, score_dacvae