Score the structurally-informed ILR single-sample scorer (ss-struct-ilr)

Scores each row of X independently with the frozen fit_ss_struct_ilr model. Each specimen is restricted to the frozen training feature universe present in X, ILR balances are computed from the specimen's own values via ws_balance_ilr (the per-sample pseudocount makes balances scale-invariant), standardised with the frozen train mean/sd, and combined with the frozen diagonal-LDA weights. No scored-batch centering, quantiles, or renormalisation is used, so the scorer is row-equivariant (singlesample_assert_row_equivariant). Specimens with no constructible balance receive the neutral score 0.

Usage

score_ss_struct_ilr(model, X, meta = NULL)

Arguments

model

A ss_struct_ilr_model from fit_ss_struct_ilr.

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 case-like.

See also

fit_ss_struct_ilr