Score specimens with a frozen Nystrom diffusion-map discriminator

Scores each row of X independently with a frozen man_nystrom_model. For each specimen, the scorer computes a per-sample rCLR vector over the model features present in X, derives a diffusion-map geometry from the frozen landmarks restricted to those same present features, embeds that specimen by its own Nystrom affinity row to the frozen landmarks, applies frozen coordinate standardization, and evaluates the frozen linear head.

Usage

score_man_nystrom(model, X, meta = NULL)

Arguments

model: A man_nystrom_model object returned by fit_man_nystrom.
X: Numeric matrix (samples $\times$ features) of non-negative abundance values. Columns must be named feature ids.
meta: Optional per-sample metadata. Accepted for interface uniformity and ignored by this method.

Value

Plain finite numeric vector of length nrow(X). Larger values are more case-like. If fewer than model$hp$min_features model features are present, the documented neutral score 0 is returned for every row.

Details

At inference, no statistic is estimated from the scored rows / test batch: no test-batch centering, renormalization, bandwidth update, graph update, or head refit is performed. When the present feature set is a strict subset of the training universe, the landmark diffusion geometry IS re-derived over the present columns – but solely from the frozen training landmarks, never from any scored row – so row-equivariance is preserved; at full feature overlap this re-derivation reproduces the fit-time geometry exactly. Partial-overlap scoring is therefore an approximate graceful degradation (the scored row's per-sample rCLR is re-centred over the present features while the frozen landmark coordinates are sub-columned, and the re-derived eigenbasis need not align with the fit-time head), not an exact projection. The rCLR representation is exactly invariant to multiplying one specimen by a positive scalar because $\log(c v_j)-mean_k\log(c v_k)=\log(v_j)-mean_k\log(v_k)$ over that specimen's own positive present features.

References

Coifman RR, Lafon S. (2006) Diffusion maps. Applied and Computational Harmonic Analysis 21: 5-30.

Bengio Y, Paiement JF, Vincent P, Delalleau O, Le Roux N, Ouimet M. (2004) Out-of-sample extensions for LLE, Isomap, MDS, Eigenmaps, and Spectral Clustering. Advances in Neural Information Processing Systems 16.

Examples

if (FALSE) { # \dontrun{
set.seed(1)
X <- matrix(stats::rgamma(60 * 30, shape = 3), nrow = 60,
            dimnames = list(NULL, paste0("miR-", seq_len(30))))
y <- rep(c(0, 1), each = 30)
X[y == 1, 1:6] <- X[y == 1, 1:6] + 20
X[y == 0, 7:12] <- X[y == 0, 7:12] + 20
model <- fit_man_nystrom(X, y)
score_man_nystrom(model, X[1, , drop = FALSE])
} # }