Score within-cohort ILR balance discriminator — score_ws_balance

Scores each row of X independently with the frozen model learned by fit_ws_balance_ilr. The scorer first restricts the specimen panel to features from the frozen training universe that are present in X. Empty or constant fitted support, absent required features, all-nondetect rows, insufficient canonical SBP features, and rows with no valid balance raise a singlesample_input_error; see score_specimen. Valid balances are summed per row and the frozen orientation sign is applied so higher scores are more case-like.

Usage

score_ws_balance_ilr(model, X, meta = NULL, strict = FALSE)

Arguments

model: A ws_balance_ilr_model object returned by fit_ws_balance_ilr.
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.
strict: Logical scalar; defaults to FALSE, preserving partial balance scoring. If TRUE, every distinct balance member present at fitting must be supplied, or MISSING_REQUIRED_FEATURES is raised. Complement and abundance-ranked balances use the entire frozen training universe as their member set. Other balances count only their named members present at fitting. Missing universe metadata fails closed with MISSING_FITTED_UNIVERSE. Existing input guards apply in both modes.

Value

Finite numeric vector of length nrow(X). Scores use only each row's own values plus the frozen model, with no scored-batch centering, quantiles, or renormalization. The support data-frame attribute has one row per score: n_available counts supplied distinct frozen balance members, n_fitted counts all such members, and full_support reports whether these counts agree. This records named-column availability, not detection, assay quality or calibration. Other training columns can still affect the default specimen pseudocount. Save attr(score, "support") before subsetting or numeric coercion, which can discard attributes.

References

Egozcue J. J., Pawlowsky-Glahn V. (2005) Groups of parts and their balances in compositional data analysis. Mathematical Geology 37(7): 795-828.

Examples

if (FALSE) { # \dontrun{
sbp <- ws_default_sbp()
features <- unique(unlist(lapply(sbp, function(b) {
  c(b$numerator, b$denominator)
}), use.names = FALSE))
features <- setdiff(features, c("ALL_NON_RBC", "ALL_NON_PLT",
                                "TOP_K_BY_ABUNDANCE", "TAIL_BY_ABUNDANCE"))
X <- matrix(stats::rgamma(40 * length(features), shape = 2), nrow = 40,
            dimnames = list(NULL, features))
y <- rep(c(0, 1), each = 20)
model <- fit_ws_balance_ilr(X, y)
score_ws_balance_ilr(model, X[1, , drop = FALSE])
} # }