Fit within-cohort ILR balance discriminator — fit_ws_balance

Wraps ws_balance_ilr as a frozen single-sample discriminator. The raw score is the row sum of valid sequential-binary-partition ILR balances computed from each specimen's own feature values. Fitting learns only a one-bit orientation from the training labels: if the mean raw score is at least as high in cases as controls, the stored sign is +1; otherwise it is -1. No test data or test-time batch statistic is used.

Usage

fit_ws_balance_ilr(X_train, y_train, meta_train = NULL, hp = list())

Arguments

X_train: Numeric matrix (samples \(\times\) features) of non-negative abundance values. Columns must be uniquely named feature ids.
y_train: Integer/numeric 0/1 labels aligned to X_train; 1 is case/disease and 0 is control.
meta_train: Optional per-sample metadata. Accepted for interface uniformity and ignored by this method.
hp: List of frozen hyperparameters: sbp sequential binary partition (default ws_default_sbp()), pseudocount (default NULL, using the primitive's per-sample default), aggregate ("gmean" or "trimmed_gmean", default "gmean"), min_balance_coverage (default 0.8), and min_sbp_features canonical SBP feature floor (default 3L).

Value

A plain list of class ws_balance_ilr_model containing the frozen sbp, pseudocount, aggregate, min_balance_coverage, min_sbp_features, orientation sign, sbp_canonical_features, feature_universe, and resolved hp.

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 <- score_ws_balance_ilr(model, X)
} # }