Fit the CoDaCoRe stagewise log-ratio-balance discriminator — fit_coda

Trains the CoDaCoRe continuous relaxation stagewise (boosting) via reticulate-python torch (CPU by default for exact reproducibility), DISCRETIZES each balance to two disjoint feature sets, FREEZES them, fits a base-R IRLS ridge logistic head over the discrete balances, EXPORTS the discrete sets + head to R, and DISCARDS the python module. The fitted model holds no external pointer.

Usage

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

Arguments

X_train: Numeric matrix (samples \(\times\) features) of non-negative abundances with unique, non-empty feature names (colnames).
y_train: Numeric / integer 0/1 labels (1 = case), length nrow(X_train), with at least one case and one control.
meta_train: Optional per-sample metadata. Accepted for interface uniformity and otherwise ignored.
hp: Optional list of hyperparameters. Allowed fields: max_balances (max stagewise balances, positive integer; default 3L), epochs (Adam epochs per stage, positive integer; default 200L), lr (positive Adam learning rate; default 1e-2), tau (positive softmax relaxation temperature; default 1.0), ridge (non-negative L2 ridge on the logistic head; default 1e-4), min_features (feature-overlap floor at scoring, positive integer; default 3L), device ("cpu" (default), "cuda", or "auto"; "cuda"/"auto" fall back to CPU with no GPU – but note the default "cpu" is what guarantees an EXACTLY reproducible discretization), and seed (integer; default 42L).

Value

Object of class coda_codacore_model: a list with feature_universe, balances (a list of list(A, B) 1-based universe-position index sets per frozen discrete balance), intercept, weights (the frozen logistic head over the balances), n_balances, device (resolved), seed, and hp. No python pointer.

References

Gordon-Rodriguez E, Quinn TP, Cunningham JP. (2021) Learning sparse log-ratios for high-throughput sequencing data. Bioinformatics 38(1):157-163.

Examples

if (FALSE) { # \dontrun{
set.seed(1)
n <- 120; p <- 16; k <- 3
X <- matrix(stats::rgamma(n * p, shape = 30, rate = 2), nrow = n,
            dimnames = list(NULL, paste0("f", sprintf("%02d", seq_len(p)))))
y <- rep(c(0, 1), each = n / 2)
X[y == 1, 1:k] <- X[y == 1, 1:k] * 1.8
X[y == 1, (p - k + 1):p] <- X[y == 1, (p - k + 1):p] / 1.8
model <- fit_coda_codacore(X, y)
score_coda_codacore(model, X[1, , drop = FALSE])
} # }