nmfkc.criterion: Compute model selection criteria for a fitted nmfkc model

In nmfkc: Non-Negative Matrix Factorization with Kernel Covariates

Compute model selection criteria for a fitted nmfkc model

Description

nmfkc.criterion computes the effective rank, clustering-quality measures (silhouette, CPCC, dist.cor), and the clustering-crispness statistic (B.prob.max.mean) from a fitted nmfkc model.

This function can be called on a model that was fitted with detail = "fast" or detail = "minimal" to compute the full set of criteria afterwards.

Usage

nmfkc.criterion(object, Y, detail = c("full", "fast", "minimal"), ...)

Arguments

object	An object of class "nmfkc" returned by nmfkc.
Y	The original observation matrix (P x N) used for fitting.
detail	Character string controlling the level of computation: "full" (default) computes all criteria including silhouette, CPCC and dist.cor; "fast" skips the expensive distance-based criteria; "minimal" skips the fit and clustering statistics.
...	Additional arguments: Y.weights (non-negative weight matrix; lm()-style loss \sum W \, r^2; default: all ones). See nmfkc for full details.

Value

A list with components:

r.squared

R-squared between Y and XB.

sigma

Residual standard deviation.

mae

Mean absolute error.

B.prob

Column-normalized coefficient matrix (soft-clustering probabilities).

B.cluster

Hard clustering labels (argmax of B.prob per column).

X.prob

Row-normalized basis matrix.

X.cluster

Hard clustering labels per row of X.

criterion

Named list: B.prob.max.mean, effective.rank, silhouette, CPCC, dist.cor.

See Also

nmfkc, nmfkc.rank

Examples

Y <- t(iris[, -5])
res <- nmfkc(Y, rank = 3, detail = "fast")
crit <- nmfkc.criterion(res, Y)
crit$criterion$silhouette

nmfkc documentation built on July 14, 2026, 1:07 a.m.