Element-wise cross-validation for nmfkc.signed

Element-wise k-fold CV (Wold's CV): held-out elements are masked via Y.weights = 0 during fitting, and the RMSE on those elements is reported. Loops over candidate rank values.

Usage

nmfkc.signed.ecv(Y, A, rank = 1:3, ...)

Arguments

Y: Real-valued \(Q_{\mathrm{obs}} \times N\) response matrix (signed entries allowed).
A: Real-valued \(D \times N\) covariate matrix (signed).
rank: Integer vector of candidate ranks (default 1:3).
...: Passed to nmfkc.signed; also accepts nfolds (default 5; div alias), seed (default 123).

Value

A list with objfunc (MSE per rank), sigma (RMSE), objfunc.fold (per-fold per-rank), folds, Q.grid.

Lifecycle

This function is experimental.

References

Ding, C. H. Q., Li, T., & Jordan, M. I. (2010). Convex and semi-nonnegative matrix factorizations. IEEE Transactions on Pattern Analysis and Machine Intelligence, 32(1), 45–55.