Element-wise Cross-Validation for Signed-Bottleneck NMF-AE

Element-wise k-fold cross-validation for nmfae.signed to select the decoder / encoder ranks \((Q, R)\). Mirrors nmfae.ecv but uses the weighted Signed-Bottleneck NMF-AE fit path (Y1.weights): test-fold elements are zero-weighted during fitting, and held-out MSE is computed on those elements.

Usage

nmfae.signed.ecv(Y1, Y2 = Y1, rank = 1:2, rank.encoder = NULL, ...)

Arguments

Y1

Output matrix (P1 x N).

Y2

Input matrix (P2 x N). Default Y1.

rank

Integer vector of candidate Q values. Default 1:2.

rank.encoder

Integer vector of candidate R values, or NULL (default: pair R = Q, diagonal grid).

...

Additional arguments:

nfolds / div: Number of folds. Default 5.
seed: RNG seed for fold assignment. Default 123.
nstart: Number of random restarts per fit. Default 1. Signed models have more local minima (the bottleneck can carry both signs), so nstart >= 10 is recommended for reproducible rank selection.
Other args: epsilon, maxit, warm.start, etc.\ are passed to nmfae.signed.

Value

An object of class c("nmfae.signed.ecv", "nmfae.ecv") with objfunc (MSE per pair), sigma (RMSE), objfunc.fold (per-fold MSE), folds, QR, paired.

Lifecycle

This function is experimental. The interface may change in future versions.

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.