predict.nmfae: Predict method for nmfae objects

In nmfkc: Non-Negative Matrix Factorization with Kernel Covariates

Predict method for nmfae objects

Description

predict.nmfae computes fitted or predicted values from a three-layer NMF model. Without newY2, returns the in-sample fitted values X_1 \Theta X_2 Y_2. With newY2, computes out-of-sample predictions X_1 \Theta X_2 \cdot \mathrm{newY2}.

When type = "class", each column is classified to the row with the maximum predicted value (useful when Y_1 is a one-hot class matrix from nmfkc.class).

If Y1 (actual values) is provided, it is stored as an attribute so that plot.predict.nmfae can produce an observed-vs-predicted scatter plot (for type = "response") or a confusion matrix heatmap (for type = "class").

Usage

## S3 method for class 'nmfae'
predict(object, newY2 = NULL, Y1 = NULL, type = c("response", "class"), ...)

Arguments

object	An object of class "nmfae" returned by nmfae.
newY2	Optional new input matrix (P2 x M) for prediction. If NULL, returns in-sample fitted values.
Y1	Optional actual output matrix for comparison plotting.
type	Character. "response" (default) returns the predicted matrix. "class" returns a factor of predicted class labels (row with max value).
...	Not used.

Value

For type = "response": a matrix of class "predict.nmfae". For type = "class": a factor of class "predict.nmfae" with predicted class labels. If Y1 was provided, actual classes are stored in attr(result, "actual").

See Also

nmfae, plot.predict.nmfae, nmfkc.class

Examples

set.seed(1)
Y <- matrix(runif(20), nrow = 4)
res <- nmf.rrr(Y, rank1 = 2)
pred <- predict(res)

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