mse: Mean Squared Error (MSE) of a factorization

In zdebruine/LSMF: Least squares matrix factorization

Description

Mean squared error of factor models "W" and "H" given "A"

Usage

mse(A, W, H, detail = TRUE)

Arguments

A	dgCMatrix of samples (columns) by features (rows)
W	matrix of class "matrix" with factors (columns) by features (rows)
H	matrix of class "matrix" with samples (columns) by factors (rows)
detail	Whether to calculate residual and imputed error for both entire and zero-masked inputs

Value

If detail = FALSE, returns only mean squared error. If detail = TRUE, returns a list of:

• mse: mean squared error, mean((A - W * H)^2)

• err.imp: imputed signal in W and H not in A, given by (A - W * H)_>0

• err.res: residual signal in A not in W and H, given by (A - W * H)_>0

• zm.err.mse: mean squared error for all non-zero values in A

• zm.err.imp: imputed signal for all non-zero values in A, regardless of whether they are non-zero in W * H

• zm.err.res: residual signal for all non-zero values in A, regardless of whether they are non-zero in W * H

Examples

## Not run: 
data(moca7k)
model <- lsmf(moca7k, 5, rel.tol = 1e-2)
err.summary <- mse(moca7k, model$W, model$H)
err.summary$mse == model$mse
# [1] TRUE

## End(Not run)

zdebruine/LSMF documentation built on Jan. 1, 2021, 1:50 p.m.