nndsvd: Non-negative double SVD

In zdebruine/LSMF: Least squares matrix factorization

Description

NNDSVD with multiple methods for handling zeros

Usage

nndsvd(A, k, method = "nndsvdar")

Arguments

A	sparse matrix
k	rank
method	One of "nndsvd", "nndsvda", or "nndsvdar" (default). See details.

Details

Several methods are available for handling zeros:

• nndsvd: Leave zero values as zero

• nndsvda: Replace zero values with the mean of u or v

• nndsvdar: Replace zero values with positive random values in the range (0.01 * mean, 1 * mean)

Armadillo SpMat truncated svd is used

Value

list of components:

• W: nndsvd result for feature loadings

• H: nndsvd result for sample factors

• u: initial svd result for U

• v: initial svd result for v

• d: initial svd result for d

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