mpmd: Multifactor Penalized Matrix Decomposition

In schoonees/spca: Lasso-Penalized Singular Value Decomposition and Sparse Principal Components Analysis

Description

Multifactor version of the rank-1 L1-penalized matrix decomposition in pmd. Components are found by applying pmd to the original matrix after deflation by deducting the components already found.

Usage

mpmd(
  Z,
  k = 2,
  c1 = 1,
  c2 = 1,
  maxit = 20,
  eps = sqrt(.Machine$double.eps),
  centre = FALSE
)

Arguments

Z	Matrix to be decomposed
k	Required rank of the result
c1	L1-norm bound for U (greater than or equal to 1), either length-1, or with k entries (one for each component).
c2	L1-norm bound for V (greater than or equal to 1), either length-1, or with k entries (one for each component)
maxit	Maximum number of iterations
eps	Stopping criterion, an absolute error tolerance on the mean squared reconstruction error
centre	Logical indicating whether to centre the matrix Z using the overall mean before analysis

Value

A list with the penalized singular value decomposition (d, U, V).

Examples

## Simple random test matrix
set.seed(1)
Z <- matrix(rnorm(100), nrow = 20, ncol = 5)

## Ordinary SVD (equivalent up to changes in sign)
mpmd(Z, c1 = 5, c2 = 3, k = 5)
svd(Z)

## Test with constant c1 and c2
mpmd(Z, c1 = 2, c2 = 1.25, k = 5)

## Test with different c1 and c2 for different components
mpmd(Z, c1 = c(3, 1.25), c2 = c(2, 1.25), k = 2)

schoonees/spca documentation built on Jan. 31, 2021, 6:21 p.m.