cv.softSVD: cross-validation for softSVD

In mengchen18/omic3plus: Different methods for the integrative analysis multiple omics data

Description

This function use k-fold cross-valiation method to optimize the sparsity of right singular values

Usage

cv.softSVD(x, nf = 1, kv.opt = c(0.3, 0.5, 0.8), ku.opt = c(0.3, 0.5,
  0.8), wv = 1, wu = 1, pos = FALSE, maxiter = 50, init = c("svd",
  "average")[2], tol = sqrt(.Machine$double.eps), verbose = FALSE,
  ncores = 1, fold = 5, nstart = 1, seed = NULL, loorss = FALSE,
  scan = FALSE, nsd = 2)

Arguments

x	input matrix
nf	number of component
kv.opt	optional value for sparsity on right singular value
ku.opt	optional value for sparsity on left singular value
wv	weight for columns
wu	weight for rows
pos	whether retein non-negative results
tol	convergence tolerance
verbose	if print the progress
ncores	the number of cores used, passed to mclapply
fold	fold number in cross validation
nstart	how many time the k-fold cross validation should be done
seed	set seed
loorss	if the Leave-one-out procedure should be used in matrix reconstruction
scan	if the sum of PRESS should be plotted
nsd	the n*sd for selecting k automatically
maiter	maximum number of iteration

Value

list consist of two matrix. - cvv - The PRESS for right singular vector, mean and sd - cvu - the PRESS for left singular vector, mean and sd - sel.v - the selected v - sel.u - the selected u

Examples

###
seed <- rnorm(8)
noise1 <- matrix(rnorm(8*15, sd = 0.3), 15, 8)
noise2 <- matrix(rnorm(8*20, sd = 0.3), 8, 20)
noise1[1:4, ] <- noise1[1:4, ] + rbind(seed, seed, seed, seed)
noise2[, 1:6] <- noise2[, 1:6] + seed
mat <- noise1 %*% noise2
dim(mat)

cv <- cv.softSVD(x = mat, nf = 1, kv.opt = 1:10, ku.opt = 1:8)
boxplot(cv$cvu$press)
boxplot(cv$cvv$press)

#' cv <- cv.softSVD(x = mat, nf = 1, kv.opt = 1:10, ku.opt = 1:8, scan = TRUE)

mengchen18/omic3plus documentation built on May 6, 2019, 4:59 p.m.