corrmat_rand: Strength Preserving Correlation Matrix Randomizer

In GatesLab/evalClust: Random Perturbation of Count Matrices

Description

This function implements the Masuda, Kojaku & Sano (2018) "Configuration model for correlation matrices preserving the node strength" algorithm for generating correlation matrices that have the same strength distribution as the original matrix.

Usage

corrmat_rand(org_corr, sampleSize, n, tol = 1e-04, stepSize = 0.001,
  verbose = F)

Arguments

org_corr	A p \times p positive definite covariance (which will be transformed into a correlation matrix) or a correlation matrix
sampleSize	The sample size for a correlation matrix, lower values for more sampling error
n	Number of random correlations matrices to return
tol	Tolerance for configuration model convergence. Default to .0001
stepSize	Step size for configuration model NR solver. Increase to increase convergence speed. Default to .001
verbose	Print convergence information to screen.

Value

A list containing: A p \times p \times n array of rewired correlation matrices and a p \times p matrix of the configuration model.

Examples

org_corr = matrix(0, 20, 20)

org_corr[1:10, 1:10] = .5
org_corr[11:20, 11:20] = .5

noise = t(sapply(as.list(1:100),FUN= function(x){return(rnorm(20,0,1))}))

org_corr = cov2cor(org_corr + cov(noise))

rand_corr = corrmat_rand(org_corr, 10000, 1)[[1]]

org_strength = colSums(org_corr)
rand_strength = colSums(rand_corr)

cor(org_strength, rand_strength)

GatesLab/evalClust documentation built on Nov. 17, 2019, 3:59 a.m.