soft_thresh: Soft Thresholding Function for Signed Correlation Networks

In abnormally-distributed/rsfcNet: Analyze resting state functional connectivity networks

Description

This function implements the soft thresholding function from WGCNA. This is used to downweight negative edges while at the same time shifting the matrix entries to be >= 0. This is accomplished by adding 1 to the matrix entries, then dividing by two. Then, the resulting values are raised to a power, denoted as lambda. Recommended values for lambda are 4 to 8 for partial correlation networks, and 8 to 12 for correlation networks. Larger values result in stronger thresholding. The formula used is given below.

w_{ij}^{\text {signed }}=≤ft[\frac{ρ≤ft(i,j\right)+1}{2}\right]^{λ}

Usage

soft_thresh(x, lambda = 8, zero.diag = T)

Arguments

x	a correlation or partial correlation matrix
lambda	the power to which the matrix will be raised. defaults to 8. setting lambda to "auto" will automatically choose a value based on the operator norm of the transformed matrix. A small value is added to lambda at each iteration, and the algorithm stops when the change in operator norm is <= than 0.01.
zero.diag	if TRUE (the default) the diagonal of the returned matrix will be set to zero. otherwise, it will be set to 1.

Value

a matrix

abnormally-distributed/rsfcNet documentation built on March 8, 2020, 5:32 p.m.