screen_cv.glasso: Cross-validated glasso with additional thresholding

In FrankD/nethet: A bioconductor package for high-dimensional exploration of biological network heterogeneity

Description

Cross-validated glasso with additional thresholding

Usage

screen_cv.glasso(x, include.mean = FALSE, folds = min(10, dim(x)[1]),
  length.lambda = 20, lambdamin.ratio = ifelse(ncol(x) > nrow(x), 0.01,
  0.001), penalize.diagonal = FALSE, trunc.method = "linear.growth",
  trunc.k = 5, plot.it = FALSE, se = FALSE, use.package = "huge",
  verbose = FALSE)

Arguments

x	The input data. Needs to be a num.samples by dim.samples matrix.
include.mean	Include mean in likelihood. TRUE / FALSE (default).
folds	Number of folds in the cross-validation (default=10).
length.lambda	Length of lambda path to consider (default=20).
lambdamin.ratio	Ratio lambda.min/lambda.max.
penalize.diagonal	If TRUE apply penalization to diagonal of inverse covariance as well. (default=FALSE)
trunc.method	None / linear.growth (default) / sqrt.growth
trunc.k	truncation constant, number of samples per predictor (default=5)
plot.it	TRUE / FALSE (default)
se	default=FALSE.
use.package	'glasso' or 'huge' (default).
verbose	If TRUE, output la.min, la.max and la.opt (default=FALSE).

Details

Run glasso on a single dataset, using cross-validation to estimate the penalty parameter lambda. Performs additional thresholding (optionally).

Value

Returns a list with named elements 'rho.opt', 'w', 'wi', 'wi.orig', 'mu'. Variable rho.opt is the optimal (scaled) penalization parameter (rho.opt=2*la.opt/n). Variable w is the estimated covariance matrix. The variables wi and wi.orig are matrices of size dim.samples by dim.samples containing the truncated and untruncated inverse covariance matrix. Variable mu is the mean of the input data.

Author(s)

n.stadler

Examples

n=50
p=5
x=matrix(rnorm(n*p),n,p)
wihat=screen_cv.glasso(x,folds=2)$wi

FrankD/nethet documentation built on Oct. 5, 2020, 8:22 a.m.