nethet (version 1.4.2)

het_cv_glasso: Cross-validated glasso on heterogeneous dataset with grouping

Description

Run glasso on a heterogeneous dataset to obtain networks (inverse covariance matrices) of the variables in the dataset for each pre-specified group of samples.

Usage

het_cv_glasso(data, grouping = rep(1, dim(data)[1]), mc.flag = FALSE,
  use.package = "huge", normalise = FALSE, verbose = FALSE, ...)

Arguments

data

The heterogenous network data. Needs to be a num.samples by dim.samples matrix or dataframe.

grouping

The grouping of samples; a vector of length num.samples, with num.groups unique elements.

mc.flag

Whether to use parallel processing via package mclapply to distribute the glasso estimation over different groups.

use.package

'glasso' for glasso package, or 'huge' for huge package (default)

normalise

If TRUE, normalise the columns of the data matrix before running glasso.

verbose

If TRUE, output progress.

...

Further parameters to be passed to screen_cv.lasso.

Value

• Returns a list with named elements 'Sig', 'SigInv', 'Mu', 'Sigma.diag', 'group.names' and 'var.names. The variables Sig and SigInv are arrays of size dim.samples by dim.samples by num.groups, where the first two dimensions contain the (inverse) covariance matrix for the network obtained by running glasso on group k. Variables Mu and Sigma.diag contain the mean and variance of the input data, and group.names and var.names contains the names for the groups and variables in the data (if specified as colnames of the input data matrix).

Details

This function runs the graphical lasso with cross-validation to determine the best parameter lambda for each group of samples. Note that this function defaults to using package huge (rather than package glasso) unless otherwise specified, as it tends to be more numerically stable.

Examples

n = 100
p = 25

# Generate networks with random means and covariances.
sim.result = sim_mix_networks(n, p, n.comp=3)

test.data = sim.result$data
test.labels = sim.result$comp

# Reconstruct networks for each component
networks = het_cv_glasso(data=test.data, grouping=test.labels)