scio (version 0.9.0)

sciopath: Compute the SCIO estimates for a grid of penalty values

Description

Estimates a sparse inverse covariance matrix using a Sparse Column-wise Inverse Operator, path-following a grid of values for the regularization parameter

Usage

sciopath(S, lambdalist=NULL, thr = 1e-4, maxit = 1e4, pen.diag=F, sym=T)

Arguments

S

Input covariance matrix of size p by p (symmetric).

lambdalist

Vector of non-negative regularization parameters for the lasso penalty. The path is computed from the largest to the smallest value of this vector. If not given, 10 values are generated.

thr

Threshold for convergence. Iterations stop when the maximum change in two successive updates is less than thr. Default 1e-4.

maxit

Maximum number of iterations for each column computation. Default 10,000.

pen.diag

Whether the diagonal should be penalized. Default False.

sym

Whether the return values should be symmetrized. Default True.

Value

A list with components:

wlist

Estimated covariance matrices, an array of dimension (nrow(s),ncol(n), length(lambdalist))

lambdalist

Regularization parameters used

Details

This is a fast, nonparametric approach to estimate sparse inverse covariance matrices, with possibly really large dimensions. Details of this procedure are described in the reference.

References

Weidong Liu and Xi Luo (2012). Fast and Adaptive Sparse Precision Matrix Estimation in High Dimensions. arXiv:1203.3896.

Examples

# NOT RUN {
set.seed(100)
x <- matrix(rnorm(50*20),ncol=4)
s <- var(x)
a <- sciopath(s)
# }