rsggm: rsggm (Robust Sparse Gaussian Graphical Modeling via the Gamma-Divergence)

Description

Robust estimation of sparse inverse covariance matrix via the gamma-divergence

Usage

rsggm(x, gamma, lambda, nlambda = 10, delta = 0.2, penalty.offdiag = FALSE, 
        method = "glasso", maxit = 100, tol.plogL = 1e-8, msg = TRUE, 
        Omega.init, mu.init)

Arguments

A data matrix.

gamma

(non-negative) A numeric vector of tuning parameters for the gamma-divergence.

lambda

(Optional) A matrix of tuning parameters for the lasso. Each column corresponds to gamma, so that the number of column must be equal to the length of gamma.

nlambda

The length of lambda if lambda is not specified. Default is 10

delta

The ratio of maximum value of lambda and the minimum value of lambda if lambda is not specified. Default is 0.2.

penalty.offdiag

If FALSE, the diagonal elements of inverse covariance are not penalized. Dafault is FALSE.

method

Estimation method of the graphical lasso. Default is glasso. It can be QUIC.

maxit

The maximum number of iteration.

tol.plogL

Tolerance of the maximum value of penalized likelihood estimation.

msg

If TRUE, a messege is shown.

Omega.init

(Optional) The initial value of the inverse covariance matrix.

mu.init

(Optional) The initial value of the mean vector.

Value

Omega

inverse covariance matrix

mean vector

weight

weight obtained by the gamma-lasso algorithm

nedges

number of edges

lambda

gamma

References

Hirose, K. and Fujisawa, H. (2017). Robust sparse Gaussian graphical modeling, Journal of Multivariate Analysis, 161, 172-190. Fujisawa, H., and Eguchi, S. (2008). Robust parameter estimation with a small bias against heavy contamination, Journal of Multivariate Analysis, 99(9), 2053-2081.

Examples

Run this code

# NOT RUN {
#generate data
set.seed(1234)
x <- rsggm.generator(N=200,p=20)

#fit the robust sparse Gaussian graphical modeling
fit <- rsggm(x,gamma=c(0.05,0.1),delta=0.2)

#output the result for specitic gamma and lambda
out(fit,gamma=0.05,lambda=0.2)
# }