Multiscale GLasso • mglasso

Inference of Multiscale Gaussian Graphical Models.

Installation

remotes::install_github("desanou/mglasso")
library(mglasso)

library(mglasso)
install_conesta()

Basic Usage

Simulate some block diagonal model

library(Matrix)
n = 50
K = 3
p = 9
rho = 0.85
blocs <- list()

for (j in 1:K) {
  bloc <- matrix(rho, nrow = p/K, ncol = p/K)
  for(i in 1:(p/K)) { bloc[i,i] <- 1 }
  blocs[[j]] <- bloc
}

mat.covariance <- Matrix::bdiag(blocs)
Matrix::image(mat.covariance)

1.1 True cluster partition

rep(1:3, each = 3)

1.2. Simulate gaussian data from the covariance matrix

set.seed(11)
X <- mvtnorm::rmvnorm(n, mean = rep(0,p), sigma = as.matrix(mat.covariance))

Launch algorithm

X <- scale(X)    
res <- mglasso(X, lambda1 = 0.1, lambda2_start = 0.1, fuse_thresh = 1e-3)

Plot results compact version

3.1 Estimated regression vectors

plot_mglasso(res)

level9 denotes a partition with 9 clusters. We observe a shrinkage effect in the estimated coefficients due to the fuse-group lasso penalty parameter.

3.2 Estimated clustering partitions

res$out$level9$clusters
res$out$level7$clusters
res$out$level4$clusters
res$out$level3$clusters
res$out$level1$clusters

The uncovered partition obtained while increasing \(\lambda_2\) is a hierarchical partition under some constraints.

Reference

Edmond, Sanou; Christophe, Ambroise; Geneviève, Robin; (2022): Inference of Multiscale Gaussian Graphical Model. ArXiv. Preprint. https://doi.org/10.48550/arXiv.2202.05775