What is GGMselect?

GGMselect is a R package dedicated to graph estimation in Gaussian Graphical Models. The main functions return the adjacency matrix of an undirected graph estimated from a data matrix.

The function selectFast selects a graph within the (data-driven) families of graphs EW, C01, and LA. The function selectQE selects a graph within the family of graphs QE and the function selectMyFam within a given family of graphs.
Other functions:
- simulateGraph generates a sparse (non-uniform) Gaussian Graphical Model; The covariance matrix can be used to generate input for the main functions.
- penalty, the penalty function.
- convertGraph converts graphs into adjacency matrices.

This package is developed in the MaIAGE Lab. of INRA - Jouy-en-Josas, France.

Authors

Bouvier A. (MIA, INRA, Jouy-en-Josas, France), Giraud C. (Ecole Polytechnique, CMAP, UMR 7641, Palaiseau, France), Huet S. (MIA, INRA, Jouy-en-Josas, France), Verzelen N. (Université Paris Sud, UMR 8628, Orsay, France).

References

C. Giraud, S. Huet, and N. Verzelen. Graph selection with GGMselect. 2009. http://fr.arxiv.org/abs/0907.0619

Example

# ----------------------------------------
# Random graph generator: use of simulateGraph
# ----------------------------------------
p=30
n=30
eta=0.11
Gr <- simulateGraph(p,eta)
# simulate data
X <- rmvnorm(n, mean=rep(0,p), sigma=Gr$C)
# ----------------------------------------
# Graph selection with family QE:  use of selectQE
# ----------------------------------------
GQE <- selectQE(X)
# ----------------------------------------
# Plot the result with the help of the package network
# ----------------------------------------
library(network)
gV <- network(Gr$G)
g <- network(GQE$G)
jpeg("example.jpg", width=680, height=580)
par(mfrow=c(1,2), pty = "s", mai=c(par()$mai[1:2], 0, par()$mai[4]))
a <- plot(gV, usearrows = FALSE)
title(sub="Simulated graph")
plot(g,coord=a, usearrows = FALSE)
title(sub="Graph selected with QE family")
dev.off()

Licence
GGMselect is an open source, freely available software released under the GPL-3 Licence.

Requirements
R system, -- including C and Fortran compilers --, and the R-packages: mvtnorm, lars, mvbutils (>= 2.0), gtools.

Documentation
An introduction to GGMselect [pdf]

Download
Download the last version (0.1-12) here or from the CRAN site.

Contact
Sylvie.Huet at inra.fr