em.litree: em.litree

In cambroise/LITree: Latent Incomplete Tree Inference via Graphical Gaussian Model

Description

The function em.litree implements the algorithm of Gaussian Graphical Model Inference with missing variable described in Robin et. al (2017). The underlying model is based on the aggregation of spanning trees, and the estimation procedure on the Expectation-Maximization algorithm. We treat the graph structure and the unobserved nodes as missing variables and compute posterior probabilities of edge appearance. To provide a complete methodology, we also propose three model selection criteria to estimate the number of missing nodes.

Usage

em.litree(X, k = 0:2, criterion = "ICL_T", max.iter = 20, eps = 0.1)

Arguments

X	X is a data matrix
k	is a vector of missing variable number. It should be an integer greater than 0 or a vector of such integers
criterion	is the name of the criterion used for selecting the best model ("ICL_T", "ICL_ZT", "BIC" ). ICL_T by default
max.iter	is the maximum number of iterations (20 by default)
eps	is the precision used for stopping the algorithm

Value

A list of three items, criteria a dataframe whose columns are the three critera and each lines corresponds to a given number of missing variables, a list of models and the best model according the specified criterion

References

Genevi<c3><a8>ve Robin, Christophe Ambroise, St<c3><a9>phane Robin (Submitted on 26 May 2017). Graphical model inference with unobserved variable via latent tree aggregation. Arxiv Paper. https://arxiv.org/abs/1705.09464

Examples

data(cyto)
res.raf.full <- em.litree(X.raf,0:2)

cambroise/LITree documentation built on May 6, 2019, 8:32 p.m.