blossomTree: Bossom tree graph estimation and density estimation

Description Usage Arguments Details Value References Examples

Description

The main function for bossom tree graph estimation and density estimation.

Usage

1
blossomTree(data, lambda = NULL, refit = FALSE, verbose = TRUE)

Arguments

data

An n by d data matrix, where n is the sample size and d is the dimension.

lambda

A sequence of positive numbers to control the regularization of the graphical lasso.

refit

If refit = TRUE, the refit graphical lasso is applied. The default value is FALSE.

verbose

If verbose = FALSE, tracing information printing is disabled. The default value is TRUE.

Details

The function partitions the data matrix into two subsamples with equal size by random allocation. The first subsample is used to construct a family of blossom tree graphs and the corresponding density estimators, while the second subsample is then used to determine the optimal blossom tree by maximizing the held-out log-likelihood.

The refit graphical lasso is a two-step procedure: in the first step, a sparse inverse covariance matrix is obtained by the graphical lasso; in the second step, a Gaussian model is refit without regularization, but enforcing the sparsity pattern obtained in the first step.

Value

loglike

Held-out log-likelihood for a family of blossom tree density estimators with m=1,...,d-1 forest edges.

adj

Adjacency matrices of a family of blossom tree graphs with m=1,...,d-1 forest edges.

best.loglike

Maximum of the held-out log-likelihood.

best.adj

Adjacency matrix of the optimal blossom tree graph corresponding to the maximum of the held-out log-likelihood.

References

Zhe Liu and John Lafferty. Blossom tree graphical models. Advances in Neural Information Processing Systems, 27:1458–1465, 2014.

Examples

1
2
3
4
5
library(igraph)

fit <- blossomTree(data)
bt <- fit$best.adj
plot(bt)

zhejosephliu/blossomTree documentation built on May 4, 2019, 10:17 p.m.