PCG: Power Chain Graph Structure Learning

In vthorrf/PCG: Procedure for learning Power Chain Graphs

Description

Learns a power chain graph following Franco, Barros, Wiberg & Laros (under review).

Usage

PCG(data, clusters=NULL, alpha=.05, fineTuning=F)

Arguments

data	A data frame of the variables to be included in the PCG.
clusters	A integer vector of expected classes for each variable in data. If an exploratory approach is preferred, use "ega" for learning the cluster by the EGA method or "cgmm" learning the cluster by the CGMM method.
alpha	The significance level for testing the directional relations with the PC-stable algorithm.
fineTuning	Boolean indicating if fine-tuning should be performed. Defaults to FALSE.

Value

A list containing the following component:

PCG	The estimated PCG.
clusters	The integer vector of expected classes for each variable in data.
CG_U	The undirected part of the chain graph induced by the PCG.
CG_D	The directed part of the chain graph induced by the PCG.
CG_PC	The directed part of the chain graph implied by the PCG, fine-tuned with the PC-stable algorithm.
CG_HC	The directed part of the chain graph implied by the PCG, fine-tuned with the hill-climbing algorithm.
CG_MMCC	The directed part of the chain graph implied by the PCG, fine-tuned with the Min-Max hill-climbing algorithm.

Examples

### Use Big Five data
require(psych)
data("bfi")
data <- bfi[complete.cases(bfi),1:25]

### Run the analysis
fit <- PCG(data, clusters="ega", alpha=.05, fineTuning=F)

### Extract the results
pcg <- fit$PCG
ug  <- fit$CG_U
dg  <- fit$CG_D

### Visualize
require(qgraph)
require(EGAnet)
par(mfrow=c(2,1))
qgraph::qgraph(pcg)
qgraph::qgraphMixed(ug * qgraph::cor_auto(data, forcePD=T),
                    dg * EGAnet::EGA(data, plot=F)$network,
                    groups = as.factor(fit$clusters),
                    labels = colnames(data))

vthorrf/PCG documentation built on April 4, 2021, 4:10 a.m.