MCMC.step: Main step of the MCMC.GBN algorithm

In andreamrau/GBNcausal: Causal estimation for Gaussian Bayesian networks

Description

This function is used by the MCMC.GBN algorithm every step : it uses the newMallowsProposal function and accepts or rejects the proposed GBN.

Usage

MCMC.step(GBN, data, alpha = 0.05, alpha2 = 0.05, lambda, listblocks = list(), str = matrix(1, length(GBN@resSigma), length(GBN@resSigma)))

Arguments

GBN	GBN - An object of class GBN. If this is a step of the MCMC.GBN function, this argument is the maximum likelihood estimation of the previous iteration.
data	data - can be obtained by the function dataFormat or dataCreate.
alpha	double - First Mallows temperature.
alpha2	double - Second Mallows temperature. If listblocks is empty, this argument won't be used.
lambda	logarithmic - Coefficient of the penalty Ridge.
listblocks	list - A list of nodes of the form (c("N1","N2"),c("N3","N4","N5")) where "N1","N2","N3","N4" and "N5" are elements of rownames and colnames of firstGBN elements and data elements.
str	matrix - To improve the efficiency of the algorithm, a structure can be add. Colnames and rownames are not needed.

Value

newGBN	If accept equals 1, new GBN is the maximum likelihood estimation of the previous GBN based on a new order of the nodes and the data. If accept equals 0, newGBN is the GBN of the previous step.
accept	If 1, the new GBN is the new estimation. If 0, the new GBN is the old GBN.

Examples

# Data creation

seed = 1990
n = 3000
p <- 10
m<-rep(0,10)
sigma<-rep(0.1,10) 

W <- 1*upper.tri(matrix(0,p,p))

data <- dataCreate(nbData = 2*p, p = 10,KO = list(1,9), nbKO = c(p,p), W = W , m = m,sigma = sigma, seed = seed)$data

# Initial Value

W1=1*upper.tri(matrix(0,p,p)) 
m1=rep(0,p)
s1=rep(10e-4,p)
colnames(W1)=names(m1)=names(s1)=rownames(W1)=paste("N",1:p,sep="")

firstGBN = new("GBNetwork",WeightMatrix=W1,resMean=m1,resSigma=s1)
firstGBN = GBNmle(firstGBN,data,lambda= 0,sigmapre=s1)$GBN	  

# Algorithm

MCMC.step(firstGBN, data, alpha=0.05, lambda = 0)

andreamrau/GBNcausal documentation built on May 12, 2019, 3:34 a.m.