GBNmle: Maximum likelihood estimator

In andreamrau/GBNcausal: Causal estimation for Gaussian Bayesian networks

Description

This function computes the maximum likelihood estimator of the object GBN calculated according to the data.

The system to solve can be rewritten as : AW=b, where W is the weightmatrix of the GBN.

Usage

GBNmle(GBN, data, lambda = 0, sigmapre = rep(0, dim(data$x)[2]))

Arguments

GBN	An object of the class GBN.
data	data - Data is a format obtained by the function dataFormat.
lambda	logarithmic - coefficient of the penalty Ridge.
sigmapre	vector - In the algorithm MCMC.GBN, sigmapre is the variance of the GBN of the previous iteration. If there is no previous iteration, ie if the function is used alone, sigmapre is a null vector.

Value

GBN	The estimator of the initial GBN
A	A member of the equation AW = b solved by the function.
b	b member of the equation AW = b solved by the function.
y	centered observation (data$x).

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=10e10,sigmapre=rep(0.1,p))$GBN

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