newMallowsProposal: Maximum likelihood estimation of GBN with new node order.
In andreamrau/GBNcausal: Causal estimation for Gaussian Bayesian networks

Description Usage Arguments Value Examples

This function is used by the MCMC.GBN algorithm every step : it propose a new order, order nodes of the GBN according to the new order, computes the maximum likelihood estimator based on the data.

1	newMallowsProposal(GBN, data, alpha, alpha2, lambda = 0, listblocks = list(), str = matrix(1, length(GBN@resSigma), length(GBN@resSigma)))

`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.
`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.

`GBN`	The maximum likelihood estimation of the first GBN calculated according to the data and the new node order.
`data`	The data are ordered according to the new node order proposed by the rmallows or rmallowsBlocks function using dataOrder.

# 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)  

newMallowsProposal(firstGBN, data, alpha = 0.5, alpha2 = 0.01, lambda=0)