Update the beta weights according to the gradient ascent

Source: R/F_inference.R

Update the beta weights according to the gradient ascent

FitBeta(
  beta.init,
  psi,
  maxIter = 50,
  eps = 1e-06,
  unlinked = NULL,
  sum.weights
)

Arguments

beta.init

Initial beta weight matrix
psi

Psi matrix, filled with ratios of bivariate probabilities over marginals, which can in the Gaussian case be deduced from the correlation matrix.
maxIter

Maximum number of iterations
eps

Precision parameter controlling the convergence of weights beta
unlinked

An optional vector of nodes which are not linked with each other
sum.weights

Sum constraint for the weight matrix

Value

  • edges_prob: p x p matrix of edges probabilities

  • edges_weight: p x p matrix of edges weights for any spanning tree

  • logpY: vector of log-likelihoods

  • maxIter: final number of iterations EMtree has ran

  • timeEM: EMtree computation time

Examples

set.seed(1)
n=50
p=10
Y=data_from_scratch("tree",p=p,n=n)$data
mean.val=exp((-(p-2)*log(p))/(p-1))
beta = matrix(mean.val, p, p);  diag(beta)=0
psi=Psi_alpha(cor(Y), n)$psi
FitEM = FitBeta(beta.init=beta, psi=psi, maxIter = 6, sum.weights=sum(beta))