Nothing
# Simulation Functions
Gibbs_Sampler <- function(GERGM_Object,
theta,
MCMC.burnin,
num.draws,
thin = 1,
start = NULL,
num.nodes = NULL,
directed,
possible.stats) {
# formula specifies which variables to use MCMC.burnin is the number of
# discarded draws n is the total number of draws to be reported dh is a
# function that takes as arguments, i, j, theta, net and returns the partial
# of the hamiltonian theta is the vector-valued parameter thin reduces
# autocorrelation in the simulations, every thinth iteration is returned start
# is the initial network, if not supplied, a random uniform nodesXnodes
# network is used num.nodes is the number of nodes in the network dir is a
# logical indicator of whether the network is directed
net <- GERGM_Object@network
if (is.null(num.nodes) == TRUE) {
num.nodes <- GERGM_Object@num_nodes
}
sims <- num.draws * thin
netarray <- array(NA, dim = c(num.nodes, num.nodes, num.draws + 1))
if (is.null(start))
start <- matrix(rdisp(num.nodes * num.nodes), num.nodes, num.nodes)
net <- start
diag(net) = 0
netarray[, , 1] <- net
for (t in 1:(num.draws + MCMC.burnin)) {
if (directed == TRUE) {
for (i in 1:num.nodes) {
for (j in (1:num.nodes)[-i]) {
net[i, j] <- rtexp(1, t(theta) %*% ex_dh(GERGM_Object, i, j, net))
}
}
}
if (directed == FALSE) {
for (i in 2:num.nodes) {
for (j in (1:(i - 1))) {
net[i, j] <- rtexp(1, t(theta) %*% ex_dh(GERGM_Object, i, j, net))
}
}
}
if (t > MCMC.burnin) {
netarray[, , t - MCMC.burnin] <- net
}
}
return(netarray[, , round(seq(1, sims, length = num.draws))])
}
Any scripts or data that you put into this service are public.
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.