mcmc: Metropolis-Hastings Simulation

In simecek/mhnet: Metropolis-Hastings type simulation and inference for Bayesian Networks

Description

Perform Metropolis-Hastings simulation over space of directed acyclic graphs.

Usage

mcmc(start.nw, dataset, reporter, Nsim = 1000, burn = 0.1, verbose = TRUE,
  trace = FALSE, score = BICbnnet, whitelist = NULL, blacklist = NULL,
  lprior = log.prior, move.score = NULL, move.lprior = NULL,
  move.proposals = NULL)

Arguments

start.nw	Starting network
dataset	Data frame with observations
reporter	Statistic to be generated
Nsim	Number of simulations
burn	Proportion of simulations to be used as a burn-in
verbose	If TRUE then report messanges and plots will be given
trace	If TRUE then all reporter values are given, not just the average
score	A function to calculate BIC
whitelist	For move proposal, 2-column matrix or data.frame with compulsory edges (vertices as numbers)
blacklist	For move proposal, 2-column matrix or data.frame with forbidden edges (vertices as numbers)
lprior	Log of prior distribution of acyclic directed graph
move.score	A function to adjust the score to addition / removal / reversal of an edge
move.lprior	A function to adjust the log-prior to addition / removal / reversal of an edge
move.proposals	A function to adjust the proposals to addition / removal / reversal of an edge

Value

Either average or all reporter values based on trace settings.

Examples

rg <- tabu(gaussian.test)
nw <- import.bnlearn(rg)

number.of.edges <- function(nw) sum(nw$adjmat)
ne.dist <- mcmc(nw, gaussian.test, number.of.edges, verbose=FALSE, trace=TRUE)
table(do.call("c", ne.dist$trace)[101:1000]) / 900

all.edges <- function(nw) nw$adjmat
A <- mcmc(nw, gaussian.test, all.edges, verbose=FALSE, trace=FALSE)
strength.plot(export.bnlearn(nw), matrix2strength(A, nw$nodes))

simecek/mhnet documentation built on May 29, 2019, 10:01 p.m.