dmgbayes.gaussiansample(dmgbayes)R Documentation

MCMC sampler for the parameters of a Gaussian directed mixed graph models

Description

A Gibbs sampling procedure to sample parameters from the posterior distribution of a Gaussian model that is parameterized by a Gaussian directed mixed graph model. The output is stored in a data file.

Usage

   dmgbayes.gaussiansample(var.names, directed, bidirected, train, islatent,
                           prior.V, dmg.df, b.priormean, b.priorvar, b.fixed,
                           mcmc.num.samples, mcmc.burn.in, mcmc.step,
                           output.filename, mcmc.options, test = NULL)

Arguments

var.names Array of string naming all variables (observed and latents)
directed A m x 2 array, where each row (i, j) encodes a directed edge from the jth variable into the ith variable. All variables should be encoded as integers (1, 2, 3, ...), and their number should match the corresponding entry in var.names
bidirected A m' x 2 array, m' being the total number of bi-directed edges. As above, each two-dimensional row (i, j) encodes a bi-directed edge connecting the ith and jth variables
train Data which we condition on when computing the posterior
islatent Binary array, indicating with '1' which variables in var.names are unobserved variables
prior.V A p x p matrix with the matrix parameter of the G-Inverse Wishart prior for the covariance matrix of the error terms
dmg.df Degrees of freedom of the same prior
b.priormean A (m + p) dimensional array contaning the mean parameter of the Gaussian prior for each coefficient associated with the respective directed edge. *IMPORTANT*: remember that we have to consider the intercept terms too. The intercept term for each variable $Y_i$ should be located just after the other coefficients for $Y_i$. This explains why this afrray has m + p dimensions
b.priorvar The respective variances
b.fixed A (m + p) dimensional binary array indicating with '1' those coefficients that are to be fixed to a constant value. The constant to be used is the respective entry in b.priormean. *REMEMBER*: intercepts should be considered
mcmc.num.samples Total number of samples to be generated in the Markov chain
mcmc.burn.in Total number of initial samples to be skipped when storing the results of the chain (minimum value: 0)
mcmc.step Interval between points that will be stored in the output file
output.filename Name of the file to which samples points will be stored
mcmc.options A numeric array contain miscellaneous options: mcmc.options[1] is a number between 1 and 4 indicating verbosity level of output messages returned to the R console; mcmc.options[2] should be set to '1' if you want to generate a file with sampled latents. This file will be called 'output.filename'.latents
test Optional matrix with test data to be evaluated by predictive log-likelihood

Author(s)

Ricardo Silva, Statistical Laboratory, University of Cambridge

References

Silva, R. (2008). "A Tutorial on dmgBayes: Bayesian Analysis for Directed Mixed Graph Models"

See Also

dmgbayes.gaussian.marginal.bidirected


[Package dmgbayes version 1.0 Index]