clusterProportions: Compute the Proportion of Items in Each Cluster for All...

In sams: Merge-Split Samplers for Conjugate Bayesian Nonparametric Models

Compute the Proportion of Items in Each Cluster for All Partitions

Description

Compute the Proportion of Items in Each Cluster for All Partitions

Usage

clusterProportions(partitions)

Arguments

partitions

A matrix, with each row representing a set partition of the integers 1, ..., n as cluster labels

Value

A matrix whose columns represent the cumulative proportion of the data that correspond to that cluster.

Examples

# Neal (2000) model and data
nealData <- c(-1.48, -1.40, -1.16, -1.08, -1.02, 0.14, 0.51, 0.53, 0.78)
mkLogPosteriorPredictiveDensity <- function(data = nealData,
                                            sigma2 = 0.1^2,
                                            mu0 = 0,
                                            sigma02 = 1) {
  function(i, subset) {
    posteriorVariance <- 1 / ( 1/sigma02 + length(subset)/sigma2 )
    posteriorMean <- posteriorVariance * ( mu0/sigma02 + sum(data[subset])/sigma2 )
    posteriorPredictiveSD <- sqrt(posteriorVariance + sigma2)
    dnorm(data[i], posteriorMean, posteriorPredictiveSD, log=TRUE)
  }
}

logPostPredict <- mkLogPosteriorPredictiveDensity()

nSamples <- 500L
partitions <- matrix(0, nrow=nSamples, ncol=length(nealData))
for ( i in 2:nSamples ) {
  partitions[i,] <- nealAlgorithm3(partitions[i-1,], logPostPredict, mass = 1.0, nUpdates = 2)
}
clusterProportions(partitions)

sams documentation built on April 20, 2022, 1:06 a.m.