shallot (version 0.4.8)

sampling.model: Specifying a Sampling Model

Description

These functions set the mass, discount, and temperature parameters and, in the case of them being random, specify the parameters of their distribution.

Usage

sampling.model(sample.parameter, log.density)

Arguments

sample.parameter

A function taking two arguments with names indices and parameter having default values c() and NULL, respectively. With those default values, the function should return a sample from the centering distribution as an object of any type. Otherwise, indices is a vector of integers indicating the cluster elements and parameter is the current value of the parameter for the cluster, and the function should return an updated value for the cluster parameter based on a valid MCMC update.

log.density

A function taking three arguments with names i, indices, and parameter and returning a double giving the log of the likelihood contribution of item i to a clustering with members indices and parameter parameter.

Value

An object of class shallot.distribution.data.

Examples

# NOT RUN {
## Model inputs.
data <- c(-1.48, -1.40, -1.16, -1.08, -1.02, 0.14, 0.51, 0.53, 0.78)
sigma  <- 0.1
mu0    <- 0.0
sigma0 <- 1.0

## Derived values.
s2 <- sigma * sigma
s02 <- sigma0 * sigma0
s02Inv <- 1.0 / s02
c <- -1.0 / (2.0 * s2)

## Sampling model of Neal (JCGS, 2009)
## Function to perform an MCMC update of the parameter.
sample.parameter <- function(indices=c(), parameter=NULL) {
  sum <- sum(data[indices])
  variance <- 1 / (s02Inv + length(indices) / s2)
  mean <- variance * (mu0 / s02 + sum / s2)
  rnorm(1, mean=mean, sd=sqrt(variance))
}

## Function to evaluate the likelihood contribution for an observation.
log.density <- function(i, indices, parameter) {
  resid <- data[i] - parameter
  c * resid * resid
}

sm <- sampling.model(sample.parameter, log.density)
sm

# }