Block Gibbs Sampler with Incomplete Multinomial Distribution
Description
Implement an efficient block Gibbs sampler with incomplete
data from a multinomial distribution taking values from the k
categories 1,2,...,k, where data are assumed to miss at random
and each missing datum belongs to one and only one of m
distinct non-empty proper subsets A1, A2,..., Am of 1,2,...,k
and the k categories are labelled such that only consecutive
A's may overlap.