MCMC algorithm for posterior inference under hierarchical Dirichlet process mixtures
From Cohen Courses
Jump to navigationJump to searchThe Chinese restaurant franchise can yield a Gibbs sampling scheme for posterior sampling given observations x. Rather than dealing with the 's and 's, we shall sample their index variables and instead.
- Sampling t. The prior probability that takes on a particular previously used value t is proportional to , whereas the probability that it takes on a new value (say ) is proportional to . The likelihood due to given for some previously used t is , then
If the sampled value of is , we obtain a sample of by sampling from
If as a result of updating some table t becomes unoccupied, i.e., , then the probability that this table will be reoccupied in the future will be zero, since this is always proportional to . As a result, we may delete the corresponding from the data structure. If as a result of deleting some mixture component k becomes unallocated, we delete this mixture component as well.
- Sampling k. Since changing actually changes the component membership of all data items in table t, the likelihood obtained by setting is given by , so that the conditional
probability of is