org.knowceans.dirichlet.sandbox
Class StickBreakingProcess

java.lang.Object
  extended by org.knowceans.dirichlet.sandbox.StickBreakingProcess

public class StickBreakingProcess
extends java.lang.Object

StickBreakingProcess implements Sethurahman's stick breaking process (SBP), which can be described as follows: A stick of length b is broken into two parts whose lengths are distributed according to a beta distribution b'_k ~ Beta(1, gamma) with a concentration parameter gamma.

The SBP is a metaphor for the masses of the clusters that a Dirichlet process samples into: Sampling from a G ~ DP(gamma, G0) can be imagined as chosing either a known sample theta_k from the base measure G0 or, depending on the concentration parameter gamma, picking a new sample theta_K+1 ~ G0 from the base measure, with the sample index k and the total number of distinct observed states K. The likelihoods of the resulting states (infinitely growing in number as infinitely many new data points are observed) are described by the SBP: G = sum_k beta_k delta(theta_k) with the Kronecker (or Dirac?) delta representing a spike in the pdf at theta_k.

Teh at al.'s paper "Hierarchical Dirichlet Processes" in NIPS17 has a good overview presentation of the DP and its metaphors. Further information on the properties can be found in Navarro et al.'s paper on "Modelling individual differences using Dirichlet processes" (preprint for J. Math. Psych ., 2005), aside from the original, but rather complex derivations by Blackwell and MacQueen (1973), Fergusson (1973) and Antoniak (1974).

Author:
heinrich

Constructor Summary
StickBreakingProcess()
           
 
Method Summary
 
Methods inherited from class java.lang.Object
clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait
 

Constructor Detail

StickBreakingProcess

public StickBreakingProcess()