org.knowceans.sandbox.hlda
Class ChineseRestaurantProcess

java.lang.Object
  org.knowceans.sandbox.hlda.ChineseRestaurantProcess

public class ChineseRestaurantProcess
extends java.lang.Object

CrpNode models a nested Chinese restaurant process (CRP)

Author:

heinrich

Nested Class Summary

(package private) class

ChineseRestaurantProcess.CrpNode

Field Summary

private int	datasize total occupation number
private double	gamma concentration parameter of the CRP
private java.util.Vector<ChineseRestaurantProcess.CrpNode>	nodes list of occupied tables in the CRP

Constructor Summary

ChineseRestaurantProcess(double gamma)
initialise CRP.

Method Summary

static void	main(java.lang.String[] args)
static <T> void	print(java.util.Collection<T> a)
static <T> void	print(T[] a)
private int	sampleCrp() sample a cluster according to the CRP scheme: CRP example: index : [0] [1] [2] draw 1: 1 0 0 e.g., -> [0] draw 2: 1/(1+gamma) gamma/(1+gamma) 0 -> [1] draw 3: 1/(2+gamma) 1/(2+gamma) gamma/(2+gamma) -> [0] draw 4: 2/(3+gamma) 1/(3+gamma) gamma/(3+gamma) -> [1] i.e., the probability of drawing [2] becomes lower with every draw, leading to an aggregated probability of cluster number that is p(clusters) ~ log(datasize).

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Field Detail

nodes

private java.util.Vector<ChineseRestaurantProcess.CrpNode> nodes

list of occupied tables in the CRP

gamma

private double gamma

concentration parameter of the CRP

datasize

private int datasize

total occupation number

Constructor Detail

ChineseRestaurantProcess

public ChineseRestaurantProcess(double gamma)

initialise CRP.

Parameters:

gamma - concentration parameter of the underlying CRP (p=1/nodes if gamma is equal to the mean occupation number per child node). A new cluster is introduced in distances of log(data) items.

Method Detail

main

public static void main(java.lang.String[] args)

sampleCrp

private int sampleCrp()

sample a cluster according to the CRP scheme:

CRP example: index : [0] [1] [2] draw 1: 1 0 0 e.g., -> [0] draw 2: 1/(1+gamma) gamma/(1+gamma) 0 -> [1] draw 3: 1/(2+gamma) 1/(2+gamma) gamma/(2+gamma) -> [0] draw 4: 2/(3+gamma) 1/(3+gamma) gamma/(3+gamma) -> [1] i.e., the probability of drawing [2] becomes lower with every draw, leading to an aggregated probability of cluster number that is p(clusters) ~ log(datasize).

Returns:

print

public static <T> void print(T[] a)

Parameters:

a -

print

public static <T> void print(java.util.Collection<T> a)

Parameters:

a -