com.mapr.stats

Class BetaWalk

java.lang.Object

com.mapr.stats.BetaWalk

public class BetaWalk
extends Object

Follow a Metropolis random walk that should converge to a beta distribution. Since the point is to have slowly changing probabilities for simulating non-stationary conversion processes, having substantial sample to sample correlation is good here.

The probabilities returned will be beta distributed if you take enough steps. Steps are proposed using a normally distributed step in soft-max space which gives a random walk bounded to (0,1) in probability space. The proposal distribution winds up taking very small steps near the boundaries with larger steps in the middle. Steps are accepted or rejected according to the Metropolis algorithm. Computing the probabilities for acceptance or rejection in the probability space while taking the step in log-odds space is OK since the proposal probability is still symmetrical.

Constructor Summary

Constructors

Constructor and Description
BetaWalk(double alpha, double beta, double stepSize)

Method Summary

Methods

Modifier and Type	Method and Description
double	step()

Methods inherited from class java.lang.Object

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

Constructor Detail

BetaWalk

public BetaWalk(double alpha,
        double beta,
        double stepSize)

Method Detail

step

public double step()