com.mapr.stats.bandit

Class EpsilonGreedy

java.lang.Object

com.mapr.stats.bandit.BayesianBandit

com.mapr.stats.bandit.EpsilonGreedy

All Implemented Interfaces:

Iterable<AbstractBayesianDistribution>

public class EpsilonGreedy
extends BayesianBandit

Solves a bandit problem using an epsilon greedy algorithm. In this algorithm, a fixed proportion of trials are allocated to uniform exploration while all others are devoted to the current best bandit alternative.

Constructor Summary

Constructors

Constructor and Description
EpsilonGreedy(int bandits, double epsilon, Random gen)

Method Summary

Methods

Modifier and Type	Method and Description
boolean	addModelDistribution(AbstractBayesianDistribution distribution)
int	sample() Samples probability estimates from each bandit and picks the apparent best
void	train(int bandit, double reward) Apply feedback to the bandit we chose.

Methods inherited from class com.mapr.stats.bandit.BayesianBandit

getMean, iterator, rank

Methods inherited from class java.lang.Object

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

Constructor Detail

EpsilonGreedy

public EpsilonGreedy(int bandits,
             double epsilon,
             Random gen)

Method Detail

sample

public int sample()

Samples probability estimates from each bandit and picks the apparent best

Overrides:

sample in class BayesianBandit

Returns:

The index of the chosen bandit

train

public void train(int bandit,
         double reward)

Apply feedback to the bandit we chose.

Overrides:

train in class BayesianBandit

Parameters:

bandit - Which bandit got the impression

reward - Did it pay off?

addModelDistribution

public boolean addModelDistribution(AbstractBayesianDistribution distribution)

Overrides:

addModelDistribution in class BayesianBandit