com.mwt.explorers

Class EpsilonGreedyExplorer<T>

java.lang.Object

com.mwt.explorers.EpsilonGreedyExplorer<T>

All Implemented Interfaces:

ConsumePolicy<T>, Explorer<T>

Direct Known Subclasses:

VariableActionEpsilonGreedyExplorer

public class EpsilonGreedyExplorer<T>
extends Object
implements Explorer<T>, ConsumePolicy<T>

The epsilon greedy exploration algorithm. This is a good choice if you have no idea which actions should be preferred. Epsilon greedy is also computationally cheap.

Constructor Summary

Constructors

Constructor and Description
EpsilonGreedyExplorer(Policy<T> defaultPolicy, float epsilon, int numActions) The constructor

Method Summary

Modifier and Type	Method and Description
DecisionTuple	chooseAction(long saltedSeed, T context) Determines the action to take and the probability with which it was chosen, for a given context.
void	enableExplore(boolean explore)
protected int	getNumActions(T context)
void	updatePolicy(Policy<T> newPolicy)

Methods inherited from class java.lang.Object

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

Constructor Detail

EpsilonGreedyExplorer

public EpsilonGreedyExplorer(Policy<T> defaultPolicy,
                             float epsilon,
                             int numActions)

The constructor

Parameters:

defaultPolicy - A default function which outputs an action given a context.

epsilon - The probability of a random exploration.

numActions - The number of actions to randomize over.

Method Detail

getNumActions

protected int getNumActions(T context)

updatePolicy

public void updatePolicy(Policy<T> newPolicy)

Specified by:

updatePolicy in interface ConsumePolicy<T>

chooseAction

public DecisionTuple chooseAction(long saltedSeed,
                                  T context)

Description copied from interface: Explorer

Determines the action to take and the probability with which it was chosen, for a given context.

Specified by:

chooseAction in interface Explorer<T>

Parameters:

saltedSeed - A PRG seed based on a unique id information provided by the user

context - A user-defined context for the decision

Returns:

The action to take, the probability it was chosen, and a flag indicating whether to record this decision

enableExplore

public void enableExplore(boolean explore)

Specified by:

enableExplore in interface Explorer<T>