burlap.behavior.stochasticgames.agents.naiveq.history

Class SGQWActionHistoryFactory

java.lang.Object

burlap.behavior.stochasticgames.agents.naiveq.history.SGQWActionHistoryFactory

All Implemented Interfaces:

AgentFactory

public class SGQWActionHistoryFactory
extends java.lang.Object
implements AgentFactory

An agent factory for Q-learning with history agents.

Author:

James MacGlashan

Field Summary

Fields

Modifier and Type	Field and Description
protected double	discount The discount rate the Q-learning algorithm will use
protected SGDomain	domain The stochastic games domain in which the agent will act
protected double	epsilon The epislon value for epislon greedy policy.
protected int	historySize How much history the agent should remember
protected double	learningRate The learning rate the Q-learning algorithm will use
protected int	maxPlayers The maximum number of players that can be in the game
protected QFunction	qinit A default Q-value initializer
protected HashableStateFactory	stateHash The state hashing factory the Q-learning algorithm will use

Constructor Summary

Constructors

Constructor and Description
SGQWActionHistoryFactory(SGDomain d, double discount, double learningRate, HashableStateFactory stateHash, int historySize) Initializes the factory

Method Summary

Modifier and Type	Method and Description
SGAgent	generateAgent(java.lang.String agentName, SGAgentType type) Generates a new SGAgent
void	setEpsilon(double epsilon) Sets the epislon parmaeter (for epsilon greedy policy).
void	setQValueInitializer(QFunction qinit) Sets the Q-value initialization function that will be used by the agent.

Methods inherited from class java.lang.Object

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

Field Detail

domain

protected SGDomain domain

The stochastic games domain in which the agent will act

discount

protected double discount

The discount rate the Q-learning algorithm will use

learningRate

protected double learningRate

The learning rate the Q-learning algorithm will use

stateHash

protected HashableStateFactory stateHash

The state hashing factory the Q-learning algorithm will use

historySize

protected int historySize

How much history the agent should remember

maxPlayers

protected int maxPlayers

The maximum number of players that can be in the game

qinit

protected QFunction qinit

A default Q-value initializer

epsilon

protected double epsilon

The epislon value for epislon greedy policy. If negative, then the policy of the created agent will not be different than its default.

Constructor Detail

SGQWActionHistoryFactory

public SGQWActionHistoryFactory(SGDomain d,
                                double discount,
                                double learningRate,
                                HashableStateFactory stateHash,
                                int historySize)

Initializes the factory

Parameters:

d - the stochastic games domain in which the agent will act

discount - The discount rate the Q-learning algorithm will use

learningRate - The learning rate the Q-learning algorithm will use

stateHash - The state hashing factory the Q-learning algorithm will use

historySize - How much history the agent should remember

Method Detail

setQValueInitializer

public void setQValueInitializer(QFunction qinit)

Sets the Q-value initialization function that will be used by the agent.

Parameters:

qinit - the Q-value initialization function.

setEpsilon

public void setEpsilon(double epsilon)

Sets the epislon parmaeter (for epsilon greedy policy). If set to a negative, then the default policy of the create agent will be used.

Parameters:

epsilon - the epsilon value to use

generateAgent

public SGAgent generateAgent(java.lang.String agentName,
                             SGAgentType type)

Description copied from interface: AgentFactory

Generates a new SGAgent

Specified by:

generateAgent in interface AgentFactory

Parameters:

agentName - the name for the agent

type - the SGAgentType for the agent

Returns:

a new SGAgent