simuLCS
Class AgentClassifierLearning

java.lang.Object
  simuLCS.Entity
      simuLCS.Agent
          simuLCS.AgentClassifier
              simuLCS.AgentClassifierLearning

public class AgentClassifierLearning

extends AgentClassifier

Agent with a Behaviour and a Learning Classifier System plugged in.

Author:

Benoit

Field Summary

protected ClassifierSet	behaviourLearning
protected ClassifierSet	behaviourTemp
protected ClassifierSet	behaviourToShow This behaviour is just a copy of behaviourLearning, with the classifiers ordered to present them to the user.
protected javax.swing.JButton	buttonUpdate
protected java.awt.geom.Point2D	coordGhost
static int	EXPECTED_BEHAVIOUR
static int	EXPECTED_BEHAVIOUR_TO_SHOW
protected boolean	isGhostPainted
protected boolean	isLearning
private static double	MAX_ERROR
protected int	nbOfStepsLearning
protected int	nbOfStepsLearningFromLastPlot
protected java.awt.geom.Point2D	oldcoordGhost
static int	REAL_BEHAVIOUR
protected double	totalRewardFromLastPlot
protected double	totalRewardUntilNow
private static int	UPDATE_BEHAVIOUR_TO_SHOW
protected boolean	updateBehaviourTemp
protected boolean	updateBehaviourToShow

Fields inherited from class simuLCS.AgentClassifier

behaviour, DETECTION_DISTANCE, isDetectionLimited, lastMovVector, template

Fields inherited from class simuLCS.Agent

arena, heading, oldcoord

Fields inherited from class simuLCS.Entity

color_ext, color_int, coord, generator, name, paintName, size

Constructor Summary

	AgentClassifierLearning(java.util.Random gen, Arena a, java.lang.String name) Creates an ACL with a default behaviour "Avoid danger".
protected	AgentClassifierLearning(java.util.Random gen, Arena a, java.lang.String name, Template t, ClassifierSet b) Creates an ACL with a given Template and a given Behaviour

Method Summary

protected void	clearGhost(java.awt.Graphics2D g, java.awt.geom.Point2D c) Clear the drawing of the ghost
double	getAverageRewardFromLastPlot() Get the average reward from the last plot, and reset the counters to start again to record
double	getAverageRewardUntilNow()
ClassifierSet	getBehaviour(int whichOne) Returns the given behaviour (real one, expected one, or expected one ordered)
java.awt.geom.Point2D	getCoordGhost()
int	getNbOfStepsLearning()
java.awt.geom.Point2D	getOldcoordGhost()
protected double	getReward(java.awt.geom.Point2D real, java.awt.geom.Point2D expected) Get the reward from the environnement.
double	getTotalRewardUntilNow()
boolean	isGhostPainted()
boolean	isLearning()
private boolean	isUpdateBehaviourTemp()
void	move(Arena arena, Entity[] others, int nbEntities, java.awt.Graphics2D g) Moves the agent.
protected void	moveAndLearn(java.awt.geom.Point2D nextCoord, Vector2D nextMovVector, Arena arena, Entity[] others, int nbEntities, java.awt.Graphics2D g) Moves the Ghost and updates the rules with the reward
protected ZClassifierSet	moveLearner(Arena arena, Entity[] others, int nbEntities, java.awt.Graphics2D g) From the actual position of the agent, the system will try to guess where it is going to be at the next time step, by calculating its response according to its expected behaviour (ie the current behaviour provided by the Learning System).
void	paint(java.awt.Graphics2D g) Removes the drawing of the agent for the old coordinates and repaint the agent at its new coordinates.
protected void	paintGhost(java.awt.Graphics2D g) Paint the ghost (ie the Agent at the expected position)
void	requestUpdateBehaviourToShow(boolean simuIsRunning, javax.swing.JButton b) Asks for the behaviour to be updated.
void	setGhostPainted(boolean b)
void	setLearning(boolean b)
void	setTotalRewardUntilNow(double d)
private void	setUpdateBehaviourTemp(boolean b)
protected void	updateBehaviourToShow() Create a copy of the current population and order the classifiers so that they can be shown ranked.

Methods inherited from class simuLCS.AgentClassifier

canDetect, drawVisibilityCircle, getBehaviour, getLastMovVector, getNewMovVectorAccordingTo, getTemplate, isDetectionLimited, setBehaviour, setDetectionLimited, setLastMovVector, setTemplate

Methods inherited from class simuLCS.Agent

clear, clear, getArena, getCoordNearTo, getRadius, isDangerous, isMoving, move, toString

Methods inherited from class simuLCS.Entity

collideWithMe, distanceTo, getColor_ext, getColor_int, getCoord, getId, getName, getRadiusBody, getSize, isPaintName, setColor_ext, setColor_int, setName, setPaintName, setSize

Methods inherited from class java.lang.Object

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

Field Detail

REAL_BEHAVIOUR

public static final int REAL_BEHAVIOUR

See Also:

Constant Field Values

EXPECTED_BEHAVIOUR

public static final int EXPECTED_BEHAVIOUR

See Also:

Constant Field Values

EXPECTED_BEHAVIOUR_TO_SHOW

public static final int EXPECTED_BEHAVIOUR_TO_SHOW

See Also:

Constant Field Values

behaviourToShow

protected ClassifierSet behaviourToShow

This behaviour is just a copy of behaviourLearning, with the classifiers ordered to present them to the user.

behaviourTemp

protected ClassifierSet behaviourTemp

updateBehaviourTemp

protected boolean updateBehaviourTemp

updateBehaviourToShow

protected boolean updateBehaviourToShow

buttonUpdate

protected javax.swing.JButton buttonUpdate

coordGhost

protected java.awt.geom.Point2D coordGhost

oldcoordGhost

protected java.awt.geom.Point2D oldcoordGhost

isLearning

protected boolean isLearning

isGhostPainted

protected boolean isGhostPainted

MAX_ERROR

private static final double MAX_ERROR

See Also:

Constant Field Values

UPDATE_BEHAVIOUR_TO_SHOW

private static final int UPDATE_BEHAVIOUR_TO_SHOW

See Also:

Constant Field Values

nbOfStepsLearning

protected int nbOfStepsLearning

totalRewardUntilNow

protected double totalRewardUntilNow

totalRewardFromLastPlot

protected double totalRewardFromLastPlot

nbOfStepsLearningFromLastPlot

protected int nbOfStepsLearningFromLastPlot

behaviourLearning

protected ClassifierSet behaviourLearning

Constructor Detail

AgentClassifierLearning

protected AgentClassifierLearning(java.util.Random gen,
                                  Arena a,
                                  java.lang.String name,
                                  Template t,
                                  ClassifierSet b)

Creates an ACL with a given Template and a given Behaviour

Parameters:

gen -

a -

name -

t -

b -

AgentClassifierLearning

public AgentClassifierLearning(java.util.Random gen,
                               Arena a,
                               java.lang.String name)

Creates an ACL with a default behaviour "Avoid danger".

Parameters:

gen -

a -

name -

Method Detail

moveLearner

protected ZClassifierSet moveLearner(Arena arena,
                                     Entity[] others,
                                     int nbEntities,
                                     java.awt.Graphics2D g)

From the actual position of the agent, the system will try to guess where it is going to be at the next time step, by calculating its response according to its expected behaviour (ie the current behaviour provided by the Learning System).

Parameters:

arena -

g -

Returns:

the total set of Classifiers that lead this response (Set of all the ActionSets involved in the process)

See Also:

behaviourLearning

moveAndLearn

protected void moveAndLearn(java.awt.geom.Point2D nextCoord,
                            Vector2D nextMovVector,
                            Arena arena,
                            Entity[] others,
                            int nbEntities,
                            java.awt.Graphics2D g)

Moves the Ghost and updates the rules with the reward

Parameters:

nextCoord - the REAL next position (to compare with the expected one)

nextMovVector - the REAL next Movement Vector

arena -

others -

nbEntities -

g -

getReward

protected double getReward(java.awt.geom.Point2D real,
                           java.awt.geom.Point2D expected)

Get the reward from the environnement.

Parameters:

expected -

Returns:

move

public void move(Arena arena,
                 Entity[] others,
                 int nbEntities,
                 java.awt.Graphics2D g)

Moves the agent. Moves the ghost and applies the learning system if isLearning == true.

Overrides:

move in class AgentClassifier

See Also:

isLearning

clearGhost

protected void clearGhost(java.awt.Graphics2D g,
                          java.awt.geom.Point2D c)

Clear the drawing of the ghost

Parameters:

g -

c -

paintGhost

protected void paintGhost(java.awt.Graphics2D g)

Paint the ghost (ie the Agent at the expected position)

Parameters:

g -

updateBehaviourToShow

protected void updateBehaviourToShow()

Create a copy of the current population and order the classifiers so that they can be shown ranked.

requestUpdateBehaviourToShow

public void requestUpdateBehaviourToShow(boolean simuIsRunning,
                                         javax.swing.JButton b)

Asks for the behaviour to be updated. In order to avoid Concurrency problems, the update will be done at a thread-safe stage.

Parameters:

simuIsRunning -

b -

paint

public void paint(java.awt.Graphics2D g)

Removes the drawing of the agent for the old coordinates and repaint the agent at its new coordinates. Paints the ghost as well if isGhostPainted==true. (This avoids a blinking effect)

Overrides:

paint in class AgentClassifier

See Also:

Agent#paint(), isGhostPainted

getCoordGhost

public java.awt.geom.Point2D getCoordGhost()

isGhostPainted

public boolean isGhostPainted()

isLearning

public boolean isLearning()

getNbOfStepsLearning

public int getNbOfStepsLearning()

getOldcoordGhost

public java.awt.geom.Point2D getOldcoordGhost()

setGhostPainted

public void setGhostPainted(boolean b)

Parameters:

b -

setLearning

public void setLearning(boolean b)

getBehaviour

public ClassifierSet getBehaviour(int whichOne)

Returns the given behaviour (real one, expected one, or expected one ordered)

Overrides:

getBehaviour in class Agent

Parameters:

whichOne -

Returns:

isUpdateBehaviourTemp

private boolean isUpdateBehaviourTemp()

setUpdateBehaviourTemp

private void setUpdateBehaviourTemp(boolean b)

getTotalRewardUntilNow

public double getTotalRewardUntilNow()

getAverageRewardUntilNow

public double getAverageRewardUntilNow()

setTotalRewardUntilNow

public void setTotalRewardUntilNow(double d)

getAverageRewardFromLastPlot

public double getAverageRewardFromLastPlot()

Get the average reward from the last plot, and reset the counters to start again to record

Returns: