Class Agent

java.lang.Object
  |
  +--Agent

All Implemented Interfaces:

java.io.Serializable

public class Agent

extends java.lang.Object

implements java.io.Serializable

This class is an agent, containing the main loop of the XCS and the routines handling communications. The XCS is implemented following Martin V. Butz and Steward Wilson (2001) "An algorithmic description of XCS". Moreover, the performance evaluation of each agent is computed here.

Since:

JDK 1.3

See Also:

Serialized Form

Field Summary

private Book	addressBook The address book of the agent
private double	averErr initialization of the average performance, fitness and prediction error of the agent's population of classifiers
private double	averExp initialization of the average performance, fitness and prediction error of the agent's population of classifiers
private double	averFit initialization of the average performance, fitness and prediction error of the agent's population of classifiers
private double	averPerf initialization of the average performance, fitness and prediction error of the agent's population of classifiers
private XClassifier	bestClassifier The classifier selected for communication
private static XCSConstants	cons Stores the relevant constants for XCS.
private double	distLevelOne threshold distance between the central competencies of agents and both signals from the environment and offsprings generated by GA
private double	distLevelTwo threshold distance from the core competence to accept communication
private XClassifierSet	gpop Stores the current population of classifiers in XCS.
private int	name The name of the agent
private int	nbCommunications Number of Communications done by each agent for each problem
private double	noAnswer reward for the manager in case there is no solution found by the agent
private double	perC The percentage of agents asked at random in the whole population of agents
private java.util.Vector	randomArray A vector used to store agents picked up at random from the total population of agents during communications
private double	reward The reward the agent receives from the environment for a given answer
(package private) boolean	solution Has the agent found a solution to the current problem?
private static int	totalNumberOfAgents The total number of agents involved in the experiment

Constructor Summary

Agent(double cc, double consultRatio)
Constructs the XCS system of the agent.

Method Summary

private XClassifierSet	answer(java.lang.String state, XClassifierSet collectedAnswersArray, int time) The method used by agents to answer the question.
private XClassifierSet	askAgentsAtRandom(AgentsPopulations population, java.lang.String state, Environment env, XClassifierSet collectedAnswersArray, int time) This method asks agents randomly chosen in the population to answer the question. 10% of the whole population is selected.
private XClassifierSet	askAgentsInAddressBook(AgentsPopulations population, Environment env, java.lang.String state, XClassifierSet collectedAnswersArray, int time) This method asks to each agent in the address book to try to answer the question.
private boolean	checkAgent(int ran) Checks that 1) the agent does not choose itself 2) the agents does not choose an agent already in its address book and then computes whether or not a signal is in the core competence of the agent.
private boolean	checkRand(int[] t, int i) Checks that a random number is not drawn twice in one single process
private java.util.Vector	chooseAgents(int size, AgentsPopulations population) Build an array of randomly chosen agents.
boolean	doOneSingleStepProblemExplore(java.lang.String state, int counter, AgentsPopulations wholePopulation, int selector) Executes one main learning loop for a single step problem.
Book	getAddressBook() return the address book of the agent
boolean	getCloseState(java.lang.String state, AgentsPopulations wholePopulation) The agent computes whether or not the signal from the environment is in its realm of specialisation or not before to tell it to the manager.
int	getName() Return the name of the agent
int	getNbCommunications() Returns the number of communications made by the agent
double	getPerformance() return the reward received by the agent from the environment once it has fired an action
XClassifierSet	getPopulation() Return the population of rules of an agent
private java.util.Vector	randomPopulation(AgentsPopulations population) Construct a random population of agents picked up from the global population.
private XClassifier	selectBestAnswer(XClassifierSet collectedAnswersArray, java.lang.String state, int time) THis method is used by the agent who asks a question to select the best answer from the ones it recieved
void	setAddressBook(Book addBook) Set an address book for the agent
void	setName(int i) Set the name of the agent
void	setXClassifierSet(XClassifierSet aXClassifierSet) Set the initial population of classifiers within an agent
private void	startCommunication(java.lang.String state, Environment env, XClassifierSet matchSet, AgentsPopulations wholePopulation, int time) This method is the communication itself.

Methods inherited from class java.lang.Object

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

Field Detail

totalNumberOfAgents

private static int totalNumberOfAgents

The total number of agents involved in the experiment

gpop

private XClassifierSet gpop

Stores the current population of classifiers in XCS.

bestClassifier

private XClassifier bestClassifier

The classifier selected for communication

name

private int name

The name of the agent

solution

boolean solution

Has the agent found a solution to the current problem?

distLevelOne

private double distLevelOne

threshold distance between the central competencies of agents and both signals from the environment and offsprings generated by GA

distLevelTwo

private double distLevelTwo

threshold distance from the core competence to accept communication

noAnswer

private double noAnswer

reward for the manager in case there is no solution found by the agent

addressBook

private Book addressBook

The address book of the agent

randomArray

private java.util.Vector randomArray

A vector used to store agents picked up at random from the total population of agents during communications

cons

private static XCSConstants cons

Stores the relevant constants for XCS.

See Also:

XCSConstants

perC

private double perC

The percentage of agents asked at random in the whole population of agents

averPerf

private double averPerf

initialization of the average performance, fitness and prediction error of the agent's population of classifiers

averFit

private double averFit

initialization of the average performance, fitness and prediction error of the agent's population of classifiers

averExp

private double averExp

initialization of the average performance, fitness and prediction error of the agent's population of classifiers

averErr

private double averErr

initialization of the average performance, fitness and prediction error of the agent's population of classifiers

reward

private double reward

The reward the agent receives from the environment for a given answer

nbCommunications

private int nbCommunications

Number of Communications done by each agent for each problem

Constructor Detail

Agent

public Agent(double cc,
             double consultRatio)

Constructs the XCS system of the agent.

Parameters:

cc - value of the threshold distance between the central competencies of agents and signals and offsprings generated by GA

consultRatio - value of the percentage of the population of agents asked at random during communications

See Also:

XCSConstants.XCSConstants()

Method Detail

setXClassifierSet

public void setXClassifierSet(XClassifierSet aXClassifierSet)

Set the initial population of classifiers within an agent

Parameters:

aXClassifierSet - An object of the class XClassifierSet attached to each agent

setName

public void setName(int i)

Set the name of the agent

Parameters:

i - The agent's name

setAddressBook

public void setAddressBook(Book addBook)

Set an address book for the agent

Parameters:

addBook - Address book of the agent

getAddressBook

public Book getAddressBook()

return the address book of the agent

getPerformance

public double getPerformance()

return the reward received by the agent from the environment once it has fired an action

getPopulation

public XClassifierSet getPopulation()

Return the population of rules of an agent

getName

public int getName()

Return the name of the agent

doOneSingleStepProblemExplore

public boolean doOneSingleStepProblemExplore(java.lang.String state,
                                             int counter,
                                             AgentsPopulations wholePopulation,
                                             int selector)

Executes one main learning loop for a single step problem. In addition, it checks whether a communication loop should be started and, if yes, it starts the communication process. if the agent has no answer, or a "bad" answer in terms of prediction or fitness to the signal from the environment, or if the signal is remote from its core competencies, then it asks for help. If despite the communication process, the agent has still no answer, then he passes the question through to the next agent in the list. Else, it answers with the rule it has, doing some exploration.

Parameters:

state - The actual problem instance.

counter - The number of problems observed so far in exploration.

wholePopulation - The whole set of agents in the population

selector - Which mode of rule selection is used (bestAction or roulette wheel)

See Also:

XClassifierSet.XClassifierSet(String,XClassifierSet,int,int), PredictionArray.PredictionArray(XClassifierSetManager, int), PredictionArray.bestActionWinner(), PredictionArray.rouletteActionWinner(), XClassifierSet.XClassifierSet(XClassifierSet,int), Environment.executeAction(int), XClassifierSet.updateSet(double, double), XClassifierSet.runGA(int, Environment, java.lang.String, int, double, double), XClassifierSet.testPerformance(double), XClassifierSet.testFitness(double), XClassifierSet.computeMinDistance(Environment, java.lang.String, double), startCommunication(java.lang.String, Environment, XClassifierSet, AgentsPopulations, int)

startCommunication

private void startCommunication(java.lang.String state,
                                Environment env,
                                XClassifierSet matchSet,
                                AgentsPopulations wholePopulation,
                                int time)

This method is the communication itself. First, the agent ask agents present in its address book. Second, it asks agents taken at random in the whole population. Third, it chooses the best answer among the ones collected and inserts it in its own population and in its matchSet.

Parameters:

state - the current problem submitted by the environment

env - the environment or problem space

matchSet - the set of classifiers with conditions matchning the problem

wholePopulation - the population of agents in the experiment

time - the time counter of the experiment

See Also:

XClassifierSet.XClassifierSet(int), askAgentsInAddressBook(AgentsPopulations, Environment, java.lang.String, XClassifierSet, int), askAgentsAtRandom(AgentsPopulations, java.lang.String, Environment, XClassifierSet, int), selectBestAnswer(XClassifierSet, java.lang.String, int), XClassifierSet.addBestAnswerToPopulation(XClassifier, XClassifierSet)

askAgentsInAddressBook

private XClassifierSet askAgentsInAddressBook(AgentsPopulations population,
                                              Environment env,
                                              java.lang.String state,
                                              XClassifierSet collectedAnswersArray,
                                              int time)

This method asks to each agent in the address book to try to answer the question. The answers are collected in an array collectedAnswersArray. Moreover, the status of agents in the address book are actualized.

Parameters:

population - the population of agents

env - the problem space

state - the signal from the environment

collectedAnswersArray - The array storing the answers received by the asking agent

time - The time counter of the experiment

See Also:

Slot.actualizeTime(), Slot.actualizeCounter(), answer(java.lang.String, XClassifierSet, int), Slot.trackExchange(XClassifier), Book.deleteAddress(Slot), Book.ordering(), Book.transferAddressBook(Agent)

askAgentsAtRandom

private XClassifierSet askAgentsAtRandom(AgentsPopulations population,
                                         java.lang.String state,
                                         Environment env,
                                         XClassifierSet collectedAnswersArray,
                                         int time)

This method asks agents randomly chosen in the population to answer the question. 10% of the whole population is selected. First, an array of randomly chosen agents is built. Second each of these agents are asked to answer and they are added to the address book if they do so. Answers are added to the collectedAnswersArray.

Parameters:

population - the population of agents

state - the signal from the environment

env - the problem space

collectedAnswersArray - The array storing the answers received by the asking agent

time - The time counter of the experiment

See Also:

randomPopulation(AgentsPopulations), XClassifierSet.computeMinDistance(Environment, java.lang.String, double), answer(java.lang.String, XClassifierSet, int), Book.addToAddressBook(Agent)

answer

private XClassifierSet answer(java.lang.String state,
                              XClassifierSet collectedAnswersArray,
                              int time)

The method used by agents to answer the question. First, the asked agent compares the question and the rules it has in its population of classifiers. If some of these rules match, it stores them in an array. Next, if it has some potential answers to the question, it picks up the best one. Then the best answer is transmitted to the questioning agent by being put in the array collecting the answers from all the asked agents.

Parameters:

state - Signal from the environment

time - The time counter of the experiment

See Also:

XClassifier.test(java.lang.String)

selectBestAnswer

private XClassifier selectBestAnswer(XClassifierSet collectedAnswersArray,
                                     java.lang.String state,
                                     int time)

THis method is used by the agent who asks a question to select the best answer from the ones it recieved

Parameters:

collectedAnswersArray - Array Where answers are stored

time - The time counter of the experiment

randomPopulation

private java.util.Vector randomPopulation(AgentsPopulations population)

Construct a random population of agents picked up from the global population. We must consider the case of an address book of size < (total number of agents in population - number of agents asked at random)and address book of size >= (total number of agents in population - number of agents asked at random).

Parameters:

population - The population of agents in the experiment

See Also:

chooseAgents(int,AgentsPopulations)

chooseAgents

private java.util.Vector chooseAgents(int size,
                                      AgentsPopulations population)

Build an array of randomly chosen agents. However, agents must not already be in the address book, must not be itself and an agent can not be drawn twice.

Parameters:

size - The number of agents to be picked up

population - The whole population from which agents will be drawn

See Also:

#checkAgent(in), #checkRand(int,int)

checkRand

private boolean checkRand(int[] t,
                          int i)

Checks that a random number is not drawn twice in one single process

Parameters:

t - The list of agents randomly drawn

i - the latest drawn agent. the method checks that i is not already in the list.

checkAgent

private boolean checkAgent(int ran)

Checks that 1) the agent does not choose itself 2) the agents does not choose an agent already in its address book and then computes whether or not a signal is in the core competence of the agent.

Parameters:

ran - The random number drawn

getCloseState

public boolean getCloseState(java.lang.String state,
                             AgentsPopulations wholePopulation)

The agent computes whether or not the signal from the environment is in its realm of specialisation or not before to tell it to the manager.

Parameters:

state - The current problem from the environment

wholePopulation - the instance of the class AgentsPopulation from where the classs Environment can be reached

getNbCommunications

public int getNbCommunications()

Returns the number of communications made by the agent