Class AgentsPopulations

java.lang.Object
  |
  +--AgentsPopulations

All Implemented Interfaces:

java.io.Serializable

public class AgentsPopulations

extends java.lang.Object

implements java.io.Serializable

This is the class where the population of agents are initialized. It is also here that the main loop of each experiment is located. In addition, various computations for graphs analysis are carried out here.

Since:

JDK 1.3

See Also:

Serialized Form

Field Summary

private Agent[]	agentSet The list of agents
private java.util.Vector	agSet The created agents list for purpose of communication
double	averClique Average cliquishness in the Agents population, for each problem
int	communication Communication between agents is allowed (=1) or not (=0)
private static XCSConstants	cons Stores the relevant constants for XCS.
double	consultRatio Percentage of the population to be consulted by the agents
double	coreCompetence Competence range of the agents (determines their core specialisations)
Environment	env Stores the posed problem.
private int	maxProblems Specifies the number of exploration problems/trials to solve in one experiment by one agent.
private int	nrExps Specifies the number of investigated experiments for each agent.
(package private) int[]	orderAgentsTable The order in which agents receive the problem
(package private) java.lang.String	orgaType Type of the organization studied m= manager c= community
int	payoffType Structure of the reward function in the environment
(package private) double	performance Performance one agent gains in answering a signal from the problem space
private double	performanceCP the cumulated performance of the system
java.lang.String	problemType The type of the environment, i.e. the problem space
(package private) java.io.PrintWriter	pwCP The output file
(package private) java.util.Random	R Random generator
(package private) int	run Number of runs of the total simulation
(package private) double	saveProb Probabilities for data to be saved for statistical treatment
int	selectorAgent Selection in the PredictionArray for each Agent: 1: bestActionWinner 0: rouletteActionWinner
int	totalCommunications The total number of communications in the population for each problem.
private static int	totalNumberOfAgents the number of agents constituting the population
private static AgentsPopulations	wholePopulation The total population of agents

Constructor Summary

AgentsPopulations(int nbAgents, double competence, double consultRatio0, int selectAgent0, int communicate, int payoffType0, java.lang.String problem, int run0, double saveProb0, java.io.PrintWriter pwCP0, int maxProblems0)
Constructor for the initial organisation

Method Summary

private void	affiche(int[][] matrix, java.io.PrintWriter pW1) Prints the matrix of relationships into the outfile
private int[][]	buildRelationships(int[][] mat) Constructs the matrix of the relationships between agents stored in the out file at the end of each experiment
private double	computeCliquishness(int[][] mat) Computes the average cliquishness of the graph derived from the matrix of relationships
private double	computePerformance(double[] perfArray) Computes the average performance of the population of agents based on the performance of the responding agents at each signal.
private int	computeSize(int[] neighborhood) Computes the size of the neignborhood of an agent
private int	countEdges(int[][] mat, int[] clus, int size) Count the number of links existing between the agents belonging to the same neignborhood
(package private) void	doOneSingleStepExperiment() Executes one single-step experiment.
Agent[]	getAgentSet() return the population of agents considered
java.util.Vector	getSet() return the population of agents considered during a communication process
private int[]	listNeighbors(int[][] mat, int[] neighborhood, int i) Builds the list of first neighbors of an agent.
void	run() The loop where the population of agents is generated and the experiment is launched
void	runExperiment(Environment envt) Runs the posed problem with XCS.
void	setNumberOfExperiments(int exps) Resets the number of experiments for each agent.
void	setNumberOfTrials(int trials) Resets the maximal number of trials in one experiment for an agent.
private void	shakeAgentsOrder(int nbAgents) Method used to build at each experiment a random list.
private void	startExperiments() This function runs the number of experiments specified.

Methods inherited from class java.lang.Object

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

Field Detail

maxProblems

private int maxProblems

Specifies the number of exploration problems/trials to solve in one experiment by one agent.

nrExps

private int nrExps

Specifies the number of investigated experiments for each agent.

wholePopulation

private static AgentsPopulations wholePopulation

The total population of agents

totalNumberOfAgents

private static int totalNumberOfAgents

the number of agents constituting the population

performanceCP

private double performanceCP

the cumulated performance of the system

agentSet

private Agent[] agentSet

The list of agents

agSet

private java.util.Vector agSet

The created agents list for purpose of communication

env

public Environment env

Stores the posed problem.

cons

private static XCSConstants cons

Stores the relevant constants for XCS.

See Also:

XCSConstants

communication

public int communication

Communication between agents is allowed (=1) or not (=0)

coreCompetence

public double coreCompetence

Competence range of the agents (determines their core specialisations)

consultRatio

public double consultRatio

Percentage of the population to be consulted by the agents

selectorAgent

public int selectorAgent

Selection in the PredictionArray for each Agent: 1: bestActionWinner 0: rouletteActionWinner

payoffType

public int payoffType

Structure of the reward function in the environment

problemType

public java.lang.String problemType

The type of the environment, i.e. the problem space

totalCommunications

public int totalCommunications

The total number of communications in the population for each problem.

averClique

public double averClique

Average cliquishness in the Agents population, for each problem

performance

double performance

Performance one agent gains in answering a signal from the problem space

run

int run

Number of runs of the total simulation

orgaType

java.lang.String orgaType

Type of the organization studied m= manager c= community

pwCP

java.io.PrintWriter pwCP

The output file

orderAgentsTable

int[] orderAgentsTable

The order in which agents receive the problem

saveProb

double saveProb

Probabilities for data to be saved for statistical treatment

R

java.util.Random R

Random generator

Constructor Detail

AgentsPopulations

public AgentsPopulations(int nbAgents,
                         double competence,
                         double consultRatio0,
                         int selectAgent0,
                         int communicate,
                         int payoffType0,
                         java.lang.String problem,
                         int run0,
                         double saveProb0,
                         java.io.PrintWriter pwCP0,
                         int maxProblems0)

Constructor for the initial organisation

Parameters:

nbAgents - Number of Agents (except manager) used in the experiments

competence - Size of the domain of specialisation of agents

consultRatio0 - Percentage of agents' population that is asked at random

selectAgent0 - Mode of rule selection used by agents

communicate - Whether (1) or not (0) communication among agents is allowed

payoffType0 - Structure of the reward function of the environment

problem - The type of problem space

run0 - Number of runs

saveProb0 - Probability of data to be saved

pwCP0 - Name of the first file where results are stored

maxProblems0 - Number of experiments in one run

Method Detail

run

public void run()

The loop where the population of agents is generated and the experiment is launched

getAgentSet

public Agent[] getAgentSet()

return the population of agents considered

getSet

public java.util.Vector getSet()

return the population of agents considered during a communication process

setNumberOfTrials

public void setNumberOfTrials(int trials)

Resets the maximal number of trials in one experiment for an agent. (The default is set to 15000)

Parameters:

trials - The number of problems submitted to agents during one experiment

setNumberOfExperiments

public void setNumberOfExperiments(int exps)

Resets the number of experiments for each agent. (The default is set to 500)

Parameters:

exps - The number of experiments

runExperiment

public void runExperiment(Environment envt)

Runs the posed problem with XCS. The function essentially initializes the output File and then runs the experiments.

Parameters:

envt - The problem space

See Also:

startExperiments()

startExperiments

private void startExperiments()

This function runs the number of experiments specified. After the initialization of the empty population and of the manager, the problem solving routine by the organisation is executed. The problem addressed here is actually a one-step problem.

See Also:

Agent.Agent(double,double), Book#Book(int, int), XClassifierSet.XClassifierSet(int, int), Agent.setAddressBook(Book), Agent.setXClassifierSet(XClassifierSet), Agent.setName(int), doOneSingleStepExperiment()

doOneSingleStepExperiment

void doOneSingleStepExperiment()

Executes one single-step experiment. This is the loop actually monitoring the performance. In addition, performances, efficiencies and some indicators of the graphs produced are computed after matrix of relationships between agents have been built.

See Also:

Square.resetState(), Agent.doOneSingleStepProblemExplore(String, int, AgentsPopulations, int), buildRelationships(int[][]), affiche(int[][], PrintWriter), computeCliquishness(int[][]), computePerformance(double[]), XClassifierSet.printCharacteristics(PrintWriter)

buildRelationships

private int[][] buildRelationships(int[][] mat)

Constructs the matrix of the relationships between agents stored in the out file at the end of each experiment

Parameters:

mat - the matrix initialized with 0s

affiche

private void affiche(int[][] matrix,
                     java.io.PrintWriter pW1)

Prints the matrix of relationships into the outfile

Parameters:

matrix - The matrix of relationships

pW1 - The output file

computeCliquishness

private double computeCliquishness(int[][] mat)

Computes the average cliquishness of the graph derived from the matrix of relationships

Parameters:

mat - the matrix of links (or edges)

See Also:

#listNeighbors(int[][],int,int), #computeSize(int), #countEdges(int[][],int,int)

listNeighbors

private int[] listNeighbors(int[][] mat,
                            int[] neighborhood,
                            int i)

Builds the list of first neighbors of an agent. That is, for an agent i, the routine builds the set of j such that d(i,j)=1 or d(j,i)=1

Parameters:

mat - the matrix of relationships

neighborhood - the initialized neighborhood with all values set to -1

i - the agent of which we seek the neighborhood

computeSize

private int computeSize(int[] neighborhood)

Computes the size of the neignborhood of an agent

Parameters:

neighborhood - The list of first neignbors of the agent considered

countEdges

private int countEdges(int[][] mat,
                       int[] clus,
                       int size)

Count the number of links existing between the agents belonging to the same neignborhood

Parameters:

mat - The matrix of relationships

clus - The list of first neighbors considered

size - The number of neighbors in the neignborhood considered

computePerformance

private double computePerformance(double[] perfArray)

Computes the average performance of the population of agents based on the performance of the responding agents at each signal.

Parameters:

perfArray - The array storing the performance of agents

shakeAgentsOrder

private void shakeAgentsOrder(int nbAgents)

Method used to build at each experiment a random list. When an agent cannot answer a problem (by itself or by communicating), it passes it to the next agent in this list.