Class Manager

java.lang.Object
  |
  +--Manager

All Implemented Interfaces:

java.io.Serializable

public class Manager

extends java.lang.Object

implements java.io.Serializable

This class is a manager, containing the main loop of the manager's XCS. The XCS is implemented following Martin V. Butz and Steward Wilson (2001) "An algorithmic description of XCS". Moreover, it handles the evaluation of performance.

Since:

JDK 1.3

See Also:

Serialized Form

Field Summary

(package private) int	actionWinner The index of the last best agent
private static XCSConstantsManager	cons Stores the relevant constants for XCS.
private XClassifierSetManager	gpop Stores the current population of XCS.
private XClassifierSetManager	matchSet The Match set for each problem
private static int	nrActions The number of actions of the manager (=total number of agents)
private double	reward The reward the manager receives from the environment for a given answer
private java.lang.String	state the environment state corresponding to the actual problem

Constructor Summary

Manager(int conditionLength, int nrAgents)
Constructs the XCS system of the manager.

Method Summary

int	getActualAgent(java.lang.String actualState, int time, AgentsPopulations wholePopulation, int selector0) The task of the manager is to choose the most qualified agent to fulfill the task submitted by the environment (to answer the problem submitted by the environment).
double	getPerformance() return the reward received by the agent
XClassifierSetManager	getPopulation() Return the population of rules of an agent
int	pickUpAgentAtRandom(AgentsPopulations wholePop, java.lang.String actualState, int time, int selector0) Choose an agent to perform the task at hand.
void	showPopStructure() Display the proportion of each advocated action in the set of rules of the manager
void	updatePop(double reward, int counter) The manager learns from the reward it received from the environment after the chosen agent has answered.

Methods inherited from class java.lang.Object

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

Field Detail

nrActions

private static int nrActions

The number of actions of the manager (=total number of agents)

gpop

private XClassifierSetManager gpop

Stores the current population of XCS.

matchSet

private XClassifierSetManager matchSet

The Match set for each problem

actionWinner

int actionWinner

The index of the last best agent

cons

private static XCSConstantsManager cons

Stores the relevant constants for XCS.

See Also:

XCSConstantsManager

reward

private double reward

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

state

private java.lang.String state

the environment state corresponding to the actual problem

Constructor Detail

Manager

public Manager(int conditionLength,
               int nrAgents)

Constructs the XCS system of the manager.

Parameters:

conditionLength - The bit-string length of the condition parts of classifiers

nrAgents - Total number of agents in the population; also the number of possible actions for the manager.

See Also:

XCSConstantsManager.XCSConstantsManager(), XClassifierSetManager.XClassifierSetManager(int, int)

Method Detail

getPerformance

public double getPerformance()

return the reward received by the agent

getPopulation

public XClassifierSetManager getPopulation()

Return the population of rules of an agent

getActualAgent

public int getActualAgent(java.lang.String actualState,
                          int time,
                          AgentsPopulations wholePopulation,
                          int selector0)

The task of the manager is to choose the most qualified agent to fulfill the task submitted by the environment (to answer the problem submitted by the environment). In this routine, the manager tries to find in its memory the best suited agents (the best classifier).

Parameters:

actualState - The current problem

time - A time counter (nb of experiments)

wholePopulation - Total number of Agents in the system

selector0 - mode of classifier selection by the manager (1->bestAction; 0->rouletteWheel)

See Also:

XClassifierSetManager.XClassifierSetManager(String, XClassifierSetManager, int, AgentsPopulations, int), PredictionArray.PredictionArray(XClassifierSetManager, int), PredictionArray.bestActionWinner(), PredictionArray.rouletteActionWinner()

updatePop

public void updatePop(double reward,
                      int counter)

The manager learns from the reward it received from the environment after the chosen agent has answered.

Parameters:

reward - The reward received from the environment after an answer to the problem has been fired. It is used to update both manager's and agent's classifier

counter - A time counter

See Also:

XClassifierSetManager.XClassifierSetManager(XClassifierSetManager, int), XClassifierSetManager.updateSet(double, double), XClassifierSetManager.runGAManager(int, String, int)

pickUpAgentAtRandom

public int pickUpAgentAtRandom(AgentsPopulations wholePop,
                               java.lang.String actualState,
                               int time,
                               int selector0)

Choose an agent to perform the task at hand. The agent is chosen among those that are unused (low average performance of the agent) and that, at the same time, declares that the problem is in their core domains (closeState=true) . If an agent is chosen, a new rule is inserted in the manager's memory (signal:name of the agent). This routine is thus a sort of directed covering. It is used by the manager when getActualAgent() failed to provide an answer.

Parameters:

wholePop - The total population of agents

actualState - the current problem

time - A time counter

selector0 - mode of classifiers selection

See Also:

Agent.getCloseState(String, AgentsPopulations), XClassifierManager#XClassifierManager(int, int, String, int), XClassifierSetManager.addXClassifierToPopulation(XClassifierManager), XClassifierSetManager.deleteFromPopulation(), XClassifierSetManager.XClassifierSetManager(String, XClassifierSetManager, int, AgentsPopulations, int), PredictionArray.PredictionArray(XClassifierSetManager, int), PredictionArray.bestActionWinner(), PredictionArray#roultetteActionWinner()

showPopStructure

public void showPopStructure()

Display the proportion of each advocated action in the set of rules of the manager