Java EGreedyMaxWellfare类代码示例

import burlap.behavior.stochasticgames.madynamicprogramming.policies.EGreedyMaxWellfare; //导入依赖的package包/类
/**
 * Initializes this Q-learning agent. This agent's Q-source will use a {@link HashBackedQSource} q-source and the learning policy is defaulted
 * to an epsilon = 0.1 maximum wellfare ({@link EGreedyMaxWellfare}) derived policy. If queryOtherAgentsForTheirQValues is set to true, then this agent will
 * only store its own Q-values and will use the other agent's stored Q-values to determine theirs.
 * @param d the domain in which to perform learing
 * @param discount the discount factor
 * @param learningRate the constant learning rate
 * @param hashFactory the hashing factory used to index states and Q-values
 * @param qInit the default Q-value to which all initial Q-values will be initialized
 * @param backupOperator the backup operator to use that defines the solution concept being learned
 * @param queryOtherAgentsForTheirQValues it true, then the agent uses the Q-values for other agents that are stored by them; if false then the agent stores a Q-value for each other agent in the world.
 */
public MultiAgentQLearning(SGDomain d, double discount, double learningRate, HashableStateFactory hashFactory, double qInit, SGBackupOperator backupOperator, boolean queryOtherAgentsForTheirQValues){
	this.init(d);
	this.discount = discount;
	this.learningRate = new ConstantLR(learningRate);
	this.hashingFactory = hashFactory;
	this.qInit = new ValueFunctionInitialization.ConstantValueFunctionInitialization(qInit);
	this.backupOperator = backupOperator;
	this.queryOtherAgentsQSource = queryOtherAgentsForTheirQValues;
	
	this.myQSource = new QSourceForSingleAgent.HashBackedQSource(this.hashingFactory, this.qInit);
	
	this.learningPolicy = new PolicyFromJointPolicy(new EGreedyMaxWellfare(this, 0.1));
}
 

import burlap.behavior.stochasticgames.madynamicprogramming.policies.EGreedyMaxWellfare; //导入依赖的package包/类
/**
 * Initializes. The policy will be defaulted to a epsilon-greedy max welfare policy.
 * @param d the domain in which to perform learing
 * @param discount the discount factor
 * @param learningRate the constant learning rate
 * @param hashFactory the hashing factory used to index states and Q-values
 * @param qInit the default Q-value to which all initial Q-values will be initialized
 * @param backupOperator the backup operator to use that defines the solution concept being learned
 * @param queryOtherAgentsForTheirQValues it true, then the agent uses the Q-values for other agents that are stored by them; if false then the agent stores a Q-value for each other agent in the world.
 */
public MAQLFactory(SGDomain d, double discount, double learningRate, HashableStateFactory hashFactory, double qInit, SGBackupOperator backupOperator, boolean queryOtherAgentsForTheirQValues){
	this.domain = d;
	this.discount = discount;
	this.learningRate = new ConstantLR(learningRate);
	this.hashingFactory = hashFactory;
	this.qInit = new ValueFunctionInitialization.ConstantValueFunctionInitialization(qInit);
	this.backupOperator = backupOperator;
	this.queryOtherAgentsQSource = queryOtherAgentsForTheirQValues;
	this.learningPolicy = new PolicyFromJointPolicy(new EGreedyMaxWellfare(0.1));
}
 

import burlap.behavior.stochasticgames.madynamicprogramming.policies.EGreedyMaxWellfare; //导入依赖的package包/类
/**
 * Initializes. The policy will be defaulted to a epsilon-greedy max welfare policy.
 * @param d the domain in which to perform learing
 * @param discount the discount factor
 * @param learningRate the constant learning rate
 * @param hashFactory the hashing factory used to index states and Q-values
 * @param qInit the default Q-value to which all initial Q-values will be initialized
 * @param backupOperator the backup operator to use that defines the solution concept being learned
 * @param queryOtherAgentsForTheirQValues it true, then the agent uses the Q-values for other agents that are stored by them; if false then the agent stores a Q-value for each other agent in the world.
 */
public MAQLFactory(SGDomain d, double discount, double learningRate, HashableStateFactory hashFactory, double qInit, SGBackupOperator backupOperator, boolean queryOtherAgentsForTheirQValues){
	this.domain = d;
	this.discount = discount;
	this.learningRate = new ConstantLR(learningRate);
	this.hashingFactory = hashFactory;
	this.qInit = new ConstantValueFunction(qInit);
	this.backupOperator = backupOperator;
	this.queryOtherAgentsQSource = queryOtherAgentsForTheirQValues;
	this.learningPolicy = new PolicyFromJointPolicy(new EGreedyMaxWellfare(0.1));
}
 

import burlap.behavior.stochasticgames.madynamicprogramming.policies.EGreedyMaxWellfare; //导入依赖的package包/类
public CoCoQLearningFactory(SGDomain d, double discount, LearningRate learningRate, HashableStateFactory hashFactory, ValueFunctionInitialization qInit, boolean queryOtherAgentsForTheirQValues, double epsilon){
	this.init(d, discount, learningRate, hashFactory, qInit, new CoCoQ(), queryOtherAgentsForTheirQValues, new PolicyFromJointPolicy(new EGreedyMaxWellfare(epsilon)));
}
 

import burlap.behavior.stochasticgames.madynamicprogramming.policies.EGreedyMaxWellfare; //导入依赖的package包/类
public static void VICoCoTest(){

		//grid game domain
		GridGame gridGame = new GridGame();
		final OOSGDomain domain = gridGame.generateDomain();

		final HashableStateFactory hashingFactory = new SimpleHashableStateFactory();

		//run the grid game version of prisoner's dilemma
		final State s = GridGame.getPrisonersDilemmaInitialState();

		//define joint reward function and termination conditions for this game
		JointRewardFunction rf = new GridGame.GGJointRewardFunction(domain, -1, 100, false);
		TerminalFunction tf = new GridGame.GGTerminalFunction(domain);

		//both agents are standard: access to all actions
		SGAgentType at = GridGame.getStandardGridGameAgentType(domain);

		//create our multi-agent planner
		MAValueIteration vi = new MAValueIteration(domain, rf, tf, 0.99, hashingFactory, 0., new CoCoQ(), 0.00015, 50);

		//instantiate a world in which our agents will play
		World w = new World(domain, rf, tf, s);


		//create a greedy joint policy from our planner's Q-values
		EGreedyMaxWellfare jp0 = new EGreedyMaxWellfare(0.);
		jp0.setBreakTiesRandomly(false); //don't break ties randomly

		//create agents that follows their end of the computed the joint policy
		MultiAgentDPPlanningAgent a0 = new MultiAgentDPPlanningAgent(domain, vi, new PolicyFromJointPolicy(0, jp0), "agent0", at);
		MultiAgentDPPlanningAgent a1 = new MultiAgentDPPlanningAgent(domain, vi, new PolicyFromJointPolicy(1, jp0), "agent1", at);

		w.join(a0);
		w.join(a1);

		//run some games of the agents playing that policy
		GameEpisode ga = null;
		for(int i = 0; i < 3; i++){
			ga = w.runGame();
		}

		//visualize results
		Visualizer v = GGVisualizer.getVisualizer(9, 9);
		new GameSequenceVisualizer(v, domain, Arrays.asList(ga));


	}
 

import burlap.behavior.stochasticgames.madynamicprogramming.policies.EGreedyMaxWellfare; //导入依赖的package包/类
public CoCoQLearningFactory(SGDomain d, double discount, LearningRate learningRate, HashableStateFactory hashFactory, QFunction qInit, boolean queryOtherAgentsForTheirQValues, double epsilon){
	this.init(d, discount, learningRate, hashFactory, qInit, new CoCoQ(), queryOtherAgentsForTheirQValues, new PolicyFromJointPolicy(new EGreedyMaxWellfare(epsilon)));
}
 

import burlap.behavior.stochasticgames.madynamicprogramming.policies.EGreedyMaxWellfare; //导入依赖的package包/类
/**
 * Initializes this Q-learning agent. This agent's Q-source will use a {@link HashBackedQSource} q-source and the learning policy is defaulted
 * to an epsilon = 0.1 maximum wellfare ({@link EGreedyMaxWellfare}) derived policy. If queryOtherAgentsForTheirQValues is set to true, then this agent will
 * only store its own Q-values and will use the other agent's stored Q-values to determine theirs.
 * @param d the domain in which to perform learing
 * @param discount the discount factor
 * @param learningRate the constant learning rate
 * @param hashFactory the hashing factory used to index states and Q-values
 * @param qInit the default Q-value to which all initial Q-values will be initialized
 * @param backupOperator the backup operator to use that defines the solution concept being learned
 * @param queryOtherAgentsForTheirQValues it true, then the agent uses the Q-values for other agents that are stored by them; if false then the agent stores a Q-value for each other agent in the world.
 * @param agentName the name of the agent
 * @param agentType the {@link SGAgentType} for the agent defining its action space
 */
public MultiAgentQLearning(SGDomain d, double discount, double learningRate, HashableStateFactory hashFactory, double qInit, SGBackupOperator backupOperator, boolean queryOtherAgentsForTheirQValues, String agentName, SGAgentType agentType){
	this.init(d, agentName, agentType);
	this.discount = discount;
	this.learningRate = new ConstantLR(learningRate);
	this.hashingFactory = hashFactory;
	this.qInit = new ConstantValueFunction(qInit);
	this.backupOperator = backupOperator;
	this.queryOtherAgentsQSource = queryOtherAgentsForTheirQValues;
	
	this.myQSource = new QSourceForSingleAgent.HashBackedQSource(this.hashingFactory, this.qInit);
	
	this.learningPolicy = new PolicyFromJointPolicy(new EGreedyMaxWellfare(this, 0.1));
}