burlap.domain.stochasticgames.gridgame

Class GridGame.GGJointRewardFunction

java.lang.Object

burlap.domain.stochasticgames.gridgame.GridGame.GGJointRewardFunction

All Implemented Interfaces:

JointRewardFunction

Enclosing class:

GridGame

public static class GridGame.GGJointRewardFunction
extends java.lang.Object
implements JointRewardFunction

Specifies goal rewards and default rewards for agents. Defaults rewards to 0 reward everywhere except transition to unviersal or personal goals which return a reward 1.

Author:

James MacGlashan

Constructor Summary

Constructors

Constructor and Description
GGJointRewardFunction(OODomain ggDomain) Initializes for a given domain.
GGJointRewardFunction(OODomain ggDomain, double stepCost, double goalReward, boolean noopIncursStepCost) Initializes for a given domain, step cost reward and goal reward.
GGJointRewardFunction(OODomain ggDomain, double stepCost, double universalGoalReward, boolean noopIncursStepCost, java.util.Map<java.lang.Integer,java.lang.Double> personalGoalRewards) Initializes for a given domain, step cost reward, universal goal reward, and unique personal goal reward for each player.
GGJointRewardFunction(OODomain ggDomain, double stepCost, double personalGoalReward, double universalGoalReward, boolean noopIncursStepCost) Initializes for a given domain, step cost reward, personal goal reward, and universal goal reward.

Method Summary

Modifier and Type	Method and Description
protected double	defaultCost(int aid, JointAction ja) Returns a default cost for an agent assuming the agent didn't transition to a goal state.
protected double	getPersonalGoalReward(OOState s, java.lang.String agentName) Returns the personal goal rewards.
double[]	reward(State s, JointAction ja, State sp) Returns the reward received by each agent specified in the joint action.

Methods inherited from class java.lang.Object

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

Constructor Detail

GGJointRewardFunction

public GGJointRewardFunction(OODomain ggDomain)

Initializes for a given domain. Defaults rewards to 0 reward everywhere except transition to unviersal or personal goals which return a reward 1.

Parameters:

ggDomain - the domain

GGJointRewardFunction

public GGJointRewardFunction(OODomain ggDomain,
                             double stepCost,
                             double goalReward,
                             boolean noopIncursStepCost)

Initializes for a given domain, step cost reward and goal reward.

Parameters:

ggDomain - the domain

stepCost - the reward returned for all transitions except transtions to goal locations

goalReward - the reward returned for transitioning to a personal or universal goal

noopIncursStepCost - if true, then noop actions also incur the stepCost reward; if false, then noops always return 0 reward.

GGJointRewardFunction

public GGJointRewardFunction(OODomain ggDomain,
                             double stepCost,
                             double personalGoalReward,
                             double universalGoalReward,
                             boolean noopIncursStepCost)

Initializes for a given domain, step cost reward, personal goal reward, and universal goal reward.

Parameters:

ggDomain - the domain

stepCost - the reward returned for all transitions except transtions to goal locations

personalGoalReward - the reward returned for transitions to a personal goal

universalGoalReward - the reward returned for transitions to a universal goal

noopIncursStepCost - if true, then noop actions also incur the stepCost reward; if false, then noops always return 0 reward.

GGJointRewardFunction

public GGJointRewardFunction(OODomain ggDomain,
                             double stepCost,
                             double universalGoalReward,
                             boolean noopIncursStepCost,
                             java.util.Map<java.lang.Integer,java.lang.Double> personalGoalRewards)

Initializes for a given domain, step cost reward, universal goal reward, and unique personal goal reward for each player.

Parameters:

ggDomain - the domain

stepCost - the reward returned for all transitions except transtions to goal locations

universalGoalReward - the reward returned for transitions to a universal goal

noopIncursStepCost - if true, then noop actions also incur the stepCost reward; if false, then noops always return 0 reward.

personalGoalRewards - a map from player numbers to their personal goal reward (the first player number is 0)

Method Detail

reward

public double[] reward(State s,
                       JointAction ja,
                       State sp)

Description copied from interface: JointRewardFunction

Returns the reward received by each agent specified in the joint action.

Specified by:

reward in interface JointRewardFunction

Parameters:

s - that state in which the joint action was taken.

ja - the joint action taken.

sp - the resulting state from taking the joint action

Returns:

a Map from agent names to the reward that they received.

defaultCost

protected double defaultCost(int aid,
                             JointAction ja)

Returns a default cost for an agent assuming the agent didn't transition to a goal state. If noops incur step cost, then this is always the step cost. If noops do not incur step costs and the agent took a noop, then 0 is returned.

Parameters:

aid - the agent of interest

ja - the joint action set

Returns:

the default reward; either step cost or 0.

getPersonalGoalReward

protected double getPersonalGoalReward(OOState s,
                                       java.lang.String agentName)

Returns the personal goal rewards. If a single common personal goal reward was set then that is returned. If different personal goal rewards were defined for each player number, then that is queried and returned instead.

Parameters:

s - the state in which the agent player numbers are defined

agentName - the agent name for which the person goal reward is to be returned

Returns:

the personal goal reward for the specified agent.