Package ai.libs.jaicore.search.algorithms.mdp.mcts.spuct

Class SPUCBPolicy<N,A>

java.lang.Object

ai.libs.jaicore.search.algorithms.mdp.mcts.uct.AUpdatingPolicy<T,A>

ai.libs.jaicore.search.algorithms.mdp.mcts.uct.UCBPolicy<N,A>

ai.libs.jaicore.search.algorithms.mdp.mcts.spuct.SPUCBPolicy<N,A>

All Implemented Interfaces:

IPathUpdatablePolicy<N,A,java.lang.Double>, IPolicy<N,A>, org.api4.java.common.control.ILoggingCustomizable

public class SPUCBPolicy<N,A>
extends UCBPolicy<N,A>
implements org.api4.java.common.control.ILoggingCustomizable

Constructor Summary

Constructors

Constructor	Description
SPUCBPolicy(double gamma, boolean maximize, double bigD)
SPUCBPolicy(double gamma, double bigD)

Method Summary

Modifier and Type	Method	Description
java.lang.String	getLoggerName()
double	getScore(N node, A action)
void	setLoggerName(java.lang.String name)
void	updatePath(org.api4.java.datastructure.graph.ILabeledPath<N,A> path, java.util.List<java.lang.Double> scores)	Note that this is a transposition-based and hence, only partially path-dependent, update.

Methods inherited from class ai.libs.jaicore.search.algorithms.mdp.mcts.uct.UCBPolicy

getActionBasedOnScores, getEmpiricalMean, getExplorationConstant, getExplorationTerm, setExplorationConstant

Methods inherited from class ai.libs.jaicore.search.algorithms.mdp.mcts.uct.AUpdatingPolicy

getAction, getBehaviorWhenActionForNotFullyExploredStateIsRequested, getGamma, getLabelOfNode, isMaximize, setBehaviorWhenActionForNotFullyExploredStateIsRequested

Methods inherited from class java.lang.Object

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

Constructor Detail

SPUCBPolicy

public SPUCBPolicy(double gamma,
                   double bigD)

SPUCBPolicy

public SPUCBPolicy(double gamma,
                   boolean maximize,
                   double bigD)

Method Detail

getLoggerName

public java.lang.String getLoggerName()

Specified by:

getLoggerName in interface org.api4.java.common.control.ILoggingCustomizable

Overrides:

getLoggerName in class UCBPolicy<N,A>

setLoggerName

public void setLoggerName(java.lang.String name)

Specified by:

setLoggerName in interface org.api4.java.common.control.ILoggingCustomizable

Overrides:

setLoggerName in class UCBPolicy<N,A>

updatePath

public void updatePath(org.api4.java.datastructure.graph.ILabeledPath<N,A> path,
                       java.util.List<java.lang.Double> scores)

Description copied from class: AUpdatingPolicy

Note that this is a transposition-based and hence, only partially path-dependent, update. The labels are associated to nodes of the original MDP (states) and not to nodes in the MCTS search tree (paths)! This means that, in fact, several paths are (partially) updated simultanously. However, on all other paths crossing the nodes on the updated paths, only those situations are updated and not the situations in higher nodes of the search tree.

Specified by:

updatePath in interface IPathUpdatablePolicy<N,A,java.lang.Double>

Overrides:

updatePath in class AUpdatingPolicy<N,A>

getScore

public double getScore(N node,
                       A action)

Overrides:

getScore in class UCBPolicy<N,A>