tracks.singlePlayer.advanced.olets

Class SingleTreeNode

java.lang.Object

tracks.singlePlayer.advanced.olets.SingleTreeNode

public class SingleTreeNode
extends java.lang.Object

Code written by Adrien Couetoux, acouetoux@ulg.ac.be. Date: 15/12/2015

Author:

Adrien Couëtoux

Field Summary

Fields

Modifier and Type	Field and Description
SingleTreeNode[]	children The array that stores the children nodes.
double	epsilon Epsilon used for breaking ties
double	K Constant used in exploration part of UCB formulas
int	nodeDepth Depth of the node in the tree
int	num_actions Number of actions in this node.
SingleTreeNode	parent The parent node (is null for the root node of a tree)

Constructor Summary

Constructors

Constructor and Description
SingleTreeNode(int num_actions) Public constructor for nodes with no declared parent node (eg. for a root node)
SingleTreeNode(SingleTreeNode parent, int depth, int actionIndex, double tabooBias, int num_actions) Public constructor for nodes with a parent node

Method Summary

Modifier and Type	Method and Description
void	backUp(SingleTreeNode node, double result) Backtracks along the visited branch of the tree, to update the stored data, including the expectimax values
int	getActionIndex()
int	getNbGenerated()
int	getNodeDepth()
int	mostVisitedAction() Finds the action that was selected the most times
boolean	notFullyExpanded() Checks if the current node is fully expanded, i.e. if all actions have been selected at least once.
void	refreshTree() Updates nodes attributes in a tree; mostly used to reset number of simulations to 1 to reduce the weight of past simulations when salvaging a tree branch from one time step to the next
SingleTreeNode	selectChild() Selects a child node, from the current node.
void	setNodeDepth(int newDepth)
void	setTabooBias(double tabooBias)

Methods inherited from class java.lang.Object

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

Field Detail

K

public final double K

Constant used in exploration part of UCB formulas

parent

public SingleTreeNode parent

The parent node (is null for the root node of a tree)

children

public final SingleTreeNode[] children

The array that stores the children nodes. It should be of fixed size equal to the number of actions

nodeDepth

public int nodeDepth

Depth of the node in the tree

epsilon

public double epsilon

Epsilon used for breaking ties

num_actions

public int num_actions

Number of actions in this node.

Constructor Detail

SingleTreeNode

public SingleTreeNode(int num_actions)

Public constructor for nodes with no declared parent node (eg. for a root node)

SingleTreeNode

public SingleTreeNode(SingleTreeNode parent,
                      int depth,
                      int actionIndex,
                      double tabooBias,
                      int num_actions)

Public constructor for nodes with a parent node

Parameters:

parent - the parent node

depth - the tree depth at which the node is added

actionIndex - the index of the action that was chosen immediately before creating this node

tabooBias - the location bias of this node, computed based on the avatar location

Method Detail

setNodeDepth

public void setNodeDepth(int newDepth)

getNodeDepth

public int getNodeDepth()

getActionIndex

public int getActionIndex()

getNbGenerated

public int getNbGenerated()

setTabooBias

public void setTabooBias(double tabooBias)

refreshTree

public void refreshTree()

Updates nodes attributes in a tree; mostly used to reset number of simulations to 1 to reduce the weight of past simulations when salvaging a tree branch from one time step to the next

backUp

public void backUp(SingleTreeNode node,
                   double result)

Backtracks along the visited branch of the tree, to update the stored data, including the expectimax values

Parameters:

node - the initial node of the backup (usually a tree leaf)

result - the value measured before back tracking (eg. the score when the simulation ended)

selectChild

public SingleTreeNode selectChild()

Selects a child node, from the current node. It currently selects based on an epsilon-greedy, the greedy part being made according to adjusted expectimax values

Returns:

the selected child node

mostVisitedAction

public int mostVisitedAction()

Finds the action that was selected the most times

Returns:

the most selected action from the current node

notFullyExpanded

public boolean notFullyExpanded()

Checks if the current node is fully expanded, i.e. if all actions have been selected at least once.

Returns:

true if there is an action that has not been tried yet, false otherwise.