weka.classifiers.trees.m5
Class RuleNode

java.lang.Object
  weka.classifiers.Classifier
      weka.classifiers.trees.m5.RuleNode

All Implemented Interfaces:

java.lang.Cloneable, OptionHandler, java.io.Serializable

public class RuleNode

extends Classifier

Constructs a node for use in an m5 tree or rule

Version:

$Revision: 1.8 $

Author:

Mark Hall (mhall@cs.waikato.ac.nz)

See Also:

Serialized Form

Field Summary

private int	m_classIndex the class index
private double	m_devFraction a node will not be split if its class standard deviation is less than 5% of the class standard deviation of all the instances
private double	m_globalAbsDeviation the absolute deviation of the global class
private double	m_globalDeviation a node will not be split if the class deviation of its instances is less than m_devFraction of the deviation of the global class
private int	m_id Node id.
private int[]	m_indices Indices of the attributes to be used in generating a linear model at this node
private Instances	m_instances instances reaching this node
private boolean	m_isLeaf Node is a leaf
private int	m_leafModelNum the number assigned to the linear model if this node is a leaf
protected RuleNode	m_left child nodes
private PreConstructedLinearModel	m_nodeModel the linear model at this node
private int	m_numAttributes the number of attributes
protected int	m_numInstances the number of instances reaching this node
int	m_numParameters the number of paramters in the chosen model for this node---either the subtree model or the linear model.
private RuleNode	m_parent the parent of this node
private double	m_pruningMultiplier
private boolean	m_regressionTree Make a regression tree instead of a model tree
protected RuleNode	m_right
private double	m_rootMeanSquaredError the mean squared error of the model at this node (either linear or subtree)
private boolean	m_saveInstances Save the instances at each node (for visualizing in the Explorer's treevisualizer.
private int	m_splitAtt attribute this node splits on
private double	m_splitNum a node will not be split if it contains less then m_splitNum instances
private double	m_splitValue the value of the split attribute
private static double	SMOOTHING_CONSTANT Constant used in original m5 smoothing calculation

Fields inherited from class weka.classifiers.Classifier

m_Debug

Constructor Summary

RuleNode(double globalDev, double globalAbsDev, RuleNode parent)
Creates a new RuleNode instance.

Method Summary

protected int	assignIDs(int lastID) Assigns a unique identifier to each node in the tree
private boolean[]	attsTestedAbove() Returns an array containing the indexes of attributes used in tests above this node
private boolean[]	attsTestedBelow() Returns an array containing the indexes of attributes used in tests below this node
void	buildClassifier(Instances data) Build this node (find an attribute and split point)
private void	buildLinearModel(int[] indices) Build a linear model for this node using those attributes specified in indices.
double	classifyInstance(Instance inst) Classify an instance using this node.
void	findBestLeaf(double[] maxCoverage, RuleNode[] bestLeaf) Find the leaf with greatest coverage
double	getMinNumInstances() Get the minimum number of instances to allow at a leaf node
PreConstructedLinearModel	getModel() Get the linear model at this node
int	getNumInstances() Return the number of instances that reach this node.
boolean	getRegressionTree() Get the value of regressionTree.
void	graph(java.lang.StringBuffer text) Assign a unique identifier to each node in the tree and then calls graphTree
protected void	graphTree(java.lang.StringBuffer text) Return a dotty style string describing the tree
void	installLinearModels() Traverses the tree and installs linear models at each node.
void	installSmoothedModels()
boolean	isLeaf() Return true if this node is a leaf
RuleNode	leftNode() Get the left child of this node
java.lang.String	nodeToString() Returns a description of this node (debugging purposes)
int	numberOfLinearModels() Get the number of linear models in the tree
int	numLeaves(int leafCounter) Sets the leaves' numbers
private int	numParameters() Get the number of parameters in the model at this node
RuleNode	parentNode() Get the parent of this node
void	printAllModels() Print all the linear models at the learf (debugging purposes)
java.lang.String	printLeafModels() print all leaf models
java.lang.String	printNodeLinearModel() print the linear model at this node
void	prune() Recursively prune the tree
private double	pruningFactor(int num_instances, int num_params) Compute the pruning factor
void	returnLeaves(FastVector[] v) Return a list containing all the leaves in the tree
RuleNode	rightNode() Get the right child of this node
protected double	rootMeanSquaredError() Get the root mean squared error at this node
void	setMinNumInstances(double minNum) Set the minumum number of instances to allow at a leaf node
void	setRegressionTree(boolean newregressionTree) Set the value of regressionTree.
protected void	setSaveInstances(boolean save) Set whether to save instances for visualization purposes.
protected static double	smoothingOriginal(double n, double pred, double supportPred) Applies the m5 smoothing procedure to a prediction
void	split() Finds an attribute and split point for this node
int	splitAtt() Get the index of the splitting attribute for this node
double	splitVal() Get the split point for this node
java.lang.String	toString() print the linear model at this node
java.lang.String	treeToString(int level) Recursively builds a textual description of the tree

Methods inherited from class weka.classifiers.Classifier

debugTipText, distributionForInstance, forName, getDebug, getOptions, listOptions, makeCopies, setDebug, setOptions

Methods inherited from class java.lang.Object

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

Field Detail

m_instances

private Instances m_instances

instances reaching this node

m_classIndex

private int m_classIndex

the class index

m_numInstances

protected int m_numInstances

the number of instances reaching this node

m_numAttributes

private int m_numAttributes

the number of attributes

m_isLeaf

private boolean m_isLeaf

Node is a leaf

m_splitAtt

private int m_splitAtt

attribute this node splits on

m_splitValue

private double m_splitValue

the value of the split attribute

m_nodeModel

private PreConstructedLinearModel m_nodeModel

the linear model at this node

m_numParameters

public int m_numParameters

the number of paramters in the chosen model for this node---either the subtree model or the linear model. The constant term is counted as a paramter---this is for pruning purposes

m_rootMeanSquaredError

private double m_rootMeanSquaredError

the mean squared error of the model at this node (either linear or subtree)

m_left

protected RuleNode m_left

child nodes

m_right

protected RuleNode m_right

m_parent

private RuleNode m_parent

the parent of this node

m_splitNum

private double m_splitNum

a node will not be split if it contains less then m_splitNum instances

m_devFraction

private double m_devFraction

a node will not be split if its class standard deviation is less than 5% of the class standard deviation of all the instances

m_pruningMultiplier

private double m_pruningMultiplier

m_leafModelNum

private int m_leafModelNum

the number assigned to the linear model if this node is a leaf. = 0 if this node is not a leaf

m_globalDeviation

private double m_globalDeviation

a node will not be split if the class deviation of its instances is less than m_devFraction of the deviation of the global class

m_globalAbsDeviation

private double m_globalAbsDeviation

the absolute deviation of the global class

m_indices

private int[] m_indices

Indices of the attributes to be used in generating a linear model at this node

SMOOTHING_CONSTANT

private static final double SMOOTHING_CONSTANT

Constant used in original m5 smoothing calculation

See Also:

Constant Field Values

m_id

private int m_id

Node id.

m_saveInstances

private boolean m_saveInstances

Save the instances at each node (for visualizing in the Explorer's treevisualizer.

m_regressionTree

private boolean m_regressionTree

Make a regression tree instead of a model tree

Constructor Detail

RuleNode

public RuleNode(double globalDev,
                double globalAbsDev,
                RuleNode parent)

Creates a new RuleNode instance.

Parameters:

globalDev - the global standard deviation of the class

globalAbsDev - the global absolute deviation of the class

parent - the parent of this node

Method Detail

buildClassifier

public void buildClassifier(Instances data)
                     throws java.lang.Exception

Build this node (find an attribute and split point)

Specified by:

buildClassifier in class Classifier

Parameters:

data - the instances on which to build this node

Throws:

java.lang.Exception - if an error occurs

classifyInstance

public double classifyInstance(Instance inst)
                        throws java.lang.Exception

Classify an instance using this node. Recursively calls classifyInstance on child nodes.

Overrides:

classifyInstance in class Classifier

Parameters:

inst - the instance to classify

Returns:

the prediction for this instance

Throws:

java.lang.Exception - if an error occurs

smoothingOriginal

protected static double smoothingOriginal(double n,
                                          double pred,
                                          double supportPred)
                                   throws java.lang.Exception

Applies the m5 smoothing procedure to a prediction

Parameters:

n - number of instances in selected child of this node

pred - the prediction so far

supportPred - the prediction of the linear model at this node

Returns:

the current prediction smoothed with the prediction of the linear model at this node

Throws:

java.lang.Exception - if an error occurs

split

public void split()
           throws java.lang.Exception

Finds an attribute and split point for this node

Throws:

java.lang.Exception - if an error occurs

buildLinearModel

private void buildLinearModel(int[] indices)
                       throws java.lang.Exception

Build a linear model for this node using those attributes specified in indices.

Parameters:

indices - an array of attribute indices to include in the linear model

Throws:

java.lang.Exception

attsTestedAbove

private boolean[] attsTestedAbove()

Returns an array containing the indexes of attributes used in tests above this node

Returns:

an array of attribute indexes

attsTestedBelow

private boolean[] attsTestedBelow()

Returns an array containing the indexes of attributes used in tests below this node

Returns:

an array of attribute indexes

numLeaves

public int numLeaves(int leafCounter)

Sets the leaves' numbers

Parameters:

leafCounter - the number of leaves counted

Returns:

the number of the total leaves under the node

toString

public java.lang.String toString()

print the linear model at this node

printNodeLinearModel

public java.lang.String printNodeLinearModel()

print the linear model at this node

printLeafModels

public java.lang.String printLeafModels()

print all leaf models

nodeToString

public java.lang.String nodeToString()

Returns a description of this node (debugging purposes)

Returns:

a string describing this node

treeToString

public java.lang.String treeToString(int level)

Recursively builds a textual description of the tree

Parameters:

level - the level of this node

Returns:

string describing the tree

installLinearModels

public void installLinearModels()
                         throws java.lang.Exception

Traverses the tree and installs linear models at each node. This method must be called if pruning is not to be performed.

Throws:

java.lang.Exception - if an error occurs

installSmoothedModels

public void installSmoothedModels()
                           throws java.lang.Exception

Throws:

java.lang.Exception

prune

public void prune()
           throws java.lang.Exception

Recursively prune the tree

Throws:

java.lang.Exception - if an error occurs

pruningFactor

private double pruningFactor(int num_instances,
                             int num_params)

Compute the pruning factor

Parameters:

num_instances - number of instances

num_params - number of parameters in the model

Returns:

the pruning factor

findBestLeaf

public void findBestLeaf(double[] maxCoverage,
                         RuleNode[] bestLeaf)

Find the leaf with greatest coverage

Parameters:

maxCoverage - the greatest coverage found so far

bestLeaf - the leaf with the greatest coverage

returnLeaves

public void returnLeaves(FastVector[] v)

Return a list containing all the leaves in the tree

Parameters:

v - a single element array containing a vector of leaves

parentNode

public RuleNode parentNode()

Get the parent of this node

Returns:

the parent of this node

leftNode

public RuleNode leftNode()

Get the left child of this node

Returns:

the left child of this node

rightNode

public RuleNode rightNode()

Get the right child of this node

Returns:

the right child of this node

splitAtt

public int splitAtt()

Get the index of the splitting attribute for this node

Returns:

the index of the splitting attribute

splitVal

public double splitVal()

Get the split point for this node

Returns:

the split point for this node

numberOfLinearModels

public int numberOfLinearModels()

Get the number of linear models in the tree

Returns:

the number of linear models

isLeaf

public boolean isLeaf()

Return true if this node is a leaf

Returns:

true if this node is a leaf

rootMeanSquaredError

protected double rootMeanSquaredError()

Get the root mean squared error at this node

Returns:

the root mean squared error

getModel

public PreConstructedLinearModel getModel()

Get the linear model at this node

Returns:

the linear model at this node

getNumInstances

public int getNumInstances()

Return the number of instances that reach this node.

Returns:

the number of instances at this node.

numParameters

private int numParameters()

Get the number of parameters in the model at this node

Returns:

the number of parameters in the model at this node

getRegressionTree

public boolean getRegressionTree()

Get the value of regressionTree.

Returns:

Value of regressionTree.

setMinNumInstances

public void setMinNumInstances(double minNum)

Set the minumum number of instances to allow at a leaf node

Parameters:

minNum - the minimum number of instances

getMinNumInstances

public double getMinNumInstances()

Get the minimum number of instances to allow at a leaf node

Returns:

a double value

setRegressionTree

public void setRegressionTree(boolean newregressionTree)

Set the value of regressionTree.

Parameters:

newregressionTree - Value to assign to regressionTree.

printAllModels

public void printAllModels()

Print all the linear models at the learf (debugging purposes)

assignIDs

protected int assignIDs(int lastID)

Assigns a unique identifier to each node in the tree

Parameters:

lastID - last id number used

Returns:

ID after processing child nodes

graph

public void graph(java.lang.StringBuffer text)

Assign a unique identifier to each node in the tree and then calls graphTree

Parameters:

text - a StringBuffer value

graphTree

protected void graphTree(java.lang.StringBuffer text)

Return a dotty style string describing the tree

Parameters:

text - a StringBuffer value

setSaveInstances

protected void setSaveInstances(boolean save)

Set whether to save instances for visualization purposes. Default is to save memory.

Parameters:

save - a boolean value