weka.classifiers.trees

Class REPTree.Tree

java.lang.Object

weka.classifiers.trees.REPTree.Tree

All Implemented Interfaces:

java.io.Serializable, RevisionHandler

Enclosing class:

REPTree

protected class REPTree.Tree
extends java.lang.Object
implements java.io.Serializable, RevisionHandler

An inner class for building and storing the tree structure

See Also:

Serialized Form

Field Summary

Fields

Modifier and Type	Field and Description
protected int	m_Attribute The attribute to split on.
protected double[]	m_ClassProbs Class probabilities from the training data in the nominal case.
protected double[]	m_Distribution The (unnormalized) class distribution in the nominal case.
protected double[]	m_HoldOutDist Class distribution of hold-out set at node in the nominal case.
protected double	m_HoldOutError The hold-out error of the node.
protected Instances	m_Info The header information (for printing the tree).
protected double[]	m_Prop The proportions of training instances going down each branch.
protected double	m_SplitPoint The split point.
protected REPTree.Tree[]	m_Successors The subtrees of this tree.

Constructor Summary

Constructors

Modifier	Constructor and Description
protected	Tree()

Method Summary

Modifier and Type	Method and Description
protected void	backfitHoldOutSet() Backfits data from holdout set.
protected void	buildTree(int[][][] sortedIndices, double[][][] weights, Instances data, double totalWeight, double[] classProbs, Instances header, double minNum, double minVariance, int depth, int maxDepth) Recursively generates a tree.
protected double	distribution(double[][] props, double[][][] dists, int att, int[] sortedIndices, double[] weights, double[][] subsetWeights, Instances data) Computes class distribution for an attribute.
protected double[]	distributionForInstance(Instance instance) Computes class distribution of an instance using the tree.
protected void	doSmoothing() Smoothes class probabilities stored at node.
protected double	gain(double[][] dist, double priorVal) Computes value of splitting criterion after split.
java.lang.String	getRevision() Returns the revision string.
protected void	insertHoldOutInstance(Instance inst, double weight, REPTree.Tree parent) Inserts an instance from the hold-out set into the tree.
protected void	insertHoldOutSet(Instances data) Inserts hold-out set into tree.
protected java.lang.String	leafString(REPTree.Tree parent) Outputs description of a leaf node.
protected double	numericDistribution(double[][] props, double[][][] dists, int att, int[] sortedIndices, double[] weights, double[][] subsetWeights, Instances data, double[] vals) Computes class distribution for an attribute.
protected int	numNodes() Computes size of the tree.
protected double	priorVal(double[][] dist) Computes value of splitting criterion before split.
protected double	reducedErrorPrune() Prunes the tree using the hold-out data (bottom-up).
protected double	singleVariance(double s, double sS, double weight) Computes the variance for a single set
java.lang.String	sourceExpression(int index) Returns a string containing java source code equivalent to the test made at this node.
protected void	splitData(int[][][][] subsetIndices, double[][][][] subsetWeights, int att, double splitPoint, int[][] sortedIndices, double[][] weights, Instances data) Splits instances into subsets.
protected int	toGraph(java.lang.StringBuffer text, int num, REPTree.Tree parent) Outputs one node for graph.
java.lang.StringBuffer[]	toSource(java.lang.String className, REPTree.Tree parent) Returns source code for the tree as if-then statements.
protected java.lang.String	toString(int level, REPTree.Tree parent) Recursively outputs the tree.
protected double	variance(double[] s, double[] sS, double[] sumOfWeights) Computes variance for subsets.

Methods inherited from class java.lang.Object

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

Field Detail

m_Info

protected Instances m_Info

The header information (for printing the tree).

m_Successors

protected REPTree.Tree[] m_Successors

The subtrees of this tree.

m_Attribute

protected int m_Attribute

The attribute to split on.

m_SplitPoint

protected double m_SplitPoint

The split point.

m_Prop

protected double[] m_Prop

The proportions of training instances going down each branch.

m_ClassProbs

protected double[] m_ClassProbs

Class probabilities from the training data in the nominal case. Holds the mean in the numeric case.

m_Distribution

protected double[] m_Distribution

The (unnormalized) class distribution in the nominal case. Holds the sum of squared errors and the weight in the numeric case.

m_HoldOutDist

protected double[] m_HoldOutDist

Class distribution of hold-out set at node in the nominal case. Straight sum of weights plus sum of weighted targets in the numeric case (i.e. array has only two elements).

m_HoldOutError

protected double m_HoldOutError

The hold-out error of the node. The number of miss-classified instances in the nominal case, the sum of squared errors in the numeric case.

Constructor Detail

Tree

protected Tree()

Method Detail

distributionForInstance

protected double[] distributionForInstance(Instance instance)
                                    throws java.lang.Exception

Computes class distribution of an instance using the tree.

Parameters:

instance - the instance to compute the distribution for

Returns:

the distribution

Throws:

java.lang.Exception - if computation fails

sourceExpression

public final java.lang.String sourceExpression(int index)

Returns a string containing java source code equivalent to the test made at this node. The instance being tested is called "i". This routine assumes to be called in the order of branching, enabling us to set the >= condition test (the last one) of a numeric splitpoint to just "true" (because being there in the flow implies that the previous less-than test failed).

Parameters:

index - index of the value tested

Returns:

a value of type 'String'

toSource

public java.lang.StringBuffer[] toSource(java.lang.String className,
                                         REPTree.Tree parent)
                                  throws java.lang.Exception

Returns source code for the tree as if-then statements. The class is assigned to variable "p", and assumes the tested instance is named "i". The results are returned as two stringbuffers: a section of code for assignment of the class, and a section of code containing support code (eg: other support methods).

TODO: If the outputted source code encounters a missing value for the evaluated attribute, it stops branching and uses the class distribution of the current node to decide the return value. This is unlike the behaviour of distributionForInstance().

Parameters:

className - the classname that this static classifier has

parent - parent node of the current node

Returns:

an array containing two stringbuffers, the first string containing assignment code, and the second containing source for support code.

Throws:

java.lang.Exception - if something goes wrong

toGraph

protected int toGraph(java.lang.StringBuffer text,
                      int num,
                      REPTree.Tree parent)
               throws java.lang.Exception

Outputs one node for graph.

Parameters:

text - the buffer to append the output to

num - the current node id

parent - the parent of the nodes

Returns:

the next node id

Throws:

java.lang.Exception - if something goes wrong

leafString

protected java.lang.String leafString(REPTree.Tree parent)
                               throws java.lang.Exception

Outputs description of a leaf node.

Parameters:

parent - the parent of the node

Returns:

the description of the node

Throws:

java.lang.Exception - if generation fails

toString

protected java.lang.String toString(int level,
                                    REPTree.Tree parent)

Recursively outputs the tree.

Parameters:

level - the current level

parent - the current parent

Returns:

the generated substree

buildTree

protected void buildTree(int[][][] sortedIndices,
                         double[][][] weights,
                         Instances data,
                         double totalWeight,
                         double[] classProbs,
                         Instances header,
                         double minNum,
                         double minVariance,
                         int depth,
                         int maxDepth)
                  throws java.lang.Exception

Recursively generates a tree.

Parameters:

sortedIndices - the sorted indices of the instances

weights - the weights of the instances

data - the data to work with

totalWeight -

classProbs - the class probabilities

header - the header of the data

minNum - the minimum number of instances in a leaf

minVariance -

depth - the current depth of the tree

maxDepth - the maximum allowed depth of the tree

Throws:

java.lang.Exception - if generation fails

doSmoothing

protected void doSmoothing()

Smoothes class probabilities stored at node.

numNodes

protected int numNodes()

Computes size of the tree.

Returns:

the number of nodes

splitData

protected void splitData(int[][][][] subsetIndices,
                         double[][][][] subsetWeights,
                         int att,
                         double splitPoint,
                         int[][] sortedIndices,
                         double[][] weights,
                         Instances data)
                  throws java.lang.Exception

Splits instances into subsets.

Parameters:

subsetIndices - the sorted indices in the subset

subsetWeights - the weights of the subset

att - the attribute index

splitPoint - the split point for numeric attributes

sortedIndices - the sorted indices of the whole set

weights - the weights of the whole set

data - the data to work with

Throws:

java.lang.Exception - if something goes wrong

distribution

protected double distribution(double[][] props,
                              double[][][] dists,
                              int att,
                              int[] sortedIndices,
                              double[] weights,
                              double[][] subsetWeights,
                              Instances data)
                       throws java.lang.Exception

Computes class distribution for an attribute.

Parameters:

props -

dists -

att - the attribute index

sortedIndices - the sorted indices of the instances

weights - the weights of the instances

subsetWeights - the weights of the subset

data - the data to work with

Returns:

the split point

Throws:

java.lang.Exception - if computation fails

numericDistribution

protected double numericDistribution(double[][] props,
                                     double[][][] dists,
                                     int att,
                                     int[] sortedIndices,
                                     double[] weights,
                                     double[][] subsetWeights,
                                     Instances data,
                                     double[] vals)
                              throws java.lang.Exception

Computes class distribution for an attribute.

Parameters:

props -

dists -

att - the attribute index

sortedIndices - the sorted indices of the instances

weights - the weights of the instances

subsetWeights - the weights of the subset

data - the data to work with

vals -

Returns:

the split point

Throws:

java.lang.Exception - if computation fails

variance

protected double variance(double[] s,
                          double[] sS,
                          double[] sumOfWeights)

Computes variance for subsets.

Parameters:

s -

sS -

sumOfWeights -

Returns:

the variance

singleVariance

protected double singleVariance(double s,
                                double sS,
                                double weight)

Computes the variance for a single set

Parameters:

s -

sS -

weight - the weight

Returns:

the variance

priorVal

protected double priorVal(double[][] dist)
                   throws java.lang.InterruptedException

Computes value of splitting criterion before split.

Parameters:

dist -

Returns:

the splitting criterion

Throws:

java.lang.InterruptedException

gain

protected double gain(double[][] dist,
                      double priorVal)

Computes value of splitting criterion after split.

Parameters:

dist -

priorVal - the splitting criterion

Returns:

the gain after splitting

reducedErrorPrune

protected double reducedErrorPrune()
                            throws java.lang.Exception

Prunes the tree using the hold-out data (bottom-up).

Returns:

the error

Throws:

java.lang.Exception - if pruning fails for some reason

insertHoldOutSet

protected void insertHoldOutSet(Instances data)
                         throws java.lang.Exception

Inserts hold-out set into tree.

Parameters:

data - the data to insert

Throws:

java.lang.Exception - if something goes wrong

insertHoldOutInstance

protected void insertHoldOutInstance(Instance inst,
                                     double weight,
                                     REPTree.Tree parent)
                              throws java.lang.Exception

Inserts an instance from the hold-out set into the tree.

Parameters:

inst - the instance to insert

weight - the weight of the instance

parent - the parent of the node

Throws:

java.lang.Exception - if insertion fails

backfitHoldOutSet

protected void backfitHoldOutSet()
                          throws java.lang.Exception

Backfits data from holdout set.

Throws:

java.lang.Exception - if insertion fails

getRevision

public java.lang.String getRevision()

Returns the revision string.

Specified by:

getRevision in interface RevisionHandler

Returns:

the revision