weka.classifiers.trees.lmt

Class LMTNode

java.lang.Object

weka.classifiers.AbstractClassifier

weka.classifiers.trees.lmt.LogisticBase

weka.classifiers.trees.lmt.LMTNode

All Implemented Interfaces:

java.io.Serializable, java.lang.Cloneable, Classifier, BatchPredictor, CapabilitiesHandler, CapabilitiesIgnorer, CommandlineRunnable, OptionHandler, RevisionHandler, WeightedInstancesHandler

public class LMTNode
extends LogisticBase

Class for logistic model tree structure.

Version:

$Revision: 11566 $

Author:

Niels Landwehr, Marc Sumner

See Also:

Serialized Form

Field Summary

Fields

Modifier and Type	Field and Description
double	m_alpha Alpha-value (for pruning) at the node
double	m_numIncorrectModel Weighted number of training examples currently misclassified by the logistic model at the node
double	m_numIncorrectTree Weighted number of training examples currently misclassified by the subtree rooted at the node

Fields inherited from class weka.classifiers.AbstractClassifier

BATCH_SIZE_DEFAULT, NUM_DECIMAL_PLACES_DEFAULT

Constructor Summary

Constructors

Constructor and Description
LMTNode(ModelSelection modelSelection, int numBoostingIterations, boolean fastRegression, boolean errorOnProbabilities, int minNumInstances, double weightTrimBeta, boolean useAIC, NominalToBinary ntb, int numDecimalPlaces) Constructor for logistic model tree node.

Method Summary

Modifier and Type	Method and Description
int	assignIDs(int lastID) Assigns unique IDs to all nodes in the tree
int	assignLeafModelNumbers(int leafCounter) Assigns numbers to the logistic regression models at the leaves of the tree
void	buildClassifier(Instances data) Method for building a logistic model tree (only called for the root node).
void	buildTree(Instances data, SimpleLinearRegression[][] higherRegressions, double totalInstanceWeight, double higherNumParameters, Instances numericDataHeader) Method for building the tree structure.
void	calculateAlphas() Updates the alpha field for all nodes.
double[]	distributionForInstance(Instance instance) Returns the class probabilities for an instance given by the logistic model tree.
java.lang.String	getModelParameters() Returns a string describing the number of LogitBoost iterations performed at this node, the total number of LogitBoost iterations performed (including iterations at higher levels in the tree), and the number of training examples at this node.
java.util.Vector<LMTNode>	getNodes() Return a list of all inner nodes in the tree
void	getNodes(java.util.Vector<LMTNode> nodeList) Fills a list with all inner nodes in the tree
int	getNumInnerNodes() Method to count the number of inner nodes in the tree
int	getNumLeaves() Returns the number of leaves in the tree.
java.lang.String	getRevision() Returns the revision string.
java.lang.String	graph() Returns graph describing the tree.
boolean	hasModels() Returns true if the logistic regression model at this node has changed compared to the one at the parent node.
double[]	modelDistributionForInstance(Instance instance) Returns the class probabilities for an instance according to the logistic model at the node.
java.lang.String	modelsToString() Returns a string describing the logistic regression function at the node.
int	numLeaves() Returns the number of leaves (normal count).
int	numNodes() Returns the number of nodes.
void	prune(double alpha) Prunes a logistic model tree using the CART pruning scheme, given a cost-complexity parameter alpha.
int	prune(double[] alphas, double[] errors, Instances test) Method for performing one fold in the cross-validation of the cost-complexity parameter.
java.lang.String	toString() Returns a description of the logistic model tree (tree structure and logistic models)
void	treeErrors() Updates the numIncorrectTree field for all nodes.

Methods inherited from class weka.classifiers.trees.lmt.LogisticBase

cleanup, getMaxIterations, getNumRegressions, getUseAIC, getUsedAttributes, getWeightTrimBeta, percentAttributesUsed, setHeuristicStop, setMaxIterations, setUseAIC, setWeightTrimBeta

Methods inherited from class weka.classifiers.AbstractClassifier

batchSizeTipText, classifyInstance, debugTipText, distributionsForInstances, doNotCheckCapabilitiesTipText, forName, getBatchSize, getCapabilities, getDebug, getDoNotCheckCapabilities, getNumDecimalPlaces, getOptions, implementsMoreEfficientBatchPrediction, listOptions, makeCopies, makeCopy, numDecimalPlacesTipText, postExecution, preExecution, run, runClassifier, setBatchSize, setDebug, setDoNotCheckCapabilities, setNumDecimalPlaces, setOptions

Methods inherited from class java.lang.Object

equals, getClass, hashCode, notify, notifyAll, wait, wait, wait

Methods inherited from interface weka.core.OptionHandler

makeCopy

Field Detail

m_alpha

public double m_alpha

Alpha-value (for pruning) at the node

m_numIncorrectModel

public double m_numIncorrectModel

Weighted number of training examples currently misclassified by the logistic model at the node

m_numIncorrectTree

public double m_numIncorrectTree

Weighted number of training examples currently misclassified by the subtree rooted at the node

Constructor Detail

LMTNode

public LMTNode(ModelSelection modelSelection,
               int numBoostingIterations,
               boolean fastRegression,
               boolean errorOnProbabilities,
               int minNumInstances,
               double weightTrimBeta,
               boolean useAIC,
               NominalToBinary ntb,
               int numDecimalPlaces)

Constructor for logistic model tree node.

Parameters:

modelSelection - selection method for local splitting model

numBoostingIterations - sets the numBoostingIterations parameter

fastRegression - sets the fastRegression parameter

errorOnProbabilities - Use error on probabilities for stopping criterion of LogitBoost?

minNumInstances - minimum number of instances at which a node is considered for splitting

Method Detail

buildClassifier

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

Method for building a logistic model tree (only called for the root node). Grows an initial logistic model tree and prunes it back using the CART pruning scheme.

Specified by:

buildClassifier in interface Classifier

Overrides:

buildClassifier in class LogisticBase

Parameters:

data - the data to train with

Throws:

java.lang.Exception - if something goes wrong

buildTree

public void buildTree(Instances data,
                      SimpleLinearRegression[][] higherRegressions,
                      double totalInstanceWeight,
                      double higherNumParameters,
                      Instances numericDataHeader)
               throws java.lang.Exception

Method for building the tree structure. Builds a logistic model, splits the node and recursively builds tree for child nodes.

Parameters:

data - the training data passed on to this node

higherRegressions - An array of regression functions produced by LogitBoost at higher levels in the tree. They represent a logistic regression model that is refined locally at this node.

totalInstanceWeight - the total number of training examples

higherNumParameters - effective number of parameters in the logistic regression model built in parent nodes

Throws:

java.lang.Exception - if something goes wrong

prune

public void prune(double alpha)
           throws java.lang.Exception

Prunes a logistic model tree using the CART pruning scheme, given a cost-complexity parameter alpha.

Parameters:

alpha - the cost-complexity measure

Throws:

java.lang.Exception - if something goes wrong

prune

public int prune(double[] alphas,
                 double[] errors,
                 Instances test)
          throws java.lang.Exception

Method for performing one fold in the cross-validation of the cost-complexity parameter. Generates a sequence of alpha-values with error estimates for the corresponding (partially pruned) trees, given the test set of that fold.

Parameters:

alphas - array to hold the generated alpha-values

errors - array to hold the corresponding error estimates

test - test set of that fold (to obtain error estimates)

Throws:

java.lang.Exception - if something goes wrong

getNumInnerNodes

public int getNumInnerNodes()

Method to count the number of inner nodes in the tree

Returns:

the number of inner nodes

getNumLeaves

public int getNumLeaves()

Returns the number of leaves in the tree. Leaves are only counted if their logistic model has changed compared to the one of the parent node.

Returns:

the number of leaves

treeErrors

public void treeErrors()

Updates the numIncorrectTree field for all nodes. This is needed for calculating the alpha-values.

calculateAlphas

public void calculateAlphas()
                     throws java.lang.Exception

Updates the alpha field for all nodes.

Throws:

java.lang.Exception

getNodes

public java.util.Vector<LMTNode> getNodes()

Return a list of all inner nodes in the tree

Returns:

the list of nodes

getNodes

public void getNodes(java.util.Vector<LMTNode> nodeList)

Fills a list with all inner nodes in the tree

Parameters:

nodeList - the list to be filled

hasModels

public boolean hasModels()

Returns true if the logistic regression model at this node has changed compared to the one at the parent node.

Returns:

whether it has changed

modelDistributionForInstance

public double[] modelDistributionForInstance(Instance instance)
                                      throws java.lang.Exception

Returns the class probabilities for an instance according to the logistic model at the node.

Parameters:

instance - the instance

Returns:

the array of probabilities

Throws:

java.lang.Exception

distributionForInstance

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

Returns the class probabilities for an instance given by the logistic model tree.

Specified by:

distributionForInstance in interface Classifier

Overrides:

distributionForInstance in class LogisticBase

Parameters:

instance - the instance

Returns:

the array of probabilities

Throws:

java.lang.Exception - if distribution can't be computed successfully

numLeaves

public int numLeaves()

Returns the number of leaves (normal count).

Returns:

the number of leaves

numNodes

public int numNodes()

Returns the number of nodes.

Returns:

the number of nodes

toString

public java.lang.String toString()

Returns a description of the logistic model tree (tree structure and logistic models)

Overrides:

toString in class LogisticBase

Returns:

describing string

getModelParameters

public java.lang.String getModelParameters()

Returns a string describing the number of LogitBoost iterations performed at this node, the total number of LogitBoost iterations performed (including iterations at higher levels in the tree), and the number of training examples at this node.

Returns:

the describing string

assignIDs

public int assignIDs(int lastID)

Assigns unique IDs to all nodes in the tree

assignLeafModelNumbers

public int assignLeafModelNumbers(int leafCounter)

Assigns numbers to the logistic regression models at the leaves of the tree

modelsToString

public java.lang.String modelsToString()

Returns a string describing the logistic regression function at the node.

graph

public java.lang.String graph()
                       throws java.lang.Exception

Returns graph describing the tree.

Throws:

java.lang.Exception - if something goes wrong

getRevision

public java.lang.String getRevision()

Returns the revision string.

Specified by:

getRevision in interface RevisionHandler

Overrides:

getRevision in class LogisticBase

Returns:

the revision