edu.udo.cs.yale.operator.learner.meta
Class BayBoostModel

java.lang.Object
  edu.udo.cs.yale.operator.AbstractIOObject
      edu.udo.cs.yale.operator.ResultObjectAdapter
          edu.udo.cs.yale.operator.AbstractModel
              edu.udo.cs.yale.operator.IOModel
                  edu.udo.cs.yale.operator.learner.PredictionModel
                      edu.udo.cs.yale.operator.learner.meta.BayBoostModel

All Implemented Interfaces:

IOObject, Model, ResultObject, Saveable

public class BayBoostModel
extends PredictionModel

A model for the Bayesian Boosting algorithm by Martin Scholz.

Version:

$Id: BayBoostModel.java,v 1.62 2006/09/30 00:05:30 ingomierswa Exp $

Author:

Martin Scholz

Field Summary

private static java.lang.String	CONV_TO_CRISP
static java.lang.String	ID
private static java.lang.String	MAX_MODEL_NUMBER
private int	maxModelNumber
private java.util.List<BayBoostBaseModelInfo>	modelInfo
private double[]	priors
private double	threshold

Constructor Summary

BayBoostModel()

BayBoostModel(Attribute label)
Needed for creation of IOModel (when loading from file).

BayBoostModel(Attribute label, java.util.List<BayBoostBaseModelInfo> modelInfos, double[] priors)

Method Summary

static boolean	adjustIntermediateProducts(double[] products, double[] liftFactors) Helper method to adjust the intermediate products during model application.
private void	cleanUpSpecialAttributes(ExampleSet exampleSet, Attribute[] specialAttributes) Removes the provided special labels from the exampleSet and exampleTable.
private Attribute[]	createSpecialAttributes(ExampleSet exampleSet) Creates a special attribute for each label to store intermediate results.
BayBoostBaseModelInfo	getBayBoostBaseModelInfo(int index)
ContingencyMatrix	getContingencyMatrix(int index) Getter method for a specific confusion matrix
private double[]	getFactorsForModel(int modelNr, int predicted) Gets factors for models in the case of general nominal class labels.
java.lang.String	getIdentifier()
Model	getModel(int index) Getter method for embedded models
double[]	getModelWeights() This method is only supported for boolean target attributes.
int	getNumberOfModels()
private double	getPriorOfClass(int classIndex) Getter method for prior class probabilities estimated as the relative frequencies in the training set.
double[]	getPriors() Getter for the prior array
private void	initIntermediateResultAttributes(ExampleSet exampleSet, Attribute[] specAttrib)
void	performPrediction(ExampleSet exampleSet, Attribute predictedLabel) Iterates over all models and returns the class with maximum likelihood.
void	readPredictionModelData(java.io.ObjectInputStream in) Reads all models from file.
void	setMaxModelNumber(int numModels) Using this setter with a positive value makes the model discard all but the specified number of base models.
void	setParameter(java.lang.String name, java.lang.String value) Setting the parameter MAX_MODEL_NUMBER allows to discard all but the first n models for specified n.
java.lang.String	toString()
private void	translateOddsIntoPredictions(Example example, Attribute[] specAttrib, Attribute exampleSetLabel)
private void	updateEstimates(ExampleSet exampleSet, ContingencyMatrix cm, Attribute[] specialAttributes)
void	writePredictionModelData(java.io.ObjectOutputStream out) Writes the models subsequently to the output stream.

Methods inherited from class edu.udo.cs.yale.operator.learner.PredictionModel

apply, createPredictedLabel, getLabel, readData, removePredictedLabel, setLabel, writeData

Methods inherited from class edu.udo.cs.yale.operator.AbstractModel

equals, getExtension, getFileDescription, readModel, readModel, save, setParameter, writeModel, writeModel

Methods inherited from class edu.udo.cs.yale.operator.ResultObjectAdapter

addAction, getActions, getName, getVisualisationComponent, toHTML, toResultString

Methods inherited from class edu.udo.cs.yale.operator.AbstractIOObject

copy, read, read, read, write, write

Methods inherited from class java.lang.Object

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

Methods inherited from interface edu.udo.cs.yale.operator.ResultObject

getActions, getName, getVisualisationComponent, toResultString

Methods inherited from interface edu.udo.cs.yale.operator.IOObject

copy, write, write

Field Detail

ID

public static final java.lang.String ID

See Also:

Constant Field Values

modelInfo

private java.util.List<BayBoostBaseModelInfo> modelInfo

priors

private double[] priors

maxModelNumber

private int maxModelNumber

MAX_MODEL_NUMBER

private static final java.lang.String MAX_MODEL_NUMBER

See Also:

Constant Field Values

CONV_TO_CRISP

private static final java.lang.String CONV_TO_CRISP

See Also:

Constant Field Values

threshold

private double threshold

Constructor Detail

BayBoostModel

public BayBoostModel()

BayBoostModel

public BayBoostModel(Attribute label)

Needed for creation of IOModel (when loading from file).

Parameters:

label - the class label

BayBoostModel

public BayBoostModel(Attribute label,
                     java.util.List<BayBoostBaseModelInfo> modelInfos,
                     double[] priors)

Parameters:

label - the class label

modelInfos - a List of Object[2] arrays, each entry holding a model and a double[][] array containing weights for all prediction/label combinations.

priors - an array of the prior probabilities of labels

Method Detail

getBayBoostBaseModelInfo

public BayBoostBaseModelInfo getBayBoostBaseModelInfo(int index)

setParameter

public void setParameter(java.lang.String name,
                         java.lang.String value)
                  throws OperatorException

Setting the parameter MAX_MODEL_NUMBER allows to discard all but the first n models for specified n. CONV_TO_CRISP allows to set another threshold than 0.5 for boolean prediction problems.

Throws:

OperatorException

setMaxModelNumber

public void setMaxModelNumber(int numModels)

Using this setter with a positive value makes the model discard all but the specified number of base models. A value of -1 turns off this option.

getIdentifier

public java.lang.String getIdentifier()

Returns:

the static class identifier of this model

readPredictionModelData

public void readPredictionModelData(java.io.ObjectInputStream in)
                             throws java.io.IOException

Reads all models from file.

Specified by:

readPredictionModelData in class PredictionModel

Throws:

java.io.IOException

writePredictionModelData

public void writePredictionModelData(java.io.ObjectOutputStream out)
                              throws java.io.IOException

Writes the models subsequently to the output stream.

Specified by:

writePredictionModelData in class PredictionModel

Throws:

java.io.IOException

toString

public java.lang.String toString()

Overrides:

toString in class PredictionModel

Returns:

a String representation of this boosting model.

getNumberOfModels

public int getNumberOfModels()

Returns:

the number of embedded models

getFactorsForModel

private double[] getFactorsForModel(int modelNr,
                                    int predicted)

Gets factors for models in the case of general nominal class labels. The indices are not in YALE format, so add Attribute.FIRST_CLASS_INDEX before calling this method and before reading from the returned array.

Parameters:

modelNr - the number of the model

predicted - the predicted label

Returns:

a double[] object with the factors to be applied for each class if the corresponding rule yields predicted.

getPriorOfClass

private double getPriorOfClass(int classIndex)

Getter method for prior class probabilities estimated as the relative frequencies in the training set.

Parameters:

classIndex - the index of a class, starting with 0

Returns:

the prior probability of the specified class

getPriors

public double[] getPriors()

Getter for the prior array

getModel

public Model getModel(int index)

Getter method for embedded models

Parameters:

index - the number of a model part of this boost model

Returns:

binary or nominal decision model for the given classification index.

getContingencyMatrix

public ContingencyMatrix getContingencyMatrix(int index)

Getter method for a specific confusion matrix

Parameters:

index - the number of the model for which to read the confusion matrix

Returns:

a ConfusionMatrix object

performPrediction

public void performPrediction(ExampleSet exampleSet,
                              Attribute predictedLabel)
                       throws OperatorException

Iterates over all models and returns the class with maximum likelihood.

Specified by:

performPrediction in class PredictionModel

Parameters:

exampleSet - the set of examples to be classified

predictedLabel - the label that finally holds the predictions

Throws:

OperatorException

createSpecialAttributes

private Attribute[] createSpecialAttributes(ExampleSet exampleSet)
                                     throws OperatorException

Creates a special attribute for each label to store intermediate results.

Throws:

OperatorException

cleanUpSpecialAttributes

private void cleanUpSpecialAttributes(ExampleSet exampleSet,
                                      Attribute[] specialAttributes)
                               throws OperatorException

Removes the provided special labels from the exampleSet and exampleTable.

Throws:

OperatorException

initIntermediateResultAttributes

private void initIntermediateResultAttributes(ExampleSet exampleSet,
                                              Attribute[] specAttrib)

translateOddsIntoPredictions

private void translateOddsIntoPredictions(Example example,
                                          Attribute[] specAttrib,
                                          Attribute exampleSetLabel)

updateEstimates

private void updateEstimates(ExampleSet exampleSet,
                             ContingencyMatrix cm,
                             Attribute[] specialAttributes)

adjustIntermediateProducts

public static boolean adjustIntermediateProducts(double[] products,
                                                 double[] liftFactors)

Helper method to adjust the intermediate products during model application.

Parameters:

products - the intermediate products, these values are changed by the method

liftFactors - the factor vector that applies for the prediction for the current example

Returns:

true iff the class is deterministically known after applying this method

getModelWeights

public double[] getModelWeights()
                         throws OperatorException

This method is only supported for boolean target attributes. It computes a flattened version of model weights. In constrast to the original version the final predictions are additive logarithms of the lift ratios, additively rescaled so that the prediction false of model i produces -i if true produces weight i. This means that only one weight per model is required. The first component of the returned array is the part that is independent of any prediction, the i-th component is the weight of model i. The (log-)linear model predicts depending on whether the linear combination of predictions (either -1 or 1) is greater than 0 or not. Infinite values are problematic, so a min/max value is used.

Returns:

the flattened weights of all models

Throws:

OperatorException