edu.udo.cs.yale.operator.learner.igss.hypothesis
Class Hypothesis

java.lang.Object
  edu.udo.cs.yale.operator.learner.igss.hypothesis.Hypothesis

Direct Known Subclasses:

Rule

public abstract class Hypothesis
extends java.lang.Object

Abstract superclass for all possible kinds of hypothesis.

Version:

$Id: Hypothesis.java,v 1.1 2006/10/02 16:07:49 ingomierswa Exp $

Author:

Dirk Dach

Field Summary

protected double	coveredWeight Stores the total weight of all examples covered by this hypothesis.
protected static boolean	createAllHypothesis Create h->Y+/Y- or h->Y+ only.
static int	FIRST_TYPE_INDEX
static java.lang.String[]	HYPOTHESIS_SPACE_TYPES
private static Attribute	label The label attribute.
static int	LAST_TYPE_INDEX
static int	NEGATIVE_CLASS
static int	POSITIVE_CLASS
protected double	positiveWeight Stores the weight of examples covered by this hypothesis with correct prediction.
protected static boolean	rejectionSampling Indicates if rejection sampling is used to draw the examples that this rule is applied to.
static int	TYPE_RULE

Constructor Summary

Hypothesis()

Hypothesis(Attribute[] regulars, Attribute l, boolean rs, boolean createAll)
Create a new dummy hypothesis to allow calling the 'init' method, initialize the regularAttributes, label and p0 fields.

Method Summary

abstract boolean	applicable(Example e) Tests if the hypothesis is applicable to the example without updating the internal statistics of the hypothesis.
abstract void	apply(Example e) Hypothesis is applied to the example and internal statistics of the hypothesis are updated.
abstract boolean	canBeRefined() Returns true only if this hypothesis can still be refined.
abstract Hypothesis	clone() Clone method.
abstract int	getComplexity() Returns complexity of the hypothesis.
double	getCoveredWeight() Returns the covered weight of this hypothesis.
static Attribute	getLabel() Returns the label.
double	getPositiveWeight() Returns the covered positive weight of this hypothesis.
abstract int	getPrediction() Returns the label index the hypothesis predicts.
abstract java.util.LinkedList<Hypothesis>	init(int minComplexity) Used to generate the first hypothesis or the first group of hypothesis.
abstract java.util.LinkedList<Hypothesis>	refine() Generates new hypothesis by adding one degree of complexity and creating all rules that are possible now.
void	reset() Sets 'coveredWeight' and 'positiveWeight' back to 0.0d.
void	setCoveredWeight(double value) Sets the covered weight of this hypothesis.
void	setPositiveWeight(double value) Sets the covered positive weight of this hypothesis.

Methods inherited from class java.lang.Object

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

Field Detail

HYPOTHESIS_SPACE_TYPES

public static final java.lang.String[] HYPOTHESIS_SPACE_TYPES

FIRST_TYPE_INDEX

public static final int FIRST_TYPE_INDEX

See Also:

Constant Field Values

TYPE_RULE

public static final int TYPE_RULE

See Also:

Constant Field Values

LAST_TYPE_INDEX

public static final int LAST_TYPE_INDEX

See Also:

Constant Field Values

POSITIVE_CLASS

public static final int POSITIVE_CLASS

See Also:

Constant Field Values

NEGATIVE_CLASS

public static final int NEGATIVE_CLASS

See Also:

Constant Field Values

label

private static Attribute label

The label attribute.

rejectionSampling

protected static boolean rejectionSampling

Indicates if rejection sampling is used to draw the examples that this rule is applied to. In this case (positive)example counter is incremented by one for every (positive)example the rule is applicable to. In the other case(weights are used directly) the weight of the example is added to the counters.

coveredWeight

protected double coveredWeight

Stores the total weight of all examples covered by this hypothesis.

positiveWeight

protected double positiveWeight

Stores the weight of examples covered by this hypothesis with correct prediction.

createAllHypothesis

protected static boolean createAllHypothesis

Create h->Y+/Y- or h->Y+ only.

Constructor Detail

Hypothesis

public Hypothesis(Attribute[] regulars,
                  Attribute l,
                  boolean rs,
                  boolean createAll)

Create a new dummy hypothesis to allow calling the 'init' method, initialize the regularAttributes, label and p0 fields.

Hypothesis

public Hypothesis()

Method Detail

clone

public abstract Hypothesis clone()

Clone method.

Overrides:

clone in class java.lang.Object

getPrediction

public abstract int getPrediction()

Returns the label index the hypothesis predicts.

getLabel

public static Attribute getLabel()

Returns the label.

reset

public void reset()

Sets 'coveredWeight' and 'positiveWeight' back to 0.0d.

getCoveredWeight

public double getCoveredWeight()

Returns the covered weight of this hypothesis.

setCoveredWeight

public void setCoveredWeight(double value)

Sets the covered weight of this hypothesis.

getPositiveWeight

public double getPositiveWeight()

Returns the covered positive weight of this hypothesis.

setPositiveWeight

public void setPositiveWeight(double value)

Sets the covered positive weight of this hypothesis.

apply

public abstract void apply(Example e)

Hypothesis is applied to the example and internal statistics of the hypothesis are updated.

applicable

public abstract boolean applicable(Example e)

Tests if the hypothesis is applicable to the example without updating the internal statistics of the hypothesis.

refine

public abstract java.util.LinkedList<Hypothesis> refine()

Generates new hypothesis by adding one degree of complexity and creating all rules that are possible now. Classes extending this abstract class define the way the hypothesis are generated by the IGSS alsgorithm by implementing the refine() method. Be sure to avoid the creation of duplicate hypothesis, for example by overwriting and using the equals method. For example refine could increase the allowed depth of a decision tree or the number of literals of a conjunctive rule. Must return null if the hypothesis cannot be refined any more.

canBeRefined

public abstract boolean canBeRefined()

Returns true only if this hypothesis can still be refined.

init

public abstract java.util.LinkedList<Hypothesis> init(int minComplexity)

Used to generate the first hypothesis or the first group of hypothesis.

getComplexity

public abstract int getComplexity()

Returns complexity of the hypothesis.