edu.udo.cs.yale.operator.learner.igss.utility
Class AbstractUtility

java.lang.Object
  edu.udo.cs.yale.operator.learner.igss.utility.AbstractUtility

All Implemented Interfaces:

Utility

Direct Known Subclasses:

Binomial, InstanceAveraging, Linear, Squared

public abstract class AbstractUtility
extends java.lang.Object
implements Utility

Abstract superclass for all utility functions.

Version:

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

Author:

Dirk Dach

Field Summary

protected int	large The number of covered examples before normal approximation is used.
protected double[]	priors The prior probability of the two classes of the label.

Fields inherited from interface edu.udo.cs.yale.operator.learner.igss.utility.Utility

FIRST_TYPE_INDEX, LAST_TYPE_INDEX, TYPE_ACCURACY, TYPE_BINOMIAL, TYPE_LINEAR, TYPE_SQUARED, TYPE_WRACC, UTILITY_TYPES

Constructor Summary

AbstractUtility(double[] priors, int large)
Constructor for all utilities.

Method Summary

double	calculateM(double delta, double epsilon) Calculates the M-value needed for the GSS algorithm.
abstract double	conf(double totalWeight, double delta) Calculates the normal approximation of the confidence intervall.
abstract double	conf(double totalWeight, double totalPositiveWeight, Hypothesis hypo, double delta) Calculates the normal approximation of the confidence intervall for a specific hypothesis.
double	confidenceIntervall(double totalWeight, double delta) Calculates the the unspecific confidence intervall.
double	confidenceIntervall(double totalWeight, double totalPositiveWeight, Hypothesis hypo, double delta) Calculates the the confidence intervall for a specific hypothesis.
abstract double	confSmallM(double totalWeight, double delta) Calculates the confidence intervall for small numbers of examples.
double	inverseNormal(double p) Calculates the inverse of the normal distribution, e.g.inverseNormal(0.95)==1.64.

Methods inherited from class java.lang.Object

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

Methods inherited from interface edu.udo.cs.yale.operator.learner.igss.utility.Utility

getUpperBound, utility

Field Detail

priors

protected double[] priors

The prior probability of the two classes of the label.

large

protected int large

The number of covered examples before normal approximation is used.

Constructor Detail

AbstractUtility

public AbstractUtility(double[] priors,
                       int large)

Constructor for all utilities.

Method Detail

calculateM

public double calculateM(double delta,
                         double epsilon)

Calculates the M-value needed for the GSS algorithm.

Specified by:

calculateM in interface Utility

confidenceIntervall

public double confidenceIntervall(double totalWeight,
                                  double delta)

Calculates the the unspecific confidence intervall. Uses Chernoff bounds if the number of random experiments is too small and normal approximatione otherwise. Considers the number of examples as the number of random experiments. problematic for g*(p-p0)) hypothesis, that only cover a small amount of examples. No normal approximation should be used in this case.

Specified by:

confidenceIntervall in interface Utility

confidenceIntervall

public double confidenceIntervall(double totalWeight,
                                  double totalPositiveWeight,
                                  Hypothesis hypo,
                                  double delta)

Calculates the the confidence intervall for a specific hypothesis. Uses Chernoff bounds if the number of random experiments is too small and normal approximation otherwise. This method is adapted for g*(p-p0) utility types. Every example for that the rule is applicable is one random experiment. Should be overwritten by subclasses if they make a different random experiment.

Specified by:

confidenceIntervall in interface Utility

confSmallM

public abstract double confSmallM(double totalWeight,
                                  double delta)

Calculates the confidence intervall for small numbers of examples.

conf

public abstract double conf(double totalWeight,
                            double delta)

Calculates the normal approximation of the confidence intervall.

conf

public abstract double conf(double totalWeight,
                            double totalPositiveWeight,
                            Hypothesis hypo,
                            double delta)

Calculates the normal approximation of the confidence intervall for a specific hypothesis.

inverseNormal

public double inverseNormal(double p)

Calculates the inverse of the normal distribution, e.g.inverseNormal(0.95)==1.64.