edu.udo.cs.yale.generator
Class SinusFactory

java.lang.Object
  edu.udo.cs.yale.generator.SinusFactory

public class SinusFactory
extends java.lang.Object

Factory class to produce new attributes based on the fourier synthesis of the label mapped on an attribute dimension.

Version:

$Id: SinusFactory.java,v 2.17 2006/09/29 21:09:37 ingomierswa Exp $

Author:

Ingo Mierswa

Field Summary

static java.lang.String[]	ADAPTION_TYPES
private int	adaptionType Indicates the type of frequency adaption.
private int	attributesPerPeak Generates this number of peaks in a range of epsilon * frequency.
private double	epsilon Generates this peaksPerPeak peaks in the range of epsilon * frequency.
private FastFourierTransform	fft The fast fourier transformation calculator.
private SpectrumFilter	filter The spectrum filter type which should be applied on the spectrum after the fourier transformation.
static int	GAUSSIAN
private int	maxPeaks The maximal number of generated attributes for each possible attribute.
private static double	MIN_EVIDENCE Indicates the min evidence factor.
private PeakFinder	peakFinder The algorithm to find the peaks in the frequency spectrum.
static int	UNIFORMLY
static int	UNIFORMLY_WITHOUT_NU

Constructor Summary

SinusFactory(int maxPeaks)
Creates a new sinus factory which creates maxPeaks new peaks.

Method Summary

private java.util.List<Attribute>	generateAttribute(ExampleSet exampleSet, FeatureGenerator generator)
void	generateSinusFunctions(ExampleSet exampleSet, java.util.List<AttributePeak> attributes, java.util.Random random) Generates a new sinus function attribute for all given attribute peaks.
java.util.List<AttributePeak>	getAttributePeaks(ExampleSet exampleSet, Attribute first, Attribute second) Calculates the fourier transformation from the first attribute on the second and delivers the maxPeaks highest peaks.
void	setAdaptionType(int type)
void	setAttributePerPeak(int attributesPerPeak) Must be bigger than 2!
void	setEpsilon(double epsilon)

Methods inherited from class java.lang.Object

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

Field Detail

MIN_EVIDENCE

private static final double MIN_EVIDENCE

Indicates the min evidence factor.

See Also:

Constant Field Values

ADAPTION_TYPES

public static final java.lang.String[] ADAPTION_TYPES

UNIFORMLY

public static final int UNIFORMLY

See Also:

Constant Field Values

UNIFORMLY_WITHOUT_NU

public static final int UNIFORMLY_WITHOUT_NU

See Also:

Constant Field Values

GAUSSIAN

public static final int GAUSSIAN

See Also:

Constant Field Values

attributesPerPeak

private int attributesPerPeak

Generates this number of peaks in a range of epsilon * frequency. Necessary because the FT does not deliver the correct frequency (aliasing, leakage) in all cases. In later releases this should be replaced by a gradient search or a evolutionary search for the correct value.

epsilon

private double epsilon

Generates this peaksPerPeak peaks in the range of epsilon * frequency. Necessary because the FT does not deliver the correct frequency (aliasing, leakage) in all cases. In later releases this should be replaced by a gradient search or a evolutionary search for the correct value.

adaptionType

private int adaptionType

Indicates the type of frequency adaption.

maxPeaks

private int maxPeaks

The maximal number of generated attributes for each possible attribute. Corresponds to the highest peaks in the frequency spectrum of the label in the source attribute's space.

fft

private FastFourierTransform fft

The fast fourier transformation calculator.

filter

private SpectrumFilter filter

The spectrum filter type which should be applied on the spectrum after the fourier transformation.

peakFinder

private PeakFinder peakFinder

The algorithm to find the peaks in the frequency spectrum.

Constructor Detail

SinusFactory

public SinusFactory(int maxPeaks)

Creates a new sinus factory which creates maxPeaks new peaks. Uses Blackman-Harris window function and no spectrum filter as default. The adaption type is gaussian with an epsilon of 0.1. The factory produces three attributes for each highest peak as default.

Method Detail

setAdaptionType

public void setAdaptionType(int type)

setEpsilon

public void setEpsilon(double epsilon)

setAttributePerPeak

public void setAttributePerPeak(int attributesPerPeak)

Must be bigger than 2!

getAttributePeaks

public java.util.List<AttributePeak> getAttributePeaks(ExampleSet exampleSet,
                                                       Attribute first,
                                                       Attribute second)
                                                throws OperatorException

Calculates the fourier transformation from the first attribute on the second and delivers the maxPeaks highest peaks. Returns a list with the highest attribute peaks.

Throws:

OperatorException

generateSinusFunctions

public void generateSinusFunctions(ExampleSet exampleSet,
                                   java.util.List<AttributePeak> attributes,
                                   java.util.Random random)
                            throws GenerationException

Generates a new sinus function attribute for all given attribute peaks. Since the frequency cannot be calculated exactly (leakage, aliasing), several new attribute may be added for each peak. These additional attributes are randomly chosen (uniformly in epsilon range, uniformly without nu, gaussian with epsilon as standard deviation)

Throws:

GenerationException

generateAttribute

private java.util.List<Attribute> generateAttribute(ExampleSet exampleSet,
                                                    FeatureGenerator generator)
                                             throws GenerationException

Throws:

GenerationException