edu.udo.cs.yale.operator.learner.kernel.evosvm
Class Kernel

java.lang.Object
  edu.udo.cs.yale.operator.learner.kernel.evosvm.Kernel

Direct Known Subclasses:

DotKernel, EpanechnikovKernel, GaussianCombinationKernel, MultiquadricKernel, PolynomialKernel, RBFKernel, SigmoidKernel

public abstract class Kernel
extends java.lang.Object

Returns the distance of two examples. The method init(ExampleSet) must be invoked before the correct distances can be returned. Please note that subclasses must provide an empty constructor to allow kernel creation via reflection (for reading kernels from disk).

Version:

$Id: Kernel.java,v 1.8 2006/09/30 00:05:30 ingomierswa Exp $

Author:

Ingo Mierswa

Field Summary

private double[][]	distances The complete distance matrix for this kernel and a given example set.
static int	KERNEL_DOT Indicates a linear kernel.
static int	KERNEL_EPANECHNIKOV Indicates a Epanechnikov kernel.
static int	KERNEL_GAUSSIAN_COMBINATION Indicates a Gaussian combination kernel.
static int	KERNEL_MULTIQUADRIC Indicates a multiquadric kernel.
static int	KERNEL_POLYNOMIAL Indicates a polynomial kernel.
static int	KERNEL_RADIAL Indicates a rbf kernel.
static int	KERNEL_SIGMOID Indicates a sigmoid kernel.
static java.lang.String[]	KERNEL_TYPES The kernels which can be used for the EvoSVM.

Constructor Summary

Kernel()

Method Summary

abstract double	calculateDistance(double[] x1, double[] x2) Subclasses must implement this method.
static Kernel	createKernel(int kernelType)
static Kernel	createKernel(int kernelType, EvoSVM operator)
double	getDistance(int x1, int x2) Returns the distance between the examples with the given indices.
double	getSum(java.util.Collection supportVectors, double[] currentX) Calculates w*x from the given support vectors using this kernel function.
abstract int	getType() Must return one out of KERNEL_DOT, KERNEL_RADIAL, KERNEL_POLYNOMIAL, or KERNEL_SIGMOID.
void	init(ExampleSet exampleSet) Calculates all distances and store them in a matrix to speed up optimization.
double	innerProduct(double[] x1, double[] x2) Calculates the inner product of the given vectors.
double	norm2(double[] x1, double[] x2) Calculates the L2-norm, i.e.
static Kernel	readKernelData(java.io.ObjectInputStream in) Reads the kernel type and generates a new kernel of this type.
abstract void	readKernelParameters(java.io.ObjectInputStream in) Subclasses must implement this method.
void	writeKernelData(java.io.ObjectOutputStream out) Writes the kernel type into the given stream.
abstract void	writeKernelParameters(java.io.ObjectOutputStream out) Subclasses must implement this method.

Methods inherited from class java.lang.Object

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

Field Detail

KERNEL_TYPES

public static final java.lang.String[] KERNEL_TYPES

The kernels which can be used for the EvoSVM.

KERNEL_DOT

public static final int KERNEL_DOT

Indicates a linear kernel.

See Also:

Constant Field Values

KERNEL_RADIAL

public static final int KERNEL_RADIAL

Indicates a rbf kernel.

See Also:

Constant Field Values

KERNEL_POLYNOMIAL

public static final int KERNEL_POLYNOMIAL

Indicates a polynomial kernel.

See Also:

Constant Field Values

KERNEL_SIGMOID

public static final int KERNEL_SIGMOID

Indicates a sigmoid kernel.

See Also:

Constant Field Values

KERNEL_EPANECHNIKOV

public static final int KERNEL_EPANECHNIKOV

Indicates a Epanechnikov kernel.

See Also:

Constant Field Values

KERNEL_GAUSSIAN_COMBINATION

public static final int KERNEL_GAUSSIAN_COMBINATION

Indicates a Gaussian combination kernel.

See Also:

Constant Field Values

KERNEL_MULTIQUADRIC

public static final int KERNEL_MULTIQUADRIC

Indicates a multiquadric kernel.

See Also:

Constant Field Values

distances

private double[][] distances

The complete distance matrix for this kernel and a given example set.

Constructor Detail

Kernel

public Kernel()

Method Detail

getType

public abstract int getType()

Must return one out of KERNEL_DOT, KERNEL_RADIAL, KERNEL_POLYNOMIAL, or KERNEL_SIGMOID.

calculateDistance

public abstract double calculateDistance(double[] x1,
                                         double[] x2)

Subclasses must implement this method.

writeKernelParameters

public abstract void writeKernelParameters(java.io.ObjectOutputStream out)
                                    throws java.io.IOException

Subclasses must implement this method.

Throws:

java.io.IOException

readKernelParameters

public abstract void readKernelParameters(java.io.ObjectInputStream in)
                                   throws java.io.IOException

Subclasses must implement this method.

Throws:

java.io.IOException

init

public void init(ExampleSet exampleSet)

Calculates all distances and store them in a matrix to speed up optimization.

getDistance

public double getDistance(int x1,
                          int x2)

Returns the distance between the examples with the given indices.

innerProduct

public double innerProduct(double[] x1,
                           double[] x2)

Calculates the inner product of the given vectors.

norm2

public double norm2(double[] x1,
                    double[] x2)

Calculates the L2-norm, i.e. ||x-y||^2.

getSum

public double getSum(java.util.Collection supportVectors,
                     double[] currentX)

Calculates w*x from the given support vectors using this kernel function.

writeKernelData

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

Writes the kernel type into the given stream. Invokes writeKernelParameters(ObjectOutputStream) afterwards.

Throws:

java.io.IOException

readKernelData

public static Kernel readKernelData(java.io.ObjectInputStream in)
                             throws java.io.IOException

Reads the kernel type and generates a new kernel of this type. After that the method readKernelParameters(ObjectInputStream) is invoked.

Throws:

java.io.IOException

createKernel

public static Kernel createKernel(int kernelType)

createKernel

public static Kernel createKernel(int kernelType,
                                  EvoSVM operator)
                           throws UndefinedParameterError

Throws:

UndefinedParameterError