de.lmu.ifi.dbs.elki.distance.similarityfunction.kernel

Class KernelMatrix

java.lang.Object

de.lmu.ifi.dbs.elki.distance.similarityfunction.kernel.KernelMatrix

public class KernelMatrix
extends java.lang.Object

Kernel matrix representation.

Since:

0.1

Author:

Simon Paradies

Nested Class Summary

Nested Classes

Modifier and Type	Class and Description
private static interface	KernelMatrix.DBIDMap Map a DBID to its offset TODO: move to shared code.
private class	KernelMatrix.RangeMap Map a DBID to an integer offset, DBIDRange version.
private class	KernelMatrix.SortedArrayMap Map a DBID to an integer offset, Version to support arbitrary DBIDs.

Field Summary

Fields

Modifier and Type	Field and Description
(package private) KernelMatrix.DBIDMap	idmap Static mapping from DBIDs to indexes.
(package private) double[][]	kernel The kernel matrix

Constructor Summary

Constructors

Constructor and Description
KernelMatrix(double[][] matrix) Makes a new kernel matrix from matrix (with data copying).
KernelMatrix(PrimitiveSimilarityFunction<? super O> kernelFunction, Relation<? extends O> relation, DBIDs ids) Provides a new kernel matrix.
KernelMatrix(SimilarityQuery<? super O> kernelFunction, Relation<? extends O> relation, DBIDs ids) Provides a new kernel matrix.

Method Summary

Modifier and Type	Method and Description
static double[][]	centerKernelMatrix(KernelMatrix kernelMatrix) Centers the Kernel Matrix in Feature Space according to Smola et.
static double[][]	centerMatrix(double[][] matrix) Centers the matrix in feature space according to Smola et Schoelkopf, Learning with Kernels p. 431 Alters the input matrix.
double	getDistance(DBIDRef o1, DBIDRef o2) Returns the kernel distance between the two specified objects.
double[][]	getKernel() Get the kernel matrix.
double	getSimilarity(DBIDRef id1, DBIDRef id2) Get the kernel similarity for the given objects.
double	getSquaredDistance(DBIDRef id1, DBIDRef id2) Returns the squared kernel distance between the two specified objects.

Methods inherited from class java.lang.Object

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

Field Detail

kernel

double[][] kernel

The kernel matrix

idmap

KernelMatrix.DBIDMap idmap

Static mapping from DBIDs to indexes.

Constructor Detail

KernelMatrix

public KernelMatrix(PrimitiveSimilarityFunction<? super O> kernelFunction,
                    Relation<? extends O> relation,
                    DBIDs ids)

Provides a new kernel matrix.

Parameters:

kernelFunction - the kernel function used to compute the kernel matrix

relation - the database that holds the objects

ids - the IDs of those objects for which the kernel matrix is computed

KernelMatrix

public KernelMatrix(SimilarityQuery<? super O> kernelFunction,
                    Relation<? extends O> relation,
                    DBIDs ids)

Provides a new kernel matrix.

Parameters:

kernelFunction - the kernel function used to compute the kernel matrix

relation - the database that holds the objects

ids - the IDs of those objects for which the kernel matrix is computed

KernelMatrix

public KernelMatrix(double[][] matrix)

Makes a new kernel matrix from matrix (with data copying).

Parameters:

matrix - a matrix

Method Detail

getDistance

public double getDistance(DBIDRef o1,
                          DBIDRef o2)

Returns the kernel distance between the two specified objects.

Parameters:

o1 - first ObjectID

o2 - second ObjectID

Returns:

the distance between the two objects

getKernel

public double[][] getKernel()

Get the kernel matrix.

Returns:

kernel

getSquaredDistance

public double getSquaredDistance(DBIDRef id1,
                                 DBIDRef id2)

Returns the squared kernel distance between the two specified objects.

Parameters:

id1 - first ObjectID

id2 - second ObjectID

Returns:

the distance between the two objects

centerMatrix

public static double[][] centerMatrix(double[][] matrix)

Centers the matrix in feature space according to Smola et Schoelkopf, Learning with Kernels p. 431 Alters the input matrix. If you still need the original matrix, use centeredMatrix = centerKernelMatrix(uncenteredMatrix.copy()) {

Parameters:

matrix - the matrix to be centered

Returns:

centered matrix (for convenience)

centerKernelMatrix

public static double[][] centerKernelMatrix(KernelMatrix kernelMatrix)

Centers the Kernel Matrix in Feature Space according to Smola et. Schoelkopf, Learning with Kernels p. 431 Alters the input matrix. If you still need the original matrix, use centeredMatrix = centerKernelMatrix(uncenteredMatrix.copy()) {

Parameters:

kernelMatrix - the kernel matrix to be centered

Returns:

centered kernelMatrix (for convenience)

getSimilarity

public double getSimilarity(DBIDRef id1,
                            DBIDRef id2)

Get the kernel similarity for the given objects.

Parameters:

id1 - First object

id2 - Second object

Returns:

Similarity.