org.openimaj.math.matrix

Class ThinSingularValueDecomposition

java.lang.Object

org.openimaj.math.matrix.ThinSingularValueDecomposition

public class ThinSingularValueDecomposition
extends Object

Thin SVD based on Adrian Kuhn's wrapper around ARPACK. This can scale to really large matrices (bigger than RAM), given an implementation of Matrix that is backed by disk.

Note that the current version of (Java)ARPACK is not thread-safe. Allowances have been made in this implementation to synchronize the call to SingularValues.decompose() against the ThinSingularValueDecomposition class. Care must be taken if you are using JARPACK outside this class in a multi-threaded application.

Author:

Jonathon Hare (jsh2@ecs.soton.ac.uk)

Field Summary

Fields

Modifier and Type	Field and Description
double[]	S The singular values
Jama.Matrix	U The U matrix
Jama.Matrix	Vt The transpose of the V matrix

Constructor Summary

Constructors

Constructor and Description
ThinSingularValueDecomposition(Jama.Matrix matrix, int ndims) Perform thin SVD on matrix, calculating at most ndims dimensions.
ThinSingularValueDecomposition(Matrix matrix, int ndims) Perform thin SVD on matrix, calculating at most ndims dimensions.

Method Summary

Modifier and Type	Method and Description
Jama.Matrix	getSmatrix()
Jama.Matrix	getSmatrixSqrt()
static Jama.Matrix	reduceRank(Jama.Matrix m, int rank) Reduce the rank of the input matrix using the thin SVD to get a lower rank least-squares estimate of the input.
protected double[]	reverse(double[] vector)
protected Jama.Matrix	vectorArrayToMatrix(Vector[] vectors, boolean rows)

Methods inherited from class java.lang.Object

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

Field Detail

U

public Jama.Matrix U

The U matrix

S

public double[] S

The singular values

Vt

public Jama.Matrix Vt

The transpose of the V matrix

Constructor Detail

ThinSingularValueDecomposition

public ThinSingularValueDecomposition(Jama.Matrix matrix,
                                      int ndims)

Perform thin SVD on matrix, calculating at most ndims dimensions.

Parameters:

matrix - the matrix

ndims - the number of singular values/vectors to calculate; actual number may be less.

ThinSingularValueDecomposition

public ThinSingularValueDecomposition(Matrix matrix,
                                      int ndims)

Perform thin SVD on matrix, calculating at most ndims dimensions.

Parameters:

matrix - the matrix

ndims - the number of singular values/vectors to calculate; actual number may be less.

Method Detail

reverse

protected double[] reverse(double[] vector)

vectorArrayToMatrix

protected Jama.Matrix vectorArrayToMatrix(Vector[] vectors,
                                          boolean rows)

getSmatrix

public Jama.Matrix getSmatrix()

Returns:

The S matrix

getSmatrixSqrt

public Jama.Matrix getSmatrixSqrt()

Returns:

The sqrt of the singular vals as a matrix.

reduceRank

public static Jama.Matrix reduceRank(Jama.Matrix m,
                                     int rank)

Reduce the rank of the input matrix using the thin SVD to get a lower rank least-squares estimate of the input.

Parameters:

m - matrix to reduce the rank of

rank - the desired rank

Returns:

the rank-reduced matrix