edu.ucla.sspace.matrix

Class LocalityPreservingProjection

java.lang.Object

edu.ucla.sspace.matrix.LocalityPreservingProjection

public class LocalityPreservingProjection
extends Object

A Java wrapper around the Matlab implementation of the Spectral Regression form of Locality Preserving Projection (LPP), which is a linear-time subspace projection. Details on LPP may be found in the paper by He and Niyogi:

• Xiaofei He and Partha Niyogi, "Locality Preserving Projections," in Proceedings of Advances in Neural Information Processing Systems 16 (NIPS 2003). Vancouver Canada. 2003. Available here

Further information about the Spectral Regression version of the algorithm may be found on Deng Cai's webpage. This class requires that the following Matlab files availabe on the aforementioned webpage be made available in the root directory of the calling code: SR_caller.m, SR.m, lsqr2.m, and lars.m.

This class requires the availability of Matlab or Octave.

Each of the projection methods requires the prior calculation of the affinity matrix. This matrix identifies which rows in the input matrix have some degree of local proximity. A non-zero value in the affinity matrix specifies the weight of the connection between the two rows. Weights may be of any positive value. See AffinityMatrixCreator for programatic ways of creating an affinity matrix based on similarity methods.

See Also:

SVD, AffinityMatrixCreator

Method Summary

Methods

Modifier and Type	Method and Description
static MatrixFile	project(MatrixFile inputMatrix, MatrixFile affinityMatrix, int dimensions) Projects the rows of the input matrix into a lower dimensional subspace using the Locality Preserving Projection (LPP) algorithm and the affinity matrix as a guide to locality.
static Matrix	project(Matrix m, MatrixFile affinityMatrix, int dimensions) Projects the rows of the input matrix into a lower dimensional subspace using the Locality Preserving Projection (LPP) algorithm and the affinity matrix as a guide to locality.
static Matrix	project(Matrix m, Matrix affinityMatrix, int dimensions) Projects the rows of the input matrix into a lower dimensional subspace using the Locality Preserving Projection (LPP) algorithm and the affinity matrix as a guide to locality.

Methods inherited from class java.lang.Object

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

Method Detail

project

public static MatrixFile project(MatrixFile inputMatrix,
                 MatrixFile affinityMatrix,
                 int dimensions)

Projects the rows of the input matrix into a lower dimensional subspace using the Locality Preserving Projection (LPP) algorithm and the affinity matrix as a guide to locality.

Parameters:

inputMatrix - a matrix file whose rows will be projected

affinityMatrix - a square matrix whose entries denote locality between the rows of the inputMatrix. Note that this matrix's dimensions must be equal to the number of rows in the input matrix.

dimensions - the number of dimensions into which the inputMatrix should be projected

Returns:

a file containing the LPP-reduced data in DENSE_TEXT format.

Throws:

IOError - if any exception occurrs during processing

project

public static Matrix project(Matrix m,
             MatrixFile affinityMatrix,
             int dimensions)

Projects the rows of the input matrix into a lower dimensional subspace using the Locality Preserving Projection (LPP) algorithm and the affinity matrix as a guide to locality.

Parameters:

m - a matrix whose rows will be projected

affinityMatrix - a square matrix whose entries denote locality between the rows of the inputMatrix. Note that this matrix's dimensions must be equal to the number of rows in the input matrix.

dimensions - the number of dimensions into which the inputMatrix should be projected

Returns:

a Matrix that contains the rows of m projected into the specified number of dimensions

Throws:

IOError - if any exception occurrs during processing

project

public static Matrix project(Matrix m,
             Matrix affinityMatrix,
             int dimensions)

Projects the rows of the input matrix into a lower dimensional subspace using the Locality Preserving Projection (LPP) algorithm and the affinity matrix as a guide to locality.

Parameters:

m - a matrix whose rows will be projected

affinityMatrix - a square matrix whose entries denote locality between the rows of the inputMatrix. Note that this matrix's dimensions must be equal to the number of rows in the input matrix.

dimensions - the number of dimensions into which the inputMatrix should be projected

Returns:

a Matrix that contains the rows of m projected into the specified number of dimensions

Throws:

IOError - if any exception occurrs during processing