edu.ucla.sspace.matrix

Interface AffinityMatrixCreator

All Known Implementing Classes:

MinSimilarityAffinityMatrixCreator, NearestNeighborAffinityMatrixCreator

public interface AffinityMatrixCreator

A functional class that given a matrix or matrix file, computes the affinity matrix for the rows in that matrix. For an n x m matrix, the affinity matrix is an n x n matrix specifying which of the rows are connected; viewed as a graph, the affinity matrix specifies which rows have edges between them and what the weights of those edges are.

This class provides multiple options for specifying how the affinity matrix should be computed both in terms of the number of neighbors and the weights of the edges to those neighbors.

Due to the full density of the affinity matrix, the matrix is always written to disk, rather than stored in memory. This class also provides support for computing the affinity matrix off-core, i.e. without needing to load the matrix entirely into memory. If the off-core method is specified, the returned matrix will also be on disk.

This class is thread safe.

See Also:

LocalityPreservingProjection

Method Summary

Methods

Modifier and Type	Method and Description
MatrixFile	calculate(Matrix input) Computes the affinity matrix for the input matrix according to the specified similarity metrics, returning the result as a file on disk.
MatrixFile	calculate(MatrixFile input) Computes the affinity matrix for the input matrix according to the specified similarity metrics, returning the result as a file on disk.
MatrixFile	calculate(MatrixFile input, boolean useColumns) Computes the affinity matrix for the input matrix according to the specified similarity metrics, optionally treating the columns as data points, and returning the result as a file on disk.
void	setFunctions(SimilarityFunction edgeSim, SimilarityFunction kernelSim) Sets the SimilarityFunctions to be used for selecting edges and then weighting edges.
void	setParams(double... params) Sets any numeric parameters, such as thresholds, for this AffinityMatrixCreator.

Method Detail

setParams

void setParams(double... params)

Sets any numeric parameters, such as thresholds, for this AffinityMatrixCreator. Each AffinityMatrixCreator should specify how many parameters it expects and in what order.

setFunctions

void setFunctions(SimilarityFunction edgeSim,
                SimilarityFunction kernelSim)

Sets the SimilarityFunctions to be used for selecting edges and then weighting edges. edgeSim will be computed over every possible edge. The returned scores will be used to filter out valid edges in the graph. kernelSim will be used to compute the final edge weight between two valid edges in the graph, this is also called a kernel function.

Parameters:

edgeSim - The SimilarityFunction used to filter edges. This metric is assumed to be symmetric.

kernelSim - The SimilarityFunction used to weight valid edges. This metric may or may not be symmetric.

calculate

MatrixFile calculate(Matrix input)

Computes the affinity matrix for the input matrix according to the specified similarity metrics, returning the result as a file on disk. Because the input matrix is provided already in memory, all similarity comparisons are performed in memory, which offers a significant speed-up compared to the off-core overloads of this method that operate on a MatrixFile as input.

Parameters:

input - The Matrix that contains data points that are to be formed into an affinity matrix.

Returns:

the affinity matrix as a file on disk. The result may be read back into memory with MatrixFile.read().

calculate

MatrixFile calculate(MatrixFile input)

Computes the affinity matrix for the input matrix according to the specified similarity metrics, returning the result as a file on disk. The similarity comparisons are all performed off-core, which ensures that the input matrix does not need to be loaded in memory in its entirety in order to compute the affinity matrix.

Parameters:

input - The MatrixFile that contains data points that are to be formed into an affinity matrix.

Returns:

the affinity matrix as a file on disk. The result may be read back into memory with MatrixFile.read().

calculate

MatrixFile calculate(MatrixFile input,
                   boolean useColumns)

Computes the affinity matrix for the input matrix according to the specified similarity metrics, optionally treating the columns as data points, and returning the result as a file on disk. The similarity comparisons are all performed off-core, which ensures that the input matrix does not need to be loaded in memory in its entirety in order to compute the affinity matrix.

This method is primiarily intended to support computing the affinity matrix for the SVDLIBC_SPARSE_BINARY format which is stored in column-major order.

Parameters:

input - The MatrixFile that contains data points that are to be formed into an affinity matrix.

useColumns - true if the affinity matrix should be calculated for the columns of the input matrix, not the rows.

Returns:

the affinity matrix as a file on disk. The result may be read back into memory with MatrixFile.read().