org.openimaj.knn.pq

Class ShortProductQuantiserUtilities

java.lang.Object

org.openimaj.knn.pq.ShortProductQuantiserUtilities

public final class ShortProductQuantiserUtilities
extends Object

Utility methods for easily creating a ByteProductQuantiser using (Exact) K-Means.

Author:

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

Method Summary

Modifier and Type	Method and Description
static ShortProductQuantiser	train(short[][] data, int numAssigners, int nIter) Learn a ShortProductQuantiser by applying exact K-Means to sub-vectors extracted from the given data.
static ShortProductQuantiser	train(short[][] data, int numAssigners, int K, int nIter) Learn a ShortProductQuantiser by applying exact K-Means to sub-vectors extracted from the given data.

Methods inherited from class java.lang.Object

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

Method Detail

train

public static ShortProductQuantiser train(short[][] data,
                                          int numAssigners,
                                          int K,
                                          int nIter)

Learn a ShortProductQuantiser by applying exact K-Means to sub-vectors extracted from the given data. The length of the subvectors is determined by dividing the vector length by the requested number of assigners. The number of clusters per vector subset is constant, and must be less than 256.

Parameters:

data - the data to train the ShortProductQuantiser on.

numAssigners - the number of sub-quantisers to learn

K - the number of centroids per sub-quantiser

nIter - the maximum number of iterations for each k-means clustering

Returns:

a trained ShortProductQuantiser.

train

public static ShortProductQuantiser train(short[][] data,
                                          int numAssigners,
                                          int nIter)

Learn a ShortProductQuantiser by applying exact K-Means to sub-vectors extracted from the given data. The length of the subvectors is determined by dividing the vector length by the requested number of assigners. The number of clusters per vector subset is constant, and set at 256.

Parameters:

data - the data to train the ShortProductQuantiser on.

numAssigners - the number of sub-quantisers to learn

nIter - the maximum number of iterations for each k-means clustering

Returns:

a trained ShortProductQuantiser.