org.openimaj.ml.clustering.kmeans

Class KMeansConfiguration<NN extends NearestNeighbours<DATA,?,?>,DATA>

java.lang.Object

org.openimaj.ml.clustering.kmeans.KMeansConfiguration<NN,DATA>

Type Parameters:

NN - The type of NearestNeighbours to use

DATA - The type of data

All Implemented Interfaces:

Cloneable

public class KMeansConfiguration<NN extends NearestNeighbours<DATA,?,?>,DATA>
extends Object
implements Cloneable

Configuration for the K-Means algorithm.

Author:

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

Field Summary

Fields

Modifier and Type	Field and Description
protected int	blockSize The size of processing blocks for each thread
static int	DEFAULT_BLOCK_SIZE The default number of samples per parallel assignment instance.
static int	DEFAULT_NUMBER_ITERATIONS The default number of iterations.
protected NearestNeighboursFactory<? extends NN,DATA>	factory The factory for producing the NearestNeighbours objects used in assignment.
protected int	K The number of clusters
protected int	niters The max number of iterations
protected ExecutorService	threadpool The threadpool for parallel processing

Constructor Summary

Constructors

Constructor and Description
KMeansConfiguration() A completely default configuration used primarily as a convenience function for reading.
KMeansConfiguration(int K, NearestNeighboursFactory<? extends NN,DATA> nnFactory) Create configuration for data that will create K clusters.
KMeansConfiguration(int K, NearestNeighboursFactory<? extends NN,DATA> nnFactory, int niters) Create configuration for data that will create K clusters.
KMeansConfiguration(int K, NearestNeighboursFactory<? extends NN,DATA> nnFactory, int niters, ExecutorService threadpool) Create configuration for data that will create K clusters.
KMeansConfiguration(int K, NearestNeighboursFactory<? extends NN,DATA> nnFactory, int niters, int blockSize, ExecutorService threadpool) Create configuration for data with M dimensions that will create K clusters.

Method Summary

Methods

Modifier and Type	Method and Description
KMeansConfiguration<NN,DATA>	clone()
int	getBlockSize() Get the number of samples processed in a batch by a thread.
int	getK() Get the number of clusters
int	getMaxIterations() Get the maximum allowed number of iterations.
NearestNeighboursFactory<? extends NN,DATA>	getNearestNeighbourFactory() Get the factory that produces the NearestNeighbours during clustering.
int	numClusters() Get the number of clusters
void	setBlockSize(int blockSize) Set the number of samples processed in a batch by a thread.
void	setK(int k) Set the number of clusters
void	setMaxIterations(int niters) Set the maximum allowed number of iterations.
void	setNearestNeighbourFactory(NearestNeighboursFactory<? extends NN,DATA> factory) Set the factory that produces the NearestNeighbours during clustering.
void	setNumClusters(int k) Set the number of clusters

Methods inherited from class java.lang.Object

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

Field Detail

DEFAULT_BLOCK_SIZE

public static final int DEFAULT_BLOCK_SIZE

The default number of samples per parallel assignment instance.

See Also:

Constant Field Values

DEFAULT_NUMBER_ITERATIONS

public static final int DEFAULT_NUMBER_ITERATIONS

The default number of iterations.

See Also:

Constant Field Values

K

protected int K

The number of clusters

factory

protected NearestNeighboursFactory<? extends NN extends NearestNeighbours<DATA,?,?>,DATA> factory

The factory for producing the NearestNeighbours objects used in assignment.

blockSize

protected int blockSize

The size of processing blocks for each thread

niters

protected int niters

The max number of iterations

threadpool

protected ExecutorService threadpool

The threadpool for parallel processing

Constructor Detail

KMeansConfiguration

public KMeansConfiguration(int K,
                   NearestNeighboursFactory<? extends NN,DATA> nnFactory)

Create configuration for data that will create K clusters. The algorithm will run for a maximum of DEFAULT_NUMBER_ITERATIONS iterations, and make use of all available processors, processing with blocks of DEFAULT_BLOCK_SIZE vectors.

The specified NearestNeighboursFactory determines the actual type of k-means that will be performed; it could be exact nearest-neighbours, or it could be an approximate method, for example based on an ensemble of kd-trees.

Parameters:

K - number of clusters to be found

nnFactory - the factory for producing the NearestNeighbours.

KMeansConfiguration

public KMeansConfiguration(int K,
                   NearestNeighboursFactory<? extends NN,DATA> nnFactory,
                   int niters)

Create configuration for data that will create K clusters. The algorithm will run for a maximum of the given number of iterations, and will make use of all available processors, processing with blocks of DEFAULT_BLOCK_SIZE vectors.

The specified NearestNeighboursFactory determines the actual type of k-means that will be performed; it could be exact nearest-neighbours, or it could be an approximate method, for example based on an ensemble of kd-trees.

Parameters:

K - number of clusters to be found

nnFactory - the factory for producing the NearestNeighbours.

niters - number of iterations

KMeansConfiguration

public KMeansConfiguration(int K,
                   NearestNeighboursFactory<? extends NN,DATA> nnFactory,
                   int niters,
                   ExecutorService threadpool)

Create configuration for data that will create K clusters. The algorithm will run for a maximum of the given number of iterations, and will make use of the provided threadpool, processing with blocks of DEFAULT_BLOCK_SIZE vectors.

The specified NearestNeighboursFactory determines the actual type of k-means that will be performed; it could be exact nearest-neighbours, or it could be an approximate method, for example based on an ensemble of kd-trees.

Parameters:

K - number of clusters to be found

nnFactory - the factory for producing the NearestNeighbours.

threadpool - threadpool to use for parallel processing

niters - number of training iterations

KMeansConfiguration

public KMeansConfiguration(int K,
                   NearestNeighboursFactory<? extends NN,DATA> nnFactory,
                   int niters,
                   int blockSize,
                   ExecutorService threadpool)

Create configuration for data with M dimensions that will create K clusters. The algorithm will run for a maximum of the given number of iterations, and will make use of nThreads processors, processing with blocks of DEFAULT_BLOCK_SIZE vectors.

The specified NearestNeighboursFactory determines the actual type of k-means that will be performed; it could be exact nearest-neighbours, or it could be an approximate method, for example based on an ensemble of kd-trees.

Parameters:

K - number of clusters to be found

nnFactory - the factory for producing the NearestNeighbours.

threadpool - threadpool to use for parallel processing

blockSize - number of samples per parallel thread

niters - number of training iterations

KMeansConfiguration

public KMeansConfiguration()

A completely default configuration used primarily as a convenience function for reading. The number of dimensions, number of clusters and nearest-neighbours factory must be set before the configuration is used.

Method Detail

clone

public KMeansConfiguration<NN,DATA> clone()

Overrides:

clone in class Object

getK

public int getK()

Get the number of clusters

Returns:

the number of clusters

setK

public void setK(int k)

Set the number of clusters

Parameters:

k - the number of clusters

numClusters

public int numClusters()

Get the number of clusters

Returns:

the number of clusters

setNumClusters

public void setNumClusters(int k)

Set the number of clusters

Parameters:

k - the number of clusters

getBlockSize

public int getBlockSize()

Get the number of samples processed in a batch by a thread. This needs to be small enough that that the memory isn't exhausted, but big enough for the thread to have enough data to work for a while.

Returns:

the the number of samples processed in a batch by a thread

setBlockSize

public void setBlockSize(int blockSize)

Set the number of samples processed in a batch by a thread. This needs to be small enough that that the memory isn't exhausted, but big enough for the thread to have enough data to work for a while.

Parameters:

blockSize - the number of samples processed in a batch by a thread

getMaxIterations

public int getMaxIterations()

Get the maximum allowed number of iterations.

Returns:

the maximum allowed number of iterations.

setMaxIterations

public void setMaxIterations(int niters)

Set the maximum allowed number of iterations.

Parameters:

niters - the maximum allowed number of iterations.

getNearestNeighbourFactory

public NearestNeighboursFactory<? extends NN,DATA> getNearestNeighbourFactory()

Get the factory that produces the NearestNeighbours during clustering.

Returns:

the factory

setNearestNeighbourFactory

public void setNearestNeighbourFactory(NearestNeighboursFactory<? extends NN,DATA> factory)

Set the factory that produces the NearestNeighbours during clustering.

Parameters:

factory - the factory to set