Module org.elasticsearch.server

Package org.elasticsearch.index.codec.vectors.cluster

Class HierarchicalKMeans

java.lang.Object

org.elasticsearch.index.codec.vectors.cluster.HierarchicalKMeans

public class HierarchicalKMeans
extends Object

An implementation of the hierarchical k-means algorithm that better partitions data than naive k-means

Field Summary

Fields

Modifier and Type	Field	Description
static final float	DEFAULT_SOAR_LAMBDA
static final int	MAX_ITERATIONS_DEFAULT
static final int	MAXK
static final int	NO_SOAR_ASSIGNMENT
static final int	SAMPLES_PER_CLUSTER_DEFAULT

Method Summary

Modifier and Type	Method	Description
KMeansResult	cluster(org.apache.lucene.index.FloatVectorValues vectors, int targetSize)	clusters the set of vectors by starting with a rough number of partitions and then recursively refining those lastly a pass is made to adjust nearby neighborhoods and add an extra assignment per vector to nearby neighborhoods
static HierarchicalKMeans	ofConcurrent(int dimension, org.apache.lucene.search.TaskExecutor executor, int numWorkers)
static HierarchicalKMeans	ofConcurrent(int dimension, org.apache.lucene.search.TaskExecutor executor, int numWorkers, int maxIterations, int samplesPerCluster, int clustersPerNeighborhood, float soarLambda)
static HierarchicalKMeans	ofSerial(int dimension)
static HierarchicalKMeans	ofSerial(int dimension, int maxIterations, int samplesPerCluster, int clustersPerNeighborhood, float soarLambda)

Methods inherited from class java.lang.Object

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

Field Details

MAXK

public static final int MAXK

See Also:

• Constant Field Values

MAX_ITERATIONS_DEFAULT

public static final int MAX_ITERATIONS_DEFAULT

See Also:

• Constant Field Values

SAMPLES_PER_CLUSTER_DEFAULT

public static final int SAMPLES_PER_CLUSTER_DEFAULT

See Also:

• Constant Field Values

DEFAULT_SOAR_LAMBDA

public static final float DEFAULT_SOAR_LAMBDA

See Also:

• Constant Field Values

NO_SOAR_ASSIGNMENT

public static final int NO_SOAR_ASSIGNMENT

See Also:

• Constant Field Values

Method Details

ofSerial

public static HierarchicalKMeans ofSerial(int dimension)

ofSerial

public static HierarchicalKMeans ofSerial(int dimension,
 int maxIterations,
 int samplesPerCluster,
 int clustersPerNeighborhood,
 float soarLambda)

ofConcurrent

public static HierarchicalKMeans ofConcurrent(int dimension,
 org.apache.lucene.search.TaskExecutor executor,
 int numWorkers)

ofConcurrent

public static HierarchicalKMeans ofConcurrent(int dimension,
 org.apache.lucene.search.TaskExecutor executor,
 int numWorkers,
 int maxIterations,
 int samplesPerCluster,
 int clustersPerNeighborhood,
 float soarLambda)

cluster

public KMeansResult cluster(org.apache.lucene.index.FloatVectorValues vectors,
 int targetSize)
                     throws IOException

clusters the set of vectors by starting with a rough number of partitions and then recursively refining those lastly a pass is made to adjust nearby neighborhoods and add an extra assignment per vector to nearby neighborhoods

Parameters:

vectors - the vectors to cluster

targetSize - the rough number of vectors that should be attached to a cluster

Returns:

the centroids and the vectors assignments and SOAR (spilled from nearby neighborhoods) assignments

Throws:

IOException - is thrown if vectors is inaccessible