jsat.linear.vectorcollection.lsh

Class RandomProjectionLSH<V extends Vec>

java.lang.Object

jsat.linear.vectorcollection.lsh.RandomProjectionLSH<V>

All Implemented Interfaces:

java.io.Serializable, java.lang.Cloneable, VectorCollection<V>

public class RandomProjectionLSH<V extends Vec>
extends java.lang.Object
implements VectorCollection<V>

An implementation of Locality Sensitive Hashing for the CosineDistance using random projections. This forms a vector collection that performs a linear search of the data set, but does so in a more efficient manner by comparing hamming distances in a byte array. However, the reported distances are only approximations - and may not be correct. For this reason the results are also approximate.

See:

• Charikar, M. S. (2002). Similarity estimation techniques from rounding algorithms. Proceedings of the thiry-fourth annual ACM symposium on Theory of computing - STOC ’02 (pp. 380–388). New York, New York, USA: ACM Press. doi:10.1145/509907.509965
• Durme, B. Van,&Lall, A. (2010). Online Generation of Locality Sensitive Hash Signatures. Proceedings of the ACL 2010 Conference Short Papers (pp. 231–235). Stroudsburg, PA, USA.

See Also:

Serialized Form

Constructor Summary

Constructors

Constructor and Description
RandomProjectionLSH(java.util.List<V> vecs, int ints, boolean inMemory) Creates a new Random Projection LSH object that uses a full matrix of normally distributed values.
RandomProjectionLSH(java.util.List<V> vecs, int ints, int poolSize) Creates a new Random Projection LSH object that uses a pool of normally distributed values to approximate a full matrix with considerably less memory storage.

Method Summary

Modifier and Type	Method and Description
void	build(boolean parallel, java.util.List<V> collection, DistanceMetric dm) Builds this metric index from the given collection of points using the given distance metric.
VectorCollection<V>	clone()
V	get(int indx) Accesses a vector from this collection via index.
java.util.List<java.lang.Double>	getAccelerationCache()
DistanceMetric	getDistanceMetric()
int	getSignatureBitLength() Returns the signature or encoding length in bits.
void	search(Vec query, double range, java.util.List<java.lang.Integer> neighbors, java.util.List<java.lang.Double> distances) Performs a range search of the current collection.
void	search(Vec query, int numNeighbors, java.util.List<java.lang.Integer> neighbors, java.util.List<java.lang.Double> distances) Performs k-Nearest Neighbor search of the current collection.
void	setDistanceMetric(DistanceMetric dm) Sets the distance metric used for this collection.
int	size() Returns the number of vectors stored in the collection

Methods inherited from class java.lang.Object

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

Methods inherited from interface jsat.linear.vectorcollection.VectorCollection

build, build, build, getVecs, search, search, search, search, search, search

Constructor Detail

RandomProjectionLSH

public RandomProjectionLSH(java.util.List<V> vecs,
                           int ints,
                           boolean inMemory)

Creates a new Random Projection LSH object that uses a full matrix of normally distributed values.

Parameters:

vecs - the list of vectors to form a collection for

ints - the number of integers to use for the encoding

inMemory - true to construct the full matrix in memory, or false to construct the needed values on demand. This reduces memory use at increased CPU usage.

RandomProjectionLSH

public RandomProjectionLSH(java.util.List<V> vecs,
                           int ints,
                           int poolSize)

Creates a new Random Projection LSH object that uses a pool of normally distributed values to approximate a full matrix with considerably less memory storage.

Parameters:

vecs - the list of vectors to form a collection for

ints - the number of integers to use for the encoding

poolSize - the number of normally distributed random variables to store. Matrix values will be pulled on demand from an index in the pool of values.

Method Detail

getAccelerationCache

public java.util.List<java.lang.Double> getAccelerationCache()

Specified by:

getAccelerationCache in interface VectorCollection<V extends Vec>

build

public void build(boolean parallel,
                  java.util.List<V> collection,
                  DistanceMetric dm)

Description copied from interface: VectorCollection

Builds this metric index from the given collection of points using the given distance metric.

Specified by:

build in interface VectorCollection<V extends Vec>

Parameters:

parallel - true if the index should be built in parallel, or false if it should be done in a single thread.

collection - the list of vectors to put into the index

dm - the distance metric to build the index using.

search

public void search(Vec query,
                   double range,
                   java.util.List<java.lang.Integer> neighbors,
                   java.util.List<java.lang.Double> distances)

Description copied from interface: VectorCollection

Performs a range search of the current collection. The index and distance of each found neighbor will be placed into the given Lists.

Specified by:

search in interface VectorCollection<V extends Vec>

Parameters:

query - the point to search for the neighbors within a given radius.

range - the radius to search for all the neighbors with a distance ≤ range.

neighbors - the list to store the index of the neighbors in. Will be sorted by distance to the query, and paired with the values in distances.

distances - the list to store the distance of the neighbors to the query in. Will be sorted, and paired with the values in neighbors.

search

public void search(Vec query,
                   int numNeighbors,
                   java.util.List<java.lang.Integer> neighbors,
                   java.util.List<java.lang.Double> distances)

Description copied from interface: VectorCollection

Performs k-Nearest Neighbor search of the current collection. The index and distance of each found neighbor will be placed into the given Lists.

Specified by:

search in interface VectorCollection<V extends Vec>

Parameters:

query - the point to search for the k-nearest neighbors of

numNeighbors - the number of neighbors k to search for.

neighbors - the list to store the index of the neighbors in. Will be sorted by distance to the query, and paired with the values in distances.

distances - the list to store the distance of the neighbors to the query in. Will be sorted, and paired with the values in neighbors.

getSignatureBitLength

public int getSignatureBitLength()

Returns the signature or encoding length in bits.

Returns:

the signature length in bits

size

public int size()

Description copied from interface: VectorCollection

Returns the number of vectors stored in the collection

Specified by:

size in interface VectorCollection<V extends Vec>

Returns:

the size of the collection

get

public V get(int indx)

Description copied from interface: VectorCollection

Accesses a vector from this collection via index.

Specified by:

get in interface VectorCollection<V extends Vec>

Parameters:

indx - the index in [0, VectorCollection.size()) of the vector to access

Returns:

the vector from the collection

clone

public VectorCollection<V> clone()

Specified by:

clone in interface VectorCollection<V extends Vec>

Overrides:

clone in class java.lang.Object

setDistanceMetric

public void setDistanceMetric(DistanceMetric dm)

Description copied from interface: VectorCollection

Sets the distance metric used for this collection.

Specified by:

setDistanceMetric in interface VectorCollection<V extends Vec>

Parameters:

dm - the distance metric to use

getDistanceMetric

public DistanceMetric getDistanceMetric()

Specified by:

getDistanceMetric in interface VectorCollection<V extends Vec>

Returns:

the distance metric to use