com.lancedb.lance.index.vector

Class VectorIndexParams

java.lang.Object

com.lancedb.lance.index.vector.VectorIndexParams

public class VectorIndexParams
extends Object

Parameters for creating a vector index.

Nested Class Summary

Nested Classes

Modifier and Type	Class and Description
static class	VectorIndexParams.Builder

Method Summary

Modifier and Type	Method and Description
DistanceType	getDistanceType()
Optional<HnswBuildParams>	getHnswParams()
IvfBuildParams	getIvfParams()
Optional<PQBuildParams>	getPqParams()
Optional<SQBuildParams>	getSqParams()
static VectorIndexParams	ivfFlat(int numPartitions, DistanceType distanceType) Create a new IVF index with flat quantizer.
static VectorIndexParams	ivfPq(int numPartitions, int numBits, int numSubVectors, DistanceType distanceType, int maxIterations) Create a new IVF index with PQ quantizer.
String	toString()
static VectorIndexParams	withIvfHnswPqParams(DistanceType distanceType, IvfBuildParams ivf, HnswBuildParams hnsw, PQBuildParams pq) Create a new IVF HNSW index with PQ quantizer.
static VectorIndexParams	withIvfHnswSqParams(DistanceType distanceType, IvfBuildParams ivf, HnswBuildParams hnsw, SQBuildParams sq) Create a new IVF HNSW index with SQ quantizer.
static VectorIndexParams	withIvfPqParams(DistanceType distanceType, IvfBuildParams ivf, PQBuildParams pq) Create a new IVF index with PQ quantizer.

Methods inherited from class java.lang.Object

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

Method Detail

ivfFlat

public static VectorIndexParams ivfFlat(int numPartitions,
                                        DistanceType distanceType)

Create a new IVF index with flat quantizer.

Parameters:

numPartitions - the number of partitions of IVF (Inverted File Index)

distanceType - the distance type for calculating the distance between vectors

Returns:

the VectorIndexParams

ivfPq

public static VectorIndexParams ivfPq(int numPartitions,
                                      int numBits,
                                      int numSubVectors,
                                      DistanceType distanceType,
                                      int maxIterations)

Create a new IVF index with PQ quantizer.

Parameters:

numPartitions - the number of partitions of IVF (Inverted File Index)

numBits - maps the float vectors to integer vectors, each integer is of num_bits. Now only 8 bits are supported

numSubVectors - the number of sub-vectors for PQ (Product Quantization)

distanceType - the distance type for calculating the distance between vectors

maxIterations - K-means max iterations. This will run k-means clustering on each subvector to determine the centroids that will be used to quantize the subvectors.

Returns:

the VectorIndexParams

withIvfPqParams

public static VectorIndexParams withIvfPqParams(DistanceType distanceType,
                                                IvfBuildParams ivf,
                                                PQBuildParams pq)

Create a new IVF index with PQ quantizer.

Parameters:

distanceType - the distance type for calculating the distance between vectors

ivf - the IVF build parameters

pq - the PQ build parameters

Returns:

the VectorIndexParams

withIvfHnswPqParams

public static VectorIndexParams withIvfHnswPqParams(DistanceType distanceType,
                                                    IvfBuildParams ivf,
                                                    HnswBuildParams hnsw,
                                                    PQBuildParams pq)

Create a new IVF HNSW index with PQ quantizer. The dataset is partitioned into IVF partitions, and each partition builds an HNSW graph.

Parameters:

distanceType - the distance type for calculating the distance between vectors

ivf - the IVF build parameters

hnsw - the HNSW build parameters

pq - the PQ build parameters

Returns:

the VectorIndexParams

withIvfHnswSqParams

public static VectorIndexParams withIvfHnswSqParams(DistanceType distanceType,
                                                    IvfBuildParams ivf,
                                                    HnswBuildParams hnsw,
                                                    SQBuildParams sq)

Create a new IVF HNSW index with SQ quantizer. The dataset is partitioned into IVF partitions, and each partition builds an HNSW graph.

Parameters:

distanceType - the distance type for calculating the distance between vectors

ivf - the IVF build parameters

hnsw - the HNSW build parameters

sq - the SQ build parameters

Returns:

the VectorIndexParams

getDistanceType

public DistanceType getDistanceType()

getIvfParams

public IvfBuildParams getIvfParams()

getPqParams

public Optional<PQBuildParams> getPqParams()

getHnswParams

public Optional<HnswBuildParams> getHnswParams()

getSqParams

public Optional<SQBuildParams> getSqParams()

toString

public String toString()

Overrides:

toString in class Object