Package org.apache.lucene.util.hnsw
Class HnswGraphBuilder
java.lang.Object
org.apache.lucene.util.hnsw.HnswGraphBuilder
- All Implemented Interfaces:
HnswBuilder
- Direct Known Subclasses:
InitializedHnswGraphBuilder,MergingHnswGraphBuilder
Builder for HNSW graph. See
HnswGraph for a gloss on the algorithm and the meaning of the
hyper-parameters.
Thread-safety: This class is NOT thread safe, it cannot be shared across threads, however, it IS safe for multiple HnswGraphBuilder to build the same graph, if the graph's size is known in the beginning (like when doing merge)
-
Nested Class Summary
Nested ClassesModifier and TypeClassDescriptionstatic final classA restricted, specialized knnCollector that can be used when building a graph. -
Field Summary
FieldsModifier and TypeFieldDescriptionprotected final HnswGraphBuilder.GraphBuilderKnnCollectorstatic final intDefault number of the size of the queue maintained while searching during a graph construction.static final intDefault number of maximum connections per nodeprotected booleanprotected final HnswGraphSearcherprotected final OnHeapHnswGraphstatic final StringA name for the HNSW component for the info-stream *protected final org.apache.lucene.util.hnsw.HnswLockprotected InfoStreamprotected final intstatic longRandom seed for level generation; public to expose for testing *protected final UpdateableRandomVectorScorer -
Constructor Summary
ConstructorsModifierConstructorDescriptionprotectedHnswGraphBuilder(RandomVectorScorerSupplier scorerSupplier, int M, int beamWidth, long seed, int graphSize) Reads all the vectors from vector values, builds a graph connecting them by their dense ordinals, using the given hyperparameter settings, and returns the resulting graph.protectedHnswGraphBuilder(RandomVectorScorerSupplier scorerSupplier, int M, int beamWidth, long seed, OnHeapHnswGraph hnsw) protectedHnswGraphBuilder(RandomVectorScorerSupplier scorerSupplier, int M, int beamWidth, long seed, OnHeapHnswGraph hnsw, org.apache.lucene.util.hnsw.HnswLock hnswLock, HnswGraphSearcher graphSearcher) Reads all the vectors from vector values, builds a graph connecting them by their dense ordinals, using the given hyperparameter settings, and returns the resulting graph.protectedHnswGraphBuilder(RandomVectorScorerSupplier scorerSupplier, int beamWidth, long seed, OnHeapHnswGraph hnsw) -
Method Summary
Modifier and TypeMethodDescriptionvoidaddGraphNode(int node) Inserts a doc with a vector value to the graphvoidaddGraphNode(int node, IntHashSet eps0) Inserts a doc with a vector value to the graph, searching on level 0 with provided entry pointsprotected voidaddVectors(int minOrd, int maxOrd) add vectors in range [minOrd, maxOrd)build(int maxOrd) Adds all nodes to the graph up to the providedmaxOrd.static HnswGraphBuildercreate(RandomVectorScorerSupplier scorerSupplier, int M, int beamWidth, long seed) static HnswGraphBuildercreate(RandomVectorScorerSupplier scorerSupplier, int M, int beamWidth, long seed, int graphSize) Once this method is called, no further updates to the graph are accepted (addGraphNode will throw IllegalStateException).getGraph()voidsetInfoStream(InfoStream infoStream) Set info-stream to output debugging information
-
Field Details
-
DEFAULT_MAX_CONN
public static final int DEFAULT_MAX_CONNDefault number of maximum connections per node- See Also:
-
DEFAULT_BEAM_WIDTH
public static final int DEFAULT_BEAM_WIDTHDefault number of the size of the queue maintained while searching during a graph construction.- See Also:
-
HNSW_COMPONENT
A name for the HNSW component for the info-stream *- See Also:
-
randSeed
public static long randSeedRandom seed for level generation; public to expose for testing * -
M
protected final int M -
scorer
-
graphSearcher
-
beamCandidates
-
hnsw
-
hnswLock
protected final org.apache.lucene.util.hnsw.HnswLock hnswLock -
infoStream
-
frozen
protected boolean frozen
-
-
Constructor Details
-
HnswGraphBuilder
protected HnswGraphBuilder(RandomVectorScorerSupplier scorerSupplier, int M, int beamWidth, long seed, int graphSize) throws IOException Reads all the vectors from vector values, builds a graph connecting them by their dense ordinals, using the given hyperparameter settings, and returns the resulting graph.- Parameters:
scorerSupplier- a supplier to create vector scorer from ordinals.M- – graph fanout parameter used to calculate the maximum number of connections a node can have – M on upper layers, and M * 2 on the lowest level.beamWidth- the size of the beam search to use when finding nearest neighbors.seed- the seed for a random number generator used during graph construction. Provide this to ensure repeatable construction.graphSize- size of graph, if unknown, pass in -1- Throws:
IOException
-
HnswGraphBuilder
protected HnswGraphBuilder(RandomVectorScorerSupplier scorerSupplier, int M, int beamWidth, long seed, OnHeapHnswGraph hnsw) throws IOException - Throws:
IOException
-
HnswGraphBuilder
protected HnswGraphBuilder(RandomVectorScorerSupplier scorerSupplier, int beamWidth, long seed, OnHeapHnswGraph hnsw) throws IOException - Throws:
IOException
-
HnswGraphBuilder
protected HnswGraphBuilder(RandomVectorScorerSupplier scorerSupplier, int M, int beamWidth, long seed, OnHeapHnswGraph hnsw, org.apache.lucene.util.hnsw.HnswLock hnswLock, HnswGraphSearcher graphSearcher) throws IOException Reads all the vectors from vector values, builds a graph connecting them by their dense ordinals, using the given hyperparameter settings, and returns the resulting graph.- Parameters:
scorerSupplier- a supplier to create vector scorer from ordinals.M- – graph fanout parameter used to calculate the maximum number of connections a node can have – M on upper layers, and M * 2 on the lowest level.beamWidth- the size of the beam search to use when finding nearest neighbors.seed- the seed for a random number generator used during graph construction. Provide this to ensure repeatable construction.hnsw- the graph to build, can be previously initialized- Throws:
IOException
-
-
Method Details
-
create
public static HnswGraphBuilder create(RandomVectorScorerSupplier scorerSupplier, int M, int beamWidth, long seed) throws IOException - Throws:
IOException
-
create
public static HnswGraphBuilder create(RandomVectorScorerSupplier scorerSupplier, int M, int beamWidth, long seed, int graphSize) throws IOException - Throws:
IOException
-
build
Description copied from interface:HnswBuilderAdds all nodes to the graph up to the providedmaxOrd.- Specified by:
buildin interfaceHnswBuilder- Parameters:
maxOrd- The maximum ordinal (excluded) of the nodes to be added.- Throws:
IOException
-
setInfoStream
Description copied from interface:HnswBuilderSet info-stream to output debugging information- Specified by:
setInfoStreamin interfaceHnswBuilder
-
getCompletedGraph
Description copied from interface:HnswBuilderOnce this method is called, no further updates to the graph are accepted (addGraphNode will throw IllegalStateException). Final modifications to the graph (eg patching up disconnected components, re-ordering node ids for better delta compression) may be triggered, so callers should expect this call to take some time.- Specified by:
getCompletedGraphin interfaceHnswBuilder- Throws:
IOException
-
getGraph
- Specified by:
getGraphin interfaceHnswBuilder
-
addVectors
add vectors in range [minOrd, maxOrd)- Throws:
IOException
-
addGraphNode
Description copied from interface:HnswBuilderInserts a doc with a vector value to the graph- Specified by:
addGraphNodein interfaceHnswBuilder- Throws:
IOException
-
addGraphNode
Description copied from interface:HnswBuilderInserts a doc with a vector value to the graph, searching on level 0 with provided entry points- Specified by:
addGraphNodein interfaceHnswBuilder- Throws:
IOException
-