Package org.apache.lucene.search.suggest.document

Class NRTSuggesterBuilder

java.lang.Object

org.apache.lucene.search.suggest.document.NRTSuggesterBuilder

final class NRTSuggesterBuilder
extends java.lang.Object

Builder for NRTSuggester

Nested Class Summary

Nested Classes

Modifier and Type	Class	Description
private static class	NRTSuggesterBuilder.Entry

Field Summary

Fields

Modifier and Type	Field	Description
private BytesRefBuilder	analyzed
private Builder<PairOutputs.Pair<java.lang.Long,BytesRef>>	builder
static int	END_BYTE	Marks end of the analyzed input and start of dedup byte.
private int	endByte
private java.util.PriorityQueue<NRTSuggesterBuilder.Entry>	entries
private int	maxAnalyzedPathsPerOutput
private PairOutputs<java.lang.Long,BytesRef>	outputs
static int	PAYLOAD_SEP	Label used to separate surface form and docID in the output
private int	payloadSep
private IntsRefBuilder	scratchInts

Constructor Summary

Constructors

Constructor	Description
NRTSuggesterBuilder()	Create a builder for NRTSuggester

Method Summary

Modifier and Type	Method	Description
void	addEntry(int docID, BytesRef surfaceForm, long weight)	Adds an entry for the latest input term, should be called after startTerm(org.apache.lucene.util.BytesRef) on the desired input
void	finishTerm()	Writes all the entries for the FST input term
private static int	maxNumArcsForDedupByte(int currentNumDedupBytes)	Num arcs for nth dedup byte: if n <= 5: 1 + (2 * n) else: (1 + (2 * n)) * n
void	startTerm(BytesRef analyzed)	Initializes an FST input term to add entries against
boolean	store(DataOutput output)	Builds and stores a FST that can be loaded with NRTSuggester.load(IndexInput, CompletionPostingsFormat.FSTLoadMode))}

Methods inherited from class java.lang.Object

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

Field Detail

PAYLOAD_SEP

public static final int PAYLOAD_SEP

Label used to separate surface form and docID in the output

See Also:

Constant Field Values

END_BYTE

public static final int END_BYTE

Marks end of the analyzed input and start of dedup byte.

See Also:

Constant Field Values

outputs

private final PairOutputs<java.lang.Long,BytesRef> outputs

builder

private final Builder<PairOutputs.Pair<java.lang.Long,BytesRef>> builder

scratchInts

private final IntsRefBuilder scratchInts

analyzed

private final BytesRefBuilder analyzed

entries

private final java.util.PriorityQueue<NRTSuggesterBuilder.Entry> entries

payloadSep

private final int payloadSep

endByte

private final int endByte

maxAnalyzedPathsPerOutput

private int maxAnalyzedPathsPerOutput

Constructor Detail

NRTSuggesterBuilder

public NRTSuggesterBuilder()

Create a builder for NRTSuggester

Method Detail

startTerm

public void startTerm(BytesRef analyzed)

Initializes an FST input term to add entries against

addEntry

public void addEntry(int docID,
                     BytesRef surfaceForm,
                     long weight)
              throws java.io.IOException

Adds an entry for the latest input term, should be called after startTerm(org.apache.lucene.util.BytesRef) on the desired input

Throws:

java.io.IOException

finishTerm

public void finishTerm()
                throws java.io.IOException

Writes all the entries for the FST input term

Throws:

java.io.IOException

store

public boolean store(DataOutput output)
              throws java.io.IOException

Builds and stores a FST that can be loaded with NRTSuggester.load(IndexInput, CompletionPostingsFormat.FSTLoadMode))}

Throws:

java.io.IOException

maxNumArcsForDedupByte

private static int maxNumArcsForDedupByte(int currentNumDedupBytes)

Num arcs for nth dedup byte: if n <= 5: 1 + (2 * n) else: (1 + (2 * n)) * n

TODO: is there a better way to make the fst built to be more TopNSearcher friendly?