Package org.apache.lucene.document

Class LateInteractionField

java.lang.Object

org.apache.lucene.document.Field

org.apache.lucene.document.BinaryDocValuesField

org.apache.lucene.document.LateInteractionField

All Implemented Interfaces:

IndexableField

public class LateInteractionField
extends BinaryDocValuesField

A field for storing multi-vector values for late interaction models.

The value is stored as a binary payload, and can be retrieved using decode(BytesRef). Multi-vectors are expected to have the same dimension for each composing token vector. This is stored along with the token vectors in the first 4 bytes of the payload.

Note: This field does not ensure consistency in token vector dimensions for values across documents.

Nested Class Summary

Nested classes/interfaces inherited from class org.apache.lucene.document.Field

Field.Store

Field Summary

Fields inherited from class org.apache.lucene.document.BinaryDocValuesField

TYPE

Fields inherited from class org.apache.lucene.document.Field

fieldsData, name, type

Constructor Summary

Constructors

Constructor	Description
LateInteractionField(String name, float[][] value)	Creates a new LateInteractionField from the provided multi-vector matrix.

Method Summary

Modifier and Type	Method	Description
static float[][]	decode(BytesRef payload)	Decodes provided BytesRef into a multi-vector matrix.
static BytesRef	encode(float[][] value)	Encodes provided multi-vector matrix into a BytesRef that can be stored in the LateInteractionField.
float[][]	getValue()	Returns the multi-vector value stored in this field
void	setValue(float[][] value)	Set multi-vector value for the field.

Methods inherited from class org.apache.lucene.document.Field

binaryValue, fieldType, getCharSequenceValue, invertableType, name, numericValue, readerValue, setBytesValue, setBytesValue, setByteValue, setDoubleValue, setFloatValue, setIntValue, setLongValue, setReaderValue, setShortValue, setStringValue, setTokenStream, storedValue, stringValue, tokenStream, tokenStreamValue, toString

Methods inherited from class java.lang.Object

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

Constructor Details

LateInteractionField

public LateInteractionField(String name,
 float[][] value)

Creates a new LateInteractionField from the provided multi-vector matrix.

Parameters:

name - field name

value - multi-vector value

Method Details

setValue

public void setValue(float[][] value)

Set multi-vector value for the field.

Value should not be null or empty. Composing token vectors for provided multi-vector value should have the same dimension.

getValue

public float[][] getValue()

Returns the multi-vector value stored in this field

encode

public static BytesRef encode(float[][] value)

Encodes provided multi-vector matrix into a BytesRef that can be stored in the LateInteractionField.

Composing token vectors for the multi-vector are expected to have the same dimension, which is stored along with the token vectors in the first 4 bytes of the payload. Use decode(BytesRef) to retrieve the multi-vector.

Parameters:

value - Multi-Vector to encode

Returns:

BytesRef representation for provided multi-vector

decode

public static float[][] decode(BytesRef payload)

Decodes provided BytesRef into a multi-vector matrix.

The token vectors are expected to have the same dimension, which is stored along with the token vectors in the first 4 bytes of the payload. Meant to be used as a counterpart to encode(float[][])

Parameters:

payload - to decode into multi-vector value

Returns:

decoded multi-vector value