hybridSearch()

This operation performs multi-vector search on a collection and returns search results after reranking.

public SearchResp hybridSearch(HybridSearchReq request)

Request Syntax

hybridSearch(HybridSearchReq.builder()
    .collectionName(String collectionName)
    .databasename(String databaseName)
    .partitionNames(List<String> partitionNames)
    .searchRequests(List<AnnSearchReq> searchRequests)
    .ranker(BaseRanker ranker)
    .topK(int topK)
    .outFields(List<String> outFields)
    .roundDecimal(int roundDecimal)
    .consistencyLevel(ConsistencyLevel consistencyLevel)
    .build()
)

BUILDER METHODS:

• collectionName(String collectionName)

  The name of an existing collection.

• databasename(String databaseName)

  Target database name(optional).
  

• partitionNames(List<String> partitionNames)

  A list of partition names.

• searchRequests(List<AnnSearchReq> searchRequests)

  A list of search requests, where each request is an AnnSearchReq object. Each request corresponds to a different vector field and a different set of search parameters.

  • AnnSearchReq

    A class representing an ANN search request. Its builder methods are as follows:

    • vectorFieldName(String): The vector field to use in the request.

    • topK(int): The maximum number of results to return in the request. When performing a hybrid search with multiple ANN search requests, the top results defined by topK from each request will be combined and re-ranked before returning the final search results.

    • expr(String): The expression to filter the results(Optional).

    • vectors(List<BaseVec>): The query vector to search in the request. BaseVector is a base class for abstract vector classes. The following classes are derived from BaseVector. Choose the correct class as input according to DataType of the vector field.

      Class Name	Constructors	Description
      FloatVec	FloatVec(List data) FloatVec(float[] data)	For DataType.FloatVector type field.
      BinaryVec	BinaryVec(ByteBuffer data) BinaryVec(byte[] data)	For DataType.BinaryVector type field.
      Float16Vec	Float16Vec(ByteBuffer data) Float16Vec(byte[] data)	For DataType.Float16Vector type field.
      BFloat16Vec	BFloat16Vec(ByteBuffer data) BFloat16Vec(byte[] data)	For DataType.BFloat16Vector type field.
      SparseFloatVec	SparseFloatVec(SortedMap data)	For DataType.SparseFloatVector type field.

  • params(String)

    A JSON dictionary format string of search parameters for the request.

• topK(int topK)

  The total number of entities to return.

• outFields(List<String> outFields)

  A list of field names to include in each entity in return. The value defaults to null. If left unspecified, only the primary field is included.

• roundDecimal(int roundDecimal)

  The number of decimal places to which Milvus rounds the calculated distances. The value defaults to -1, indicating that Milvus skips rounding the calculated distances and returns the raw value.

• consistencyLevel(ConsistencyLevel consistencyLevel)

  The consistency level of the target collection. The value defaults to the one specified when you create the current collection.

RETURN TYPE:

SearchResp

RETURNS:

A SearchResp object representing specific search results with the specified output fields and relevance score.

PARAMETERS:

• searchResults (List<List<SearchResult>>)

  A list of SearchResp.SearchResult, the size of searchResults equals the number of query vectors of the search. Each List<SearchResult> is a top-K result of a query vector. Each SearchResult represents an entity hit by the search.

  Member of SearchResult:
  

EXCEPTIONS:

• MilvusClientExceptions

  This exception will be raised when any error occurs during this operation.

Example

import io.milvus.v2.client.ConnectConfig;
import io.milvus.v2.client.MilvusClientV2;
import io.milvus.v2.common.ConsistencyLevel;
import io.milvus.v2.service.vector.request.AnnSearchReq;
import io.milvus.v2.service.vector.request.HybridSearchReq;
import io.milvus.v2.service.vector.request.data.BinaryVec;
import io.milvus.v2.service.vector.request.data.FloatVec;
import io.milvus.v2.service.vector.request.data.SparseFloatVec;
import io.milvus.v2.service.vector.request.ranker.RRFRanker;
import io.milvus.v2.service.vector.response.SearchResp;

import java.nio.ByteBuffer;

// 1. Set up a client
ConnectConfig connectConfig = ConnectConfig.builder()
        .uri("http://localhost:19530")
        .token("root:Milvus")
        .build();
        
MilvusClientV2 client = new MilvusClientV2(connectConfig);

// 2. Setup input
List<Float> floatVector = generateFolatVector();
ByteBuffer binaryVector = generateBinaryVector();
SortedMap<Long, Float> sparseVector = generateSparseVector();

List<AnnSearchReq> searchRequests = new ArrayList<>();
searchRequests.add(AnnSearchReq.builder()
        .vectorFieldName("float_vector")
        .vectors(Collections.singletonList(new FloatVec(floatVector)))
        .params("{\"nprobe\": 10}")
        .topK(10)
        .build());
searchRequests.add(AnnSearchReq.builder()
        .vectorFieldName("binary_vector")
        .vectors(Collections.singletonList(new BinaryVec(binaryVector)))
        .topK(50)
        .build());
searchRequests.add(AnnSearchReq.builder()
        .vectorFieldName("sparse_vector")
        .vectors(Collections.singletonList(new SparseFloatVec(sparseVector)))
        .topK(100)
        .build());

// 3. Hybrid search
HybridSearchReq hybridSearchReq = HybridSearchReq.builder()
        .collectionName(randomCollectionName)
        .searchRequests(searchRequests)
        .ranker(new RRFRanker(20))
        .topK(10)
        .consistencyLevel(ConsistencyLevel.BOUNDED)
        .build();
SearchResp searchResp = client.hybridSearch(hybridSearchReq);

List<List<SearchResp.SearchResult>> searchResults = searchResp.getSearchResults();
List<SearchResp.SearchResult> results = searchResults.get(0); // nq = 1, searchResults size is 1
for (SearchResp.SearchResult result : results) {
    System.out.printf("ID: %d, Score: %f, %s\n", (long)result.getId(), result.getScore(), result.getEntity().toString());
}