hybrid_search()

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

📘Notes

This method applies only to dedicated serving clusters and on-demand compute.

• For this operation in a collection of a serving cluster, please create MilvusClient with the cluster endpoint.

  • Free & Serverless

    https://{cluster-id}.serverless.{region}.vectordb.zillizcloud.com

  • Dedicated

    https://{cluster-id}.{region}.vectordb.zillizcloud.com:19530

• For this operation in a collection for on-demand compute, create MilvusClient with the project endpoints, and then create a session to attach to an on-demand cluster for searches.

  https://{project-id}.{region}.api.zillizcloud.com

Request Syntax

python

hybrid_search(
    collection_name: str,
    reqs: List[AnnSearchRequest],
    ranker: Union[BaseRanker, Function],
    limit: int = 10,
    output_fields: Optional[List[str]] = None,
    timeout: Optional[float] = None,
    partition_names: Optional[List[str]] = None,
    **kwargs,
) -> SearchResult

PARAMETERS:

• collection_name (str) -
  [REQUIRED]
  The name of the collection to search.

• reqs (List[AnnSearchRequest]) -
  [REQUIRED]
  The ANN search requests combined by the hybrid search. Construct each request with AnnSearchRequest(data, anns_field, param, limit, expr=None, expr_params=None, filter=None).

  • data (Union[List, SparseMatrixInputType]) -
    [REQUIRED]
    The query vectors or sparse matrix used for this ANN search request.

  • anns_field (str) -
    [REQUIRED]
    The name of the vector field to search.

  • param (Dict) -
    [REQUIRED]
    The ANN search parameters, such as the metric type and search-specific settings.

  • limit (int) -
    [REQUIRED]
    The maximum number of matches returned by this ANN search request.

  • expr (Optional[str]) -
    Default: None
    The Boolean filtering expression applied before the ANN search. Do not provide both expr and filter.

  • expr_params (Optional[dict]) -
    Default: None
    The values substituted into expression-template placeholders.

  • filter (Optional[str]) -
    Default: None
    The alias for expr. Do not provide both values. The resolved expression is available through the read-only filter property as request.filter.

• ranker (Union[BaseRanker, Function]) -
  [REQUIRED]
  The ranker used to combine and order results from the search requests.

• limit (int) -
  Default: 10
  The maximum number of records to return, also known as topk.

• output_fields (Optional[List[str]]) -
  Default: None
  The scalar fields to include in each search result.

• timeout (Optional[float]) -
  Default: None
  The maximum time, in seconds, to wait for the RPC. When omitted, the client waits until the server responds or an error occurs.

• partition_names (Optional[List[str]]) -
  Default: None
  The names of the partitions to search.

• kwargs (Any) -
  The additional search options, including the pagination offset and consistency level.

RETURN TYPE:

SearchResult

RETURNS:

Search results for the combined ANN requests after applying each request's expression or filter.

EXCEPTIONS:

• MilvusException
  Raised when the server rejects the request or the RPC fails. Inspect the server error message for exact failure details.

Examples

The example constructs an ANN request and runs a hybrid search.

python

from pymilvus import AnnSearchRequest, MilvusClient

client = MilvusClient(uri="YOUR_CLUSTER_ENDPOINT")
request = AnnSearchRequest(
    data=[[0.1, 0.2, 0.3]],
    anns_field="vector",
    param={"metric_type": "COSINE"},
    limit=10,
    filter='category == "paper"',
)
results = client.hybrid_search(
    collection_name="book_chunks",
    reqs=[request],
    ranker=None,
    limit=10,
)
print(request.filter)
print(results)