[説明] このページは機械翻訳された日本語版です。内容に誤りがございましたら、報告していただけると助かります。

search()

Addedv2.3.xModifiedv2.6.x

This operation conducts a vector similarity search with an optional scalar filtering expression.

search(data): Promise<ResStatus>

Request Syntax

milvusClient.search({
  db_name?: string,
  collection_name: string,
  partition_names?: string[];
  anns_field?: string; 
  data?: SearchDataType;
  output_fields?: string[];
  limit?: number;
  offset?: number;
  filter?: string;
  exprValues?: keyValueObj;
  params?: keyValueObj;
  metric_type?: string;
  consistency_level?: ConsistencyLevelEnum;
  ignore_growing?: boolean;
  group_by_field?: string;
  group_size?: number;
  strict_group_size?: boolean;
  hints?: string;
  round_decimal?: number;
  transformers?: OutputTransformers;
  rerank?: RankerObj | FunctionObject | FunctionScore;
})

PARAMETERS:

db_name (string) -

The name of the database to which the target collection belongs.
collection_name (string) -

[REQUIRED]

The name of the collection to search
partition_names (string[]) -

A list of the names of the partitions to search.
anns_field (string) -

The name of the target vector field for this operation. It is mandatory if you are searching in a collection with multiple vector fields.
data (number[] | number[][]) -

A list of vector embeddings.

Zilliz Cloud searches for the most similar vector embeddings to the specified ones.
output_fields (string[]) -

A list of field names to include in each entity in return.

The value defaults to None. If left unspecified, only the primary field is included.
limit (number) -

The total number of entities to return.

You can use this parameter in combination with offset in param to enable pagination.

The sum of this value and offset in param should be less than 16,384.

In a grouping search, however, limit specifies the maximum number of groups to return, rather than individual entities. Each group is formed based on the specified group_by_field.
offset (number) -

The number of records to skip in the search result.

You can use this parameter in combination with limit to enable pagination.

The sum of this value and limit should be less than 16,384.
filter (string) -

A scalar filtering condition to filter matching entities.

The value defaults to an empty string, indicating that no condition applies.

You can set this parameter to an empty string to skip scalar filtering. To build a scalar filtering condition, refer to Boolean Expression Rules.
exprValues (keyValueObj) -

If you choose to use placeholders in filter as stated in Filtering Templating, then you can specify the actual values for these placeholders as key-value pairs as the value of this parameter.
params (KeyValueObj) -

The additional search parameters in key-value pairs.
- radius (number) -
  
  Determines the threshold of least similarity. When setting metric_type to L2, ensure that this value is greater than that of range_filter. Otherwise, this value should be lower than that of range_filter.
- range_filter (number) -
  
  Refines the search to vectors within a specific similarity range. When setting metric_type to IP or COSINE, ensure that this value is greater than that of radius. Otherwise, this value should be lower than that of radius.
- level (number)
  
  Zilliz Cloud uses a unified parameter to simplify search parameter tuning instead of leaving you to work with a bunch of search parameters specific to various index algorithms.
  
  The value defaults to 1, and ranges from 1 to 5. Increasing the value results in a higher recall rate with degraded search performance.
- page_retain_order (bool) -
  
  Whether to retain the order of the search result when offset is provided.
  
  This parameter applies only when you also set radius.

metric_type (string) -

The metric type used to measure similarity between vectors. The value varies with the vector field type. The following table lists the mapping between vector field types and their supported metric types.

Field Type	Dimension Range	Supported Metric Types	Default Metric Type
`FLOAT_VECTOR`	2-32,768	`COSINE`, `L2`, `IP`	`COSINE`
`FLOAT16_VECTOR`	2-32,768	`COSINE`, `L2`, `IP`	`COSINE`
`BFLOAT16_VECTOR`	2-32,768	`COSINE`, `L2`, `IP`	`COSINE`
`INT8_VECTOR`	2-32,768	`COSINE`, `L2`, `IP`	`COSINE`
`SPARSE_FLOAT_VECTOR`	No need to specify the dimension.	`IP`, `BM25` (used only for full text search)	`IP`
`BINARY_VECTOR`	8-32,768*8	`HAMMING`, `JACCARD`, `MHJACCARD`	`HAMMING`

consistency_level (ConsistencyLevelEnum) -

The consistency level of the target collection. The value defaults to Bounded (1) with options of Strong (0), Bounded (1), Session (2), and Eventually (3).
ignore_growing (boolean) -

A boolean value indicating whether to skip the search in growing segments.
group_by_field (string) -

Groups search results by a specified field to ensure diversity and avoid returning multiple results from the same group.
group_size (number) -

The target number of entities to return within each group in a grouping search. For example, setting group_size=2 instructs the system to return up to 2 of the most similar entities (e.g., document passages or vector representations) within each group. Without setting group_size, the system defaults to returning only 1 entity per group.
strict_group_size (boolean) -

This Boolean parameter dictates whether group_size should be strictly enforced. When group_size=true, the system will attempt to fill each group with exactly group_size results, as long as sufficient data exists within each group. If there is an insufficient number of entities in a group, it will return only the available entities, ensuring that groups with adequate data meet the specified group_size.
hints (string) -

A hints string to improve search performance.
round_decimal (number) -

The number of decimal places to keep in the final results.
transformers (OutputTransformers) -

A custom function to convert data for the following data types:
- BFloat16Vector ((bf16bytes: Uint8Array) => BFloat16Vector;)
- Float16Vector ((f16: Uint8Array) => Float16Vector;)
- SparseFloatVector ((sparse: SparseVectorDic) => SparseFloatVector;)
rerank (RerankerObj | FunctionObject \ FunctionScore) -

A reranking strategy with its custom parameters. You can either use a RerankerObj, a FunctionObject, or a FunctionScore.

A RerankerObj has the following parameters:
- strategy (string) -
  
  A re-ranking strategy. Possible values are:
  - RRF ("rrf")
    
    This strategy is recommended when there is no specific emphasis. The RRF can effectively balance the importance of each vector field.
  - WEIGHTED ("weighted")
    
    This strategy is recommended if you require the results to emphasize a particular vector field. The WeightedRanker allows you to assign higher weights to certain vector fields, emphasizing them more. For instance, in multimodal searches, textual descriptions of an image might be considered more important than the colors in this image.
- params (keyValueObj) -
  
  The parameters are specific to reranking strategies.
  - When using the RRFRanker strategy, you need to input the parameter value k into the RRFRanker. The default value of k is 60. This parameter helps to determine how the ranks are combined from different ANN searches, aiming to balance and blend the importance across all searches.
  - When using the WeightedRanker strategy, you need to input weight values into the WeightedRanker function. The number of basic ANN searches in a Hybrid Search corresponds to the number of values that need to be inputted. The input values should be in the range [0,1], with values closer to 1 indicating greater importance.
A FunctionObject has the following structure.
- name (string)
  
  The name of the function. This identifier is used to reference the function within queries and collections.
- description (string)
  
  A brief description of the function’s purpose. This can be useful for documentation or clarity in larger projects and defaults to an empty string.
- type (FunctionType)
  
  The type of function for processing raw data. Possible values for this parameter isFunctionType.RERANK.
- input_field_names (string[])
  
  Leave this parameter value as an empty array.
A FunctionScore has the following structure.
- functions (FunctionObject[]) -
  
  A list of FunctionObject objects.
- params (keyValueObj) -
  
  Specifies how the specified functions work together. It has the following structure:
  - boost_mode (string) -
    
    Specifies how the specified weights influence the scores of any matching entities. Possible values are:
    - Multiply
      
      Indicates that the weighted value is equal to the original score of a matching entity multiplied by the specified weight.
      
      This is the default value.
    - Sum
      
      Indicates that the weighted value is equal to the sum of the original score of a matching entity and the specified weight
  - function_mode (string) -
    
    Specifies how the weighted values from various Boost Rankers are processed. Possible values are:
    - Multiply
      
      Indicates that the final score of a matching entity is equal to the product of the weighted values from all Boost Rankers.
      
      This is the default value.
    - Sum
      
      Indicates that the final score of a matching entity is equal to the sum of the weighted values from all Boost Rankers.

RETURNS Promise<SearchResults>

This method returns a promise that resolves to a SearchResults object.

{
    status: object,
    results: list[string],
    recalls: list[number]
}

PARAMETERS:

status (object) -
- code (number) -
  
  A code that indicates the operation result. It remains 0 if this operation succeeds.
- error_code (string | number) -
  
  An error code that indicates an occurred error. It remains Success if this operation succeeds.
- reason (string) -
  
  The reason that indicates the reason for the reported error. It remains an empty string if this operation succeeds.
results (list[object]) -

Each result object has the following keys:
- id (string) -
  
  The ID of the search result
- score(number) -
  
  The similarity score of the search result.
- Plus output fields and their values.
recalls (list[number]) -

Each number indicates the recall rate of a search against a query vector.

Example

const milvusClient = new milvusClient(MILUVS_ADDRESS);
const searchResults = await milvusClient.search({
   collection_name: 'my_collection',
   vector: [1, 2, 3, 4],
});

Request Syntax​

Example​

Request Syntax

Example