Skip to main content
Version: User Guides (Cloud)

Index Vector Fields

This guide walks you through the basic operations on creating and managing indexes on vector fields in a collection.

Overview​

Leveraging the metadata stored in an index file, Zilliz Cloud organizes your data in a specialized structure, facilitating rapid retrieval of requested information during searches or queries.

Zilliz Cloud employs AUTOINDEX to enable efficient similarity searches. It also offers these metric types: Cosine Similarity (COSINE), Euclidean Distance (L2), Inner Product (IP), JACCARD, and HAMMING to measure the distances between vector embeddings. To learn more about vector field types and metrics, refer to Similarity Metrics Explained and Schema Explained.

It is recommended to create indexes for both the vector field and scalar fields that are frequently accessed.

If your collection contains more than one vector field, you can create an index for each vector field separately. For more information, refer to With multiple vector fields.

Preparations​

As explained in Manage Collections (SDKs), Zilliz Cloud automatically generates an index and loads it into memory when creating a collection if any of the following conditions are specified in the collection creation request:

  • The dimensionality of the vector field and the metric type, or

  • The schema and the index parameters.

The code snippet below repurposes the existing code to establish a connection to a Zilliz Cloud cluster and create a collection without specifying its index parameters. In this case, the collection lacks an index and remains unloaded.

from pymilvus import MilvusClient, DataType

CLUSTER_ENDPOINT = "YOUR_CLUSTER_ENDPOINT"
TOKEN = "YOUR_CLUSTER_TOKEN"

# 1. Set up a Milvus client
client = MilvusClient(
    uri=CLUSTER_ENDPOINT,
    token=TOKEN 
)

# 2. Create schema
# 2.1. Create schema
schema = MilvusClient.create_schema(
    auto_id=False,
    enable_dynamic_field=True,
)

# 2.2. Add fields to schema
schema.add_field(field_name="id", datatype=DataType.INT64, is_primary=True)
# The dim value should be an integer greater than 1
schema.add_field(field_name="vector", datatype=DataType.FLOAT_VECTOR, dim=5)

# 3. Create collection
client.create_collection(
    collection_name="customized_setup", 
    schema=schema, 
)

Index a Collection​

To create an index for a collection or index a collection, you need to set up the index parameters and call create_index().

# 4. Set up index
# 4.1. Set up the index parameters
index_params = MilvusClient.prepare_index_params()

# 4.2. Add an index on the vector field.
index_params.add_index(
    field_name="vector",
    metric_type="COSINE",
    index_type="AUTOINDEX",
    index_name="vector_index"
)

# 4.4. Create an index file
client.create_index(
    collection_name="customized_setup",
    index_params=index_params
)

# 5. Describe index
res = client.list_indexes(
    collection_name="customized_setup"
)

In the provided code snippet, we have established indexes on the vector field with the index type set to AUTOINDEX and metric type set to COSINE. Additionally, an index on a scalar field has been created with the index type AUTOINDEX. To learn more about the index type and metric types, read AUTOINDEX Explained and Similarity Metrics Explained.

📘Notes

Currently, you can create only one index file for each field in a collection.

Check Index Details​

Once you have created an index, you can check its details.

# 5. Describe index
res = client.list_indexes(
    collection_name="customized_setup"
)

print(res)

# Output
#
# [
#     "vector_index"
# ]

res = client.describe_index(
    collection_name="customized_setup",
    index_name="vector_index"
)

print(res)

# Output
#
# {
#     "index_type": "AUTOINDEX",
#     "metric_type": "COSINE",
#     "field_name": "vector",
#     "index_name": "vector_index"
# }

You can check the index file created on a specific field, and collect the statistics on the number of rows indexed using this index file.

Drop an Index​

You can simply drop an index if it is no longer needed.

📘Notes

Before dropping an index, make sure it has been released first.

# 6. Drop index
client.drop_index(
    collection_name="customized_setup",
    index_name="vector_index"
)