multi indexing

[ranjith502]

i team

i have image embeddings and text (product ) embeddings of type dense

how to store both embeddings in the collection

below are the list of things i tired

i have created schema by defining both image and text embeddings as float vector

then inserted both vectors and created index with only image embeddings

below is the error message

<MilvusException: (code=65535, message=there is no vector index on field: [product_embeddings], please create index firstly)>

then i created index for both. below is the error message
<MilvusException: (code=65535, message=CreateIndex failed: at most one distinct index is allowed per field)>,

what is the correct way to do it

below is my use case
in the collection i need to store both embeddings
i need two indexing one for text and another for image

[yhmo]

To call search(), you must call load_collection() first.
To load_collection(), you must call create_index() to define index type for each embedding field.

An example of two embedding fields in one collection, and search/hybrid_search usage.

import time

from pymilvus import (
    MilvusClient, DataType, AnnSearchRequest, RRFRanker
)

import random

client = MilvusClient(
    uri="http://localhost:19530",
)
print(client.get_server_version())

collection_name = "BBB"

# collection schema
schema = MilvusClient.create_schema()
schema.add_field(field_name="id", datatype=DataType.INT64, is_primary=True, auto_id=False)
schema.add_field(field_name="text_embedding", datatype=DataType.FLOAT_VECTOR, dim=256)
schema.add_field(field_name="image_embedding", datatype=DataType.FLOAT_VECTOR, dim=512)
schema.add_field(field_name="text_title", datatype=DataType.VARCHAR, max_length=100)
schema.add_field(field_name="image_title", datatype=DataType.VARCHAR, max_length=100)


client.drop_collection(collection_name=collection_name)
client.create_collection(
    collection_name=collection_name,
    schema=schema,
)
print(collection_name, "created")

def gen_embedding(dim):
    return [random.random() for _ in range(dim)]

# insert some data
data = [
    {"id": 1, "text_embedding": gen_embedding(256), "image_embedding": gen_embedding(512), "text_title": "text_1", "image_title": "image_1"},
    {"id": 2, "text_embedding": gen_embedding(256), "image_embedding": gen_embedding(512), "text_title": "text_2", "image_title": "image_2"},
    {"id": 3, "text_embedding": gen_embedding(256), "image_embedding": gen_embedding(512), "text_title": "text_3", "image_title": "image_3"},
    {"id": 4, "text_embedding": gen_embedding(256), "image_embedding": gen_embedding(512), "text_title": "text_4", "image_title": "image_4"},
    {"id": 5, "text_embedding": gen_embedding(256), "image_embedding": gen_embedding(512), "text_title": "text_5", "image_title": "image_5"},
]
client.insert(collection_name=collection_name, data=data)
print(collection_name, "inserted data")

# create index
index_params = client.prepare_index_params()
# index for "text_embedding" field
index_params.add_index(
    field_name="text_embedding",
    index_type="HNSW",
    metric_type="COSINE",
    params={},
)
# inde for "image_embedding" field
index_params.add_index(
    field_name="image_embedding",
    index_type="IVF_FLAT",
    metric_type="L2",
    params={},
)
client.create_index(collection_name=collection_name, index_params=index_params)
# load collection after create_index
client.load_collection(collection_name=collection_name)

# search text embeddings
vector = [random.random() for _ in range(256)]
results = client.search(collection_name=collection_name,
                data=[vector],
                anns_field="text_embedding",
                limit=3,
                output_fields=["text_title"],
                search_params={})
print("search result of text embedding")
for res in results:
    print("============================")
    for r in res:
        print(r)

print("===========================================================")
# hybrid search on two embedding fields
req_list = []
search_param = {
    "data": [[random.random() for _ in range(256)]],
    "anns_field": "text_embedding",
    "param": {"metric_type": "COSINE"},
    "limit": 5,
}
req1 = AnnSearchRequest(**search_param)
req_list.append(req1)

search_param = {
    "data": [[random.random() for _ in range(512)]],
    "anns_field": "image_embedding",
    "param": {"metric_type": "L2"},
    "limit": 5,
}
req2 = AnnSearchRequest(**search_param)
req_list.append(req2)

hybrid_res = client.hybrid_search(collection_name=collection_name,
                                  reqs=req_list,
                                  ranker=RRFRanker(10),
                                  limit=3,
                                  output_fields=["text_title", "image_title"])
print("hybrid search result of text embedding and image embedding")
for hits in hybrid_res:
    for hit in hits:
        print(hit)

[ranjith502]

thanks i tired
it worked