Using `models.Batch` with `client.update_vectors` about qdrant-client HOT 9 CLOSED

yep - at bottom of batch update section of points docs it says

To batch many points with a single operation type, please use batching functionality in that operation directly.

which from earlier in the same page (under upload points) would suggest to me that client.update_vectors should accept a models.Batch set of points.

Not a major issue if just clarified that instead it requires the record-oriented equivalent using PointVectors, like

client.update_vectors(
    collection_name="test",
    points=[
        models.PointVectors(
            id="test-point", vector={"test-vec": [np.random.rand() for _ in range(768)]},
        )
    ],
)

which works

from qdrant-client.

Hi @fitzgeraldja

Could you please provide an example?

I tried running this code and there is no error for me

from qdrant_client import QdrantClient, models
client = QdrantClient()
client.recreate_collection('test', vectors_config=models.VectorParams(size=2, distance=models.Distance.COSINE))
client.upsert('test', points=models.Batch(ids=[1,2], vectors=[[0.0, 0.2], [0.1, 0.2]]), wait=True)

My dependencies:

python3.10.12
qdrant-client==1.7.0
pydantic==2.5.3
pydantic_core==2.14.6

from qdrant-client.

I have same dependencies other than python3.10.13 - issue is specifically for client.update_vectors rather than client.upsert, which in the source code has specific handling for models.Batch type points argument. Have tested with

client.update_vectors(
    collection_name="test",
    points=models.Batch(
        ids=["test-point"],
        vectors={"test-vec": [[np.random.rand() for _ in range(768)]]},
    ),
)

for an existing collection

from qdrant-client.

Some versions of pydantic used to convert numpy arrays to lists implicitly, however it is no longer the case.
Batch structure does not accept numpy arrays on its own, you need to convert it to a list

from qdrant-client.

in example given, np.random.rand() just returns a float, so that shouldn't be the issue (i.e. vectors is of type Dict[StrictStr, List[List[StrictFloat]]] as required for models.Batch) - it's not models.Batch(...) that throws the error, but client.update_vectors itself

from qdrant-client.

I checked the documentation at qdrant.tech

https://qdrant.tech/documentation/concepts/points/#update-vectors

Example here is indeed incorrect, we will fix it!

It should be a sequence of PointVectors. Batch or PointStruct models contain payload field, which is not relevant for update_vectors operation.

from qdrant-client.

A correct example may look like this:

client.update_vectors(
    collection_name=collection_name,
    points=[
        models.PointVectors(
            id=1,
            vector={
                [0.1, 0.1,],
            },
        ),
        models.PointVectors(
            id=2,
            vector={
            [0.2, 0.2],
            },
        ),
    ],
)

from qdrant-client.

qdrant/landing_page#541

from qdrant-client.

the documentation has been updated

from qdrant-client.