-
Notifications
You must be signed in to change notification settings - Fork 122
Description
We are currently trying to get Redis-om to work with redis-search for knn search on a HashModel.
Our model is the following:
class Document(HashModel):
vector: float = Field(
vector_options=VectorFieldOptions.flat(
type=VectorFieldOptions.TYPE.FLOAT32,
dimension=512,
distance_metric=VectorFieldOptions.DISTANCE_METRIC.COSINE
)
)When using the type float we can go into that line of code, that creates the correct vector type. However, a vector is actually NOT a float, so any creation of the model will fail.
Transforming the model to a valid model, we see that we need to use:
class Document(HashModel):
vector: list[float] = Field(
vector_options=VectorFieldOptions.flat(
type=VectorFieldOptions.TYPE.FLOAT32,
dimension=512,
distance_metric=VectorFieldOptions.DISTANCE_METRIC.COSINE
)
)That would allow us to create the correct VECTOR and use it. However, it seems that there's a check in __init__ that prevent any list to be a field in the HashModel
However, when creating the schema, there's a specific clause to check that the subscribed type, is supported if this is a list or a tuple that we are clearly in.
The question is:
How can we create a HashModel with vector list[float] to be able to create a knn_vector query?
Is it a bug where we need to skip the list and check that actually we support list and tuples in HashModel? Or should we use any other container to let us do the vector query?
Thanks already for the information!