Fix TypeError: 'float' object is not subscriptable for Sqlalchemy core

[PaxPrz]

Summary

When performing a cosine distance vector search using SQLAlchemy with the following code:

query_vector = sa.literal(embedding, type_=VECTOR)
query = sa.select(
    o.embedding_table.c.embedding_vector,
    (o.embedding_table.c.embedding_vector.op('<->')(query_vector)).label("cosine_distance")
).order_by(sa.asc("cosine_distance")).limit(1)

A TypeError is raised:

TypeError: 'float' object is not subscriptable

Details

The issue occurs because SQLAlchemy attempts to process the result using Vector._from_db, which is not appropriate for a scalar float result (i.e., the cosine distance). Vector._from_db expects a structure of numpy array, but receives a list of float from the database, leading to the error.

This line appears to be the source of the issue:

# Inside result_processor or similar internal hook
Vector._from_db(<float_value>)

[ankane]

See #132.