Inserting binary vectors using SQLalchemy

[JungeAlexander]

Hello,

Thanks for the work on this, @ankane and team! It's amazing to see more vector types and distance metrics being added continuously :) I am trying to insert a binary vector using SQLalchemy and my code looks like this:

from pgvector.sqlalchemy import BIT
from sqlalchemy.orm import DeclarativeBase, Mapped, mapped_column

class Entity(DeclarativeBase):
    [...]
    embedding: Mapped[np.array] = mapped_column(BIT(2048))

I am then trying to insert embedding as a list[int] of 0s and 1s only. However, this raises the following error:

RepositoryError: An exception occurred: (psycopg.errors.DatatypeMismatch) column "embedding" is of type bit but expression is of type smallint[]
                                                             ^
HINT:  You will need to rewrite or cast the expression.

A similar approach works fine using a list[float] when inserting into an embedding of type pgvector.sqlalchemy.Vector.

My question is: which Python type do I need to cast embedding to before insertion?

Thanks for the help in advance!

[JungeAlexander]

Addendum: I tried to insert list[float] and list[bool] values as binary vectors but this did not work either.

[ankane]

Hi @JungeAlexander, it currently only accepts a bit string.

binary_embedding = [1, 0, 1]
''.join([str(v) for v in binary_embedding]) # '101'