-
Notifications
You must be signed in to change notification settings - Fork 1
Open
Description
Consistently, the website with the tutorials instructs users to create an index in postgresql using the access method lantern_hnsw.

This is directly contradicted by the jupyter notebooks in https://github.com/lanterndata/examples/tree/main

And it is unclear which, if any, is correct.
After following the tutorials, neither access method works for me.
# python3 setup_embeddings.py
INFO:root:Connecting to ourdb on localhost:5432
INFO:root:Table movies does not exist, creating it
INFO:root:Found 3883 embeddings with dimensionality 32
INFO:root:Creating L2 Norm index on movies
Traceback (most recent call last):
File "/root/projects/lantern_movie_recommender/setup_embeddings.py", line 42, in <module>
ourdb.execute(f"CREATE INDEX ON {constants.MOVIES_TABLE_NAME} USING lantern_hnsw (vector dist_l2sq_ops) WITH (dim={dimensions});")
psycopg2.errors.UndefinedObject: access method "lantern_hnsw" does not exist
# python3 setup_embeddings.py
INFO:root:Connecting to ourdb on localhost:5432
INFO:root:Table movies does not exist, creating it
INFO:root:Found 3883 embeddings with dimensionality 32
INFO:root:Creating L2 Norm index on movies
Traceback (most recent call last):
File "/root/projects/lantern_movie_recommender/setup_embeddings.py", line 46, in <module>
ourdb.execute(f"CREATE INDEX ON {constants.MOVIES_TABLE_NAME} USING hnsw (vector dist_l2sq_ops) WITH (dim={dimensions});")
psycopg2.errors.UndefinedObject: access method "hnsw" does not exist
to reproduce, feel free to reference my source code. https://github.com/aubreyyan/lantern_movie_recommender
Metadata
Metadata
Assignees
Labels
No labels