Neo4jVector.from_existing_index() requires a call to the embedding model

[tdennisliu]

Background

When loading an existing index via Neo4jVector.from_existing_index(), a singleton of Neo4jVector is created, which in init calls the embedding model to find the number of embedding dimensions..

        self.embedding_dimension = len(embedding.embed_query("foo"))

This is not necessary as right after instantiating the singleton, from_existing_index then gets the embedding dimension from the DB anyway.

        embedding_dimension, index_type = store.retrieve_existing_index()

Problem

This call to the embedding model is a headache in CICD, where you do not want to and can not call third-party embedding services during integration tests.

Proposed Solution

We can add an optional kwarg for embedding dimension during class initialisation that circumvents the need to call the embedding model:

        if embedding_dimension:
            self.embedding_dimension = embedding_dimension
        else:
            # Calculate embedding dimension
            self.embedding_dimension = len(embedding.embed_query("foo"))

Happy for feedback or thoughts and happy to open the PR myself.

[arwentwinkle]

@tdennisliu Let's open the PR and see if we can get it pushed through

[alexthomas93]

Closed by #31