No access to the similarity_search_by_vector function

[xindexer]

There's no way to call for a near_vector search function (similarity_search_by_vector_function). One of the earlier versions had a "by_text" boolean that was used to switch to a vector search. If that is put back in then you can access it again.

if self._by_text: return self._perform_search(query, k, **kwargs) else: if self._embedding is None: raise ValueError( "_embedding cannot be None for similarity_search when " "_by_text=False" ) embedding = self._embedding.embed_query(query) return self.similarity_search_by_vector(embedding, k, **kwargs)

Not the best solution in my opinion, but it is working

[hkhairy]

You can, by passing a keyword argument to the search method. It's not as straightforward as it used to be, because now it relies on a hidden keyword argument called search_method

So, if you call

weaviate_vstore.similarity_search(..., search_method = "near_vector")

It will call the near_vector in the _perform_search method. If you open the _perform_search method in the source code, you'll find this

if search_method == "hybrid":
    embedding = self._embedding.embed_query(query)
    result = collection.query.hybrid(
        query=query, vector=embedding, limit=k, **kwargs
    )
elif search_method == "near_vector":
    result = collection.query.near_vector(limit=k, **kwargs)

There's another problem though, in this block specifically

elif search_method == "near_vector":
  result = collection.query.near_vector(limit=k, **kwargs)

The embedding of the query isn't being passed. I've raised an issue and made a PR for that.

• near_vector in _perform_search isn't passed the embedding of the query #132
• fix(vectorstores.py): fetch embedding for the near_vector method #133

[xindexer]

Ok - I had tried to use the search_method but like you said it didn't pass the vector - I like your solution much better.

[hsm207]

@xindexer in this new integration, calling similarity_search is the way to do any search. This method calls weaviate's hybrid search under the hood. So, if you want to do vector search only, then either of the following will work:

weaviate_vector_store.similarity_search(query="hello", k=5, alpha=1)

weaviate_vector_store.similarity_search(query=None, vector=[1, 2, 3], k=5, alpha=1)

[hkhairy]

@xindexer
In case you don't know, hybrid search is a hybrid of 2 things:

• text search (using an algorithm called BM25, or Best Match 25),
• and vector search.

The hybrid search favors one search outcome over the other via a parameter called alpha. So, if alpha is 0.3, then the vector search will have a weight of 0.3, while the text search will have a weight of 0.7.

You can even control the weight on the level of the property. See here in the weaviate Docs

That's why, you'll find in @hsm207 answer, he provides the alpha = 1 in the first example.