mnemosine.search.Search.vector

Kind: method

Signature

Search.vector(self, vector, top_k=10, metric='cosine')

Documentation

k-NN over stored embeddings.

Delegates to mnemosine.vec.VectorStore.knn, which uses the sqlite-vec virtual table when available and a brute-force scan otherwise.

Args:

• vector (sequence of float): The query embedding.
• top_k (int): Number of nearest neighbours to return.
• metric (str): Distance metric: "cosine" (default), "l2"
• or "dot".

Returns:

• list of dict: Each item has node (Node) and distance
• (float), ordered nearest-first.

Raises:

• VectorError: If top_k < 1 or the vector dimension conflicts
• with stored embeddings.

Example:

>>> db.embed_fn = lambda t: [1.0, 0.0, 0.0]
>>> doc.add_embedding()
>>> [r["node"].id for r in db.search.vector([1.0, 0.0, 0.0], top_k=1)]
['01G...']