graphragzen.text_embedding.vector_databases.QdrantLocalVectorDatabase

class graphragzen.text_embedding.vector_databases.QdrantLocalVectorDatabase[source]

Attributes

distance_measure

vector_size

Methods

add_vectors(vectors)

Add vectors to the database

save(location)

This is specific to the Qdrant local VectorDatabae.

search(query_vectors, k[, score_threshold, ...])

Similarity search for each of the query vectors
__init__(vector_size=None, database_location=None, overwrite_existing_db=False, distance_measure='Cosine', on_disk=False)[source]

Create or load a local Qdrant vector database and a client for interaction.

Parameters:

  • vector_size (int, optional) – Length of the vectors to store. If a new database is created this must be provided. If a database if loaded this will be read from that database and the value provided here ignored.

  • database_location (str, optional) – Location to load the DB from or store a new DB. If not provided a new database will be created in qdrant/databases/. Defaults to None.

  • overwrite_existing_db (str, optional) – If True and a database is found at database_location it will be overwritten by a new database, otherwise the database found at database_location will be loaded. Defaults to False.

  • distance_measure (Literal['Cosine', 'Euclid', 'Dot', 'Manhattan'], optional) – Method to calculate distances between vectors. Defaults to ‘Cosine’

  • on_disk (bool, optional) – If true, vectors are served from disk, improving RAM usage at the cost of latency. Defaults to False.

Return type:

None

add_vectors(vectors)[source]

Add vectors to the database

Parameters:

vectors (List[dict]) – Each dict containing {“uuid”: …, “vector”: …}. Each dict may also contain the key and values {“metadata”: …}, which will be store with the vector and retrieved upon search.

Return type:

None

distance_measure: str
save(location)[source]

This is specific to the Qdrant local VectorDatabae. This function just copies the DB to a new location. Does not work for :memory: qdrant client instances since they life in RAM.

Parameters:

location (str) – path to store the DB

Return type:

None

search(query_vectors, k, score_threshold=0.0, filters={})[source]

Similarity search for each of the query vectors

Parameters:

  • query_vectors (np.ndarray) – Vectors for n queries, shaped (n x embedding_size)

  • k (int) – Max number of results to return per query vector.

  • score_threshold (float, optional) – Exclude all vector search results with a score worse than his. Defaults to 0.0

  • filters (dict, optional) – {“key”: “value_it_should_have”, “key2”: “value_it …..

Returns:

Per query List[dict] with each dict containing

{“uuid”: …, “score”: …, “metadata”: …}

Return type:

List[List[dict]]

vector_size: int