def get_nn()

in experiments/legacy/backend/nearest_neighbors.py [0:0]

• 16 lines of code
• 5 McCabe index (conditional complexity)

def get_nn(
    embeds: list[list[float]], 
    filters: list[str] = [], 
    num_neighbors: int = config.NUM_NEIGHBORS) -> list[Neighbor]:
    """Fetch nearest neigbhors in vector store.

    Neighbors are fetched independently for each embedding then unioned.

    Args:
        embeds: list of embeddings to find neareast neighbors
        filters: category prefix to restrict results to
        - example 1: ['Mens']
            will only return suggestiongs with top level category 'Mens'
        - example 2: ['Mens', 'Pants']
            will only return suggestions with top level category 'Mens'
            and second level category 'Pants'
        num_neigbhors: number of nearest neighbors to return for EACH embedding

    Returns:
        A list of named tuples containing the the following attributes
            id: unique item identifier, usually used to join to a reference DB
            distance: the embedding distance
    """
    if len(filters) > config.CATEGORY_DEPTH:
        logging.warning(f'''Number of category filters {len(filters)} is greater
         than supported category depth {config.CATEGORY_DEPTH}. Truncating''')
        filters = filters[:config.CATEGORY_DEPTH]

    filters = [Namespace(config.FILTER_CATEGORIES[i],[f]) for i,f in enumerate(filters)]
    response = index_endpoint.find_neighbors(
        deployed_index_id=config.DEPLOYED_INDEX,
        queries=embeds,
        num_neighbors=num_neighbors,
        filter=filters
    )
    return [Neighbor(r.id, r.distance) for neighbor in response for r in neighbor]