semantic_router.index.postgres

MetricPgVecOperatorMap Objects

class MetricPgVecOperatorMap(Enum)

Enum to map the metric to PostgreSQL vector operators.

dotproduct

inner product

euclidean

L2 distance

manhattan

L1 distance

parse_vector

def parse_vector(vector_str: Union[str, Any]) -> List[float]

Parses a vector from a string or other representation.

Arguments:

vector_str (Union[str, Any]): The string or object representation of a vector.

Returns:

List[float]: A list of floats representing the vector.

clean_route_name

def clean_route_name(route_name: str) -> str

Cleans and formats the route name by stripping spaces and replacing them with hyphens.

Arguments:

route_name (str): The original route name.

Returns:

str: The cleaned and formatted route name.

PostgresIndexRecord Objects

class PostgresIndexRecord(BaseModel)

Model to represent a record in the Postgres index.

init

def __init__(**data) -> None

Initializes a new Postgres index record with given data.

Arguments:

data (dict): Field values for the record.

to_dict

def to_dict() -> Dict

Converts the record to a dictionary.

Returns:

Dict: A dictionary representation of the record.

PostgresIndex Objects

class PostgresIndex(BaseIndex)

Postgres implementation of Index.

init

def __init__(connection_string: Optional[str] = None,
             index_prefix: str = "semantic_router_",
             index_name: str = "index",
             metric: Metric = Metric.COSINE,
             namespace: Optional[str] = "",
             dimensions: int | None = None,
             init_async_index: bool = False)

Initializes the Postgres index with the specified parameters.

Arguments:

connection_string (Optional[str]): The connection string for the PostgreSQL database.
index_prefix (str): The prefix for the index table name.
index_name (str): The name of the index table.
dimensions (int): The number of dimensions for the vectors.
metric (Metric): The metric used for vector comparisons.
Optional[str]0 (Optional[str]): An optional namespace for the index.
Optional[str]2 (Optional[str]3): Whether to initialize the index asynchronously.

add

def add(embeddings: List[List[float]],
        routes: List[str],
        utterances: List[str],
        metadata_list: Optional[List[Dict[str, Any]]] = None,
        sparse_embeddings: Optional[List[SparseEmbedding]] = None,
        **kwargs) -> None

Upserts records into the index, one per (route, utterance) pair. The table

is created on first use.

Arguments:

embeddings (List[List[float]]): A list of vector embeddings to add.
routes (List[str]): A list of route names corresponding to the embeddings.
utterances (List[Any]): A list of utterances corresponding to the embeddings.
metadata_list (Optional[List[Dict[str, Any]]]): A list of metadata corresponding to the embeddings.
sparse_embeddings (Optional[List[SparseEmbedding]]): Not supported, PostgresIndex is dense only.

Raises:

List[List[float]]0: If the vector embeddings being added do not match the expected dimensions.

aadd

async def aadd(embeddings: List[List[float]],
               routes: List[str],
               utterances: List[str],
               metadata_list: Optional[List[Dict[str, Any]]] = None,
               sparse_embeddings: Optional[List[SparseEmbedding]] = None,
               batch_size: int = 100,
               **kwargs) -> None

Asynchronously upserts records into the index in batches, one per (route,

utterance) pair. The table is created on first use.

Arguments:

embeddings: A list of vector embeddings to add.
routes: A list of route names corresponding to the embeddings.
utterances: A list of utterances corresponding to the embeddings.
metadata_list: (Optional) List of metadata dictionaries.
sparse_embeddings: Not supported, PostgresIndex is dense only.
batch_size: Number of records per batch insert.

Raises:

ValueError: If the vector embeddings don't match expected dimensions.

delete

def delete(route_name: str) -> None

Deletes records with the specified route name.

Arguments:

route_name (str): The name of the route to delete records for.

adelete

async def adelete(route_name: str) -> list[str]

Asynchronously delete specified route from index if it exists. Returns the IDs

of the vectors deleted.

Arguments:

route_name (str): Name of the route to delete.

Returns:

list[str]: List of IDs of the vectors deleted.

describe

def describe() -> IndexConfig

Describes the index by returning its type, dimensions, and total vector count.

Returns:

IndexConfig: An IndexConfig object containing the index's type, dimensions, and total vector count.

is_ready

def is_ready() -> bool

Checks if the index is ready to be used.

Returns:

bool: True if the index is ready, False otherwise.

ais_ready

async def ais_ready() -> bool

Checks if the index is ready to be used.

Returns:

bool: True if the index is ready, False otherwise.

query

def query(
    vector: np.ndarray,
    top_k: int = 5,
    route_filter: Optional[List[str]] = None,
    sparse_vector: dict[int, float] | SparseEmbedding | None = None
) -> Tuple[np.ndarray, List[str]]

Searches the index for the query vector and returns the top_k results.

Arguments:

vector (np.ndarray): The query vector.
top_k (int): The number of top results to return.
route_filter (Optional[List[str]]): Optional list of routes to filter the results by.
sparse_vector (dict[int, float] | SparseEmbedding | None): Not used, PostgresIndex is dense only.

Returns:

Tuple[np.ndarray, List[str]]: A tuple containing the scores and routes of the top_k results.

aquery

async def aquery(
    vector: np.ndarray,
    top_k: int = 5,
    route_filter: Optional[List[str]] = None,
    sparse_vector: dict[int, float] | SparseEmbedding | None = None
) -> Tuple[np.ndarray, List[str]]

Asynchronously search the index for the query vector and return the top_k results.

Arguments:

vector (np.ndarray): The query vector to search for.
top_k (int, optional): The number of top results to return, defaults to 5.
route_filter (Optional[List[str]], optional): A list of route names to filter the search results, defaults to None.
sparse_vector (dict[int, float] | SparseEmbedding | None): Not used, PostgresIndex is dense only.

Returns:

Tuple[np.ndarray, List[str]]: A tuple containing an array of scores and a list of route names.