pupil.db package

pupil.db.database module

class pupil.db.DataBase(vecdb: pupil.db.vector.VectorDB, metadb: pupil.db.meta.MetaDataDB)

Bases: object

add(metadata: Any, embeddings: NDArray2D)
classmethod from_encoder(data_schema: pupil.types.IsDataclass, label_col_name: str, data_col_name: str, data: pandas.core.frame.DataFrame, encoder: Callable[[...], NDArray2D])
get(i: Union[int, List[int], NDArray[Any, Int32]], return_embeddings: bool = False, return_metadata: bool = True) → GetDatabase

pupil.db.meta module

class pupil.db.PandasDB(schema: pupil.types.IsDataclass, label: str)

Bases: object

add(data: pandas.core.frame.DataFrame) → None

Add DataFrame to the database. It should contain the label column you passed to create the instance.

Parameters

data (pd.DataFrame) –

filter(field: str, value: Any) → List[pupil.types.IsDataclass]

Filter data base on columns and values

Parameters

  • field (str) – Name of the column

  • value (str) – Value to filter on

Raises

ValueError – if field not in the columns

Returns

List of schema objects

Return type

List[IsDataclass]

get(index: Union[int, Iterable[int]]) → List[pupil.types.IsDataclass]

Get data from DataFram

Parameters

index (Union[int, Iterable[int]]) – Row number

Returns

List of schema objects

Return type

List[IsDataclass]

property is_labeled
set_label(i: int, input: Any) → None

pupil.db.vector module

class pupil.db.FaissVectorDB(similarity_metric: pupil.types.Distance = Distance.COSINE, nlist: Optional[int] = None, nprobe: Optional[int] = 5)

Bases: object

add(embeddings: NDArray2D) → None

Add embeddings into the database

Parameters

embeddings (NDArray2D) – _description_

build_index(embeddings: NDArray2D) → None
search(query: NDArray2D, n_results: int = 4) → Tuple[NDArray2D, NDArray2D]

Search embeddings to get the closest embeddings to the queries.

Parameters

  • query (NDArray2D) – Vectors to search

  • n_results (int, optional) – Number of results per query. Defaults to 4.

Returns

Return (Distances, indices)

Return type

Tuple[NDArray2D, NDArray2D]