MultiModalEncoder

class falcon.tabular.processors.MultiModalEncoder(mask: List[ColumnTypes])

Applies different types of encodings on numerical, categorical, text and date/datetime features.

__init__(mask: List[ColumnTypes]) → None

Parameters: mask (List[ColumnTypes]) – provides a type for each column at a given index

fit(X: ndarray[Any, dtype[ScalarType]], y: Optional[Any] = None, *args: Any, **kwargs: Any) → None

Fits the encoder.

Parameters

X (npt.NDArray) – data to encode
_ (Any, optional) – dummy argument to keep compatibility with pipeline training

fit_pipe(X: Any, y: Any, *args: Any, **kwargs: Any) → Any

Equivalent of fit method that is used for elements chaining inisde pipeline during training.

Parameters

X (Any) – features
y (Any) – targets

Returns

usually None

Return type

Any

forward(X: ndarray[Any, dtype[object_]], *args: Any, **kwargs: Any) → ndarray[Any, dtype[ScalarType]]

Equivalent of .predict() or .transform().

Parameters: X (npt.NDArray[object]) – data to process
Returns: processed data
Return type: npt.NDArray

get_input_type() → Type

Returns: object
Return type: Type

get_output_type() → Type

Returns: Float32Array
Return type: Type

predict(X: ndarray[Any, dtype[ScalarType]], *args: Any, **kwargs: Any) → ndarray[Any, dtype[ScalarType]]

Applies the encoder.

Parameters: X (npt.NDArray) – input data
Returns: encoded data
Return type: npt.NDArray

to_onnx() → SerializedModelRepr

Serializes the encoder to onnx. Each feature in the original dataset is mapped to its own input node (float32 for numerical or string for categorical).

Return type: SerializedModelRepr

transform(X: ndarray[Any, dtype[ScalarType]], *args: Any, **kwargs: Any) → ndarray[Any, dtype[ScalarType]]

Equivalent of self.predict(X)

Parameters: X (npt.NDArray) – features
Returns: transformed features
Return type: npt.NDArray