array_crop

ArrayCropParams

Bases: PadValueParams

Mixin class containing pad value parameters needed for array crop transformers.

getArrayLength

getArrayLength()

Gets the array length parameter.

Returns:

Type	Description
int	array length.

Source code in src/kamae/spark/transformers/array_crop.py

def getArrayLength(self) -> int:
    """
    Gets the array length parameter.
    :returns: array length.
    """
    return self.getOrDefault(self.arrayLength)

setArrayLength

setArrayLength(value)

Sets the parameter array length to the given value.

Parameters:

Name	Type	Description	Default
value	int	array length.	required

Returns:

Type	Description
ArrayCropParams	Instance of class mixed in.

Source code in src/kamae/spark/transformers/array_crop.py

def setArrayLength(self, value: int) -> "ArrayCropParams":
    """
    Sets the parameter array length to the given value.
    :param value: array length.
    :returns: Instance of class mixed in.
    """
    if value < 1:
        raise ValueError("Array length must be greater than 0.")
    return self._set(arrayLength=value)

ArrayCropTransformer

ArrayCropTransformer(
    inputCol=None,
    outputCol=None,
    inputDtype=None,
    outputDtype=None,
    layerName=None,
    arrayLength=128,
    padValue=None,
)

Bases: BaseTransformer, SingleInputSingleOutputParams, ArrayCropParams

Transformer that reshapes arrays into consistent shapes by either cropping or padding.

If the tensor is shorter than the specified length, it is padded with specified pad value.

Initialises the ArrayCropTransformer

Parameters:

Name	Type	Description	Default
inputCol	Optional[str]	Input column name.	None
outputCol	Optional[str]	Output column name.	None
inputDtype	Optional[Union[str, int, float]]	Input data type to cast input column(s) to before transforming.	None
outputDtype	Optional[Union[str, int, float]]	Output data type to cast the output column to after transforming.	None
layerName	Optional[str]	Name of the layer. Used as the name of the Keras layer	None
arrayLength	Optional[int]	The length to crop or pad the arrays to. Defaults to 128.	128
padValue	Optional[Union[str, int, float]]	The value pad the arrays with. Defaults to None.	None

Returns:

Type	Description
None	None

Source code in src/kamae/spark/transformers/array_crop.py

@keyword_only
def __init__(
    self,
    inputCol: Optional[str] = None,
    outputCol: Optional[str] = None,
    inputDtype: Optional[Union[str, int, float]] = None,
    outputDtype: Optional[Union[str, int, float]] = None,
    layerName: Optional[str] = None,
    arrayLength: Optional[int] = 128,
    padValue: Optional[Union[str, int, float]] = None,
) -> None:
    """
    Initialises the ArrayCropTransformer
    :param inputCol: Input column name.
    :param outputCol: Output column name.
    :param inputDtype: Input data type to cast input column(s) to before
    transforming.
    :param outputDtype: Output data type to cast the output column to after
    transforming.
    :param layerName: Name of the layer. Used as the name of the Keras layer
    :param arrayLength: The length to crop or pad the arrays to. Defaults to 128.
    :param padValue: The value pad the arrays with. Defaults to `None`.
    :returns: None
    """
    super().__init__()
    kwargs = self._input_kwargs
    self.setParams(**kwargs)
    self._pad_type_to_valid_element_types = {
        "int": ["int", "bigint", "smallint"],
        "float": ["float", "double", "decimal(10,0)"],
        "string": ["string"],
        "boolean": ["boolean"],
    }

compatible_dtypes property

compatible_dtypes

List of compatible data types for the layer. If the computation can be performed on any data type, return None.

Returns:

Type	Description
Optional[List[DataType]]	List of compatible data types for the layer.

get_keras_layer

get_keras_layer()

Gets the Keras layer that performs the array cropping and padding.

Returns:

Type	Description
Layer	Keras layer with name equal to the layerName parameter that performs the array cropping and padding operation.

Source code in src/kamae/spark/transformers/array_crop.py

def get_keras_layer(self) -> keras.layers.Layer:
    """
    Gets the Keras layer that performs the array cropping and padding.

    :returns: Keras layer with name equal to the layerName parameter
    that performs the array cropping and padding operation.
    """
    return ArrayCropLayer(
        name=self.getLayerName(),
        input_dtype=self.getInputKerasDtype(),
        output_dtype=self.getOutputKerasDtype(),
        array_length=self.getArrayLength(),
        pad_value=self.getPadValue(),
    )