module npy.onnx_numpy_annotation

Short summary

module mlprodict.npy.onnx_numpy_annotation

numpy annotations.

Classes

class	truncated documentation
_NDArrayAlias	Ancestor to custom signature.
NDArray	Used to annotation ONNX numpy functions.
NDArraySameType	Shortcut to simplify signature description.
NDArraySameTypeSameShape	Shortcut to simplify signature description.
NDArrayType	Shortcut to simplify signature description.
NDArrayTypeSameShape	Shortcut to simplify signature description.

Functions

function	truncated documentation
get_args_kwargs	Extracts arguments and optional parameters of a function.

Static Methods

staticmethod	truncated documentation
__class_getitem__	Overwrites this method.
_process_type	Nicknames such as floats, int, ints, all can be used to describe multiple inputs for a signature. …
_process_type	Nicknames such as floats, int, ints, all can be used to describe multiple inputs for a signature. …
_process_type	Nicknames such as floats, int, ints, all can be used to describe multiple inputs for a signature. …
_process_type	Nicknames such as floats, int, ints, all can be used to describe multiple inputs for a signature. …
_process_type	Nicknames such as floats, int, ints, all can be used to describe multiple inputs for a signature. …

Methods

method	truncated documentation
__init__
__init__
__init__
__init__
__init__	constructor
__repr__	usual
__repr__	usual
__repr__	usual
__repr__	usual
__repr__	usual
_get_output_types	Tries to infer output types.
_get_output_types	Tries to infer output types.
_get_output_types	Tries to infer output types.
_get_output_types	Tries to infer output types.
_get_output_types	Tries to infer output types.
get_inputs_outputs	Returns the list of inputs, outputs.
get_inputs_outputs	Returns the list of inputs, outputs.
get_inputs_outputs	Returns the list of inputs, outputs.
get_inputs_outputs	Returns the list of inputs, outputs.
get_inputs_outputs	Returns the list of inputs, outputs.
shape_calculator	Returns expected dimensions given the input dimensions.
shape_calculator	Returns expected dimensions given the input dimensions.
shape_calculator	Returns expected dimensions given the input dimensions.
shape_calculator	Returns expected dimensions given the input dimensions.
shape_calculator	Returns expected dimensions given the input dimensions.

Documentation

numpy annotations.

New in version 0.6.

source on GitHub

class mlprodict.npy.onnx_numpy_annotation.NDArray

Bases: ndarray, Generic[Shape, DType]

Used to annotation ONNX numpy functions.

New in version 0.6.

source on GitHub

class ShapeType(params)

Bases: object

Stores shape information.

__init__(params)

classmethod __class_getitem__(params)

Overwrites this method.

__orig_bases__ = (<class 'numpy.ndarray'>, typing.Generic[~Shape, ~DType])

__parameters__ = (~Shape, ~DType)

class mlprodict.npy.onnx_numpy_annotation.NDArraySameType(dtypes=None)

Bases: NDArrayType

Shortcut to simplify signature description.

Parameters:

dtypes – input dtypes

New in version 0.6.

source on GitHub

constructor

__init__(dtypes=None)

constructor

__repr__()

usual

class mlprodict.npy.onnx_numpy_annotation.NDArraySameTypeSameShape(dtypes=None)

Bases: NDArraySameType

Shortcut to simplify signature description.

Parameters:

dtypes – input dtypes

New in version 0.6.

source on GitHub

constructor

__init__(dtypes=None)

constructor

class mlprodict.npy.onnx_numpy_annotation.NDArrayType(dtypes=None, dtypes_out=None, n_optional=None, nvars=False)

Bases: _NDArrayAlias

Shortcut to simplify signature description.

Parameters:

• dtypes – input dtypes

• dtypes_out – output dtypes

• n_optional – number of optional parameters, 0 by default

• nvars – True if the function allows an infinite number of inputs, this is incompatible with parameter n_optional.

New in version 0.6.

source on GitHub

constructor

__init__(dtypes=None, dtypes_out=None, n_optional=None, nvars=False)

constructor

class mlprodict.npy.onnx_numpy_annotation.NDArrayTypeSameShape(dtypes=None, dtypes_out=None, n_optional=None, nvars=False)

Bases: NDArrayType

Shortcut to simplify signature description.

Parameters:

• dtypes – input dtypes

• dtypes_out – output dtypes

• n_optional – number of optional parameters, 0 by default

• nvars – True if the function allows an infinite number of inputs, this is incompatible with parameter n_optional.

New in version 0.6.

source on GitHub

constructor

__init__(dtypes=None, dtypes_out=None, n_optional=None, nvars=False)

constructor

class mlprodict.npy.onnx_numpy_annotation._NDArrayAlias(dtypes=None, dtypes_out=None, n_optional=None, nvars=False)

Bases: object

Ancestor to custom signature.

Parameters:

• dtypes – input dtypes

• dtypes_out – output dtypes

• n_optional – number of optional parameters, 0 by default

• nvars – True if the function allows an infinite number of inputs, this is incompatible with parameter n_optional.

dtypes, dtypes_out by default are a tuple of tuple:

• first dimension: type of every input

• second dimension: list of types for one input

New in version 0.6.

source on GitHub

constructor

__init__(dtypes=None, dtypes_out=None, n_optional=None, nvars=False)

constructor

__repr__()

usual

_get_output_types(key)

Tries to infer output types.

source on GitHub

static _process_type(dtypes, mapped_types, index)

Nicknames such as floats, int, ints, all can be used to describe multiple inputs for a signature. This function intreprets that.

<<<

from mlprodict.npy.onnx_numpy_annotation import _NDArrayAlias
for name in ['all', 'int', 'ints', 'floats', 'T']:
    print(name, _NDArrayAlias._process_type(name, {'T': 0}, 0))

>>>

    all (<class 'numpy.float32'>, <class 'numpy.float64'>, <class 'numpy.int32'>, <class 'numpy.int64'>, <class 'numpy.uint32'>, <class 'numpy.uint64'>)
    int (<class 'numpy.int64'>,)
    ints (<class 'numpy.int32'>, <class 'numpy.int64'>)
    floats (<class 'numpy.float32'>, <class 'numpy.float64'>)
    T ('T',)

source on GitHub

get_inputs_outputs(args, kwargs, version)

Returns the list of inputs, outputs.

Parameters:

• args – list of arguments

• kwargs – list of optional arguments

• version – required version

Returns:

tuple(inputs, kwargs, outputs, optional), inputs and outputs are tuple, kwargs are the arguments, optional is the number of optional arguments

source on GitHub

shape_calculator(dims)

Returns expected dimensions given the input dimensions.

source on GitHub

mlprodict.npy.onnx_numpy_annotation.get_args_kwargs(fct, n_optional)

Extracts arguments and optional parameters of a function.

Parameters:

• fct – function

• n_optional – number of arguments to consider as optional arguments and not parameters, this parameter skips the first n_optional paramerters

Returns:

arguments, OrderedDict

Any optional argument ending with ‘_’ is ignored.

source on GitHub