module `npy.onnx_variable`#

Short summary#

module mlprodict.npy.onnx_variable

Intermediate class between numpy and onnx.

Classes#

class	truncated documentation
`MultiOnnxVar`	Class used to return multiple `OnnxVar` at the same time.
`OnnxVar`	Variables used into onnx computation.

Properties#

property	truncated documentation
`inputs`	Returns self.onxvar.inputs.
`onnx_op`	Returns self.onxvar.onnx_op.
`onnx_op_kwargs`	Returns self.onxvar.onnx_op_kwargs.
`shape`	Shape
`size`	Size
`T`	Transpose.

Methods#

method	truncated documentation
`__add__`	Addition.
`__and__`	And.
`__eq__`	Equality.
`__floordiv__`	Division, no difference between / and //.
`__ge__`	Greater or Equal.
`__getitem__`	Returns the ith elements.
`__getitem__`	Deals with multiple scenarios.
`__gt__`	Greater.
`__init__`	constructor
`__init__`
`__invert__`	not.
`__le__`	Less or Equal.
`__lt__`	Less.
`__matmul__`	Matrix multiplication.
`__mod__`	Modulo.
`__mul__`	Multiplication.
`__ne__`	Difference.
`__neg__`	Neg.
`__or__`	And.
`__pow__`	Power.
`__radd__`	Right Addition.
`__repr__`	usual
`__rmul__`	Right multiplication.
`__rsub__`	Right subtraction.
`__rtruediv__`	Division, no difference between / and //.
`__setitem__`	Only supports vectors (1D tensor).
`__sub__`	Subtraction.
`__truediv__`	Division, no difference between / and //.
`_custom_op`	This could be handled before a call to this method but this method can change the conversion of an non-existing …
`_custom_op_filter`	This could be handled before a call to this method but this method can change the conversion of an non-existing …
`_guess_dtype`	Guesses dtype when not specified.
`_guess_dtype`	Guesses dtype when not specified.
`_make_array`	Converts y into an array if not.
`_setitem1i_`
`_setitem2i_`
`astype`	Cast
`copy`	Returns a copy of self (use of Identity node).
`flatten`	Flattens a matrix (see numpy.ndarray.flatten).
`not_`	Not.
`reshape`	Reshape
`set_onnx_name`	Forces this variable to get this name during
`to_algebra`	Converts the variable into an operator.
`to_algebra`	Converts the variable into an operator.

Documentation#

Intermediate class between numpy and onnx.

New in version 0.6.

source on GitHub

class mlprodict.npy.onnx_variable.MultiOnnxVar(*inputs, op=None, dtype=None, **kwargs)#

Bases: object

Class used to return multiple OnnxVar at the same time.

source on GitHub

constructor

__getitem__(index)#

Returns the ith elements.

source on GitHub

__init__(*inputs, op=None, dtype=None, **kwargs)#: constructor

_guess_dtype(dtype)#: Guesses dtype when not specified.

property inputs#: Returns self.onxvar.inputs.

property onnx_op#: Returns self.onxvar.onnx_op.

property onnx_op_kwargs#: Returns self.onxvar.onnx_op_kwargs.

to_algebra(op_version=None)#

Converts the variable into an operator.

source on GitHub

class mlprodict.npy.onnx_variable.OnnxVar(*inputs, op=None, select_output=None, dtype=None, **kwargs)#

Bases: object

Variables used into onnx computation.

Parameters:

inputs – variable name or object
op – ONNX operator
select_output – if multiple output are returned by ONNX operator op, it takes only one specifed by this argument
dtype – specifies the type of the variable held by this class (op is None) in that case
kwargs – addition argument to give operator op

New in version 0.6.

source on GitHub

property T#: Transpose.

__add__(y)#: Addition.

__and__(y)#: And.

__array_ufunc__ = None#

__eq__(y)#: Equality.

__floordiv__(y)#: Division, no difference between / and //.

__ge__(y)#: Greater or Equal.

__getitem__(index)#

Deals with multiple scenarios.

index is an integer or a slice, a tuple of integers and slices, example: [0, 1], [:5, :6], [::2] (scenario 1)
index is an ONNX object (more precisely an instance of OnnxVar), then the method assumes it is an array of boolean to select a subset of the tensor along the first axis, example: mat[mat == 0] (scenario 2)

source on GitHub

__gt__(y)#: Greater.

__hash__ = None#

__init__(*inputs, op=None, select_output=None, dtype=None, **kwargs)#

__invert__()#: not.

__le__(y)#: Less or Equal.

__lt__(y)#: Less.

__matmul__(y)#: Matrix multiplication.

__mod__(y)#: Modulo.

__mul__(y)#: Multiplication.

__ne__(y)#: Difference.

__neg__()#: Neg.

__or__(y)#: And.

__pow__(y)#: Power.

__radd__(y)#: Right Addition.

__repr__()#: usual

__rmul__(y)#: Right multiplication.

__rsub__(y)#: Right subtraction.

__rtruediv__(y)#: Division, no difference between / and //.

__setitem__(index, value)#

Only supports vectors (1D tensor).

index is an integer or a slice, a tuple of integers and slices, example: [0], [:5], [::2] (scenario 1)
index is an ONNX object (more precisely an instance of OnnxVar), then the method assumes it is an array of boolean to select a subset of the tensor along the first axis, example: mat[mat == 0] (scenario 2)

This processing is applied before the operator it contains. A copy should be made (Identity node or copy method).

source on GitHub

__sub__(y)#: Subtraction.

__truediv__(y)#: Division, no difference between / and //.

_custom_op(*args, op_version=None, runtime=None, **kwargs)#

This could be handled before a call to this method but this method can change the conversion of an non-existing operator depending on the given opset.

source on GitHub

_custom_op_filter(*args, op_version=None, runtime=None, **kwargs)#

This could be handled before a call to this method but this method can change the conversion of an non-existing operator depending on the given opset.

source on GitHub