def Input(shape, dtype=default_override_or(np.float32), needs_gradient=True, is_sparse=False,
dynamic_axes=Axis.default_input_variable_dynamic_axes(), name=''):
'''
Constructs an Input variable.
'''
dtype = get_default_override(Input, dtype=dtype)
return input_variable(shape=shape, dtype=dtype, needs_gradient=needs_gradient, is_sparse=is_sparse,
dynamic_axes=dynamic_axes, name=name)
def Input(shape, dtype=default_override_or(np.float32), needs_gradient=True, is_sparse=False,
dynamic_axes=Axis.default_input_variable_dynamic_axes(), name=''):
'''
Input(shape, dtype=np.float32, needs_gradient=True, is_sparse=False, dynamic_axes=Axis.default_input_variable_dynamic_axes(), name='')
Constructs an Input variable.
Input variables are used when explicitly constructing a graph.
In the context of the Layers library, however, the preferred method is to use the @\ :func:`~cntk.utils.Signature` pattern.
This is a wrapper around :func:`~cntk.ops.input_variable`.
Example:
>>> # an input receptacle for explicit graph building
>>> x = Input((2,3), is_sparse=True)
>>> x.is_sparse
True
>>> x.shape
(2, 3)
>>> y = sigmoid(x)
>>> y.shape
(2, 3)
>>> # but the preferred pattern is to use the @Function/@Signature pattern instead:
>>> from cntk.ops.functions import Function
>>> from cntk.layers.typing import *
>>> @Function
... @Signature(x = Tensor[2,3])
... def y(x):
... return sigmoid(x)
>>> y.shape
(2, 3)
>>> # type specifications can also be directly passed to Input:
>>> x = Input(**SparseTensor[2,3])
>>> x.is_sparse
True
>>> x.shape
(2, 3)
>>> y = sigmoid(x)
>>> y.shape
(2, 3)
Args:
shape (`int` or `tuple` of `ints`): vector or tensor dimension of the output of this layer
dtype (np.dtype, defaults to np.float32): data type
needs_gradient (bool, defaults to `True`):
is_sparse (bool, defaults to `False`):
dynamic_axes (object, `Axis.default_input_variable_dynamic_axes`):
name (str, defaults to ''): the name of the Function instance in the network
Returns:
an input Variable
'''
dtype = get_default_override(Input, dtype=dtype)
return input(shape=shape, dtype=dtype, needs_gradient=needs_gradient, is_sparse=is_sparse,
dynamic_axes=dynamic_axes, name=name)
def Input(shape, dtype=default_override_or(np.float32), needs_gradient=True, is_sparse=False,
dynamic_axes=Axis.default_input_variable_dynamic_axes(), name=''):
'''
Input(shape, dtype=np.float32, needs_gradient=True, is_sparse=False, dynamic_axes=Axis.default_input_variable_dynamic_axes(), name='')
Constructs an Input variable.
Args:
shape (`int` or `tuple` of `ints`): vector or tensor dimension of the output of this layer
dtype (np.dtype, defaults to np.float32): data type
needs_gradient (bool, defaults to `True`):
is_sparse (bool, defaults to `False`):
dynamic_axes (object, Axis.default_input_variable_dynamic_axes):
name (str, defaults to ''): the name of the Function instance in the network
'''
dtype = get_default_override(Input, dtype=dtype)
return input_variable(shape=shape, dtype=dtype, needs_gradient=needs_gradient, is_sparse=is_sparse,
dynamic_axes=dynamic_axes, name=name)
