def times(left, right, name=None):
"""
Tensor times operation. The output of this operation is the
tensor product of the two input tensors. It supports broadcasting. In
case of scalars its backward pass to left propagates right
times the received gradient and vice versa.
Example:
>>> C.eval(C.times([[1,2],[3,4]], [5,6]))
[array([[ 17., 39.]])]
>>> C.eval(C.times([[1,2],[3,4],[5,6]], [[0.5,0.25],[0.25,0.5]]))
[array([[[ 1. , 1.25],
[ 2.5 , 2.75],
[ 4. , 4.25]]])]
Args:
left: left side tensor
right: right side tensor
name: the name of the node in the network
Returns:
:class:`cntk.graph.ComputationNode`
"""
from cntk.ops.cntk2 import Times
return Times(left, right, name=name)
def times(left, right, output_rank=1, name=None):
"""
The output of this operation is the matrix product of the two input matrices.
It supports broadcasting. Sparse is supported in the right operand, if it is a matrix.
The operator '@' has been overloaded such that in Python 3.5 and later X @ W equals times(X, W).
Example:
>>> C.eval(C.times([[1,2],[3,4]], [5,6]))
[array([[ 17., 39.]])]
>>> C.eval(cntk.times(np.reshape(np.arange(8), (2,2,2)),np.reshape(np.arange(8), (2,2,2)), output_rank=1))
[array([[[ 28., 34.],
[ 76., 98.]]])]
>>> C.eval(cntk.times(np.reshape(np.arange(8), (2,2,2)),np.reshape(np.arange(8), (2,2,2)), output_rank=2))
[array([[[[[ 4., 5.],
[ 6., 7.]],
[[ 12., 17.],
[ 22., 27.]]],
[[[ 20., 29.],
[ 38., 47.]],
[[ 28., 41.],
[ 54., 67.]]]]])]
Args:
left: left side matrix or tensor
right: right side matrix or tensor
output_rank: in case we have tensors as arguemnts, output_rank represents
the number of axes to be collapsed in order to transform the tensors
into matrices, perform the operation and then reshape back (explode the axes)
name: the name of the node in the network
Returns:
:class:`cntk.graph.ComputationNode`
"""
from cntk.ops.cntk2 import Times
# CNTK uses column vectors and column major representation, thus we reverse
# params
op = Times(right, left, outputRank=output_rank, name=name)
wrap_numpy_arrays(op)
op.rank = op.x.rank + op.y.rank - 2
return op
def times(left, right, output_rank=1, name=None):
"""
The output of this operation is the matrix product of the two input matrices.
It supports broadcasting. Sparse is supported in the right operand, if it is a matrix.
The operator '@' has been overloaded such that in Python 3.5 and later X @ W equals times(X, W).
Example:
>>> C.eval(C.times([[1,2],[3,4]], [5,6]))
[array([[ 17., 39.]])]
>>> C.eval(cntk.times(np.reshape(np.arange(8), (2,2,2)),np.reshape(np.arange(8), (2,2,2)), output_rank=1))
[array([[[ 28., 34.],
[ 76., 98.]]])]
>>> C.eval(cntk.times(np.reshape(np.arange(8), (2,2,2)),np.reshape(np.arange(8), (2,2,2)), output_rank=2))
[array([[[[[ 4., 5.],
[ 6., 7.]],
[[ 12., 17.],
[ 22., 27.]]],
[[[ 20., 29.],
[ 38., 47.]],
[[ 28., 41.],
[ 54., 67.]]]]])]
Args:
left: left side matrix or tensor
right: right side matrix or tensor
output_rank (int): in case we have tensors as arguemnts, output_rank represents
the number of axes to be collapsed in order to transform the tensors
into matrices, perform the operation and then reshape back (explode the axes)
name (str): the name of the node in the network
Returns:
:class:`cntk.graph.ComputationNode`
"""
from cntk.ops.cntk2 import Times
# CNTK uses column vectors and column major representation, thus we reverse
# params
op = Times(right, left, outputRank=output_rank, name=name)
wrap_numpy_arrays(op)
op.rank = op.x.rank + op.y.rank - 2
return op
def times(left, right, output_rank=1, name=None):
"""
The output of this operation is the matrix product of the two input matrices.
It supports broadcasting. In case of scalars its backward pass to left propagates right
times the received gradient and vice versa.
The operator (@) has been overloaded and can equally be used instead of times().
However, it is supported in python versions 3.5 or above.
Example:
>>> C.eval(C.times([[1,2],[3,4]], [5,6]))
[array([[ 17., 39.]])]
>>> C.eval(cntk.times(np.reshape(np.arange(8), (2,2,2)),np.reshape(np.arange(8), (2,2,2)), output_rank=1))
[array([[[ 28., 34.],
[ 76., 98.]]])]
>>> C.eval(cntk.times(np.reshape(np.arange(8), (2,2,2)),np.reshape(np.arange(8), (2,2,2)), output_rank=2))
[array([[[[[ 4., 5.],
[ 6., 7.]],
[[ 12., 17.],
[ 22., 27.]]],
[[[ 20., 29.],
[ 38., 47.]],
[[ 28., 41.],
[ 54., 67.]]]]])]
Args:
left: left side matrix or tensor
right: right side matrix or tensor
output_rank: in case we have tensors as arguemnts, output_rank represents
the number of axes to be collapsed in order to transform the tensors
into matrices, perform the operation and then reshape back (explode the axes)
name: the name of the node in the network
Returns:
:class:`cntk.graph.ComputationNode`
"""
from cntk.ops.cntk2 import Times
return Times(left, right, outputRank=output_rank, name=name)