def transpose(a, axes=None):
"""
Reverse or permute the axes of an array; returns the modified array.
Args:
a (Tensor): a tensor to be transposed
axes Union[None, tuple, list]: the axes order, if axes is None, transpose
the entire tensor. Default is None.
Returns:
Tensor, the transposed tensor array.
Supported Platforms:
``Ascend`` ``GPU`` ``CPU``
Examples:
>>> import mindspore.numpy as np
>>> x = np.ones((1,2,3))
>>> x = np.transpose(x)
>>> print(x,shape)
(3,2,1)
"""
if axes is None:
shape = F.shape(a)
length = F.tuple_len(shape)
perm = F.make_range(0, length)
new_order = F.tuple_reversed(perm)
return P.Transpose()(a, new_order)
axes = _check_shape_compile(axes)
return P.Transpose()(a, axes)
def broadcast_params(self, optim_result):
"""
Apply Broadcast operations in the sequential order of parameter groups.
Returns:
bool, the status flag.
"""
param_group = []
key_group = []
for _ in range(self.dev_num):
param_group.append(F.make_tuple())
key_group.append(F.make_tuple())
for i in range(self.param_length):
param_group[self.param_rank[i]] = param_group[
self.param_rank[i]] + (self.parameters[i], )
key = P.MakeRefKey(self.param_names[i])()
key_group[
self.param_rank[i]] = key_group[self.param_rank[i]] + (key, )
new_param_group = []
for root in range(self.dev_num):
ops = P.Broadcast(root)
next_params = ops(param_group[root])
new_param_group.append(next_params)
for i in range(F.tuple_len(next_params)):
F.assign(key_group[root][i], next_params[i])
status = F.control_depend(optim_result, new_param_group[0][0])
for i in range(self.dev_num - 1):
status = F.depend(
F.control_depend(new_param_group[i],
new_param_group[i + 1][0]), status)
return status
def swapaxes(x, axis1, axis2):
"""
Interchange two axes of a tensor.
Args:
x (Tensor): A Tensor to be transposed.
axis1 (int): First axis.
axis2 (int): Second axis.
Returns:
Transposed Tensor. Has the same data type as the original tensor x.
Raises:
TypeError: If axis1 or axis2 is not integer.
ValueError: If axis1 or axis2 is not in the range from -ndim to ndim-1.
Supported Platforms:
``Ascend`` ``GPU`` ``CPU``
Examples:
>>> import mindspore
>>> import mindspore.numpy as mnp
>>> from mindspore import Tensor
>>> import numpy as onp
>>> input_x = Tensor(onp.ones((2,3,4)), mindspore.float32)
>>> output = mnp.swapaxes(x, 0, 2)
>>> print(output.shape)
(4,3,2)
"""
_check_is_int(axis1)
_check_is_int(axis2)
shape = F.shape(x)
ndim = F.tuple_len(shape)
axes = _check_axes_range((axis1, axis2), ndim)
axis1, axis2 = axes[0], axes[1]
if axis1 == axis2:
return x
if axis1 > axis2:
axis1, axis2 = axis2, axis1
perm = F.make_range(0, ndim)
new_perm = None
if axis2 + 1 < ndim:
new_perm = perm[0:axis1] + perm[axis2:axis2+1] + \
perm[axis1+1:axis2] + perm[axis1:axis1+1] + perm[axis2+1:]
else:
new_perm = perm[0:axis1] + perm[axis2:axis2+1] + \
perm[axis1+1:axis2] + perm[axis1:axis1+1]
return P.Transpose()(x, new_perm)
def bprop(x, out, dout):
if F.issubclass_(F.typeof(dout), mstype.tensor):
if F.issubclass_(F.dtype(dout), mstype.bool_):
return (dout, )
dx = op(dout, cast(F.scalar_to_array(divisor), dtype(dout)))
return (dx, )
dx = ()
input_nums = F.tuple_len(dout)
for i in range(input_nums):
ele_grad = op(dout[i],
cast(F.scalar_to_array(divisor), dtype(dout[i])))
dx = dx + (ele_grad, )
return (dx, )
def get_axis(self, x):
shape = F.shape(x)
length = F.tuple_len(shape)
perm = F.make_range(0, length)
return perm
def get_axis(x):
shape_op = P.Shape()
shape = shape_op(x)
length = F.tuple_len(shape)
perm = F.make_range(0, length)
return perm
def rollaxis(x, axis, start=0):
"""
Roll the specified axis backwards, until it lies in the given position.
The positions of the other axes do not change relative to one another.
Args:
x (Tensor): A Tensor to be transposed.
axis (int): The axis to be rolled.
start (int):
- When start >= 0:
- When start <= axis: the axis is rolled back until it lies in this position (start).
- When start > axis: the axis is rolled until it lies before this position (start).
- When start < 0: the start will be normalized as follows:
start ........... Normalized start
-(x.ndim+1) raise ValueError
-x.ndim 0
... ...
-1 x.ndim-1
0 0
... ...
x.ndim x.ndim
x.ndim+1 raise ValueError
Returns:
Transposed Tensor. Has the same data type as the original tensor x.
Supported Platforms:
``Ascend`` ``GPU`` ``CPU``
Raises:
TypeError: If axis or start is not integer.
ValueError: If axis is not in the range from -ndim to ndim-1 or
start is not in the range from -ndim to ndim.
Examples:
>>> import mindspore
>>> import mindspore.numpy as mnp
>>> from mindspore import Tensor
>>> import numpy as onp
>>> input_x = Tensor(onp.ones((2,3,4)), mindspore.float32)
>>> output = mnp.rollaxis(x, 0, 2)
>>> print(output.shape)
(3,2,4)
"""
_check_is_int(axis)
_check_is_int(start)
shape = F.shape(x)
ndim = F.tuple_len(shape)
axis = _check_axes_range(axis, ndim)
start = _check_start_normalize(start, ndim)
if start - axis >= 0 and start - axis <= 1:
return x
perm = F.make_range(0, ndim)
new_perm = None
if start < axis:
if axis + 1 < ndim:
new_perm = perm[0:start] + perm[axis:axis+1] + \
perm[start:axis] + perm[axis+1:]
else:
new_perm = perm[0:start] + perm[axis:axis + 1] + perm[start:axis]
if start > axis:
if start < ndim:
new_perm = perm[0:axis] + perm[axis+1:start] + \
perm[axis:axis+1] + perm[start:]
else:
new_perm = perm[0:axis] + perm[axis+1:start] + \
perm[axis:axis+1]
return P.Transpose()(x, new_perm)
