m53297601
/
mindspore
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

!14251 rm space of BatchMatMul,update examples of BNTrainingUpdate and NLLLoss

Browse Source 
From: @mind-lh
Reviewed-by: @ljl0711,@wuxuejian
Signed-off-by: @wuxuejian
pull/14251/MERGE
mindspore-ci-bot 4 years ago committed by Gitee
parent 7ad3a60568 2516e6cc23
commit 6646c4f35d
4 changed files with 60 additions and 85 deletions
Show Stats Download Patch File Download Diff File

8
mindspore/ops/operations/array_ops.py
Unescape View File

@ -3398,7 +3398,7 @@ class ResizeNearestNeighbor(PrimitiveWithInfer):
class GatherNd(PrimitiveWithInfer): class GatherNd(PrimitiveWithInfer):
""" r"""
Gathers slices from a tensor by indices. Gathers slices from a tensor by indices.
Using given indices to gather slices from a tensor with a specified shape. Using given indices to gather slices from a tensor with a specified shape.
@ -3407,10 +3407,10 @@ class GatherNd(PrimitiveWithInfer):
defines a slice of `input_x`: defines a slice of `input_x`:
.. math:: .. math::
output[(i_0, ..., i_{K-2})] = input_x[indices[(i_0, ..., i_{K-2})]] output[(i_0, ..., i_{K-2})] = input\_x[indices[(i_0, ..., i_{K-2})]]
The last dimension of `indices` can not more than the rank of `input_x`: The last dimension of `indices` can not more than the rank of `input_x`:
:math:`indices.shape[-1] <= input_x.rank`. :math:`indices.shape[-1] <= input\_x.rank`.
Inputs: Inputs:
- **input_x** (Tensor) - The target tensor to gather values. - **input_x** (Tensor) - The target tensor to gather values.
@ -5282,7 +5282,7 @@ class Range(PrimitiveWithCheck):
A 1-D Tensor, with the same type as the inputs. A 1-D Tensor, with the same type as the inputs.
Supported Platforms: Supported Platforms:
``GPU`` `CPU` ``GPU``
Examples: Examples:
>>> start = Tensor(0, mstype.int32) >>> start = Tensor(0, mstype.int32)

17
mindspore/ops/operations/math_ops.py
Unescape View File

@ -884,42 +884,25 @@ class BatchMatMul(MatMul):
>>> output = batmatmul(input_x, input_y) >>> output = batmatmul(input_x, input_y)
>>> print(output) >>> print(output)
[[[[3. 3. 3. 3.]] [[[[3. 3. 3. 3.]]
[[3. 3. 3. 3.]] [[3. 3. 3. 3.]]
[[3. 3. 3. 3.]] [[3. 3. 3. 3.]]
[[3. 3. 3. 3.]]] [[3. 3. 3. 3.]]]
[[[3. 3. 3. 3.]] [[[3. 3. 3. 3.]]
[[3. 3. 3. 3.]] [[3. 3. 3. 3.]]
[[3. 3. 3. 3.]] [[3. 3. 3. 3.]]
[[3. 3. 3. 3.]]]] [[3. 3. 3. 3.]]]]
>>>
>>> input_x = Tensor(np.ones(shape=[2, 4, 3, 1]), mindspore.float32) >>> input_x = Tensor(np.ones(shape=[2, 4, 3, 1]), mindspore.float32)
>>> input_y = Tensor(np.ones(shape=[2, 4, 3, 4]), mindspore.float32) >>> input_y = Tensor(np.ones(shape=[2, 4, 3, 4]), mindspore.float32)
>>> batmatmul = ops.BatchMatMul(transpose_a=True) >>> batmatmul = ops.BatchMatMul(transpose_a=True)
>>> output = batmatmul(input_x, input_y) >>> output = batmatmul(input_x, input_y)
>>> print(output) >>> print(output)
[[[[3. 3. 3. 3.]] [[[[3. 3. 3. 3.]]
[[3. 3. 3. 3.]] [[3. 3. 3. 3.]]
[[3. 3. 3. 3.]] [[3. 3. 3. 3.]]
[[3. 3. 3. 3.]]] [[3. 3. 3. 3.]]]
[[[3. 3. 3. 3.]] [[[3. 3. 3. 3.]]
[[3. 3. 3. 3.]] [[3. 3. 3. 3.]]
[[3. 3. 3. 3.]] [[3. 3. 3. 3.]]
[[3. 3. 3. 3.]]]] [[3. 3. 3. 3.]]]]
""" """

86
mindspore/ops/operations/nn_ops.py
Unescape View File

@ -978,14 +978,9 @@ class BNTrainingReduce(PrimitiveWithInfer):
>>> bn_training_reduce = ops.BNTrainingReduce() >>> bn_training_reduce = ops.BNTrainingReduce()
>>> output = bn_training_reduce(input_x) >>> output = bn_training_reduce(input_x)
>>> print(output) >>> print(output)
(Tensor(shape=[1, 2, 2, 2], dtype=Float32, value= (Tensor(shape=[3], dtype=Float32, value=
[[[[ 2.73200464e+00, 2,73200464e+00], [ 1.22880000e+04, 1.22880000e+04, 1.22880000e+04]), Tensor(shape=[3], dtype=Float32, value=
[ 2.73200464e+00, 2,73200464e+00]], [ 1.22880000e+04, 1.22880000e+04, 1.22880000e+04]))
[[ 2.73200464e+00, 2,73200464e+00],
[ 2.73200464e+00, 2,73200464e+00]]]]), Tensor(shape=[2], dtype=Float32, value= [ 9.24999952e-01,
9.24999952e-01]), Tensor(shape=[2], dtype=Float32, value= [ 9.24999952e-01, 9.24999952e-01]),
Tensor(shape=[2], dtype=Float32, value= [ 2.50000000e-01, 2.50000000e-01]), Tensor(shape=[2], dtype=Float32,
value= [ 1.87500000e-01, 1.87500000-01]))
""" """
@prim_attr_register @prim_attr_register
@ -1061,17 +1056,10 @@ class BNTrainingUpdate(PrimitiveWithInfer):
[[[[ 2.73200464e+00, 2.73200464e+00], [[[[ 2.73200464e+00, 2.73200464e+00],
[ 2.73200464e+00, 2.73200464e+00]], [ 2.73200464e+00, 2.73200464e+00]],
[[ 2.73200464e+00, 2.73200464e+00], [[ 2.73200464e+00, 2.73200464e+00],
[ 2.73200464e+00, 2.73200464e+00]]]]), Tensor(shape=[1, 2, 2, 2], dtype=Float32, value= [ 2.73200464e+00, 2.73200464e+00]]]]), Tensor(shape=[2], dtype=Float32, value= [9.24999952e-0.1,
[[[[ 2.73200464e+00, 2.73200464e+00], 9.24999952e-0.1]), Tensor(shape=[2], dtype=Float32, value= [ 9.24999952e-0.1, 9.24999952e-0.1]),
[ 2.73200464e+00, 2.73200464e+00]], Tensor(shape=[2], dtype=Float32, value= [ 2.50000000e-0.1, 2.50000000e-0.1]), Tensor(shape=[2], dtype=Float32,
[[ 2.73200464e+00, 2.73200464e+00], value= [ 1.87500000e-0.1, 1.87500000e-0.1]))
[ 2.73200464e+00, 2.73200464e+00]]]]), Tensor(shape=[1, 2, 2, 2], dtype=Float32, value=
[[[[ 2.73200464e+00, 2.73200464e+00],
[ 2.73200464e+00, 2.73200464e+00]],
[[ 2.73200464e+00, 2.73200464e+00],
[ 2.73200464e+00, 2.73200464e+00]]]]), Tensor(shape=[2], dtype=Float32, value=
[ 2.50000000e-01, 2.50000000e-01]), Tensor(shape=[2], dtype=Float32, value=
[ 1.87500000e-01, 1.87500000e-01]))
""" """
@prim_attr_register @prim_attr_register
@ -2223,8 +2211,8 @@ class NLLLoss(PrimitiveWithInfer):
Examples: Examples:
>>> input = Tensor(np.array([[0.5488135, 0.71518934], >>> input = Tensor(np.array([[0.5488135, 0.71518934],
>>> [0.60276335, 0.5448832], ... [0.60276335, 0.5448832],
>>> [0.4236548, 0.6458941]]).astype(np.float32)) ... [0.4236548, 0.6458941]]).astype(np.float32))
>>> target = Tensor(np.array([0, 0, 0]).astype(np.int32)) >>> target = Tensor(np.array([0, 0, 0]).astype(np.int32))
>>> weight = Tensor(np.array([0.3834415, 0.79172504]).astype(np.float32)) >>> weight = Tensor(np.array([0.3834415, 0.79172504]).astype(np.float32))
>>> nll_loss = ops.NLLLoss(reduction="mean") >>> nll_loss = ops.NLLLoss(reduction="mean")
@ -2357,7 +2345,7 @@ class SparseSoftmaxCrossEntropyWithLogits(PrimitiveWithInfer):
``GPU`` ``CPU`` ``GPU`` ``CPU``
Examples: Examples:
Please refer to the usage in nn.SoftmaxCrossEntropyWithLogits source code. Please refer to :class:`mindspore.nn.SparseSoftmaxCrossEntropyWithLogits`
""" """
@prim_attr_register @prim_attr_register
@ -4216,20 +4204,14 @@ class Adam(PrimitiveWithInfer):
... def construct(self, beta1_power, beta2_power, lr, beta1, beta2, epsilon, grad): ... def construct(self, beta1_power, beta2_power, lr, beta1, beta2, epsilon, grad):
... out = self.apply_adam(self.var, self.m, self.v, beta1_power, beta2_power, lr, beta1, beta2, ... out = self.apply_adam(self.var, self.m, self.v, beta1_power, beta2_power, lr, beta1, beta2,
... epsilon, grad) ... epsilon, grad)
...
... return out ... return out
>>> np.random.seed(0) ...
>>> net = Net() >>> net = Net()
>>> gradient = Tensor(np.random.rand(2, 2).astype(np.float32)) >>> gradient = Tensor(np.ones([2, 2]).astype(np.float32))
>>> output = net(0.9, 0.999, 0.001, 0.9, 0.999, 1e-8, gradient) >>> net(0.9, 0.999, 0.001, 0.9, 0.999, 1e-8, gradient)
>>> print(output) >>> print(net.var.asnumpy())
(Tensor(shape=[2, 2], dtype=Float32, value= [[0.9996838 0.9996838]
[[ 9.99697924e-01, 9.99692678e-01], [0.9996838 0.9996838]]
[ 9.99696255e-01, 9.99698043e-01]]), Tensor(shape=[2, 2], dtype=Float32, value=
[[ 9.54881310e-01, 9.71518934e-01],
[ 9.60276306e-01, 9.54488277e-01]]), Tensor(shape=[2, 2], dtype=Float32, value=
[[ 9.99301195e-01, 9.99511480e-01],
[ 9.99363303e-01, 9.99296904e-01]]))
""" """
@prim_attr_register @prim_attr_register
@ -4318,18 +4300,17 @@ class AdamNoUpdateParam(PrimitiveWithInfer):
>>> import mindspore.nn as nn >>> import mindspore.nn as nn
>>> from mindspore import Tensor, Parameter >>> from mindspore import Tensor, Parameter
>>> from mindspore.ops import operations as ops >>> from mindspore.ops import operations as ops
>>>
>>> class Net(nn.Cell): >>> class Net(nn.Cell):
>>> def __init__(self): ... def __init__(self):
>>> super(Net, self).__init__() ... super(Net, self).__init__()
>>> self.adam = ops.AdamNoUpdateParam() ... self.adam = ops.AdamNoUpdateParam()
>>> self.m = Parameter(Tensor(np.array([[0.1, 0.1, 0.1], [0.2, 0.2, 0.2]]).astype(np.float32)), ... self.m = Parameter(Tensor(np.array([[0.1, 0.1, 0.1], [0.2, 0.2, 0.2]]).astype(np.float32)),
>>> name="m") ... name="m")
>>> self.v = Parameter(Tensor(np.array([[0.1, 0.1, 0.1], [0.2, 0.2, 0.2]]).astype(np.float32)), ... self.v = Parameter(Tensor(np.array([[0.1, 0.1, 0.1], [0.2, 0.2, 0.2]]).astype(np.float32)),
>>> name="v") ... name="v")
>>> def construct(self, beta1_power, beta2_power, lr, beta1, beta2, epsilon, grad): ... def construct(self, beta1_power, beta2_power, lr, beta1, beta2, epsilon, grad):
>>> out = self.adam(self.m, self.v, beta1_power, beta2_power, lr, beta1, beta2, epsilon, grad) ... out = self.adam(self.m, self.v, beta1_power, beta2_power, lr, beta1, beta2, epsilon, grad)
>>> return out ... return out
>>> net = Net() >>> net = Net()
>>> beta1_power = Tensor(0.9, ms.float32) >>> beta1_power = Tensor(0.9, ms.float32)
>>> beta2_power = Tensor(0.999, ms.float32) >>> beta2_power = Tensor(0.999, ms.float32)
@ -4338,7 +4319,6 @@ class AdamNoUpdateParam(PrimitiveWithInfer):
>>> beta2 = Tensor(0.999, ms.float32) >>> beta2 = Tensor(0.999, ms.float32)
>>> epsilon = Tensor(1e-8, ms.float32) >>> epsilon = Tensor(1e-8, ms.float32)
>>> gradient = Tensor(np.array([[0.1, 0.1, 0.1], [0.1, 0.1, 0.1]]).astype(np.float32)) >>> gradient = Tensor(np.array([[0.1, 0.1, 0.1], [0.1, 0.1, 0.1]]).astype(np.float32))
>>>
>>> result = net(beta1_power, beta2_power, lr, beta1, beta2, epsilon, gradient) >>> result = net(beta1_power, beta2_power, lr, beta1, beta2, epsilon, gradient)
>>> print(result) >>> print(result)
[[-0.00010004 -0.00010004 -0.00010004] [[-0.00010004 -0.00010004 -0.00010004]
@ -4459,8 +4439,11 @@ class FusedSparseAdam(PrimitiveWithInfer):
>>> epsilon = Tensor(1e-8, mstype.float32) >>> epsilon = Tensor(1e-8, mstype.float32)
>>> gradient = Tensor(np.random.rand(2, 1, 2), mstype.float32) >>> gradient = Tensor(np.random.rand(2, 1, 2), mstype.float32)
>>> indices = Tensor([0, 1], mstype.int32) >>> indices = Tensor([0, 1], mstype.int32)
>>> output = net(beta1_power, beta2_power, lr, beta1, beta2, epsilon, gradient, indices) >>> net(beta1_power, beta2_power, lr, beta1, beta2, epsilon, gradient, indices)
>>> print(output) >>> print(net.var.asnumpy())
[[[0.9996963 0.9996977 ]]
[[0.99970144 0.9996992 ]]
[[0.99971527 0.99971527]]]
""" """
__mindspore_signature__ = ( __mindspore_signature__ = (
sig.make_sig('var', sig.sig_rw.RW_WRITE, dtype=sig.sig_dtype.T), sig.make_sig('var', sig.sig_rw.RW_WRITE, dtype=sig.sig_dtype.T),
@ -4604,8 +4587,11 @@ class FusedSparseLazyAdam(PrimitiveWithInfer):
>>> epsilon = Tensor(1e-8, mstype.float32) >>> epsilon = Tensor(1e-8, mstype.float32)
>>> gradient = Tensor(np.random.rand(2, 1, 2), mstype.float32) >>> gradient = Tensor(np.random.rand(2, 1, 2), mstype.float32)
>>> indices = Tensor([0, 1], mstype.int32) >>> indices = Tensor([0, 1], mstype.int32)
>>> output = net(beta1_power, beta2_power, lr, beta1, beta2, epsilon, gradient, indices) >>> net(beta1_power, beta2_power, lr, beta1, beta2, epsilon, gradient, indices)
>>> print(output) >>> print(net.var.asnumpy())
[[[0.9996866 0.9997078]]
[[0.9997037 0.9996869]]
[[1. 1. ]]]
""" """
__mindspore_signature__ = ( __mindspore_signature__ = (
sig.make_sig('var', sig.sig_rw.RW_WRITE, dtype=sig.sig_dtype.T), sig.make_sig('var', sig.sig_rw.RW_WRITE, dtype=sig.sig_dtype.T),

34
mindspore/ops/primitive.py
Unescape View File

@ -255,17 +255,20 @@ class PrimitiveWithCheck(Primitive):
Args: Args:
name (str): Name of the current Primitive. name (str): Name of the current Primitive.
Supported Platforms:
``Ascend`` ``GPU`` ``CPU``
Examples: Examples:
>>> # init a Primitive class with check >>> # init a Primitive class with check
>>> class Flatten(PrimitiveWithCheck): >>> class Flatten(PrimitiveWithCheck):
>>> @prim_attr_register ... @prim_attr_register
>>> def __init__(self): ... def __init__(self):
>>> pass ... pass
>>> def check_shape(self, input_x): ... def check_shape(self, input_x):
>>> validator.check_int(len(input_x), 1, Rel.GE, 'input_x rank', self.name) ... validator.check_int(len(input_x), 1, Rel.GE, 'input_x rank', self.name)
>>> >>>
>>> def check_dtype(self, input_x): ... def check_dtype(self, input_x):
>>> validator.check_subclass("input_x", input_x, mstype.tensor, self.name) ... validator.check_subclass("input_x", input_x, mstype.tensor, self.name)
>>> >>>
>>> # init a Primitive obj >>> # init a Primitive obj
>>> add = Flatten() >>> add = Flatten()
@ -332,18 +335,21 @@ class PrimitiveWithInfer(Primitive):
Args: Args:
name (str): Name of the current Primitive. name (str): Name of the current Primitive.
Supported Platforms:
``Ascend`` ``GPU`` ``CPU``
Examples: Examples:
>>> # init a Primitive class with infer >>> # init a Primitive class with infer
>>> class Add(PrimitiveWithInfer): >>> class Add(PrimitiveWithInfer):
>>> @prim_attr_register ... @prim_attr_register
>>> def __init__(self): ... def __init__(self):
>>> pass ... pass
>>> >>>
>>> def infer_shape(self, x, y): ... def infer_shape(self, x, y):
>>> return x # output shape same as first input 'x' ... return x # output shape same as first input 'x'
>>> >>>
>>> def infer_dtype(self, x, y): ... def infer_dtype(self, x, y):
>>> return x # output type same as first input 'x' ... return x # output type same as first input 'x'
>>> >>>
>>> # init a Primitive obj >>> # init a Primitive obj
>>> add = Add() >>> add = Add()

Write Preview
Loading…
Cancel
Save