Differences with torch.ger

torch.normal

torch.normal(mean, std, *, generator=None, out=None)
torch.normal(mean=0.0, std, *, out=None)
torch.normal(mean, std=1.0, *, out=None)
torch.normal(mean, std, size, *, out=None)

For more information, see torch.normal.

mindspore.ops.normal

mindspore.ops.normal(shape, mean, stddev, seed=None)

For more information, see mindspore.ops.normal.

Differences

API function of MindSpore is consistent with that of PyTorch.

PyTorch: Four interface usages are supported.

mean and std are both Tensor, requiring the same number of members for mean and std. The shape of the return value matches the shape of mean .
mean is the float type, std is Tensor. The shape of the return value matches the shape of std .
std is the float type, mean is Tensor. The shape of the return value matches the shape of mean .
mean and std are both float types. The shape of the return value matches the shape of size .

MindSpore: The data types supported by mean and std are Tensor, and the shape of the return value is broadcast by shape, mean, and stddev.

Categories	Subcategories	PyTorch	MindSpore	Differences
Parameters	Parameter 1	-	shape	This value in MindSpore is used to broadcast the shape of the return value together with `mean` and `stddev`
	Parameter 2	mean	mean	The data type supported in MindSpore is Tensor. Tensor and float are supported in PyTorch, corresponding to different usages
	Parameter 3	std	stddev	The data type supported in MindSpore is Tensor. Tensor and float are supported in PyTorch, corresponding to different usages
	Parameter 4	generator	seed	For details, see General Difference Parameter Table
	Parameter 5	size	-	The shape of the return value in PyTorch, used under the specified interface usage
	Parameter 6	out	-	For details, see General Difference Parameter Table

Code Example 1

In PyTorch, 'mean' and 'std' are both Tensor.

# PyTorch
import torch
import numpy as np

mean = torch.tensor(np.array([[3, 4], [5, 6]]), dtype=torch.float32)
stddev = torch.tensor(np.array([[0.2, 0.3], [0.4, 0.5]]), dtype=torch.float32)
output = torch.normal(mean, stddev)
print(output.shape)
# torch.Size([2, 2])

# MindSpore
import mindspore as ms
import numpy as np

shape = (2, 2)
mean = ms.Tensor(np.array([[3, 4], [5, 6]]), ms.float32)
stddev = ms.Tensor(np.array([[0.2, 0.3], [0.4, 0.5]]), ms.float32)
output = ms.ops.normal(shape, mean, stddev)
print(output.shape)
# (2, 2)

Code Example 2

In PyTorch, 'mean' is the float and 'std' is the Tensor.

# PyTorch
import torch
import numpy as np

mean = 3.0
stddev = torch.tensor(np.array([[0.2, 0.3], [0.4, 0.5]]), dtype=torch.float32)
output = torch.normal(mean, stddev)
print(output.shape)
# torch.Size([2, 2])

# MindSpore
import mindspore as ms
import numpy as np

shape = (2, 2)
mean = ms.Tensor(3.0, ms.float32)
stddev = ms.Tensor(np.array([[0.2, 0.3], [0.4, 0.5]]), ms.float32)
output = ms.ops.normal(shape, mean, stddev)
print(output.shape)
# (2, 2)

Code Example 3

In PyTorch, 'mean' is Tensor, and 'std' is the float.

# PyTorch
import torch
import numpy as np

mean = torch.tensor(np.array([[3, 4], [5, 6]]), dtype=torch.float32)
stddev = 1.0
output = torch.normal(mean, stddev)
print(output.shape)
# torch.Size([2, 2])

# MindSpore
import mindspore as ms
import numpy as np

shape = (2, 2)
mean = ms.Tensor(np.array([[3, 4], [5, 6]]), ms.float32)
stddev = ms.Tensor(1.0, ms.float32)
output = ms.ops.normal(shape, mean, stddev)
print(output.shape)
# (2, 2)

Code Example 4

In PyTorch, 'mean' and 'std' are both float.

# PyTorch
import torch
import numpy as np

mean = 3.0
stddev = 1.0
size = (2, 2)
output = torch.normal(mean, stddev, size)
print(output.shape)
# torch.Size([2, 2])

# MindSpore
import mindspore as ms
import numpy as np

shape = (2, 2)
mean = ms.Tensor(3.0, ms.float32)
stddev = ms.Tensor(1.0, ms.float32)
output = ms.ops.normal(shape, mean, stddev)
print(output.shape)
# (2, 2)