Function Differences with torch.Tensor.masked_scatter

torch.Tensor.masked_scatter

torch.Tensor.masked_scatter(mask, tensor) -> Tensor

For more information, see torch.Tensor.masked_scatter.

mindspore.Tensor.masked_scatter

mindspore.Tensor.masked_scatter(mask, tensor) -> Tensor

For more information, see mindspore.Tensor.masked_scatter.

Differences

PyTorch: Returns a Tensor. Updates the value in the “self Tensor” with the tensor value according to the mask.

MindSpore: MindSpore API Basically achieves the same function as PyTorch. But PyTorch supports bidirectional broadcast of mask and “self Tensor”, MindSpore only supports mask broadcasting to “self Tensor”.

Categories	Subcategories	PyTorch	MindSpore	Difference
Parameters	Parameter 1	mask	mask	PyTorch supports bidirectional broadcast of mask and “self Tensor”, MindSpore only supports mask broadcasting to “self Tensor”.
	Parameter 2	tensor	tensor	-

Code Example 1

# PyTorch
import torch

self = torch.tensor([0, 0, 0, 0, 0])
mask = torch.tensor([[0, 0, 0, 1, 1], [1, 1, 0, 1, 1]])
source = torch.tensor([[0, 1, 2, 3, 4], [5, 6, 7, 8, 9]])
output = self.masked_scatter(mask, source)
print(output)
# tensor([[0, 0, 0, 0, 1],
#         [2, 3, 0, 4, 5]])

# MindSpore
import mindspore
from mindspore import Tensor
import numpy as np

self = Tensor(np.array([[0, 0, 0, 0, 0], [0, 0, 0, 0, 0]]), mindspore.int32)
mask = Tensor(np.array([[False, False, False, True, True], [True, True, False, True, True]]), mindspore.bool_)
source = Tensor(np.array([[0, 1, 2, 3, 4], [5, 6, 7, 8, 9]]), mindspore.int32)
output = self.masked_scatter(mask, source)
print(output)
# [[0 0 0 0 1],
#  [2 3 0 4 5]]

Code Example 2

import torch

self = torch.tensor([[0, 0, 0, 0, 0], [0, 0, 0, 0, 0]])
mask = torch.tensor([0, 0, 0, 1, 1])
source = torch.tensor([[0, 1, 2, 3, 4], [5, 6, 7, 8, 9]])
output = self.masked_scatter(mask, source)
print(output)
# tensor([[0, 0, 0, 0, 1],
#         [0, 0, 0, 2, 3]])

# MindSpore
import mindspore
from mindspore import Tensor
import numpy as np

self = Tensor(np.array([[0, 0, 0, 0, 0], [0, 0, 0, 0, 0]]), mindspore.int32)
mask = Tensor(np.array([False, False, False, True, True]), mindspore.bool_)
source = Tensor(np.array([[0, 1, 2, 3, 4], [5, 6, 7, 8, 9]]), mindspore.int32)
output = self.masked_scatter(mask, source)
print(output)
# [[0 0 0 0 1],
#  [0 0 0 2 3]]