Differences with torch.Tensor.expand

torch.Tensor.expand

torch.Tensor.expand(*sizes) -> Tensor

For more information, see torch.Tensor.expand.

mindspore.Tensor.broadcast_to

mindspore.Tensor.broadcast_to(shape) -> Tensor

For more information, see mindspore.Tensor.broadcast_to.

Differences

MindSpore API function is consistent with PyTorch, with differences in the data types supported by the parameters.

PyTorch: sizes is the target shape after broadcasting, which can be of type torch.Size or a sequence consisting of int.

MindSpore: shape is the target shape after broadcasting, which can be of type tuple[int].

Categories	Subcategories	PyTorch	MindSpore	Differences
Parameter	Parameter 1	*sizes	shape	Both parameters have different names, but both indicate the target shape after broadcasting. The type of sizes can be torch.Size or a sequence consisting of int, and the type of shape can be tuple[int].

Code Example

# PyTorch
import torch

x = torch.tensor([1, 2, 3])
output = x.expand(3, 3)
print(output)
# tensor([[1, 2, 3],
#         [1, 2, 3],
#         [1, 2, 3]])

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

shape = (3, 3)
x = Tensor(np.array([1, 2, 3]))
output = x.broadcast_to(shape)
print(output)
# [[1 2 3]
#  [1 2 3]
#  [1 2 3]]