Function Differences with torch.cumsum

torch.cumsum

torch.cumsum(input, dim, *, dtype=None, out=None) -> Tensor

For more information, see torch.cumsum.

mindspore.ops.cumsum

mindspore.ops.cumsum(x, axis, dtype=None) -> Tensor

For more information, see mindspore.ops.cumsum.

Differences

PyTorch: Calculates the cumulative sum of the input Tensor on the specified axis.

MindSpore: MindSpore API implements functions basically same as PyTorch, but there are differences in parameter settings.

Categories	Subcategories	PyTorch	MindSpore	Difference
Parameters	Parameter 1	input	x	Same function, different parameter names
	Parameter 2	dim	axis	Same function, different parameter names
	Parameter 3	dtype	dtype	-
	Parameter 4	out	-	Not involved

Code Example 1

When the input tensor is the same and the accumulation axis is -1, the innermost layer of the tensor is accumulated from left to right, and the two APIs achieve the same function.

# PyTorch
import torch
from torch import tensor
import numpy as np
a = tensor(np.array([[3, 4, 6, 10], [1, 6, 7, 9], [4, 3, 8, 7], [1, 3, 7, 9]]).astype(np.float32))
y = torch.cumsum(a, dim=-1)
print(y.numpy())
# [[ 3.  7. 13. 23.]
#  [ 1.  7. 14. 23.]
#  [ 4.  7. 15. 22.]
#  [ 1.  4. 11. 20.]]

# MindSpore
from mindspore import Tensor
import mindspore.ops as ops
import numpy as np
x = Tensor(np.array([[3, 4, 6, 10], [1, 6, 7, 9], [4, 3, 8, 7], [1, 3, 7, 9]]).astype(np.float32))
y = ops.cumsum(x, -1)
print(y)
# [[ 3.  7. 13. 23.]
#  [ 1.  7. 14. 23.]
#  [ 4.  7. 15. 22.]
#  [ 1.  4. 11. 20.]]

Code Example 2

When the input tensor and the accumulation axis are the same, torch.cumsum and MindSpore get the same result by setting the data type of the output y to int8 through the parameter dtype.

# PyTorch
import torch
from torch import tensor
import numpy as np
a = tensor(np.array([[3, 4, 6, 10], [1, 6, 7, 9], [4, 3, 8, 7], [1, 3, 7, 9]]).astype(np.float32))
y = torch.cumsum(a, dim=0, dtype=torch.int8)
print(y.numpy())
# [[ 3  4  6 10]
#  [ 4 10 13 19]
#  [ 8 13 21 26]
#  [ 9 16 28 35]]
print(y.dtype)
# torch.int8

# MindSpore
import mindspore
from mindspore import Tensor
import mindspore.ops as ops
x = Tensor([[3, 4, 6, 10], [1, 6, 7, 9], [4, 3, 8, 7], [1, 3, 7, 9]], mindspore.float32)
y = ops.cumsum(x, 0, dtype=mindspore.int8)
print(y)
# [[ 3  4  6 10]
#  [ 4 10 13 19]
#  [ 8 13 21 26]
#  [ 9 16 28 35]]
print(y.dtype)
# Int8