代码拉取完成,页面将自动刷新
torch.Tensor.min(dim=None, keepdim=False)
For more information, see torch.Tensor.min.
mindspore.Tensor.min(axis=None, keepdims=False, *, initial=None, where=True, return_indices=False)
For more information, see mindspore.Tensor.min.
MindSpore is compatible with Numpy parameters initial
and where
based on PyTorch, added parameter return_ Indicators are used to control whether indexes are returned.
Categories | Subcategories | PyTorch | MindSpore | Differences |
---|---|---|---|---|
Inputs | Input 1 | dim | axis | Same function, different parameter names |
Input 2 | keepdim | keepdims | Same function, different parameter names | |
Input 3 | - | initial | Not involved | |
Input 4 | - | where | Not involved | |
Input 5 | - | return_indices | Not involved |
When no dimension is specified, the two APIs implement the same functionality.
import mindspore as ms
import torch
import numpy as np
np_x = np.array([[-0.0081, -0.3283, -0.7814, -0.0934],
[1.4201, -0.3566, -0.3848, -0.1608],
[-0.0446, -0.1843, -1.1348, 0.5722],
[-0.6668, -0.2368, 0.2790, 0.0453]]).astype(np.float32)
# mindspore
input_x = ms.Tensor(np_x)
output = input_x.min()
print(output)
# -1.1348
# torch
input_x = torch.tensor(np_x)
output = input_x.min()
print(output)
# tensor(-1.1348)
When specifying dimensions, MindSpore does not return an index by default and needs to be manually specified.
import mindspore as ms
import torch
import numpy as np
np_x = np.array([[-0.0081, -0.3283, -0.7814, -0.0934],
[1.4201, -0.3566, -0.3848, -0.1608],
[-0.0446, -0.1843, -1.1348, 0.5722],
[-0.6668, -0.2368, 0.2790, 0.0453]]).astype(np.float32)
# mindspore
input_x = ms.Tensor(np_x)
values, indices = input_x.min(axis=1, return_indices=True)
print(values)
# [-0.7814 -0.3848 -1.1348 -0.6668]
print(indices)
# [2 2 2 0]
# torch
input_x = torch.tensor(np_x)
values, indices = input_x.min(dim=1)
print(values)
# tensor([-0.7814, -0.3848, -1.1348, -0.6668])
print(indices)
# tensor([2, 2, 2, 0])
此处可能存在不合适展示的内容,页面不予展示。您可通过相关编辑功能自查并修改。
如您确认内容无涉及 不当用语 / 纯广告导流 / 暴力 / 低俗色情 / 侵权 / 盗版 / 虚假 / 无价值内容或违法国家有关法律法规的内容,可点击提交进行申诉,我们将尽快为您处理。