1. 背景
在torch和numpy中经常会遇到对tensor进行切片操作,如x[..., :3]
, [:, :2]
等,对于:
的操作很好理解,与python列表中操作相同,但是...
是什么操作呢
2. 示例
以torch中的切片为例,如下
import torch
x = torch.rand((2,3,4))
print(x)
y = x[..., 2:4]
z = x[..., :, -1]
k = x[:, :, -1]
j = x[1, ...]
print("y", y)
print("z:", z)
print("k", k)
print("j", j)
输出如下:
tensor([[[0.2626, 0.8649, 0.0496, 0.0013],
[0.5466, 0.1964, 0.1133, 0.5052],
[0.8863, 0.2080, 0.0276, 0.0924]],
[[0.4060, 0.8400, 0.4646, 0.3968],
[0.2729, 0.3125, 0.7060, 0.3351],
[0.3627, 0.8817, 0.9210, 0.3325]]])
y: tensor([[[0.0496, 0.0013],
[0.1133, 0.5052],
[0.0276, 0.0924]],
[[0.4646, 0.3968],
[0.7060, 0.3351],
[0.9210, 0.3325]]])
z: tensor([[0.0013, 0.5052, 0.0924],
[0.3968, 0.3351, 0.3325]])
k tensor([[0.0013, 0.5052, 0.0924],
[0.3968, 0.3351, 0.3325]])
j tensor([[0.4060, 0.8400, 0.4646, 0.3968],
[0.2729, 0.3125, 0.7060, 0.3351],
[0.3627, 0.8817, 0.9210, 0.3325]])
通过对上面案例总结发现:...
代替了他前面所有的:
,例如上面案例中y,他里面的...
代替了第0维和第1维的两个冒号,而对于z案例的情况省略号和冒号的作用相同