写 PermutePatch 时遇到一个 bug：在试图交换 PyTorch 数组的两个元素时，两个位置都变成同一个元素！具体见测试代码。本文兼测几种交换情况：
- 两个 python 变量
- list 两个元素
- numpy 数组两个元素
- pytorch 数组两个元素
import numpy as np
import torch
print("\tvar")
a, b = 1, 2
print("before:", a, b)
a, b = b, a
print("after:", a, b)
print("\tlist")
c = [1, 2, 3, 4]
print("before:", c)
c[0], c[2] = c[2], c[0]
print("after:", c)
print("\tnumpy")
d = np.array([1, 2, 3, 4])
print("before:", d)
d[0], d[2] = d[2], d[0]
print("after:", d)
print("\ttorch")
e = torch.tensor([1, 2, 3, 4])
print("before:", e)
e[0], e[2] = e[2], e[0]
print("after:", e) # 出事
f = torch.tensor([1, 2, 3, 4])
f[0], f[2] = f[2].item(), f[0].item()
print("after 2:", f) # 如期
输出：
var
before: 1 2
after: 2 1
list
before: [1, 2, 3, 4]
after: [3, 2, 1, 4]
numpy
before: [1 2 3 4]
after: [3 2 1 4]
torch
before: tensor([1, 2, 3, 4])
after: tensor([3, 2, 3, 4]) # <- 两个 `3`
after 2: tensor([3, 2, 1, 4])
其中 e 组的换法会出现两个 3；而 f 组加上 .item() 之后，结果就如预期了。
哇，呢啲乜嘢 bug 来㗎，黐线
