前言

搭建深度神经网络模型时，难免会遇到 torch.cat()函数，来进行tensor的拼接。

错误信息

代码案例如下：

import torch
import numpy as np

array1 = np.zeros((4, 1, 28, 28))
array2 = np.zeros((4, 1, 28, 28))
print("array1.shape:", array1.shape)      # (4, 1, 28, 28)

tensor1 = torch.tensor(array1)
tensor2 = torch.tensor(array2)
print("tensor1.shape", tensor1.shape)    # torch.Size([4, 1, 28, 28])

c = torch.cat(tensor1, tensor2, dim=0)    # 报错
1
2
3
4
5
6
7
8
9
10
11
12

这样执行会报错，错误信息如下：

TypeError: cat() received an invalid combination of arguments - got (Tensor, Tensor, dim=int), but expected one of:
 * (tuple of Tensors tensors, name dim, Tensor out)
      didn't match because some of the keywords were incorrect: dim
 * (tuple of Tensors tensors, int dim, Tensor out)
      didn't match because some of the keywords were incorrect: dim
1
2
3
4
5

解决方法

torch.cat() 函数进行tensor的拼接，将要拼接的tensor组合成元组，即可解决该报错¹。

代码修改如下：

import torch
import numpy as np

array1 = np.zeros((4, 1, 28, 28))
array2 = np.zeros((4, 1, 28, 28))
print("array1.shape:", array1.shape)      # (4, 1, 28, 28)

tensor1 = torch.tensor(array1)
tensor2 = torch.tensor(array2)
print("tensor1.shape", tensor1.shape)    # torch.Size([4, 1, 28, 28])

# c = torch.cat(tensor1, tensor2, dim=0)    # 报错
c = torch.cat((tensor1, tensor2), dim=0)
print("c.shape", c.shape)
c1 = torch.cat((tensor1, tensor2), dim=1)
print("c1.shape", c1.shape)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16

此时输出为：

array1.shape: (4, 1, 28, 28)
tensor1.shape torch.Size([4, 1, 28, 28])
c.shape torch.Size([8, 1, 28, 28])
c1.shape torch.Size([4, 2, 28, 28])
1
2
3
4

从该案例中，不仅可以学到dubug信息，还可以详细了解到 torch.cat()函数的具体用法，比如参数dim的含义等等。

---

如果我的这篇文章帮助到了你，那我也会感到很高兴，一个人能走多远，在于与谁同行。

---

参考

---

1. 使用torch.cat()和torch.stack()时出现的小错误 ↩︎