PyTorch支持类型

要注意， torch.Tensor 是 torch.FloatTensor 的别名。

类型转换方法

主要包括以下三种途径：

  1. 使用独立的类型函数；
  2. 使用 torch.type() 函数；
  3. 使用 type_as(tesnor) 将张量转换为给定类型的张量tensor。

使用独立函数

  1. import torch
  3. tensor = torch.randn(3, 5)
  4. print(tensor)
  6. # torch.long() 将tensor投射为long类型
  7. long_tensor = tensor.long()
  8. print(long_tensor)
  10. # torch.half()将tensor投射为半精度浮点类型
  11. half_tensor = tensor.half()
  12. print(half_tensor)
  14. # torch.int()将该tensor投射为int类型
  15. int_tensor = tensor.int()
  16. print(int_tensor)
  18. # torch.double()将该tensor投射为double类型
  19. double_tensor = tensor.double()
  20. print(double_tensor)
  22. # torch.float()将该tensor投射为float类型
  23. float_tensor = tensor.float()
  24. print(float_tensor)
  26. # torch.char()将该tensor投射为char类型
  27. char_tensor = tensor.char()
  28. print(char_tensor)
  30. # torch.byte()将该tensor投射为byte类型
  31. byte_tensor = tensor.byte()
  32. print(byte_tensor)
  34. # torch.short()将该tensor投射为short类型
  35. short_tensor = tensor.short()
  36. print(short_tensor)
  38. # 输出 ################################################################################
  39. -0.5841 -1.6370  0.1353  0.6334 -3.0761
  40. -0.2628  0.1245  0.8626  0.4095 -0.3633
  41.  1.3605  0.5055 -2.0090  0.8933 -0.6267
  42. [torch.FloatTensor of size 3x5]
  45.  0 -1  0  0 -3
  46.  0  0  0  0  0
  47.  1  0 -2  0  0
  48. [torch.LongTensor of size 3x5]
  51. -0.5840 -1.6367  0.1353  0.6333 -3.0762
  52. -0.2627  0.1245  0.8628  0.4094 -0.3633
  53.  1.3604  0.5054 -2.0098  0.8936 -0.6265
  54. [torch.HalfTensor of size 3x5]
  57.  0 -1  0  0 -3
  58.  0  0  0  0  0
  59.  1  0 -2  0  0
  60. [torch.IntTensor of size 3x5]
  63. -0.5841 -1.6370  0.1353  0.6334 -3.0761
  64. -0.2628  0.1245  0.8626  0.4095 -0.3633
  65.  1.3605  0.5055 -2.0090  0.8933 -0.6267
  66. [torch.DoubleTensor of size 3x5]
  69. -0.5841 -1.6370  0.1353  0.6334 -3.0761
  70. -0.2628  0.1245  0.8626  0.4095 -0.3633
  71.  1.3605  0.5055 -2.0090  0.8933 -0.6267
  72. [torch.FloatTensor of size 3x5]
  75.  0 -1  0  0 -3
  76.  0  0  0  0  0
  77.  1  0 -2  0  0
  78. [torch.CharTensor of size 3x5]
  81.    0  255    0    0  253
  82.    0    0    0    0    0
  83.    1    0  254    0    0
  84. [torch.ByteTensor of size 3x5]
  87.  0 -1  0  0 -3
  88.  0  0  0  0  0
  89.  1  0 -2  0  0
  90. [torch.ShortTensor of size 3x5]

使用 torch.type()

type(new_type=None, async=False) 

  1. import torch
  3. tensor = torch.randn(3, 5)
  4. print(tensor)
  6. int_tensor = tensor.type(torch.IntTensor)
  7. print(int_tensor)
  9. # 输出 ################################################################################
  10. -0.4449  0.0332  0.5187  0.1271  2.2303
  11.  1.3961 -0.1542  0.8498 -0.3438 -0.2834
  12. -0.5554  0.1684  1.5216  2.4527  0.0379
  13. [torch.FloatTensor of size 3x5]
  16.  0  0  0  0  2
  17.  1  0  0  0  0
  18.  0  0  1  2  0
  19. [torch.IntTensor of size 3x5]

使用 type_as(a)

  1. import torch
  3. tensor_1 = torch.FloatTensor(5)
  5. tensor_2 = torch.IntTensor([10, 20])
  6. tensor_1 = tensor_1.type_as(tensor_2)
  7. assert isinstance(tensor_1, torch.IntTensor)

参考链接