pytorch: tensor類型的構建與相互轉換

Summary

主要包括以下三種途徑：

1. 使用獨立的函數；
2. 使用torch.type()函數；
3. 使用type_as(tesnor)將張量轉換爲給定類型的張量。

使用獨立函數

import torch

tensor = torch.randn(3, 5)
print(tensor)

# torch.long() 將tensor投射爲long類型
long_tensor = tensor.long()
print(long_tensor)

# torch.half()將tensor投射爲半精度浮點類型
half_tensor = tensor.half()
print(half_tensor)

# torch.int()將該tensor投射爲int類型
int_tensor = tensor.int()
print(int_tensor)

# torch.double()將該tensor投射爲double類型
double_tensor = tensor.double()
print(double_tensor)

# torch.float()將該tensor投射爲float類型
float_tensor = tensor.float()
print(float_tensor)

# torch.char()將該tensor投射爲char類型
char_tensor = tensor.char()
print(char_tensor)

# torch.byte()將該tensor投射爲byte類型
byte_tensor = tensor.byte()
print(byte_tensor)

# torch.short()將該tensor投射爲short類型
short_tensor = tensor.short()
print(short_tensor)

-0.5841 -1.6370  0.1353  0.6334 -3.0761
-0.2628  0.1245  0.8626  0.4095 -0.3633
 1.3605  0.5055 -2.0090  0.8933 -0.6267
[torch.FloatTensor of size 3x5]


 0 -1  0  0 -3
 0  0  0  0  0
 1  0 -2  0  0
[torch.LongTensor of size 3x5]


-0.5840 -1.6367  0.1353  0.6333 -3.0762
-0.2627  0.1245  0.8628  0.4094 -0.3633
 1.3604  0.5054 -2.0098  0.8936 -0.6265
[torch.HalfTensor of size 3x5]


 0 -1  0  0 -3
 0  0  0  0  0
 1  0 -2  0  0
[torch.IntTensor of size 3x5]


-0.5841 -1.6370  0.1353  0.6334 -3.0761
-0.2628  0.1245  0.8626  0.4095 -0.3633
 1.3605  0.5055 -2.0090  0.8933 -0.6267
[torch.DoubleTensor of size 3x5]


-0.5841 -1.6370  0.1353  0.6334 -3.0761
-0.2628  0.1245  0.8626  0.4095 -0.3633
 1.3605  0.5055 -2.0090  0.8933 -0.6267
[torch.FloatTensor of size 3x5]


 0 -1  0  0 -3
 0  0  0  0  0
 1  0 -2  0  0
[torch.CharTensor of size 3x5]


   0  255    0    0  253
   0    0    0    0    0
   1    0  254    0    0
[torch.ByteTensor of size 3x5]


 0 -1  0  0 -3
 0  0  0  0  0
 1  0 -2  0  0
[torch.ShortTensor of size 3x5]

其中，torch.Tensor、torch.rand、torch.randn 均默認生成 torch.FloatTensor型 ：

import torch

tensor = torch.Tensor(3, 5)
assert isinstance(tensor, torch.FloatTensor)

tensor = torch.rand(3, 5)
assert isinstance(tensor, torch.FloatTensor)

tensor = torch.randn(3, 5)
assert isinstance(tensor, torch.FloatTensor)

使用torch.type()函數

type(new_type=None, async=False)

import torch

tensor = torch.randn(3, 5)
print(tensor)

int_tensor = tensor.type(torch.IntTensor)
print(int_tensor)

-0.4449  0.0332  0.5187  0.1271  2.2303
 1.3961 -0.1542  0.8498 -0.3438 -0.2834
-0.5554  0.1684  1.5216  2.4527  0.0379
[torch.FloatTensor of size 3x5]


 0  0  0  0  2
 1  0  0  0  0
 0  0  1  2  0
[torch.IntTensor of size 3x5]

使用type_as(tesnor)將張量轉換爲給定類型的張量

import torch

tensor_1 = torch.FloatTensor(5)

tensor_2 = torch.IntTensor([10, 20])
tensor_1 = tensor_1.type_as(tensor_2)
assert isinstance(tensor_1, torch.IntTensor)

---

[1] pytorch張量torch.Tensor類型的構建與相互轉換以及torch.type()和torch.type_as()的用法