最近開始仔細玩了一下pytorch，發現裡面有個BUG之前都沒有發現。

在測試torch最基本的示例的情況下，居然碰到了個pytorch無法轉化numpy為Tensor的問題，呈現的問題如下：

[email protected]:~/work/change/AI$ python
Python 3.6.1 (default, Jul 14 2017, 17:08:44) 
[GCC 4.8.4] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import torch
>>> import numpy as np
>>> a = np.ones(5)
>>> a
array([ 1.,  1.,  1.,  1.,  1.])
>>> b=torch.form_numpy(a)
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
AttributeError: module 'torch' has no attribute 'form_numpy'
>>> print(torch.__version__)
0.2.0_3
 

在torch的主頁上有這樣一句話，經過仔細分析才明白其中的意思：

Pylint isn't picking up that torch has the member function from_numpy. It's because torch.from_numpy is actually torch._C.from_numpy as far as Pylint is concerned.

本身而言，pytorch並不直接包含from_numpy這個方法，而需要通過_C這樣的名稱空間來呼叫。

因此利用._C的辦法進行測試，果然順利通過。

>>> b=torch.form_numpy(a)
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
AttributeError: module 'torch' has no attribute 'form_numpy'
>>> print(torch.__version__)
0.2.0_3
>>> c = torch.Tensor(3,3)
>>> c

1.00000e-32 *
 -4.4495  0.0000  0.0000
  0.0000  0.0000  0.0000
  0.0000  0.0000  0.0000
[torch.FloatTensor of size 3x3]

>>> b = torch._C.from_numpy(a)
>>> b

 1
 1
 1
 1
 1
[torch.DoubleTensor of size 5]
 

For reference, you can have Pylint ignore these by wrapping "problematic" calls with the following comments.

# pylint: disable=E1101
tensor = torch.from_numpy(np_array)
# pylint: enable=E1101

於是重新再次安裝一下這個工具。

pip install pylint

[email protected]:~/work/change/AI$ python
Python 3.6.1 (default, Jul 14 2017, 17:08:44) 
[GCC 4.8.4] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import torch
>>> import numpy as np
>>> a = np.ones(5)
>>> b=torch.from_numpy(a)
>>> b

 1
 1
 1
 1
 1
[torch.DoubleTensor of size 5]

>>>