在Pytorch的旧版本中，没有__init__.pyi这个文件（Pycharm就从__init__.pyi读取函数的声明以及参数类型），因此在Pycharm中torch.sum、torch.abs等函数就没有代码提示。没有代码提示很不方便，经过一番折腾，完美地解决了Pytorch在Pycharm中没有代码提示的问题。
- 把Pytorch的版本升级到1.0.1.post2。
torch.sum、torch.abs等函数应该是有提示了，但是torch.nn、torch.optim等却没有提示了。解决方案如下：
-
升级Pycharm到版本2019.1.1。
-
在
site-packages/torch/__init__.pyi中增加两行代码。
旧的__init__.pyi：from typing import Tuple, Optional, Union import pickle class Tensor: def abs(self, *, out: Optional['Tensor']=None) -> 'Tensor': ... ...新的
__init__.pyi：from typing import Tuple, Optional, Union import pickle from torch import nn, cuda, ops, functional, optim, autograd, onnx, utils from torch import contrib, distributions, for_onnx, jit, multiprocessing class Tensor: def abs(self, *, out: Optional['Tensor']=None) -> 'Tensor': ... ...
又遇到了“class Tensor does not define __sub__, so the ‘-’ cannot be used in its instances”这样的错误，翻译过来就是Tensor类没有定义__sub__， ‘-’ 就不能用在实例之间。同样的还有加减乘除大于小于等符号都有类似的提示。于是继续修改__init__.pyi
旧的Tensor定义：
class Tensor:
def abs(self, *, out: Optional['Tensor']=None) -> 'Tensor': ...
...
新的Tensor定义：
class Tensor:
def __add__(self, other: Union['Tensor', 'int', 'float']): ...
def __radd__(self, other: Union['Tensor', 'int', 'float']): ...
def __iadd__(self, other: Union['Tensor', 'int', 'float']): ...
def __sub__(self, other: Union['Tensor', 'int', 'float']): ...
def __rsub__(self, other: Union['Tensor', 'int', 'float']): ...
def __isub__(self, other: Union['Tensor', 'int', 'float']): ...
def __mul__(self, other: Union['Tensor', 'int', 'float']): ...
def __rmul__(self, other: Union['Tensor', 'int', 'float']): ...
def __imul__(self, other: Union['Tensor', 'int', 'float']): ...
def __truediv__(self, other: Union['Tensor', 'int', 'float']): ...
def __rtruediv__(self, other: Union['Tensor', 'int', 'float']): ...
def __itruediv__(self, other: Union['Tensor', 'int', 'float']): ...
def __pow__(self, power: 'int', modulo=None) -> 'Tensor': ...
def __rpow__(self, power: 'int', modulo=None) -> 'Tensor': ...
def __ipow__(self, power: 'int', modulo=None) -> 'Tensor': ...
def __ge__(self, other: Union['Tensor', 'int', 'float']): ...
def __gt__(self, other: Union['Tensor', 'int', 'float']): ...
def __le__(self, other: Union['Tensor', 'int', 'float']): ...
def __lt__(self, other: Union['Tensor', 'int', 'float']): ...
def __ne__(self, other: Union['Tensor', 'int', 'float']): ...
def __eq__(self, other: Union['Tensor', 'int', 'float']): ...
def __neg__(self, other: Union['Tensor', 'int', 'float']): ...
def __and__(self, other: Union['Tensor', 'int', 'float']): ...
def __rand__(self, other: Union['Tensor', 'int', 'float']): ...
def __iand__(self, other: Union['Tensor', 'int', 'float']): ...
def __or__(self, other: Union['Tensor', 'int', 'float']): ...
def __ror__(self, other: Union['Tensor', 'int', 'float']): ...
def __ior__(self, other: Union['Tensor', 'int', 'float']): ...
def __xor__(self, other: Union['Tensor', 'int', 'float']): ...
def __rxor__(self, other: Union['Tensor', 'int', 'float']): ...
def __ixor__(self, other: Union['Tensor', 'int', 'float']): ...
def __getitem__(self, item) -> 'Tensor': ...
def __setitem__(self, key, value) -> 'Tensor': ...
def __abs__(self): ...
def __ceil__(self): ...
def __floor__(self): ...
def __bool__(self) -> 'Tensor': ...
def size(self) -> 'torch.Size': ...
def cpu(self) -> 'Tensor': ...
def cuda(self, device:Optional['torch.device']=None, non_blocking: Optional['bool']=False) -> 'Tensor': ...
@property
def device(self): ...
def numpy(self) -> 'numpy.ndarray': ...
def dim(self) -> 'int': ...
def ndimension(self) -> 'int': ...
def double(self) -> 'Tensor': ...
def float(self) -> 'Tensor': ...
def half(self) -> 'Tensor': ...
def int(self) -> 'Tensor': ...
def long(self) -> 'Tensor': ...
def short(self) -> 'Tensor': ...
def abs(self, *, out: Optional['Tensor']=None) -> 'Tensor': ...
...
就把加减乘除还有Tensor.size()函数都给它添加上了。说了这么多，其实就是缺啥定义就在__init__.pyi补啥定义。
我修改的__init__.pyi可从项目PythonResources下载（本博客不在更新）。下载完之后，替换掉site-packages/torch/中的__init__.pyi就行。
如果还有没有代码提示的地方，欢迎fork仓库，改进并提交给我合并。
