Code autocomplete problem

somethingForNothing · October 23, 2018, 2:52pm

I just started learning pytorch, I don’t quite understand why I can’t use auto-completion for some functions in Pycharm and VS Code.Such as torch.empty(), torch.zeros(), etc.
Can you help me with this problem?
Thanks~ o(￣︶￣)o

ptrblck · October 23, 2018, 4:49pm

This issue is tracked here and fixes are actively developed.

tom · October 23, 2018, 7:30pm

Here is a pyi that PR12500 generates:

gist.github.com

https://gist.github.com/t-vi/0d0ae013072f96f50fa11fbc2287e33b

init.pyi


from typing import List, Tuple, Optional, Union, Any, ContextManager, Callable, overload

import builtins
import math
import pickle


class dtype: ...
_dtype = dtype

This file has been truncated. show original

As far as I know it works with mypy and PyCharm and has reasonable coverage.

Best regards

Thomas

binbin · November 10, 2018, 2:44am

This pyi file solves most built-in functions autocomplete problem. Many thanks to tom. However, when I try to type torch.nn, Pycharm shows warning ‘cannot find reference nn in __init__.pyi’. It seems this pyi file shadows all package of torch like torch.nn and torch.autograd. Is there any way to solve this? I will be very appreciated with that!

yannick · February 14, 2019, 6:05am

the pytorch 1.0.1 still not works. the torch.randn, torch.FloatTenser() can not be autocompleted by vscode or wingide.

tom · February 14, 2019, 7:18am

torch.FloatTensor is deprecated, so not recommending that is a feature.
There are bugs to be expected, but I see that you(?) already found issue 16996, which is a good place to report those.

Best regards

Thomas

Umar_Karim · March 9, 2020, 8:41am

I faced a similar problem with pytorch 1.4, vs code 1.42 with torch.nn as it was not showing autocomplete. I changed the language server from microsoft to jedi and now autocomplete works.