Closed patrick-kidger closed 3 years ago
Good point. Some of these were legacy code. Sending in a fix on dev
.
The setting of annotations to {} is so that we don't get a huge list of type hints when we call help
on the function/module. I personally find it helpful.
Open to suggestions that accomplish this in a better way!
Sounds good to me!
Can we pick a Python version to support? I'd suggest >=3.6.
Future statements: currently used are
absolute_import
,division
andprint_function
. As far as I know these only affect Python 2 which is out of support anyway. Should we really include them?Likewise, the setting of annotations to
{}
seems unnecessary to me. The minimum verison of Python now supported by PyTorch is 3.6, and type annotations were introduced in 3.5 I believe.I think we're already demanding at least Python 3.3 for the
@property@abstractmethod
to work.