Type annotate messengers #3309
Type annotate messengers #3309
Conversation
ordabayevy
commented
Jan 2, 2024
- plate_messenger
- reentrant_messenger
- reparam_messenger
pyro/poutine/runtime.py
Outdated
| args: Tuple | ||
| kwargs: Dict | ||
| value: Optional[torch.Tensor] | ||
| value: Optional[T] |
There was a problem hiding this comment.
I've learned this neat trick with Generic and TypeVar where the type of value can be inferred from the Callable signature. I also fixed effectful so that it gives the correct signature for the decorated function when the return type is diferent from torch.Tensor (e.g. reparam_messenger._get_init_messengers).
pyro/poutine/runtime.py
Outdated
| ) | ||
| # apply the stack and return its return value | ||
| apply_stack(msg) | ||
| assert msg["value"] is not None |
There was a problem hiding this comment.
Is this always correct? All tests have passed
pyro/poutine/runtime.py
Outdated
| ) | ||
| # apply the stack and return its return value | ||
| apply_stack(msg) | ||
| assert msg["value"] is not None |
There was a problem hiding this comment.
@fritzo can you have another look? I have addressed your comments.
| ) -> Union[T, torch.Tensor, None]: | ||
| obs: Optional[_T] = None, | ||
| **kwargs: _P.kwargs, | ||
| ) -> Optional[_T]: |
There was a problem hiding this comment.
Changed this back to return Optional and removed the assert msg["value"] is not None line. One concern I have is that if _T itself is None then it will raise an assertion error.