type annotation `NdarrayOrTensor`

[wyli]

Is your feature request related to a problem? Please describe.
NdarrayOrTensor should be replaced by a type variable such as NdarrayTensor
this will ensure advanced type checks for correctness however it might not be supported by all the tools such as pytype, mypy

Additional context

MONAI/monai/config/type_definitions.py

Lines 65 to 76 in 93d467d

	NdarrayTensor = TypeVar("NdarrayTensor", np.ndarray, torch.Tensor)
	"""NdarrayTensor
	Generic type which can represent either a numpy.ndarray or a torch.Tensor
	Unlike Union can create a dependence between parameter(s) / return(s)
	"""
	NdarrayOrTensor = Union[np.ndarray, torch.Tensor]
	"""NdarrayOrTensor
	Union of numpy.ndarray and torch.Tensor to be used for typing
	"""

this is a follow up of #2815

[Nic-Ma]

Currently, I think we used # type: ignore as a temp solution for many places, will also need to update them.

Thanks.

[rijobro]

I propose unifying the two to NdarrayTensor = Union[np.ndarray, torch.Tensor] and deleting NdarrayOrTensor. I tried this locally and there are a few changes that need to be made, but works for the most part.

@wyli what's the advantage of using TypeVar here?

[wyli]

that is mainly to ensure the consistency of the input and output, please see https://stackoverflow.com/questions/58903906/whats-the-difference-between-a-constrained-typevar-and-a-union

But I agree that we delete NdarrayOrTensor and have NdarrayTensor = Union[np.ndarray, torch.Tensor] as a workaround, it'll create less confusions and avoid spending too much effort in the type checking -- given that pep 484 is currently provisional https://www.python.org/dev/peps/pep-0484/ and type annotation is optional

[rijobro]

that is mainly to ensure the consistency of the input and output, please see https://stackoverflow.com/questions/58903906/whats-the-difference-between-a-constrained-typevar-and-a-union

Thanks for this link. If you check the second code block of the answer, an error is raised if U and T are of different types. We may want to change the types (input might be np.ndarray and output might be torch.Tensor). Maybe this is where our errors are coming from?