Transform Base Class

[clee-ai]

Hi!

Right now all of the transforms are inherited from object, I was wondering if we could create a common ABC for all the transforms to inherit from so that I can do static typechecking. If you think that's a good idea I'd be happy to do a PR.

Thanks!

[rusty1s]

Thanks for reaching out. Can you elaborate a bit more on the possible use-cases? Right now, all transforms just implement a single __call__(self, data) method, similar to how torchvision is doing it, so I'm not sure which benefits an abstract base class offers.

[clee-ai]

Hi, the main usecase would be for typechecking/linting so that we can validate that valid transforms are being passed as parameters or in config files:

def run(transform: object):
	T.Compose([transform])
	
run("string") # typechecker won't error


def run(transform: transform_base_class): # all transforms inherit from transform_base_class
	T.Compose([transform])

run("string") # throws linter error
run(T.RandomScale()) # okay

Let me know if that makes sense.

[rusty1s]

I see, thanks for clarifying. I'm okay with it if you wanna contribute this feature.