Unnecessary warning when resuming training

[eladsegal]

🐛 Bug

When training is resumed with resume_from_checkpoint, a warning is given about missing callbacks even if the expected callbacks are included.

I got the following:

UserWarning: Be aware that when using ``resume_from_checkpoint``, 
callbacks used to create the checkpoint need to be provided. 
Please, add the following callbacks: ['BetterModelCheckpoint', 'FreezeUnfreeze']

when the relevant variables had these values:

current_callbacks_type: {<class 'src.callbacks.model_checkpoint.BetterModelCheckpoint'>, 
<class 'src.callbacks.freeze_unfreeze.FreezeUnfreeze'>, ...}

saved_callbacks_type: {'BetterModelCheckpoint', 'FreezeUnfreeze'}

From a quick look, it seems that the source of the bug is that in current_callbacks_type there are types, and in saved_callbacks_type there are strings from Callback.state_id (which is .__qualname__ of the type).

https://github.com/PyTorchLightning/pytorch-lightning/blob/3096ab88ebb1d9c9c07c3230829171925c9e678a/pytorch_lightning/trainer/callback_hook.py#L277-L286

Sorry I didn't follow the template for reporting bugs. If this isn't enough, I'll be able to update the issue in a few days.

[awaelchli]

Thanks @eladsegal I noticed it too. Great you could find the issue in the code. If you wish feel free to send a PR for the fix. If not I will take a look.
The problem should only exist on master, the state_id changes are not released, so I'm setting the milestone to 1.5