Early Stopping Hook

[miguelvr]

❓ Questions and Help

I'm trying to implement an early stopping hook that interacts with other custom hooks that I have implemented, but I'm having trouble of getting a sensible way of sending a stop signal to the ClassyTrainer.

The only way I managed to get it working was this:

    def on_loss_and_meter(
        self, task: "tasks.ClassyTask", local_variables: Dict[str, Any]
    ) -> None:
        # FIXME: Ugly hack
        task.done_training = lambda: True
        logging.info("early stopping")

Is there a cleaner way of achieving this in the current state of the project?

[vreis]

Yea, I agree this is hacky. There's no better solution right now.

We could add an exception that would be used to exit training, and that would make this much cleaner. I was considering doing that for other reasons anyway -- let's try to get this on the v0.3 release.

On another note: instead of making a hook for early stopping, I would probably consider overriding ClassificationTask.on_phase_end and add whatever logic is needed there.

[miguelvr]

On another note: instead of making a hook for early stopping, I would probably consider overriding ClassificationTask.on_phase_end and add whatever logic is needed there.

You mean as a solution for now, or as a permanent solution?