Using EarlyStopping when calling Trainer.fit() multiple times

[stwerner97]

Hi, I am using a curriculum to train a model on progressively more challenging tasks. In each iteration, the model should either train for max_steps steps or until the EarlyStopping callback stops the process. So far, I could not find a way to modify the patience parameter of the EarlyStopping callback so that training is resumed after the stopping criterion has been met once.

...
model = BoringModel()
early_stopping = pl.callbacks.EarlyStopping(monitor="val_loss", patience=patience_per_iteration)
trainer = pl.Trainer(callbacks=[early_stopping], max_steps=max_steps_per_iteration)

for _ in range(num_iterations):
    trainer.fit(model, datamodule=datamodule)
    trainer.fit_loop.max_steps += max_steps_per_iteration
    # How can I update the patience parameter of the EarlyStopping callback?
    ...

Does pytorch-lightning support this use case? If so, how can I update the patience parameter or modify the callbacks after the creation of the Trainer instance and continue training? As a side note: is there some way to check what condition caused the Trainer to stop training?

python=3.8.10
pytorch=1.10.2
pytorch-lightning=1.6.4