def test_early_stopping_squeezes():
early_stopping = EarlyStopping(monitor="foo")
trainer = Trainer()
trainer.callback_metrics["foo"] = torch.tensor([[[0]]])
with mock.patch(
"pytorch_lightning.callbacks.EarlyStopping._evaluate_stopping_criteria",
return_value=(False, "")) as es_mock:
early_stopping._run_early_stopping_check(trainer)
es_mock.assert_called_once_with(torch.tensor(0))
def test_early_stopping_no_extraneous_invocations(tmpdir):
"""Test to ensure that callback methods aren't being invoked outside of the callback handler."""
model = ClassificationModel()
dm = ClassifDataModule()
early_stop_callback = EarlyStopping(monitor="train_loss")
early_stop_callback._run_early_stopping_check = Mock()
expected_count = 4
trainer = Trainer(
default_root_dir=tmpdir,
callbacks=[early_stop_callback],
limit_train_batches=4,
limit_val_batches=4,
max_epochs=expected_count,
enable_checkpointing=False,
)
trainer.fit(model, datamodule=dm)
assert trainer.early_stopping_callback == early_stop_callback
assert trainer.early_stopping_callbacks == [early_stop_callback]
assert early_stop_callback._run_early_stopping_check.call_count == expected_count
