Does ModelCheckpoint callback take priority over EarlyStopping callback?

callbacks
1

Here’s my runner code snippet:

    from pytorch_lightning.callbacks import ModelCheckpoint
    checkpoint_callback = ModelCheckpoint(
        save_top_k=5,
        verbose=True,
        monitor='avg_val_loss',
        mode='min'
    )

    from pytorch_lightning.callbacks.early_stopping import EarlyStopping
    early_stop_callback = EarlyStopping(
        monitor='avg_val_accuracy',
        min_delta=0.00,
        patience=10,
        verbose=True,
        mode='max'
    )

    model = Model()
    trainer = Trainer(gpus=1,
        callbacks=[early_stop_callback],
        checkpoint_callback=checkpoint_callback)
    
    trainer.fit(model)

My above training stops with the output:

Epoch 29: avg_val_loss was not in top 5

Does this mean the ModelCheckpoint callback also performs some kind of early stopping?
If not, why is my EarlyStopping callback not being effective?

2

this comes from ModelCheckpoint since you have set save_top_k=5 and verbose=True. So at epoch 29 the monitored value avg_val_loss wasn’t in top 5.

ref: https://github.com/PyTorchLightning/pytorch-lightning/blob/5c1eff351b035db5881d0cff81b1d9c9e150e2d0/pytorch_lightning/callbacks/model_checkpoint.py#L493-L498