Not saving checkpoint when monitor is None and save_top_k is -1

[ruotianluo]

🐛 Bug

When monitor is None, current will be None here
https://github.com/PyTorchLightning/pytorch-lightning/blob/6bc4490d01aed21c2d52f884d4afbeaa24a47ca0/pytorch_lightning/callbacks/model_checkpoint.py#L553

And check_monitor_top_k will return False because of that:
https://github.com/PyTorchLightning/pytorch-lightning/blob/6bc4490d01aed21c2d52f884d4afbeaa24a47ca0/pytorch_lightning/callbacks/model_checkpoint.py#L340

_update_best_and_save also doesn't take None current. raise error here: https://github.com/PyTorchLightning/pytorch-lightning/blob/6bc4490d01aed21c2d52f884d4afbeaa24a47ca0/pytorch_lightning/callbacks/model_checkpoint.py#L605

Currently, checkpointing is associated with validation, which is not necessarily always the case. I just want to save the checkpoint every k iterations.

Please reproduce using the BoringModel

Sorry, no time to reproduce for now.

To Reproduce

Use following BoringModel and post here

Expected behavior

Should save the checkpoint always if save_top_k == -1.

Environment

Note: Bugs with code are solved faster ! Colab Notebook should be made public !

• IDE: Please, use our python bug_report_model.py template.

• Colab Notebook: Please copy and paste the output from our environment collection script (or fill out the checklist below manually).

You can get the script and run it with:

wget https://raw.githubusercontent.com/PyTorchLightning/pytorch-lightning/master/tests/collect_env_details.py
# For security purposes, please check the contents of collect_env_details.py before running it.
python collect_env_details.py

• PyTorch Version (e.g., 1.0):
• OS (e.g., Linux):
• How you installed PyTorch (conda, pip, source):
• Build command you used (if compiling from source):
• Python version:
• CUDA/cuDNN version:
• GPU models and configuration:
• Any other relevant information:

Additional context

[carmocca]

When monitor is None, the _save_last_checkpoint function is the one to save the model (even if save_last is True), not _update_best_and_save.

I just manually checked and it seems to work properly

def test_bug(tmpdir):
    model_ckpt = ModelCheckpoint(dirpath=tmpdir, monitor=None, save_top_k=-1, save_last=False)
    model = BoringModel()
    trainer = Trainer(
        default_root_dir=tmpdir,
        logger=False,
        max_epochs=3,
        callbacks=[model_ckpt],
    )
    trainer.fit(model)

    print(os.listdir(tmpdir))
    # ['epoch=0-step=63.ckpt', 'epoch=2-step=191.ckpt', 'epoch=1-step=127.ckpt']

Can you elaborate on what is the problem?

[ruotianluo]

I see the code doing that. Interesting, let me check with my code.

[ruotianluo]

@carmocca I c. this is because you set save_last to False. If you set save_last to True, the checkpoints will always be last.ckpt.