possible bug in test_wake.py: 'int' object is not iterable

[jeff-regier]

tests/test_wake.py:157: 
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 
../../miniconda3/envs/stats507/lib/python3.7/site-packages/pytorch_lightning/trainer/trainer.py:921: in fit
    self.single_gpu_train(model)
../../miniconda3/envs/stats507/lib/python3.7/site-packages/pytorch_lightning/trainer/distrib_parts.py:171: in single_gpu_train
    self.run_pretrain_routine(model)
../../miniconda3/envs/stats507/lib/python3.7/site-packages/pytorch_lightning/trainer/trainer.py:1091: in run_pretrain_routine
    self.train()
../../miniconda3/envs/stats507/lib/python3.7/site-packages/pytorch_lightning/trainer/training_loop.py:374: in train
    self.run_training_epoch()
../../miniconda3/envs/stats507/lib/python3.7/site-packages/pytorch_lightning/trainer/training_loop.py:424: in run_training_epoch
    self.on_epoch_start()
../../miniconda3/envs/stats507/lib/python3.7/site-packages/pytorch_lightning/trainer/callback_hook.py:57: in on_epoch_start
    callback.on_epoch_start(self, self.get_model())
../../miniconda3/envs/stats507/lib/python3.7/site-packages/pytorch_lightning/callbacks/progress.py:314: in on_epoch_start
    total_val_batches = self.total_val_batches
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 

self = <pytorch_lightning.callbacks.progress.ProgressBar object at 0x7f049c112450>

    @property
    def total_val_batches(self) -> int:
        """
        The total number of training batches during validation, which may change from epoch to epoch.
        Use this to set the total number of iterations in the progress bar. Can return ``inf`` if the
        validation dataloader is of infinite size.
        """
        trainer = self.trainer
        total_val_batches = 0
        if trainer.fast_dev_run and trainer.val_dataloaders is not None:
            total_val_batches = len(trainer.val_dataloaders)
        elif not self.trainer.disable_validation:
            is_val_epoch = trainer.current_epoch % trainer.check_val_every_n_epoch == 0
            total_val_batches = trainer.num_val_batches if is_val_epoch else 0
>           total_val_batches = sum(total_val_batches)
E           TypeError: 'int' object is not iterable

../../miniconda3/envs/stats507/lib/python3.7/site-packages/pytorch_lightning/callbacks/progress.py:101: TypeError

[ismael-mendoza]

I'm running in the same error, @zzhaozhe-profolio do you know to fix this?

[zhezhaozz]

I'm running in the same error, @zzhaozhe-profolio do you know to fix this?

Ohh It's a bug within the pytorch-lightning, if the current epoch is not validation, we will have sum(0) for total_val_batches in the error message, which will throw a TypeError. Lightning-AI/pytorch-lightning#2213

I think while waiting for new version of lightning, currently we can bypass this bug by setting check_val_every_epoch to be default and it might solve the problem. @ismael2395

[ismael-mendoza]

Thanks @zzhaozhe-profolio , will try that.

[ismael-mendoza]

it works for me once I left it as default. Is there a reason we want to set check_val_every_n_epoch to 10 in the test? @zzhaozhe-profolio ? Otherwise we can just remove it and close this issue.

[zhezhaozz]

it works for me once I left it as default. Is there a reason we want to set check_val_every_n_epoch to 10 in the test? @zzhaozhe-profolio ? Otherwise we can just remove it and close this issue.

Nothing particularly important. I was strictly follow the old wake.py, which checks the encoder on the test image (set run_map = True) every 10 epochs. So it's fine we can remove it and val_dataloader and validation_step.

[ismael-mendoza]

Great the PR I'm about to merge solves this by just getting rid of it.