Pytorch-lightning: ModelCheckpoint epoch and progress bar epoch mismatch

[versatran01]

🐛 Bug

Progress bar epoch starts at 1, but in ModelCheckpoint it starts at 0.

To Reproduce

Steps to reproduce the behavior:

https://github.com/PyTorchLightning/pytorch-lightning/blob/8ca8336ce52ee7379f4d399520636143eb31018b/pytorch_lightning/callbacks/progress.py#L320

https://github.com/PyTorchLightning/pytorch-lightning/blob/8ca8336ce52ee7379f4d399520636143eb31018b/pytorch_lightning/callbacks/model_checkpoint.py#L212

Expected behavior

Shouldn't they be consistent?

[williamFalcon]

@Borda FYI.
in 0.8.0 we are setting all indexing to be from 0

[versatran01]

what is the epoch number of the val sanity check? I don't want to write the results of sanity check to my logs, is there a way to check if validation is currently in sanity check?

[versatran01]

This does not seem to be true in 0.8.0. Epoch still starts from 1.
https://github.com/PyTorchLightning/pytorch-lightning/blob/3256fe4e5a405db1ab00d4cf4d48cbbfc7730959/pytorch_lightning/trainer/training_loop.py#L350
This is even worse than before because now all my validation metrics in tensorboard are shifted by 1.

[williamFalcon]

yes just noticed that too... they should start from 0 @Borda

[Borda]

yes just noticed that too... they should start from 0 @Borda

ohhh that was my misunderstanding, I thought it shall start from 1 and because of the confusion I was asking about the diff indexing in steps and epochs https://github.com/PyTorchLightning/pytorch-lightning/pull/2206#issuecomment-644568854

well before it was indexed from 0...

Ok, reverting and moving epoch indexing starting from 0

[stale[bot]]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.