every_n_val_epochs -> every_n_epochs

[carmocca]

What does this PR do?

Deprecate every_n_val_epochs in favor of every_n_epochs

The flag is used in the on_validation_end hook, but we will also want to use it in the on_train_epoch_end hook

Part of #7724

Does your PR introduce any breaking changes ? If yes, please list them.

None

Before submitting

• Was this discussed/approved via a GitHub issue? (not for typos and docs)
• Did you read the contributor guideline, Pull Request section?
• Did you make sure your PR does only one thing, instead of bundling different changes together?
• Did you make sure to update the documentation with your changes? (if necessary)
• Did you write any new necessary tests? (not for typos and docs)
• Did you verify new and existing tests pass locally with your changes?
• Did you update the CHANGELOG? (not for typos, docs, test updates, or internal minor changes/refactorings)
• Did you list all the breaking changes introduced by this pull request?

PR review

• Is this pull request ready for review? (if not, please submit in draft mode)
• Check that all items from Before submitting are resolved
• Make sure the title is self-explanatory and the description concisely explains the PR
• Add labels and milestones (and optionally projects) to the PR so it can be classified

[pep8speaks]

Hello @carmocca! Thanks for updating this PR.

• In the file pytorch_lightning/callbacks/model_checkpoint.py:

Line 294:13: W503 line break before binary operator

Comment last updated at 2021-07-12 16:04:10 UTC

[codecov]

Codecov Report

Merging #8383 (8cbb71a) into master (4f1e7be) will decrease coverage by 0%.
The diff coverage is 100%.

@@          Coverage Diff           @@
##           master   #8383   +/-   ##
======================================
- Coverage      92%     92%   -0%     
======================================
  Files         216     216           
  Lines       14097   14099    +2     
======================================
- Hits        13012   12998   -14     
- Misses       1085    1101   +16     

[ananthsub]

@carmocca why is every_n_epochs along with a boolean flag to save on train epoch end preferable to supporting two flags like every_n_val_epochs and every_n_train_epochs?

The latter allows us to adjust the checkpoint frequency separately for training and validation. The mutual exclusion check is something we could address in the callback implementation.

[carmocca]

@ananthsub Is your question in the context that the user can set both save_on_train_epoch_end and save_on_validation_end? The current design only has the former which acts as an exclusive boolean already. So it is natural that the every_n_epochs flag follows the same pattern.

If there is a need to specify both, I think two ModelCheckpoint instances should be created. One with each boolean value, but not different flags.

Also this seems better to avoid user confusion about the flag difference

[ananthsub]

@carmocca it's more that as a user, users need to now know about the default behavior of save_on_train_epoch_end which is less explicit

• When reading the code, there's ambiguity around ModelCheckpoint(every_n_epochs=...). Users need to check the default value for save_on_train_epoch_end to know when exactly end of epoch checkpoints are being saved
• Same goes for writing the code: Users need to set either ModelCheckpoint(every_n_epochs=...) or ModelCheckpoint(every_n_epochs=..., save_on_train_epoch_end)which is less consistent thanModelCheckpoint(every_n_train_epochs=...)orModelCheckpoint(every_n_val_epochs=...)`

[carmocca]

there's ambiguity around ModelCheckpoint(every_n_epochs=...). Users need to check the default value for save_on_train_epoch_end to know when exactly end of epoch checkpoints are being saved

You could also make a similar argument that the user would need to know if save_on_train_epoch_end=True is set with every_n_train_epochs=... or save_on_train_epoch_end=False is set with every_n_val_epochs=...

which is less consistent thanModelCheckpoint(every_n_train_epochs=...)orModelCheckpoint(every_n_val_epochs=...)`

Are you saying every_n_train_epochs would implicitly set save_in_train_epoch_end=True and the opposite for every_n_val_epochs?