[2 / 3] improvements to saving and loading callback state

[awaelchli]

What does this PR do?

Fixes #6467

It is now possible to save and resume with multiple callbacks of the same time, provided they have a suitable state_id property that defines the equivalence class. See the added docs section for more information of how it works.

This is now possible without sacrifice:

Trainer(
    callbacks=[ModelCheckpoint(monitor="val_loss"), ModelCheckpoint(monitor="other", mode="max")],
)

and resuming from an existing checkpoint:

Trainer(
    resume_from_checkpoint"path/to/ckpt",
    callbacks=[ModelCheckpoint(monitor="val_loss"), ModelCheckpoint(monitor="other", mode="max")],
)

will load the internal state of BOTH these callbacks, where previously this wouldn't work with stateful callbacks.

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)

PR review

Anyone in the community is free to review the PR once the tests have passed.
Before you start reviewing make sure you have read Review guidelines. In short, see the following bullet-list:

• 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

Did you have fun?

Make sure you had fun coding 🙃

[pep8speaks]

Hello @awaelchli! Thanks for updating this PR. We checked the lines you've touched for PEP 8 issues, and found:

• In the file tests/callbacks/test_early_stopping.py:

Line 87:9: W503 line break before binary operator

• In the file tests/checkpointing/test_model_checkpoint.py:

Line 50:9: W503 line break before binary operator

Comment last updated at 2021-08-13 11:48:25 UTC

[codecov]

Codecov Report

Merging #7187 (184f412) into master (1a2468f) will decrease coverage by 4%.
The diff coverage is 100%.

@@           Coverage Diff           @@
##           master   #7187    +/-   ##
=======================================
- Coverage      92%     88%    -4%     
=======================================
  Files         178     178            
  Lines       14694   14702     +8     
=======================================
- Hits        13526   12925   -601     
- Misses       1168    1777   +609     

[ananthsub]

for a followup, we should use the state_key to disambiguate callbacks here too: https://github.com/PyTorchLightning/pytorch-lightning/blob/f87b2ef21f01d7bbda666088b8cdae3dcbe12010/pytorch_lightning/trainer/connectors/callback_connector.py#L134-L159

[tchaton]

LGTM !