Fix CSVLogger hyperparameter is logged at every write which increase latency significantly.

[duydl]

What does this PR do?

In CSVLogger, the save method of experiment ExperimentWriter is override such that hyperparameters are saved at every write, and log_hyperparams only update hparam, which is not efficient. The training run become very slow when there is a significantly big object in hparams and regular logging (like every hundred steps). So I believe this change is necessary.

Before submitting

• Was this discussed/agreed 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 list all the breaking changes introduced by this pull request?
• Did you update the CHANGELOG? (not for typos, docs, test updates, or minor internal changes/refactors)

PR review

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

Reviewer checklist

• 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

---

📚 Documentation preview 📚: https://pytorch-lightning--20594.org.readthedocs.build/en/20594/

[duydl]

I search the issues and believe this would fix #19240.

[ethanwharris]

Thanks @duydl! Could you add a test for this please? Let us know if you need a hand 🙂

[duydl]

@ethanwharris Hi, for the test there is only one line needed adjustment I added in the newest commit. Now to save hyperparams with CSVLogger we no longer need to call logger.save, just logger.log_hyperparams() is enough. It would be more consistent with other loggers.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 79%. Comparing base (b313cd9) to head (dd828c4).
Report is 112 commits behind head on master.

❗ There is a different number of reports uploaded between BASE (b313cd9) and HEAD (dd828c4). Click for more details.

HEAD has 1821 uploads less than BASE

Flag	BASE (b313cd9)	HEAD (dd828c4)
cpu	430	24
lightning	324	18
pytest	217	0
python3.9	108	6
lightning_fabric	54	0
python3.11	108	6
python3.10	54	3
python3.12.7	160	9
gpu	2	0
pytorch2.1	81	9
pytest-full	215	24
pytorch2.2.2	27	3
pytorch_lightning	54	6
pytorch2.3	27	3
pytorch2.5.1	54	6
pytorch2.4.1	26	3

Additional details and impacted files

@@            Coverage Diff            @@
##           master   #20594     +/-   ##
=========================================
- Coverage      88%      79%     -9%     
=========================================
  Files         267      264      -3     
  Lines       23369    23311     -58     
=========================================
- Hits        20478    18364   -2114     
- Misses       2891     4947   +2056     

[duydl]

@ethanwharris Hi, do you think this is ready to be merged?
Could you also look at #20593 to see if the change is welcome or if there is necessary adjustment?
Thank you.