Support val_check_interval values higher than number of training batches #11993
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
nikvaessen
requested review from
williamFalcon,
tchaton,
awaelchli,
SeanNaren,
carmocca,
Borda,
justusschock,
kaushikb11 and
rohitgr7
as code owners
|
It seems that the only tests which are failing are due to |
awaelchli
added
feature
for more information, see https://pre-commit.ci
…erval # Conflicts: # CHANGELOG.md # pytorch_lightning/callbacks/progress/base.py # pytorch_lightning/loops/epoch/training_epoch_loop.py # pytorch_lightning/trainer/connectors/data_connector.py # pytorch_lightning/trainer/trainer.py
…_val_check_interval # Conflicts: # pytorch_lightning/trainer/trainer.py
|
I've updated the PR so that merge conflicts with changes from the 1.6 release are resolved. Hopefully, the PR won't go stale again :) |
|
I've resolved your comments, and also added a test case for an iterable dataset without a defined length. |
use more sensible value to approx inf
rohitgr7
reviewed
Apr 20, 2022
rohitgr7
approved these changes
Apr 20, 2022
awaelchli
approved these changes
Apr 20, 2022
Merged
12 tasks
This was referenced Apr 21, 2022
eladsegal
added a commit
to eladsegal/pytorch-lightning
that referenced
this pull request
Jun 10, 2022
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
What does this PR do?
Fixes #8135
Fixes #4406
Fixes #5413.
It enables using
val_check_intervalwith a value higher than the amount of batches in one training data loader iteration. This is very useful when you use a cyclic learning rate, and want to validate on the extreme values of a cycle, without having to worry about the amount of samples in your dataset.I chose to implement the following option as provided by @carmocca in #8135 (comment)
Does your PR introduce any breaking changes? If yes, please list them.
I'm not sure as I'm not very familiar with the code base. Some people might rely on the fact that
val_check_intervaldoes not cause a validation loop to happen when the value is higher than the amount of training batches, but I doubt it as I don't see a reason for intentionally skipping a validation epoch.Before submitting
PR review
Anyone in the community is welcome to review the PR.
Before you start reviewing make sure you have read Review guidelines. In short, see the following bullet-list:
Did you have fun?
Make sure you had fun coding 🙃