[Tune; RLlib] Missing stopping criterion should not error (just warn). #45613
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
sven1977
requested review from
hongpeng-guo,
justinvyu,
matthewdeng,
raulchen and
woshiyyya
as code owners
sven1977
changed the title
|
Hey @justinvyu , could you take a look at this PR? |
justinvyu
reviewed
May 31, 2024
There was a problem hiding this comment.
Thanks! I think this generally makes sense.
Could we just add a small unit test here, asserting that the warning is logged?
https://github.com/ray-project/ray/blob/master/python/ray/tune/tests/test_trial.py#L5
python/ray/tune/experiment/trial.py
Outdated
| elif criterion not in result: | ||
| if log_once("tune_trial_stop_criterion_not_found"): | ||
| logger.warning( | ||
| f"Stopping criterion {criterion} not found in result dict. " |
There was a problem hiding this comment.
Add a quick note about the behavior that will follow:
"If this metric is never reported, the run will continue running until training is finished instead."
…tune_stop_criterium_not_found_erro
…tune_stop_criterium_not_found_erro
Signed-off-by: sven1977 <svenmika1977@gmail.com>
|
Hey @justinvyu , thanks for the review! Could you take another look? Added the test cases and enhanced the warning message. Thanks! :) |
justinvyu
reviewed
Jun 6, 2024
python/ray/tune/tests/test_trial.py
Outdated
Comment on lines
130
to
133
| result = _TrainingResult( | ||
| checkpoint=None, metrics={"a": 9.999, "b/c": 0.0, "some_other_key": True} | ||
| ) | ||
| assert not trial.should_stop(result.metrics) |
There was a problem hiding this comment.
Can we just pass in a dict without the _TrainingResult wrapper?
…tune_stop_criterium_not_found_erro
…tune_stop_criterium_not_found_erro
…tune_stop_criterium_not_found_erro
Signed-off-by: sven1977 <svenmika1977@gmail.com>
Signed-off-by: sven1977 <svenmika1977@gmail.com>
richardsliu
pushed a commit
to richardsliu/ray
that referenced
this pull request
Jun 12, 2024
ray-project#45613) Signed-off-by: Richard Liu <ricliu@google.com>
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.
Missing stopping criterion should not error (just warn).
Some RLlib algorithms take a few iterations in order to sample enough data from their EnvRunners to complete an RL episode. Only after that point in time, episode stats will be published in the result dicts returned by
Algorithm.train()(Algorithmis-atune.Trainable).The current hacky fix is to prophylactically add common episode stats (such as
episode_return_mean) to the initialized algorithm's result dict as NaNs, however, this is not a sustailable solution as users might add their own stats and will then run into theTuneError(Stopping criteria ... not provided)error without any idea how to fix this. A one-time warning here would be the better solution.Why are these changes needed?
Related issue number
Checks
git commit -s) in this PR.scripts/format.shto lint the changes in this PR.method in Tune, I've added it in
doc/source/tune/api/under thecorresponding
.rstfile.