[RLlib] Moving sampling coordination for batch_mode=complete_episodes to synchronous_parallel_sample.
#46321
+10
−15
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
…ot reducing workload when scaled and b) was using 'train_batch_size' neglecting 'train_batch_size_per_learner'. Signed-off-by: simonsays1980 <[email protected]>
simonsays1980
added
bug
sven1977
changed the title
synchronous_parallel_sample.synchronous_parallel_sample.
sven1977
reviewed
Jun 28, 2024
| # For complete episodes mode, sample as long as the number of timesteps | ||
| # done is smaller than the `train_batch_size`. | ||
| # For complete episodes mode, sample a single episode and | ||
| # leave coordination of sampling to `synchronous_parallel_sample`. |
There was a problem hiding this comment.
I very much like this!
Can we add a small TODO comment here that this logic, currently handled by synchronous_parallel_sample will eventually be moved fully into EnvRunnerGroup? So from the algo, you would do:
if self.config.batch_mode == "complete_episodes"
self.env_runner_group.sample(num_timesteps=[batch size], complete_episodes=True)
something like this ^. Don't have to do this in this PR!
There was a problem hiding this comment.
Awesome. I would love this move!
sven1977
approved these changes
Jun 28, 2024
There was a problem hiding this comment.
Nice elegant cleanup PR. Thanks @simonsays1980 !
Just one comment line to be added.
Signed-off-by: simonsays1980 <[email protected]>
simonsays1980
changed the title
synchronous_parallel_sample.batch_mode=complete_episodes to synchronous_parallel_sample.
…th 'complete_episodes' sampling happens multiple times until the number of timesteps for the 'train_batch_size' is reached. Signed-off-by: simonsays1980 <[email protected]>
Signed-off-by: simonsays1980 <[email protected]>
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.
Why are these changes needed
When sampling complete episodes each
EnvRunnersampledtrain_batch_sizebefore returning. This made sampling inefficient and led to long waiting times in case slow environments were used. Furthermore, scaling could not reduce the workload in sampling. This PR changes this and moves coordination of the sampling whencomplete_episodesare needed fully tosynchronous_parallel_samplethat can coordinate better across allEnvRunners. This should reduce sampling duration linearly by the number ofEnvRunners chosen.Related issue number
Closes #45826
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.