[RLlib] Mixin Layer Design Sketch Up #53850
Draft
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Outdated suggestions cannot be applied.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?
This PR contains a proposal for a mixin layer design for RLlib's
Algorithm
class and shows some simple examples of how this design could be implementedrllib/algorithms/rl_algorithm.RLAlgorithm
is the base algorithm class derived fromTrainable
.rllib/algorithms/ppo/ppo_algorithm.PPOAlgorithm
derives from the base class and uses theSyncEnvRunnerConcreteMixin
to include synchronized online sampling.rllib/algorithms/rl_algorithm_mixins/online_mixins.SyncEnvRunnerConcreteMixin
is derived from an abstractEnvRunnerMixin
and includes the definition of an (inner mixin layer)SyncEnvRunnerGroup
that samples only synchronous.Related issue number
Checks
git commit -s
) in this PR.scripts/format.sh
to lint the changes in this PR.method in Tune, I've added it in
doc/source/tune/api/
under thecorresponding
.rst
file.