[rllib] Basic port of baselines/deepq to rllib #709
Merged
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Outdated suggestions cannot be applied.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.
This is a straightforward adaptation of the baselines DQN implementation to conform to the RLlib API. Files to pay attention to are
rllib/dqn/dqn.py
andrllib/dqn/example.py
; the rest were mostly copied with linter fixes only.I also fixed up the licensing here by appending the OpenAI MIT license to the top-level LICENSE file.
I have a couple ideas on how to parallelize this with Ray in a followup PR:
train_freq
to be large enough to allow sufficient parallelism between training steps. Increasingtrain_freq
will probably also require an equivalent increase ofbatch_size
.batch_size
parameter to be increased. We might also consider multiple steps of optimization over replay buffer samples, similar to policy gradient.There is also literature on parallelizing DQN in other ways but that might be out of scope for now.
On a GPU instance the Pong example spends about equal time in training and rollouts, so both could be potentially valuable.
cc @pcmoritz @royf