-
Notifications
You must be signed in to change notification settings - Fork 5.8k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
[RLlib] Remove execution plan #25624
Conversation
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
awesome! my favorite kind of PR.
thanks a ton man.
max_remote_requests_in_flight_per_worker=1, | ||
ray_wait_timeout_s=0.03, | ||
) | ||
self._curr_learner_info = {} |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
can you help comment here that ddppo only has torch implementation, so it's ok to do torch specific work here?
rllib/execution/replay_ops.py
Outdated
@@ -1,136 +1,10 @@ | |||
from typing import List, Any, Optional | |||
from typing import List, Optional |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
we should be able to delete this whole file?
we should be using Arthur's RBs now right?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
We're still using the MixInMultiAgentReplayBuffer
for IMPALA, until Artur refactors that out
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Let me get rid of it..
rllib/execution/train_ops.py
Outdated
@@ -511,21 +511,10 @@ def __call__( | |||
class UpdateTargetNetwork: |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
writing the comment here, but it's not related to UpdateTargetNetwork.
can we delete the TrainOneStep and MultiGPUTrainOneStep classes now??
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
yeah I hadn't fully deleted these yet at the time that I had uploaded the PR, but they should be deleted now :)
lint and 1 unit test failure. |
ah forgot to remove a test from BUILD that I deleted |
ready to merge pending tests passing |
lint / doc error doesn't seem related. |
Looks great. Thanks for cleaning this up @avnishn!
Maybe we need to add these |
c8cbf77
to
a32f090
Compare
@jiaodong it seems this PR might help your rllib prototyping, since it removes a lot of complexity in the code. |
@suquark thanks for ccing me here :) 2000 less lines of code sounds great indeed |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Great stuff. I checked out the PR and found 1-2 minor occurences of "execution_plan" and changed them myself. When are we going to do MBMPO and MAML? Anything keeping us from writing training iteration functions on these?
Yeah, we still need to complete MB-MPO and MAML. We left them out for no specific reasons. |
Let's put them on my next sprint, please. |
This PR removed rllib/execution/concurrency_ops.py and rllib/execution/multi_gpu_impl.py which are referred to in the documentation, so the book-lint CI run is failing. here is the documentation source with the file names. |
Remove the final unused execution plans and now unused execution plan utilities
Why are these changes needed?
Related issue number
Checks
scripts/format.sh
to lint the changes in this PR.