[RLlib] Optionally don't drop last ts in v-trace calculations (APPO and IMPALA). #19601
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
Many tests in rllib depended on pendulum v0, however in gym 0.21, pendulum v0 was deprecated in favor of pendulum v1. This may change reward thresholds, so will have to potentially rerun all of the pendulum v1 benchmarks, or use another environment in favor. The same applies to frozen lake v0 and frozen lake v1 Lastly, all of the RLlib tests and Tune tests have been moved to python 3.7
…ce_optional_drop_last
…ce_optional_drop_last
…ce_optional_drop_last
gjoliver
reviewed
Oct 25, 2021
| values=values_time_major[:-1], # drop-last=True | ||
| rewards=make_time_major(rewards, drop_last=drop_last), | ||
| values=values_time_major[:-1] | ||
| if drop_last else values_time_major, |
There was a problem hiding this comment.
indentation, 4 spaces in front?
There was a problem hiding this comment.
Yeah, not sure. LINTer says it's ok. You mean the if stuff, right?
There was a problem hiding this comment.
yeah, since it's a continuation of last line. no idea, minor comment.
| discounts=tf.cast( | ||
| ~make_time_major(tf.cast(dones, tf.bool), drop_last=True), | ||
| ~make_time_major( |
There was a problem hiding this comment.
~ means NOT.
make_time_major transforms a tensor of shape [B, T, ...] into [T, B, ...].
…ce_optional_drop_last # Conflicts: # .buildkite/pipeline.yml # rllib/BUILD # rllib/agents/ddpg/tests/test_ddpg.py # rllib/agents/ddpg/tests/test_td3.py # rllib/agents/dqn/tests/test_dqn.py
|
Hey @gjoliver , answered your questions and removed some unrelated functionaliy/changes. Could you take another look? Thx! :) |
gjoliver
approved these changes
Nov 3, 2021
gjoliver
approved these changes
Nov 3, 2021
| values=values_time_major[:-1], # drop-last=True | ||
| rewards=make_time_major(rewards, drop_last=drop_last), | ||
| values=values_time_major[:-1] | ||
| if drop_last else values_time_major, |
There was a problem hiding this comment.
yeah, since it's a continuation of last line. no idea, minor comment.
…ce_optional_drop_last
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.
Add an option to APPO/IMPALA config to not drop the last ts in vtrace calculations.
vtrace_drop_last_ts(default True).First experiments with sparse reward and reward-at-end environments indicate that not dropping the last ts adds stability to the learning behavior.
Why are these changes needed?
Related issue number
Checks
scripts/format.shto lint the changes in this PR.