Munchausen RL #1221
Open
Munchausen RL #1221
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
Haichao-Zhang
force-pushed
the
Munchausen_RL
branch
from
749e344
to
28f1cef
Compare
|
Are there any experiment results? |
hnyu
reviewed
Mar 9, 2022
When submitting the PR, only tested on pendulum for sanity check. On pendulum, no clear difference compared to SAC. |
hnyu
reviewed
Mar 9, 2022
Haichao-Zhang
force-pushed
the
Munchausen_RL
branch
2 times, most recently
from
fd151e6
to
4866283
Compare
Haichao-Zhang
force-pushed
the
Munchausen_RL
branch
from
4866283
to
b330630
Compare
hnyu
requested changes
Mar 10, 2022
| @@ -267,6 +273,11 @@ def __init__(self, | |||
| critic_network_cls, q_network_cls) | |||
|
|
|||
| self._use_entropy_reward = use_entropy_reward | |||
| self._munchausen_reward_weight = max(0, munchausen_reward_weight) | |||
| if munchausen_reward_weight > 0: | |||
| assert not normalize_entropy_reward, ( | |||
There was a problem hiding this comment.
normalize_entropy_reward has no conflict with munchausen_reward_weight > 0?
hnyu
reviewed
Mar 10, 2022
| (sum(nest.flatten(log_pi_rollout_a[0])), | ||
| sum(nest.flatten(log_pi_rollout_a[1])))) | ||
| else: | ||
| log_pi_rollout_a = sum(nest.flatten(log_pi_rollout_a)) |
There was a problem hiding this comment.
Why do you want to first add log_pi here? (Unless for each log_alpha you have multiple log_pis)
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Outdated suggestions cannot be applied.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.
Implementation of the Munchausen RL.
Tested on Breakout. The reward-weight is an import hyper-parameter for tuning.
Currently, it seems that with a proper weight (e.g. 0.5), there are some gain initially, but eventually similar to SAC.