Skip to content
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.

Sign up for GitHub

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

Jump to bottom

[RLlib] Action masking example for new API stack. #46146

Merged
merged 9 commits into from
Jun 21, 2024
+329 −284
Merged

[RLlib] Action masking example for new API stack. #46146

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
9 commits
Select commit Hold shift + click to select a range
243d239
Implemented example for action masking. Added an 'unbatch(obs)' for '…
simonsays1980 Jun 19, 2024
00e1724
Included proposed changes from @sven1977's review. Added more docstri…
simonsays1980 Jun 20, 2024
2b10d68
Removed old-(hybrid)-stack action masking example b/c hybrid stack wi…
simonsays1980 Jun 20, 2024
b8f07e4
Merge branch 'master' into example-action-masking-new-stack
simonsays1980 Jun 20, 2024
14e9fa1
Removed old action masking example paths from rllib.tests.yml.
simonsays1980 Jun 20, 2024
79a8fb8
Changed naming for autoregressive actions module to conform with acti…
simonsays1980 Jun 20, 2024
d6918d5
Added the action masking example script after renaming and ran the li…
simonsays1980 Jun 20, 2024
d2045af
Fixed an arguments bug in 'ActionMaskingRLM._compute_values' and a wr…
simonsays1980 Jun 21, 2024
641688e
Merge branch 'master' into example-action-masking-new-stack
simonsays1980 Jun 21, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
3 changes: 2 additions & 1 deletion rllib/env/single_agent_env_runner.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -447,7 +447,7 @@ def _sample_episodes(
obs, infos = self.env.reset()
for env_index in range(self.num_envs):
episodes[env_index].add_env_reset(
observation=obs[env_index],
observation=unbatch(obs)[env_index],
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Strange: Shouldn't this already have caused a bug in the existing flatten obs example (which also uses a dict obs space)?

Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I think, the flatten obs example silently snuck around it: The unbatch(obs) was only forgotten in _sample_episodes and not in _sample_timesteps (has cost me a lot of time to figure it out thoug). Because the flatten obs does no evaluation - it never ran into it :D

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Ahhh, yes, makes perfect sense. Thanks for catching this!

Copy link

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

With ray==2.24.0, batch_mode="complete_episodes" will cause the bug. And obs = unbatch(obs) should outter the for-loop like what _sample_timesteps do:

        obs, infos = self.env.reset()
        obs = unbatch(obs)
        for env_index in range(self.num_envs):
            episodes[env_index].add_env_reset(
                observation=obs[env_index],

infos=infos[env_index],
)
self._make_on_episode_callback("on_episode_start", env_index, episodes)
Expand Down Expand Up @@ -724,6 +724,7 @@ def make_env(self) -> None:
asynchronous=self.config.remote_worker_envs,
)
)

self.num_envs: int = self.env.num_envs
assert self.num_envs == self.config.num_envs_per_env_runner

Expand Down
149 changes: 0 additions & 149 deletions rllib/examples/action_masking.py
View file
Open in desktop

This file was deleted.

Loading
Loading