Skip to content

Fix RecurrentRolloutBuffer not taking env_change into account for resetting states #290

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

Draft
wants to merge 3 commits into
base: master
Choose a base branch
from
Draft

Fix RecurrentRolloutBuffer not taking env_change into account for resetting states #290

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
c05c67a
Fix recurrent buffer not using env change for resetting state
araffin Mar 31, 2025
3ad432b
Fixes for newer pytorch version
araffin Mar 31, 2025
b144044
Merge branch 'master' into fix/recurrent-buffer
araffin Jun 16, 2025
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
1 change: 1 addition & 0 deletions pyproject.toml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -30,6 +30,7 @@ exclude = """(?x)(
| sb3_contrib/ars/ars.py$
| sb3_contrib/common/recurrent/policies.py$
| sb3_contrib/common/recurrent/buffers.py$
| sb3_contrib/common/torch_layers.py$
| tests/test_train_eval_mode.py$
)"""

Expand Down
13 changes: 11 additions & 2 deletions sb3_contrib/common/recurrent/buffers.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -228,6 +228,11 @@ def _get_samples(
lstm_states_pi = (self.to_torch(lstm_states_pi[0]).contiguous(), self.to_torch(lstm_states_pi[1]).contiguous())
lstm_states_vf = (self.to_torch(lstm_states_vf[0]).contiguous(), self.to_torch(lstm_states_vf[1]).contiguous())

# See issue GH#284
episode_starts = np.logical_or(self.episode_starts[batch_inds], env_change[batch_inds]).astype(
self.episode_starts.dtype
)

return RecurrentRolloutBufferSamples(
# (batch_size, obs_dim) -> (n_seq, max_length, obs_dim) -> (n_seq * max_length, obs_dim)
observations=self.pad(self.observations[batch_inds]).reshape((padded_batch_size, *self.obs_shape)),
Expand All @@ -237,7 +242,7 @@ def _get_samples(
advantages=self.pad_and_flatten(self.advantages[batch_inds]),
returns=self.pad_and_flatten(self.returns[batch_inds]),
lstm_states=RNNStates(lstm_states_pi, lstm_states_vf),
episode_starts=self.pad_and_flatten(self.episode_starts[batch_inds]),
episode_starts=self.pad_and_flatten(episode_starts),
mask=self.pad_and_flatten(np.ones_like(self.returns[batch_inds])),
)

Expand Down Expand Up @@ -372,6 +377,10 @@ def _get_samples(
observations = {key: self.pad(obs[batch_inds]) for (key, obs) in self.observations.items()}
observations = {key: obs.reshape((padded_batch_size,) + self.obs_shape[key]) for (key, obs) in observations.items()}

episode_starts = np.logical_or(self.episode_starts[batch_inds], env_change[batch_inds]).astype(
self.episode_starts.dtype
)

return RecurrentDictRolloutBufferSamples(
observations=observations,
actions=self.pad(self.actions[batch_inds]).reshape((padded_batch_size,) + self.actions.shape[1:]),
Expand All @@ -380,6 +389,6 @@ def _get_samples(
advantages=self.pad_and_flatten(self.advantages[batch_inds]),
returns=self.pad_and_flatten(self.returns[batch_inds]),
lstm_states=RNNStates(lstm_states_pi, lstm_states_vf),
episode_starts=self.pad_and_flatten(self.episode_starts[batch_inds]),
episode_starts=self.pad_and_flatten(episode_starts),
mask=self.pad_and_flatten(np.ones_like(self.returns[batch_inds])),
)