Skip to content

[BugFix] Fix PPO clip #786

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

Merged
merged 6 commits into from
Jan 3, 2023
Merged

[BugFix] Fix PPO clip #786

Changes from all commits
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
00424b7
init (#779)
vmoens Jan 1, 2023
c2482d4
init
vmoens Jan 3, 2023
8711ed0
amend
vmoens Jan 3, 2023
dac9435
amend
vmoens Jan 3, 2023
ffb41b8
Merge branch 'main' into fix_ppo
vmoens Jan 3, 2023
cca4841
amend
vmoens Jan 3, 2023
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
7 changes: 2 additions & 5 deletions torchrl/objectives/ppo.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -233,13 +233,10 @@ def forward(self, tensordict: TensorDictBase) -> TensorDictBase:
f"and {log_weight.shape})"
)
gain1 = log_weight.exp() * advantage
log_weight_clip = torch.empty_like(log_weight)
# log_weight_clip.data.clamp_(*self._clip_bounds)
idx_pos = advantage >= 0
log_weight_clip[idx_pos] = log_weight[idx_pos].clamp_max(self._clip_bounds[1])
log_weight_clip[~idx_pos] = log_weight[~idx_pos].clamp_min(self._clip_bounds[0])

log_weight_clip = log_weight.clamp(*self._clip_bounds)
gain2 = log_weight_clip.exp() * advantage

gain = torch.stack([gain1, gain2], -1).min(dim=-1)[0]
td_out = TensorDict({"loss_objective": -gain.mean()}, [])

Expand Down