Skip to content

Fix #508: Use per-env logging_step for episodic return logging #539

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

Open
gspeter-max wants to merge 11 commits into
base: master
Choose a base branch
from
Open

Fix #508: Use per-env logging_step for episodic return logging #539

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
11 commits
Select commit Hold shift + click to select a range
1e1d948
test(logging): add tests proving episodic return logging bug #508
peter-luminova Mar 1, 2026
a689592
fix(ppo): use per-env logging_step for episodic return logging #508
peter-luminova Mar 1, 2026
d05890b
test(logging): add tests proving episodic return logging bug #508
peter-luminova Mar 1, 2026
bf5c638
fix(ppo_procgen): remove break and use per-env logging_step #508
peter-luminova Mar 1, 2026
49e4059
fix(ppo_atari_lstm): use per-env logging_step #508
peter-luminova Mar 1, 2026
26da6c5
fix(ppo_continuous_action): use per-env logging_step #508
peter-luminova Mar 1, 2026
81d65c8
test(logging): add mock-based tests proving episodic return logging b…
peter-luminova Mar 1, 2026
58c6eb1
fix(jax): update jaxlib from 0.4.7 to 0.4.8
peter-luminova Mar 2, 2026
36fc0d1
fix(jax): remove jaxlib version pin to fix ARM64 compatibility
peter-luminova Mar 2, 2026
541ecb7
fix(ci): drop Python 3.8 support for JAX compatibility
peter-luminova Mar 2, 2026
202509d
fix(issue_508): revert JAX changes from PR #539
peter-luminova Mar 2, 2026
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: 4 additions & 3 deletions cleanrl/c51.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -187,11 +187,12 @@ def linear_schedule(start_e: float, end_e: float, duration: int, t: int):

# TRY NOT TO MODIFY: record rewards for plotting purposes
if "final_info" in infos:
for info in infos["final_info"]:
for i, info in enumerate(infos["final_info"]):
if info and "episode" in info:
logging_step = global_step - args.num_envs + i
print(f"global_step={global_step}, episodic_return={info['episode']['r']}")
writer.add_scalar("charts/episodic_return", info["episode"]["r"], global_step)
writer.add_scalar("charts/episodic_length", info["episode"]["l"], global_step)
writer.add_scalar("charts/episodic_return", info["episode"]["r"], logging_step)
writer.add_scalar("charts/episodic_length", info["episode"]["l"], logging_step)

# TRY NOT TO MODIFY: save data to reply buffer; handle `final_observation`
real_next_obs = next_obs.copy()
Expand Down
7 changes: 4 additions & 3 deletions cleanrl/c51_atari.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -210,11 +210,12 @@ def linear_schedule(start_e: float, end_e: float, duration: int, t: int):

# TRY NOT TO MODIFY: record rewards for plotting purposes
if "final_info" in infos:
for info in infos["final_info"]:
for i, info in enumerate(infos["final_info"]):
if info and "episode" in info:
logging_step = global_step - args.num_envs + i
print(f"global_step={global_step}, episodic_return={info['episode']['r']}")
writer.add_scalar("charts/episodic_return", info["episode"]["r"], global_step)
writer.add_scalar("charts/episodic_length", info["episode"]["l"], global_step)
writer.add_scalar("charts/episodic_return", info["episode"]["r"], logging_step)
writer.add_scalar("charts/episodic_length", info["episode"]["l"], logging_step)

# TRY NOT TO MODIFY: save data to reply buffer; handle `final_observation`
real_next_obs = next_obs.copy()
Expand Down
7 changes: 4 additions & 3 deletions cleanrl/c51_atari_jax.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -269,11 +269,12 @@ def get_action(q_state, obs):

# TRY NOT TO MODIFY: record rewards for plotting purposes
if "final_info" in infos:
for info in infos["final_info"]:
for i, info in enumerate(infos["final_info"]):
if info and "episode" in info:
logging_step = global_step - args.num_envs + i
print(f"global_step={global_step}, episodic_return={info['episode']['r']}")
writer.add_scalar("charts/episodic_return", info["episode"]["r"], global_step)
writer.add_scalar("charts/episodic_length", info["episode"]["l"], global_step)
writer.add_scalar("charts/episodic_return", info["episode"]["r"], logging_step)
writer.add_scalar("charts/episodic_length", info["episode"]["l"], logging_step)

# TRY NOT TO MODIFY: save data to reply buffer; handle `final_observation`
real_next_obs = next_obs.copy()
Expand Down
7 changes: 4 additions & 3 deletions cleanrl/c51_jax.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -233,11 +233,12 @@ def loss(q_params, observations, actions, target_pmfs):

# TRY NOT TO MODIFY: record rewards for plotting purposes
if "final_info" in infos:
for info in infos["final_info"]:
for i, info in enumerate(infos["final_info"]):
if info and "episode" in info:
logging_step = global_step - args.num_envs + i
print(f"global_step={global_step}, episodic_return={info['episode']['r']}")
writer.add_scalar("charts/episodic_return", info["episode"]["r"], global_step)
writer.add_scalar("charts/episodic_length", info["episode"]["l"], global_step)
writer.add_scalar("charts/episodic_return", info["episode"]["r"], logging_step)
writer.add_scalar("charts/episodic_length", info["episode"]["l"], logging_step)

# TRY NOT TO MODIFY: save data to reply buffer; handle `final_observation`
real_next_obs = next_obs.copy()
Expand Down
8 changes: 4 additions & 4 deletions cleanrl/ddpg_continuous_action.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -184,11 +184,11 @@ def forward(self, x):

# TRY NOT TO MODIFY: record rewards for plotting purposes
if "final_info" in infos:
for info in infos["final_info"]:
for i, info in enumerate(infos["final_info"]):
logging_step = global_step - args.num_envs + i
print(f"global_step={global_step}, episodic_return={info['episode']['r']}")
writer.add_scalar("charts/episodic_return", info["episode"]["r"], global_step)
writer.add_scalar("charts/episodic_length", info["episode"]["l"], global_step)
break
writer.add_scalar("charts/episodic_return", info["episode"]["r"], logging_step)
writer.add_scalar("charts/episodic_length", info["episode"]["l"], logging_step)

# TRY NOT TO MODIFY: save data to reply buffer; handle `final_observation`
real_next_obs = next_obs.copy()
Expand Down
7 changes: 4 additions & 3 deletions cleanrl/ddpg_continuous_action_jax.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -238,10 +238,11 @@ def actor_loss(params):

# TRY NOT TO MODIFY: record rewards for plotting purposes
if "final_info" in infos:
for info in infos["final_info"]:
for i, info in enumerate(infos["final_info"]):
logging_step = global_step - args.num_envs + i
print(f"global_step={global_step}, episodic_return={info['episode']['r']}")
writer.add_scalar("charts/episodic_return", info["episode"]["r"], global_step)
writer.add_scalar("charts/episodic_length", info["episode"]["l"], global_step)
writer.add_scalar("charts/episodic_return", info["episode"]["r"], logging_step)
writer.add_scalar("charts/episodic_length", info["episode"]["l"], logging_step)
break

# TRY NOT TO MODIFY: save data to replay buffer; handle `final_observation`
Expand Down
7 changes: 4 additions & 3 deletions cleanrl/dqn.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -174,11 +174,12 @@ def linear_schedule(start_e: float, end_e: float, duration: int, t: int):

# TRY NOT TO MODIFY: record rewards for plotting purposes
if "final_info" in infos:
for info in infos["final_info"]:
for i, info in enumerate(infos["final_info"]):
if info and "episode" in info:
logging_step = global_step - args.num_envs + i
print(f"global_step={global_step}, episodic_return={info['episode']['r']}")
writer.add_scalar("charts/episodic_return", info["episode"]["r"], global_step)
writer.add_scalar("charts/episodic_length", info["episode"]["l"], global_step)
writer.add_scalar("charts/episodic_return", info["episode"]["r"], logging_step)
writer.add_scalar("charts/episodic_length", info["episode"]["l"], logging_step)

# TRY NOT TO MODIFY: save data to reply buffer; handle `final_observation`
real_next_obs = next_obs.copy()
Expand Down
7 changes: 4 additions & 3 deletions cleanrl/dqn_atari.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -197,11 +197,12 @@ def linear_schedule(start_e: float, end_e: float, duration: int, t: int):

# TRY NOT TO MODIFY: record rewards for plotting purposes
if "final_info" in infos:
for info in infos["final_info"]:
for i, info in enumerate(infos["final_info"]):
if info and "episode" in info:
logging_step = global_step - args.num_envs + i
print(f"global_step={global_step}, episodic_return={info['episode']['r']}")
writer.add_scalar("charts/episodic_return", info["episode"]["r"], global_step)
writer.add_scalar("charts/episodic_length", info["episode"]["l"], global_step)
writer.add_scalar("charts/episodic_return", info["episode"]["r"], logging_step)
writer.add_scalar("charts/episodic_length", info["episode"]["l"], logging_step)

# TRY NOT TO MODIFY: save data to reply buffer; handle `final_observation`
real_next_obs = next_obs.copy()
Expand Down
7 changes: 4 additions & 3 deletions cleanrl/dqn_atari_jax.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -227,11 +227,12 @@ def mse_loss(params):

# TRY NOT TO MODIFY: record rewards for plotting purposes
if "final_info" in infos:
for info in infos["final_info"]:
for i, info in enumerate(infos["final_info"]):
if info and "episode" in info:
logging_step = global_step - args.num_envs + i
print(f"global_step={global_step}, episodic_return={info['episode']['r']}")
writer.add_scalar("charts/episodic_return", info["episode"]["r"], global_step)
writer.add_scalar("charts/episodic_length", info["episode"]["l"], global_step)
writer.add_scalar("charts/episodic_return", info["episode"]["r"], logging_step)
writer.add_scalar("charts/episodic_length", info["episode"]["l"], logging_step)

# TRY NOT TO MODIFY: save data to reply buffer; handle `final_observation`
real_next_obs = next_obs.copy()
Expand Down
7 changes: 4 additions & 3 deletions cleanrl/dqn_jax.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -197,11 +197,12 @@ def mse_loss(params):

# TRY NOT TO MODIFY: record rewards for plotting purposes
if "final_info" in infos:
for info in infos["final_info"]:
for i, info in enumerate(infos["final_info"]):
if info and "episode" in info:
logging_step = global_step - args.num_envs + i
print(f"global_step={global_step}, episodic_return={info['episode']['r']}")
writer.add_scalar("charts/episodic_return", info["episode"]["r"], global_step)
writer.add_scalar("charts/episodic_length", info["episode"]["l"], global_step)
writer.add_scalar("charts/episodic_return", info["episode"]["r"], logging_step)
writer.add_scalar("charts/episodic_length", info["episode"]["l"], logging_step)

# TRY NOT TO MODIFY: save data to reply buffer; handle `final_observation`
real_next_obs = next_obs.copy()
Expand Down
8 changes: 4 additions & 4 deletions cleanrl/ppg_procgen.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -309,12 +309,12 @@ def get_pi(self, x):
rewards[step] = torch.tensor(reward).to(device).view(-1)
next_obs, next_done = torch.Tensor(next_obs).to(device), torch.Tensor(done).to(device)

for item in info:
for i, item in enumerate(info):
if "episode" in item.keys():
logging_step = global_step - args.num_envs + i
print(f"global_step={global_step}, episodic_return={item['episode']['r']}")
writer.add_scalar("charts/episodic_return", item["episode"]["r"], global_step)
writer.add_scalar("charts/episodic_length", item["episode"]["l"], global_step)
break
writer.add_scalar("charts/episodic_return", item["episode"]["r"], logging_step)
writer.add_scalar("charts/episodic_length", item["episode"]["l"], logging_step)

# bootstrap value if not done
with torch.no_grad():
Expand Down
7 changes: 4 additions & 3 deletions cleanrl/ppo.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -208,11 +208,12 @@ def get_action_and_value(self, x, action=None):
next_obs, next_done = torch.Tensor(next_obs).to(device), torch.Tensor(next_done).to(device)

if "final_info" in infos:
for info in infos["final_info"]:
for i, info in enumerate(infos["final_info"]):
if info and "episode" in info:
logging_step = global_step - args.num_envs + i
print(f"global_step={global_step}, episodic_return={info['episode']['r']}")
writer.add_scalar("charts/episodic_return", info["episode"]["r"], global_step)
writer.add_scalar("charts/episodic_length", info["episode"]["l"], global_step)
writer.add_scalar("charts/episodic_return", info["episode"]["r"], logging_step)
writer.add_scalar("charts/episodic_length", info["episode"]["l"], logging_step)

# bootstrap value if not done
with torch.no_grad():
Expand Down
7 changes: 4 additions & 3 deletions cleanrl/ppo_atari.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -225,11 +225,12 @@ def get_action_and_value(self, x, action=None):
next_obs, next_done = torch.Tensor(next_obs).to(device), torch.Tensor(next_done).to(device)

if "final_info" in infos:
for info in infos["final_info"]:
for i, info in enumerate(infos["final_info"]):
if info and "episode" in info:
logging_step = global_step - args.num_envs + i
print(f"global_step={global_step}, episodic_return={info['episode']['r']}")
writer.add_scalar("charts/episodic_return", info["episode"]["r"], global_step)
writer.add_scalar("charts/episodic_length", info["episode"]["l"], global_step)
writer.add_scalar("charts/episodic_return", info["episode"]["r"], logging_step)
writer.add_scalar("charts/episodic_length", info["episode"]["l"], logging_step)

# bootstrap value if not done
with torch.no_grad():
Expand Down
7 changes: 4 additions & 3 deletions cleanrl/ppo_atari_envpool.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -240,11 +240,12 @@ def get_action_and_value(self, x, action=None):

for idx, d in enumerate(next_done):
if d and info["lives"][idx] == 0:
logging_step = global_step - args.num_envs + idx
print(f"global_step={global_step}, episodic_return={info['r'][idx]}")
avg_returns.append(info["r"][idx])
writer.add_scalar("charts/avg_episodic_return", np.average(avg_returns), global_step)
writer.add_scalar("charts/episodic_return", info["r"][idx], global_step)
writer.add_scalar("charts/episodic_length", info["l"][idx], global_step)
writer.add_scalar("charts/avg_episodic_return", np.average(avg_returns), logging_step)
writer.add_scalar("charts/episodic_return", info["r"][idx], logging_step)
writer.add_scalar("charts/episodic_length", info["l"][idx], logging_step)

# bootstrap value if not done
with torch.no_grad():
Expand Down
7 changes: 4 additions & 3 deletions cleanrl/ppo_atari_lstm.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -257,11 +257,12 @@ def get_action_and_value(self, x, lstm_state, done, action=None):
next_obs, next_done = torch.Tensor(next_obs).to(device), torch.Tensor(next_done).to(device)

if "final_info" in infos:
for info in infos["final_info"]:
for i, info in enumerate(infos["final_info"]):
if info and "episode" in info:
logging_step = global_step - args.num_envs + i
print(f"global_step={global_step}, episodic_return={info['episode']['r']}")
writer.add_scalar("charts/episodic_return", info["episode"]["r"], global_step)
writer.add_scalar("charts/episodic_length", info["episode"]["l"], global_step)
writer.add_scalar("charts/episodic_return", info["episode"]["r"], logging_step)
writer.add_scalar("charts/episodic_length", info["episode"]["l"], logging_step)

# bootstrap value if not done
with torch.no_grad():
Expand Down
7 changes: 4 additions & 3 deletions cleanrl/ppo_atari_multigpu.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -275,11 +275,12 @@ def get_action_and_value(self, x, action=None):
continue

if "final_info" in infos:
for info in infos["final_info"]:
for i, info in enumerate(infos["final_info"]):
if info and "episode" in info:
logging_step = global_step - args.num_envs + i
print(f"global_step={global_step}, episodic_return={info['episode']['r']}")
writer.add_scalar("charts/episodic_return", info["episode"]["r"], global_step)
writer.add_scalar("charts/episodic_length", info["episode"]["l"], global_step)
writer.add_scalar("charts/episodic_return", info["episode"]["r"], logging_step)
writer.add_scalar("charts/episodic_length", info["episode"]["l"], logging_step)

print(
f"local_rank: {local_rank}, action.sum(): {action.sum()}, iteration: {iteration}, agent.actor.weight.sum(): {agent.actor.weight.sum()}"
Expand Down
7 changes: 4 additions & 3 deletions cleanrl/ppo_continuous_action.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -223,11 +223,12 @@ def get_action_and_value(self, x, action=None):
next_obs, next_done = torch.Tensor(next_obs).to(device), torch.Tensor(next_done).to(device)

if "final_info" in infos:
for info in infos["final_info"]:
for i, info in enumerate(infos["final_info"]):
if info and "episode" in info:
logging_step = global_step - args.num_envs + i
print(f"global_step={global_step}, episodic_return={info['episode']['r']}")
writer.add_scalar("charts/episodic_return", info["episode"]["r"], global_step)
writer.add_scalar("charts/episodic_length", info["episode"]["l"], global_step)
writer.add_scalar("charts/episodic_return", info["episode"]["r"], logging_step)
writer.add_scalar("charts/episodic_length", info["episode"]["l"], logging_step)

# bootstrap value if not done
with torch.no_grad():
Expand Down
8 changes: 4 additions & 4 deletions cleanrl/ppo_continuous_action_isaacgym/ppo_continuous_action_isaacgym.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -291,15 +291,15 @@ def observation(self, obs):
if 0 <= step <= 2:
for idx, d in enumerate(next_done):
if d:
logging_step = global_step - args.num_envs + idx
episodic_return = info["r"][idx].item()
print(f"global_step={global_step}, episodic_return={episodic_return}")
writer.add_scalar("charts/episodic_return", episodic_return, global_step)
writer.add_scalar("charts/episodic_length", info["l"][idx], global_step)
writer.add_scalar("charts/episodic_return", episodic_return, logging_step)
writer.add_scalar("charts/episodic_length", info["l"][idx], logging_step)
if "consecutive_successes" in info: # ShadowHand and AllegroHand metric
writer.add_scalar(
"charts/consecutive_successes", info["consecutive_successes"].item(), global_step
"charts/consecutive_successes", info["consecutive_successes"].item(), logging_step
)
break

# bootstrap value if not done
with torch.no_grad():
Expand Down
8 changes: 4 additions & 4 deletions cleanrl/ppo_procgen.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -241,12 +241,12 @@ def get_action_and_value(self, x, action=None):
rewards[step] = torch.tensor(reward).to(device).view(-1)
next_obs, next_done = torch.Tensor(next_obs).to(device), torch.Tensor(next_done).to(device)

for item in info:
for i, item in enumerate(info):
if "episode" in item.keys():
logging_step = global_step - args.num_envs + i
print(f"global_step={global_step}, episodic_return={item['episode']['r']}")
writer.add_scalar("charts/episodic_return", item["episode"]["r"], global_step)
writer.add_scalar("charts/episodic_length", item["episode"]["l"], global_step)
break
writer.add_scalar("charts/episodic_return", item["episode"]["r"], logging_step)
writer.add_scalar("charts/episodic_length", item["episode"]["l"], logging_step)

# bootstrap value if not done
with torch.no_grad():
Expand Down
Loading
Loading