tdpeuter/2026SEL3-project-Brittle_Star_Controller_Modularity
tdpeuter/2026SEL3-project-Brittle_Star_Controller_Modularity
Archived
1
Fork 0
Code Issues Releases Activity

feat(logging): added avg episode at which terminated or truncated happened

Browse source
This commit is contained in:
vsc46589 vscuser 2026-04-10 00:11:18 +02:00
parent f7428f9d90
commit 71349af5b9
1 changed files with 18 additions and 1 deletions
Download patch file Download diff file Expand all files Collapse all files

19
src/brittle_star_project/trainers/PPOTrainer.py
View file

@ -205,6 +205,8 @@ class LossInfo:
explained_variance: float explained_variance: float
num_terminated: int num_terminated: int
num_truncated: int num_truncated: int
avg_terminated_length: Any
avg_truncated_length: Any
class PPOTrainer: class PPOTrainer:
@ -363,6 +365,8 @@ class PPOTrainer:
"charts/explained_variance": loss_info.explained_variance, "charts/explained_variance": loss_info.explained_variance,
"charts/num_terminated": loss_info.num_terminated, "charts/num_terminated": loss_info.num_terminated,
"charts/num_truncated": loss_info.num_truncated, "charts/num_truncated": loss_info.num_truncated,
"charts/avg_terminated_ep_length": loss_info.avg_terminated_length,
"charts/avg_truncated_ep_length": loss_info.avg_truncated_length,
"losses/value_loss": loss_info.v_loss[-1, -1].item(), "losses/value_loss": loss_info.v_loss[-1, -1].item(),
"losses/policy_loss": loss_info.pg_loss[-1, -1].item(), "losses/policy_loss": loss_info.pg_loss[-1, -1].item(),
"losses/entropy": loss_info.entropy_loss[-1, -1].item(), "losses/entropy": loss_info.entropy_loss[-1, -1].item(),
@ -414,10 +418,21 @@ class PPOTrainer:
terminated = next_env_state.terminated # (num_envs,) terminated = next_env_state.terminated # (num_envs,)
truncated = next_env_state.truncated # (num_envs,) truncated = next_env_state.truncated # (num_envs,)
episode_lengths = self.episode_stats.returned_episode_lengths
num_terminated = int(jnp.sum(terminated).item()) num_terminated = int(jnp.sum(terminated).item())
num_truncated = int(jnp.sum(truncated).item()) num_truncated = int(jnp.sum(truncated).item())
avg_terminated_length = (
jnp.sum(episode_lengths * terminated) /
jnp.maximum(jnp.sum(terminated), 1)
)
avg_truncated_length = (
jnp.sum(episode_lengths * truncated) /
jnp.maximum(jnp.sum(truncated), 1)
)
return ( return (
next_env_state, next_env_state,
next_obs, next_obs,
@ -431,7 +446,9 @@ class PPOTrainer:
avg_episodic_return=avg_episodic_return, avg_episodic_return=avg_episodic_return,
explained_variance=explained_var, explained_variance=explained_var,
num_terminated=num_terminated, num_terminated=num_terminated,
num_truncated=num_truncated num_truncated=num_truncated,
avg_terminated_length=avg_terminated_length,
avg_truncated_length=avg_truncated_length
), ),
) )