ruff fix

Author: wniec

agents/rl_das/trainer.py

@@ -131,8 +131,12 @@ def train(
         # Log per-epoch PPO diagnostics so training instability is visible
         # (e.g. actor_loss explosion, entropy collapse) without manual debugging.
         if epoch_diagnostics:
-            entry["actor_loss"] = float(np.mean([d["actor_loss"] for d in epoch_diagnostics]))
-            entry["critic_loss"] = float(np.mean([d["critic_loss"] for d in epoch_diagnostics]))
+            entry["actor_loss"] = float(
+                np.mean([d["actor_loss"] for d in epoch_diagnostics])
+            )
+            entry["critic_loss"] = float(
+                np.mean([d["critic_loss"] for d in epoch_diagnostics])
+            )
             entry["entropy"] = float(np.mean([d["entropy"] for d in epoch_diagnostics]))
 
         if epoch % eval_interval == 0:

das/env/das_env.py

@@ -16,7 +16,7 @@
 import gymnasium as gym
 from gymnasium import spaces
 
-from das.env.observation import (compute_observation, observation_dim, MAX_HISTORY_SAMPLE)
+from das.env.observation import compute_observation, observation_dim, MAX_HISTORY_SAMPLE
 from das.env.reward import compute_reward
 from das.optimizers.base import get_checkpoints
 
@@ -257,7 +257,9 @@ def _update_episode_state(self, result: dict, prev_best_y: float):
         # derive scale from the magnitude of the initial best fitness.
         if self._initial_range[0] == float("inf"):
             safe_worst = (
-                worst_y if np.isfinite(worst_y) else new_best_y + max(abs(new_best_y), 1.0)
+                worst_y
+                if np.isfinite(worst_y)
+                else new_best_y + max(abs(new_best_y), 1.0)
             )
             self._initial_range = (new_best_y, max(safe_worst, new_best_y + 1e-5))
 

das/training/rldas.py

@@ -24,7 +24,8 @@ def run_rl_das(args) -> None:
 
     # Local variable — avoid mutating args so the caller's namespace stays predictable.
     k_epoch = (
-        args.k_epoch if args.k_epoch is not None
+        args.k_epoch
+        if args.k_epoch is not None
         else max(1, int(0.3 * args.n_checkpoints))
     )