Sheep training flock _ improver

training/replay_config.py

@@ -44,7 +44,7 @@ def main():
                    help="Train with n_sheep randomized per episode (no curriculum). "
                         "Total train steps = steps-per-stage * max_sheep.")
     p.add_argument("--n-envs", type=int, default=8)
-    p.add_argument("--max-steps", type=int, default=1500)
+    p.add_argument("--max-steps", type=int, default=2500)
     p.add_argument("--eval-episodes", type=int, default=30)
     p.add_argument("--run-dir", type=str, default=None)
     args = p.parse_args()

training/sweep_reward.py

@@ -36,7 +36,9 @@ from herding_env import HerdingEnv
 class ProgressCallback(BaseCallback):
     """Print a one-line trial-progress summary every `freq` env steps.
     Tracks per-env returns and success directly from rollout rewards/infos
-    (no Monitor wrapper needed)."""
+    (no Monitor wrapper needed). The success window is COUNT-BASED, not
+    time-based, so successful episodes (which finish faster) don't oversample
+    the window vs truncated episodes (which take max_steps)."""
     def __init__(self, trial_id: int, stage_label: str, freq: int = 50_000):
         super().__init__()
         self.trial_id    = trial_id
@@ -45,6 +47,8 @@ class ProgressCallback(BaseCallback):
         self._last       = 0
         self._ep_returns = []
         self._ep_success = []
+        self._completed_count = 0   # total completed episodes since callback start
+        self._success_count   = 0   # total successful episodes since callback start
         self._cur_ret    = None    # per-env running return
 
     def _on_step(self) -> bool:
@@ -60,20 +64,26 @@ class ProgressCallback(BaseCallback):
             if not d: continue
             self._ep_returns.append(float(self._cur_ret[i]))
             info = infos[i] if i < len(infos) else {}
-            self._ep_success.append(
-                int(info.get("n_penned", 0) == info.get("n_sheep", -1))
-            )
+            success = int(info.get("n_penned", 0) == info.get("n_sheep", -1))
+            self._ep_success.append(success)
+            self._completed_count += 1
+            self._success_count   += success
             self._cur_ret[i] = 0.0
             if len(self._ep_returns) > 50:
                 self._ep_returns.pop(0); self._ep_success.pop(0)
         if self.num_timesteps - self._last >= self.freq:
             self._last = self.num_timesteps
-            n_eps = len(self._ep_returns)
+            n_eps  = len(self._ep_returns)
             mean_r = float(np.mean(self._ep_returns)) if n_eps else float("nan")
-            sr     = float(np.mean(self._ep_success)) if n_eps else float("nan")
+            # Window sr (biased: short eps over-represented), and cumulative sr
+            # (unbiased over the whole stage).
+            win_sr = float(np.mean(self._ep_success)) if n_eps else float("nan")
+            cum_sr = (self._success_count / self._completed_count
+                      if self._completed_count else float("nan"))
             print(f"           ... [trial {self.trial_id+1} | {self.stage_label} | "
                   f"{self.num_timesteps:>7,} steps | "
-                  f"ret(last {n_eps})={mean_r:+.2f}  sr={sr*100:.0f}%]",
+                  f"ret(last {n_eps})={mean_r:+.2f}  "
+                  f"win_sr={win_sr*100:.0f}%  cum_sr={cum_sr*100:.0f}%]",
                   flush=True)
         return True