Merge pull request #190 from jxd136/dev

smooth is not good

Author: lch24

logic/pve/GameLogic/board.py

@@ -194,11 +194,15 @@ def manhattan(self, x1: int, y1: int, x2: int, y2: int) -> int:
         return abs(x1 - x2) + abs(y1 - y2)
 
     def nearest_market(self, x: int, y: int) -> Optional[Tuple[int, int]]:
-        """Return (mx, my) of market with Manhattan distance ≤ 1, or None."""
+        """Return nearest market within Manhattan distance ≤ 1, or None."""
+        best = None
+        best_dist = 9999
         for mx, my in self.market_positions:
-            if self.manhattan(x, y, mx, my) <= 1:
-                return (mx, my)
-        return None
+            d = self.manhattan(x, y, mx, my)
+            if d <= 1 and d < best_dist:
+                best = (mx, my)
+                best_dist = d
+        return best
 
     def nearest_resource(self, x: int, y: int) -> Optional[ResourcePoint]:
         """Return nearest non-depleted resource within harvest range (≤ 2)."""

logic/pve/official_evaluator.py

@@ -30,7 +30,6 @@
 import argparse
 import importlib
 import json
-import math
 import os
 import sys
 from pathlib import Path
@@ -47,11 +46,6 @@
 from RLInterfaces import BaseAgent
 
 
-def _smooth_score(score: float, scale: float = 1000.0) -> float:
-    """Bounded score smoothing using tanh followed by sigmoid."""
-    z = math.tanh(score / scale)
-    return 1.0 / (1.0 + math.exp(-z))
-
 def load_agent(submission_dir: str, model_path: Optional[str], env: GameEnvironment) -> BaseAgent:
     """
     Dynamically load contestant's Agent class from submission_dir/agent.py.
@@ -129,8 +123,7 @@ def evaluate(
             action = agent.get_action(obs)
             obs, _reward, terminated, truncated, info = env.step(action)
             ep_len += 1
-            raw_score = info.get("score", 0.0)
-            ep_score = _smooth_score(raw_score)
+            ep_score = info.get("score", 0.0)
             done = terminated or truncated
 
         scores.append(ep_score)