1

1

src/environment/chengdu_majiang_env.py

@@ -1,34 +1,70 @@
 import gym
-from gym import spaces
 import numpy as np
+from gym import spaces
+
+
+from src.engine.calculate_fan import calculate_fan, is_seven_pairs, is_cleared, is_big_pairs
 from src.engine.chengdu_mahjong_engine import ChengduMahjongEngine
+from src.engine.scoring import calculate_score
+
 
 class MahjongEnv(gym.Env):
     def __init__(self):
         super(MahjongEnv, self).__init__()
         self.engine = ChengduMahjongEngine()
-        self.action_space = spaces.Discrete(108)  # 假设108种动作（摸牌/打牌）
+        self.scores = [100, 100, 100, 100]  # 四位玩家初始分数
+        self.base_score = 1  # 底分
+        self.action_space = spaces.Discrete(108)  # 动作空间：打牌的索引
         self.observation_space = spaces.Box(low=0, high=4, shape=(108,), dtype=np.int32)
 
     def reset(self):
         self.engine = ChengduMahjongEngine()
+        self.scores = [100, 100, 100, 100]  # 每局重置分数
         return self.engine.state.hands[self.engine.state.current_player]
 
     def step(self, action):
-        reward = 0
-        done = False
         try:
+            # 执行玩家动作
             self.engine.discard_tile(action)
-            reward = self.calculate_reward()  # 根据胡牌等状态定义奖励
+            # 检查是否胡牌
+            if self.engine.state.can_win(self.engine.state.hands[self.engine.state.current_player]):
+                reward, done = self.handle_win()
+            else:
+                reward, done = -1, False  # 默认小惩罚
         except ValueError:
-            reward = -10  # 非法操作扣分
+            reward, done = -10, False  # 非法操作扣分
+
         return self.engine.state.hands[self.engine.state.current_player], reward, done, {}
 
-    def calculate_reward(self):
-        if self.engine.state.can_win(self.engine.state.hands[self.engine.state.current_player]):
-            return 100  # 胡牌奖励
-        return -1  # 默认每步小惩罚
+    def handle_win(self):
+        """
+        处理胡牌后的分数结算和奖励。
+        """
+        winner = self.engine.state.current_player
+        hand = self.engine.state.hands[winner]
+        melds = self.engine.state.melds[winner]
+        is_self_draw = True  # 假设自摸（后续可动态判断）
+        is_cleared = is_cleared(hand)
+        conditions = {
+            "is_seven_pairs": is_seven_pairs(hand),
+            "is_big_pairs": is_big_pairs(hand),
+            # 添加其他条件...
+        }
+
+        # 动态计算番数
+        fan = calculate_fan(hand, melds, is_self_draw, is_cleared, conditions)
+
+        # 动态计算得分
+        scores = calculate_score(fan, self.base_score, is_self_draw)
+        self.scores[winner] += scores["winner"]
+        for i, score in enumerate(scores["loser"]):
+            self.scores[i] += score  # 扣分
+
+        # 奖励设置为赢家得分
+        reward = scores["winner"]
+        done = True  # 游戏结束
+        return reward, done
 
     def render(self, mode="human"):
-        print("当前状态：", self.engine.state.hands)
-
+        print("当前分数:", self.scores)
+        print("当前状态:", self.engine.state.hands[self.engine.state.current_player])