1

1

scripts/train_chengdu_mahjong_model.py

@@ -3,6 +3,7 @@ from stable_baselines3 import PPO
 from src.environment.chengdu_mahjong_env import ChengduMahjongEnv
 import torch
 from configs.log_config import setup_logging
+from loguru import logger  # 添加 logger
 
 def train_model():
     # 创建 MahjongEnv 环境实例
@@ -10,7 +11,7 @@ def train_model():
 
     # 检查是否有可用的 GPU
     device = "cuda" if torch.cuda.is_available() else "cpu"
-    print(f"使用设备: {device}")
+    logger.info(f"使用设备: {device}")  # 替换 print 为 logger.info
 
     # 使用 PPO 算法训练模型，切换到 MultiInputPolicy
     model = PPO(
@@ -22,18 +23,22 @@ def train_model():
     )
 
     # 训练模型，训练总步数为 100000
-    model.learn(total_timesteps=100000)
+    logger.info("开始训练模型...")
+    model.learn(total_timesteps=100)
+    logger.info("模型训练完成！")
 
     # 保存训练后的模型
     model.save("../models/ppo_mahjong_model")
+    logger.info("模型已保存到 '../models/ppo_mahjong_model'")
 
     # 测试模型
+    logger.info("开始测试模型...")
     obs = env.reset()
     done = False
     while not done:
         action, _states = model.predict(obs)  # 使用训练好的模型来选择动作
         obs, reward, done, info = env.step(action)  # 执行动作
-        print(f"动作: {action}, 奖励: {reward}, 是否结束: {done}, 信息: {info}")
+        logger.info(f"动作: {action}, 奖励: {reward}, 是否结束: {done}, 信息: {info}")  # 替换 print 为 logger.info
 
 if __name__ == "__main__":
     # 调用配置函数来设置日志

src/engine/chengdu_mahjong_engine.py

@@ -48,58 +48,28 @@ class ChengduMahjongEngine:
     def play_turn(self):
         current_player = self.state.current_player
 
-        if self.state.draw_counts[current_player] == 0:
+        # 玩家摸牌逻辑
-            # 判断庄家是否天胡
-            tianhu = self.state.can_win(
-                self.state.hands[current_player],
-                self.state.melds[current_player],
-                self.state.missing_suits[current_player],
-            )
-
-            if tianhu:
-                # 天胡结算
-                self.state.winners.append(current_player)
-                self.state.print_game_state(current_player)
-                for i in range(4):
-                    if i != current_player:
-                        self.state.scores[i] -= self.state.bottom_score * 2 ** 5
-                self.state.scores[current_player] += self.state.bottom_score * 2 ** 5
-            else:
-                # 判断是否可以杠
-                if self.state.hands[current_player].can_gang():
-                    # 获取杠牌类型
-                    gang_type = self.state.hands[current_player].get_gang_type()
-
-                    # AI 决策是否杠牌
-                    if should_gang(current_player, self.state, gang_type):
-                        self.state.print_game_state(current_player)
-
-                        # 计算杠牌得分
-                        if gang_type == "暗杠":
-                            # 暗杠分数结算
-                            for i in range(4):
-                                if i != current_player:
-                                    self.state.scores[i] -= self.state.bottom_score * 2 ** 2  # 扣分
-                            self.state.scores[current_player] += self.state.bottom_score * 2 ** 2 * 3  # 加分
-                        elif gang_type == "明杠":
-                            # 明杠分数结算
-                            for i in range(4):
-                                if i != current_player:
-                                    self.state.scores[i] -= self.state.bottom_score * 2 ** 1  # 扣分
-                            self.state.scores[current_player] += self.state.bottom_score * 2 ** 1 * 3  # 加分
-
-                        logger.info(f"玩家 {current_player} 杠牌，类型: {gang_type}")
-                else:
-                    # 当前玩家出牌
-                    tile = random_choice(self.state.hands[current_player], self.state.missing_suits[current_player])
-                    logger.info(f"玩家 {current_player} 打出牌: {tile}")
-
-                    # 检查其他玩家的反应
-                    if not check_other_players(self,tile):
-                        # 没有触发其他玩家操作，移动到下一个玩家
-                        self.state.current_player = (current_player + 1) % 4
         draw_tile(self)
 
+        # 玩家选择一张牌打出
+        tile = random_choice(self.state.hands[current_player], self.state.missing_suits[current_player])
+        logger.info(f"玩家 {current_player} 选择打牌: {tile}")
+
+        # 检查其他玩家是否可以对该牌进行操作
+        actions_taken = self.check_other_players(tile)
+
+        if not actions_taken:
+            # 将牌加入弃牌堆
+            self.state.discards[current_player].append(tile)
+            logger.info(f"玩家 {current_player} 打出的牌 {tile} 没有触发其他玩家的操作")
+
+            # 切换到下一位玩家
+            self.state.current_player = (current_player + 1) % 4
+            logger.info(f"轮到玩家 {self.state.current_player} 出牌")
+
+        # 检查游戏结束条件
+        self.check_game_over()
+
     def check_game_over(self):
         """
         检查游戏是否结束。

src/environment/chengdu_mahjong_env.py

@@ -41,24 +41,25 @@ class ChengduMahjongEnv(gym.Env):
         :return: obs, reward, done, info
         """
         current_player = self.engine.state.current_player
+        hand = self.engine.state.hands[current_player].tiles  # 当前玩家手牌
 
-        # **1. 检查动作是否合法并执行**
+        # **1. 执行动作并检查合法性**
-        if action < 14:  # 打牌动作
+        if action < len(hand):  # 打牌动作
-            if action >= len(self.engine.state.hands[current_player].tiles):
+            tile = hand[action]
-                raise ValueError(f"动作 {action} 超出手牌范围")
-            tile = self.engine.state.hands[current_player].tiles[action]
             logger.info(f"玩家 {current_player} 选择打牌: {tile}")
             self.engine.check_other_players(tile)
         elif action == 14:  # 碰
             tile_to_peng = self._get_tile_for_special_action("peng")
             if tile_to_peng:
                 handle_peng(self.engine, current_player, tile_to_peng)
+                logger.info(f"玩家 {current_player} 碰了牌: {tile_to_peng}")
             else:
                 logger.warning("碰动作无效，未满足条件")
         elif action == 15:  # 杠
             tile_to_gang = self._get_tile_for_special_action("gang")
             if tile_to_gang:
-                handle_gang(current_player, tile_to_gang, mode="an")  # 默认暗杠
+                handle_gang(self.engine, current_player, tile_to_gang, mode="an")
+                logger.info(f"玩家 {current_player} 杠了牌: {tile_to_gang}")
             else:
                 logger.warning("杠动作无效，未满足条件")
         elif action == 16:  # 胡
@@ -67,7 +68,8 @@ class ChengduMahjongEnv(gym.Env):
                     self.engine.state.melds[current_player],
                     self.engine.state.missing_suits[current_player]
             ):
-                handle_win(current_player, None, None)
+                handle_win(self.engine, current_player, None, None)
+                logger.info(f"玩家 {current_player} 胡牌！")
             else:
                 logger.warning("胡动作无效，未满足条件")
         else:
@@ -90,6 +92,8 @@ class ChengduMahjongEnv(gym.Env):
         }
         return obs, reward, done, info
 
+
+
     def _get_observation(self):
         """
         提取当前玩家的观察空间