吃豆人蒙特卡洛搜索代码

以下是使用Python编写的吃豆人蒙特卡洛搜索算法代码：

import random

class MonteCarloAgent:
    def __init__(self, eval_fn, num_rollouts=10, max_depth=30):
        self.eval_fn = eval_fn
        self.num_rollouts = num_rollouts
        self.max_depth = max_depth

    def get_action(self, state):
        legal_actions = state.get_legal_actions()
        if not legal_actions:
            return None

        action_scores = {}
        for action in legal_actions:
            total_score = 0
            for i in range(self.num_rollouts):
                score = self.simulate(state.generate_successor(action), depth=0)
                total_score += score
            action_scores[action] = total_score / self.num_rollouts

        best_action = max(action_scores, key=action_scores.get)
        return best_action

    def simulate(self, state, depth):
        if state.is_win() or state.is_lose() or depth == self.max_depth:
            return self.eval_fn(state)

        legal_actions = state.get_legal_actions()
        if not legal_actions:
            return 0

        action = random.choice(legal_actions)
        return self.simulate(state.generate_successor(action), depth+1)

def mc_eval_fn(state):
    return state.get_score()

agent = MonteCarloAgent(eval_fn=mc_eval_fn, num_rollouts=10, max_depth=30)