MonteCarlo code error

[beifeng1937]

train function in MonteCarlo.ipynb

agent.update(one_ep_transition) # 更新智能体 should be outside the for-loop

def train(cfg,env,agent):
    print('开始训练！')
    print(f'环境:{cfg.env_name}, 算法:{cfg.algo_name}, 设备:{cfg.device}')
    rewards = []  # 记录奖励
    for i_ep in range(cfg.train_eps):
        ep_reward = 0  # 记录每个回合的奖励
        one_ep_transition = []
        state = env.reset(seed=cfg.seed) # 重置环境,即开始新的回合
        for _ in range(cfg.max_steps):
            action = agent.sample_action(state)  # 根据算法采样一个动作
            next_state, reward, terminated, info = env.step(action)   # 与环境进行一次动作交互
            one_ep_transition.append((state, action, reward))  # 保存transitions
            state = next_state  # 更新状态
            ep_reward += reward  
            if terminated:
                break
        agent.update(one_ep_transition)  # 更新智能体
        rewards.append(ep_reward)
        print(f"回合：{i_ep+1}/{cfg.train_eps}，奖励：{ep_reward:.1f}")
    print('完成训练！')
    return {"rewards":rewards}

[beifeng1937]

There is also a bug in train function of Sarsa.ipynb . action = agent.sample(state) after while True should be deleted. Correct code is:

def train(cfg,env,agent):
    print('开始训练！')
    print(f'环境:{cfg.env_name}, 算法:{cfg.algo_name}, 设备:{cfg.device}')
    rewards = []  # 记录奖励
    for i_ep in range(cfg.train_eps):
        ep_reward = 0  # 记录每个回合的奖励
        state = env.reset()  # 重置环境,即开始新的回合
        action = agent.sample(state)
        while True:
            #action = agent.sample(state)   should be deleted
            next_state, reward, done, _ = env.step(action)  # 与环境进行一次动作交互
            next_action = agent.sample(next_state)
            agent.update(state, action, reward, next_state, next_action,done) # 算法更新
            state = next_state # 更新状态
            action = next_action
            ep_reward += reward
            if done:
                break
        rewards.append(ep_reward)
        print(f"回合：{i_ep+1}/{cfg.train_eps}，奖励：{ep_reward:.1f}，Epsilon：{agent.epsilon}")
    print('完成训练！')
    return {"rewards":rewards}