怎样用Python实现强化学习？OpenAI Gym入门

pip install gym

pip install gym[atari]

import gym

# 创建CartPole环境
env = gym.make('CartPole-v1')

# 重置环境，返回初始状态
state = env.reset()

# 渲染环境（可选，用于可视化）
env.render()

# 关闭环境
env.close()

import gym
import random

env = gym.make('CartPole-v1')
state = env.reset()

for _ in range(100):
    # 随机选择一个动作（0或1，分别代表向左或向右）
    action = env.action_space.sample()

    # 执行动作，返回新的状态、奖励、是否结束、额外信息
    next_state, reward, done, info = env.step(action)

    # 渲染环境
    env.render()

    # 如果游戏结束，重置环境
    if done:
        state = env.reset()
    else:
        state = next_state

env.close()

import gym
import numpy as np
import random

# 超参数
alpha = 0.1  # 学习率
gamma = 0.9  # 折扣因子
epsilon = 0.1 # 探索率
episodes = 1000 # 训练轮数

# 创建CartPole环境
env = gym.make('CartPole-v1')

# 初始化Q表
q_table = np.zeros([env.observation_space.shape[0], env.action_space.n])

# 训练
for i in range(episodes):
    state = env.reset()
    done = False
    while not done:
        # epsilon-greedy策略选择动作
        if random.uniform(0, 1) < epsilon:
            action = env.action_space.sample() # 探索
        else:
            action = np.argmax(q_table[int(state[0])]) # 利用

        # 执行动作
        next_state, reward, done, info = env.step(action)

        # 更新Q表
        old_value = q_table[int(state[0]), action]
        next_max = np.max(q_table[int(next_state[0])])
        new_value = (1 - alpha) * old_value + alpha * (reward + gamma * next_max)
        q_table[int(state[0]), action] = new_value

        # 更新状态
        state = next_state

env.close()

print("Q-table trained!")

import gym
import numpy as np

# 创建CartPole环境
env = gym.make('CartPole-v1')

# 加载训练好的Q表（假设已经训练好并保存到文件）
# q_table = np.load('q_table.npy')

# 评估
episodes = 10
total_reward = 0

for i in range(episodes):
    state = env.reset()
    done = False
    episode_reward = 0
    while not done:
        # 选择最优动作
        action = np.argmax(q_table[int(state[0])])

        # 执行动作
        next_state, reward, done, info = env.step(action)

        # 累加奖励
        episode_reward += reward

        # 更新状态
        state = next_state

    # 累加总奖励
    total_reward += episode_reward
    print(f"Episode {i+1}: Reward = {episode_reward}")

# 计算平均奖励
average_reward = total_reward / episodes
print(f"Average Reward: {average_reward}")

env.close()