极智AI | 解读强化学习中的Q-learning

import numpy as npimport gym
# 创建FrozenLake环境env = gym.make('FrozenLake-v1', is_slippery=False)
# 初始化参数num_states = env.observation_space.nnum_actions = env.action_space.nQ = np.zeros((num_states, num_actions))num_episodes = 1000max_steps = 100alpha = 0.1 # 学习率gamma = 0.99 # 折扣因子epsilon = 0.1 # 探索率
for episode in range(num_episodes): state = env.reset() for step in range(max_steps): # 选择动作（ε-贪心策略） if np.random.uniform(0, 1) < epsilon: action = env.action_space.sample() else: action = np.argmax(Q[state, :])
 # 执行动作，获得下一个状态和奖励 next_state, reward, done, info = env.step(action)
 # 更新Q函数 best_next_action = np.argmax(Q[next_state, :]) td_target = reward + gamma * Q[next_state, best_next_action] td_error = td_target - Q[state, action] Q[state, action] += alpha * td_error
 # 状态更新 state = next_state
 # 回合结束 if done: break
print("训练完成后的Q表：")print(Q)