强化学习 - Deep Q Network (DQN)

Deep Q Network（DQN）是一种结合深度学习和强化学习的方法，用于解决离散动作空间的强化学习问题。DQN 是由DeepMind团队提出的，首次应用于解决Atari游戏，但也被广泛用于其他领域，如机器人学和自动驾驶。

以下是一个使用Python和TensorFlow / Keras 实现简单的DQN的示例代码。请注意，这是一个基本的实现，实际应用中可能需要进行更多的优化和调整。

import numpy as np
import tensorflow as tf
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Dense
from tensorflow.keras.optimizers import Adam
from collections import deque
import random
import gym# 定义DQN Agent
class DQNAgent:def __init__(self, state_size, action_size):self.state_size = state_sizeself.action_size = action_sizeself.memory = deque(maxlen=2000)  # 经验回放内存self.gamma = 0.95  # 折扣因子self.epsilon = 1.0  # 探索概率self.epsilon_decay = 0.995  # 探索概率衰减self.epsilon_min = 0.01  # 最小探索概率self.learning_rate = 0.001self.model = self.build_model()def build_model(self):model = Sequential()model.add(Dense(24, input_dim=self.state_size, activation='relu'))model.add(Dense(24, activation='relu'))model.add(Dense(self.action_size, activation='linear'))model.compile(loss='mse', optimizer=Adam(lr=self.learning_rate))return modeldef remember(self, state, action, reward, next_state, done):self.memory.append((state, action, reward, next_state, done))def act(self, state):if np.random.rand() <= self.epsilon:return random.randrange(self.action_size)else:return np.argmax(self.model.predict(state)[0])def replay(self, batch_size):minibatch = random.sample(self.memory, batch_size)for state, action, reward, next_state, done in minibatch:target = rewardif not done:target = reward + self.gamma * np.amax(self.model.predict(next_state)[0])target_f = self.model.predict(state)target_f[0][action] = targetself.model.fit(state, target_f, epochs=1, verbose=0)if self.epsilon > self.epsilon_min:self.epsilon *= self.epsilon_decay# 初始化环境和Agent
env = gym.make('CartPole-v1')
state_size = env.observation_space.shape[0]
action_size = env.action_space.n
agent = DQNAgent(state_size, action_size)# 训练DQN
batch_size = 32
num_episodes = 1000for episode in range(num_episodes):state = env.reset()state = np.reshape(state, [1, state_size])total_reward = 0for time in range(500):  # 限制每个episode的步数，防止无限循环# env.render()  # 如果想可视化训练过程，可以取消注释此行action = agent.act(state)next_state, reward, done, _ = env.step(action)reward = reward if not done else -10  # 对于未结束的episode，奖励为1；结束则为-10total_reward += rewardnext_state = np.reshape(next_state, [1, state_size])agent.remember(state, action, reward, next_state, done)state = next_stateif done:print("Episode: {}, Total Reward: {}, Epsilon: {:.2}".format(episode + 1, total_reward, agent.epsilon))breakif len(agent.memory) > batch_size:agent.replay(batch_size)# 关闭环境
env.close()

在这个例子中，我们使用 OpenAI Gym 提供的 CartPole 环境作为示例。DQN Agent 的神经网络模型使用简单的全连接层。训练过程中，Agent通过经验回放（experience replay）来学习，并使用ε-greedy策略选择动作。通过运行多个episode，Agent逐渐学习到达得分较高的策略。

请注意，DQN的具体实现可能因问题的复杂性而有所不同，而且可能需要更多的技术来提高稳定性和性能，如双Q网络、优先级经验回放等。