使用python 写一个DQN 强化学习 的程序

好的，下面是一个基本的DQN强化学习的程序，使用Python和PyTorch实现：

首先，我们需要导入所需的库和模块：

import random
import numpy as np
import torch
import torch.nn as nn
import torch.optim as optim
import gym

接下来，我们定义一个神经网络类，用于表示Q值函数：

class QNetwork(nn.Module):
    def __init__(self, input_dim, output_dim):
        super(QNetwork, self).__init__()
        self.fc1 = nn.Linear(input_dim, 64)
        self.fc2 = nn.Linear(64, 64)
        self.fc3 = nn.Linear(64, output_dim)

    def forward(self, state):
        x = torch.relu(self.fc1(state))
        x = torch.relu(self.fc2(x))
        x = self.fc3(x)
        return x

然后，我们定义一个DQN类，用于实现DQN算法：

class DQNAgent:
    def __init__(self, env):
        self.env = env
        self.obs_dim = env.observation_space.shape[0]
        self.action_dim = env.action_space.n
        self.q_net = QNetwork(self.obs_dim, self.action_dim)
        self.target_q_net = QNetwork(self.obs_dim, self.action_dim)
        self.target_q_net.load_state_dict(self.q_net.state_dict())
        self.optimizer = optim.Adam(self.q_net.parameters(), lr=0.001)
        self.gamma = 0.99
        self.epsilon = 1.0
        self.epsilon_decay = 0.995
        self.min_epsilon = 0.01
        self.batch_size = 32
        self.replay_buffer = []
        self.replay_buffer_max_size = 10000

    def act(self, state):
        if np.random.rand() < self.epsilon:
            return self.env.action_space.sample()
        state = torch.tensor(state, dtype=torch.float32)
        q_values = self.q_net(state)
        return q_values.argmax().item()

    def update_replay_buffer(self, state, action, reward, next_state, done):
        self.replay_buffer.append((state, action, reward, next_state, done))
        if len(self.replay_buffer) > self.replay_buffer_max_size:
            self.replay_buffer.pop(0)

    def sample_from_replay_buffer(self):
        samples = random.sample(self.replay_buffer, self.batch_size)
        state, action, reward, next_state, done = map(np.array, zip(*samples))
        return state, action, reward, next_state, done

    def update_q_net(self):
        state, action, reward, next_state, done = self.sample_from_replay_buffer()
        state = torch.tensor(state, dtype=torch.float32)
        action = torch.tensor(action, dtype=torch.long).unsqueeze(1)
        reward = torch.tensor(reward, dtype=torch.float32).unsqueeze(1)
        next_state = torch.tensor(next_state, dtype=torch.float32)
        done = torch.tensor(done, dtype=torch.float32).unsqueeze(1)
        q_values = self.q_net(state).gather(1, action)
        next_q_values = self.target_q_net(next_state).max(1)[0].unsqueeze(1)
        target_q_values = reward + (1 - done) * self.gamma * next_q_values
        loss = nn.functional.smooth_l1_loss(q_values, target_q_values.detach())
        self.optimizer.zero_grad()
        loss.backward()
        self.optimizer.step()

    def update_target_q_net(self):
        self.target_q_net.load_state_dict(self.q_net.state_dict())

    def train(self, num_episodes):
        for i in range(num_episodes):
            state = self.env.reset()
            episode_reward = 0
            done = False
            while not done:
                action = self.act(state)
                next_state, reward, done, _ = self.env.step(action)
                episode_reward += reward
                self.update_replay_buffer(state, action, reward, next_state, done)
                state = next_state
                if len(self.replay_buffer) >= self.batch_size:
                    self.update_q_net()
                    if len(self.replay_buffer) % 100 == 0:
                        self.update_target_q_net()
            self.epsilon = max(self.epsilon * self.epsilon_decay, self.min_epsilon)
            print(f"Episode {i + 1}: Reward = {episode_reward}")

最后，我们可以使用上述DQNAgent类来训练我们的模型：

env = gym.make("CartPole-v1")
agent = DQNAgent(env)
agent.train(num_episodes=1000)

上述代码将创建一个CartPole-v1环境，并使用DQN算法进行训练，共进行1000个Episode的训练。