[PyTorch][chapter 64][强化学习-DQN]

明朝百晓生

已于 2023-11-24 15:54:38 修改

阅读量600

点赞数

文章标签： pytorch 人工智能 python

于 2023-11-24 15:51:59 首次发布

4AM_明朝百晓生

本文链接：https://blog.csdn.net/chengxf2/article/details/134419205

版权

前言：

DQN 就是结合了深度学习和强化学习的一种算法，最初是 DeepMind 在 NIPS 2013年提出，它的核心利润包括马尔科夫决策链以及贝尔曼公式。

Q-learning的核心在于Q表格，通过建立Q表格来为行动提供指引，但这适用于状态和动作空间是离散且维数不高时，当状态和动作空间是高维连续时Q表格将变得十分巨大，对于维护Q表格和查找都是不现实的。

1: DQN 历史

2: DQN 网络参数配置

3：DQN 网络模型搭建

一 DQN 历史

DQN 跟机器学习的时序差分学习里面的Q-Learning 算法相似

1.1 Q-Learning 算法

在Q Learning 中，我们有个Q table ，记录不同状态下，各个动作的Q 值

我们通过Q table 更新当前的策略

Q table 的作用：是我们输入S，通过查表返回能够获得最大Q值的动作A.

但是很多场景状态S 并不是离散的，很难去定义

1.2 DQN 发展史

Deep network+Q-learning = DQN

DQN 和 Q-tabel 没有本质区别：

Q-table: 内部维护 Q Tabel

DQN: 通过神经网络 $a= NN(s)$ , 替代了 Q Tabel

二网络模型

2.1 DQN 算法

2.1 模型

模型参数

三代码实现：

5.1 main.py

# -*- coding: utf-8 -*-
"""
Created on Fri Nov 17 16:53:02 2023

@author: chengxf2
"""

import numpy as np
import torch
import gym
import random 
from Replaybuffer import Replay
from Agent import DQN
import rl_utils
import matplotlib.pyplot as plt
from tqdm import tqdm  #生成进度条

lr = 5e-3
hidden_dim = 128
num_episodes = 500
minimal_size = 500
gamma = 0.98
epsilon =0.01
target_update = 10
buffer_size = 10000
mini_size = 500
batch_size = 64
device = torch.device("cuda") if torch.cuda.is_available() else torch.device(
    "cpu")




if __name__ == "__main__":
    env_name = 'CartPole-v0'
    env = gym.make(env_name)
    random.seed(0)
    np.random.seed(0)
    env.seed(0)
    torch.manual_seed(0)
    replay_buffer = Replay(buffer_size)
    state_dim = env.observation_space.shape[0]
    action_dim = env.action_space.n
    agent = DQN(state_dim, hidden_dim, action_dim, lr, gamma, epsilon,
                target_update, device)
    
    return_list = []
    for i in range(10):
        with tqdm(total=int(num_episodes / 10), desc='Iteration %d' % i) as pbar:
            for i_episode in range(int(num_episodes / 10)):
                
                episode_return = 0
                state = env.reset()
                done = False
                
                while not done:
                    action = agent.take_action(state)
                    next_state, reward, done, _ = env.step(action)
                    replay_buffer.add(state, action, reward, next_state, done)
                    state = next_state
                    episode_return += reward
                    # 当buffer数据的数量超过一定值后,才进行Q网络训练
                    if replay_buffer.size() > minimal_size:
                        b_s, b_a, b_r, b_ns, b_d = replay_buffer.sample(batch_size)
                        transition_dict = {
                            'states': b_s,
                            'actions': b_a,
                            'next_states': b_ns,
                            'rewards': b_r,
                            'dones': b_d
                        }
                        agent.update(transition_dict)
                return_list.append(episode_return)
                if (i_episode + 1) % 10 == 0:
                    pbar.set_postfix({
                        'episode':
                        '%d' % (num_episodes / 10 * i + i_episode + 1),
                        'return':
                        '%.3f' % np.mean(return_list[-10:])
                    })
                pbar.update(1)
                
    episodes_list = list(range(len(return_list)))
    plt.figure(1) 
    plt.subplot(1, 2, 1)  # fig.1是一个一行两列布局的图，且现在画的是左图
    plt.plot(episodes_list, return_list,c='r')
    plt.xlabel('Episodes')
    plt.ylabel('Returns')
    plt.title('DQN on {}'.format(env_name))
   
    plt.figure(1)  # 当前要处理的图为fig.1，而且当前图是fig.1的左图
    plt.subplot(1, 2, 2)  # 当前图变为fig.1的右图
    mv_return = rl_utils.moving_average(return_list, 9)
    plt.plot(episodes_list, mv_return,c='g')
    plt.xlabel('Episodes')
    plt.ylabel('Returns')
    plt.title('DQN on {}'.format(env_name))
    plt.show()

5.2 Agent.py

# -*- coding: utf-8 -*-
"""
Created on Fri Nov 17 16:00:46 2023

@author: chengxf2
"""

import random 
import numpy as np
from   torch import nn
import torch
import torch.nn.functional as F




class QNet(torch.nn.Module):
    
    
    def __init__(self, state_dim, hidden_dim, action_dim):
        
        
        super(QNet, self).__init__()
        
        self.net = nn.Sequential(
            nn.Linear(state_dim, hidden_dim),
            nn.Linear(hidden_dim, action_dim)
            )
    
    def forward(self, state):
        
        qvalue = self.net(state)
        
        return qvalue
    

class  DQN:
    
       def __init__(self,state_dim, hidden_dim, action_dim,learning_rate,
                    discount, epsilon, target_update, device
                    ):
           
           self.action_dim = action_dim
           self.q_net = QNet(state_dim, hidden_dim, action_dim).to(device)
           self.target_q_net = QNet(state_dim, hidden_dim, action_dim).to(device)
           
           #Adam 优化器
           self.optimizer = torch.optim.Adam(self.q_net.parameters(),
                                          lr=learning_rate)
           
           self.gamma = discount #折扣因子
           self.epsilon = epsilon  # e-贪心算法
           self.target_update = target_update  # 目标网络更新频率
           self.device = device
           self.count = 0 #计数器
       
       def  take_action(self, state):
           
           rnd = np.random.random() #产生随机数
           
           if rnd <self.epsilon:
               
               action = np.random.randint(0, self.action_dim)
           
           else:
                state = torch.tensor([state], dtype=torch.float).to(self.device)
                qvalue = self.q_net(state)
                
                action = qvalue.argmax().item()
           return action
       
        
       def update(self, data):
           
           states = torch.tensor(data['states'],
                              dtype=torch.float).to(self.device)
           actions = torch.tensor(data['actions']).view(-1, 1).to(
            self.device)
           rewards = torch.tensor(data['rewards'],
                               dtype=torch.float).view(-1, 1).to(self.device)
           next_states = torch.tensor(data['next_states'],
                                   dtype=torch.float).to(self.device)
           dones = torch.tensor(data['dones'],
                             dtype=torch.float).view(-1, 1).to(self.device)
           #从完整数据中按索引取值[64]
           #print("\n actions ",actions,actions.shape)
           
           q_value = self.q_net(states).gather(1,actions) #Q值
           
           #下一个状态的Q值
           max_next_q_values = self.target_q_net(next_states).max(1)[0].view(-1,1)
           q_targets = rewards + self.gamma * max_next_q_values * (1 - dones)
           
           loss = F.mse_loss(q_value, q_targets)
           loss = torch.mean(loss)
           
           self.optimizer.zero_grad()
           loss.backward()
           self.optimizer.step()
           
           
           if self.count %self.target_update  ==0:
               #更新目标网络
               self.target_q_net.load_state_dict(
                   self.q_net.state_dict()
                   )
           self.count +=1

5.3 Replaybuffer.py

# -*- coding: utf-8 -*-
"""
Created on Fri Nov 17 15:50:07 2023

@author: chengxf2
"""

import collections 
import random 
import numpy as np
class Replay:
    
    
    def __init__(self, capacity):
        #双向队列，可以在队列的两端任意添加或删除元素。
        self.buffer = collections.deque(maxlen = capacity)
    
    
    def add(self, state, action ,reward, next_state, done):
        #数据加入buffer
        self.buffer.append((state,action,reward, next_state, done))
        
    
    def sample(self, batch_size):
        #采样数据
        data = random.sample(self.buffer, batch_size)
        
        state,action, reward, next_state,done = zip(*data)
        
        return np.array(state), action, reward, np.array(next_state), done
    
    
    def size(self):
        
        return len(self.buffer)

5.4 rl_utils.py

from tqdm import tqdm
import numpy as np
import torch
import collections
import random

class ReplayBuffer:
    def __init__(self, capacity):
        self.buffer = collections.deque(maxlen=capacity) 

    def add(self, state, action, reward, next_state, done): 
        self.buffer.append((state, action, reward, next_state, done)) 

    def sample(self, batch_size): 
        transitions = random.sample(self.buffer, batch_size)
        state, action, reward, next_state, done = zip(*transitions)
        return np.array(state), action, reward, np.array(next_state), done 

    def size(self): 
        return len(self.buffer)

def moving_average(a, window_size):
    cumulative_sum = np.cumsum(np.insert(a, 0, 0)) 
    middle = (cumulative_sum[window_size:] - cumulative_sum[:-window_size]) / window_size
    r = np.arange(1, window_size-1, 2)
    begin = np.cumsum(a[:window_size-1])[::2] / r
    end = (np.cumsum(a[:-window_size:-1])[::2] / r)[::-1]
    return np.concatenate((begin, middle, end))

def train_on_policy_agent(env, agent, num_episodes):
    return_list = []
    for i in range(10):
        with tqdm(total=int(num_episodes/10), desc='Iteration %d' % i) as pbar:
            for i_episode in range(int(num_episodes/10)):
                episode_return = 0
                transition_dict = {'states': [], 'actions': [], 'next_states': [], 'rewards': [], 'dones': []}
                state = env.reset()
                done = False
                while not done:
                    action = agent.take_action(state)
                    next_state, reward, done, _ = env.step(action)
                    transition_dict['states'].append(state)
                    transition_dict['actions'].append(action)
                    transition_dict['next_states'].append(next_state)
                    transition_dict['rewards'].append(reward)
                    transition_dict['dones'].append(done)
                    state = next_state
                    episode_return += reward
                return_list.append(episode_return)
                agent.update(transition_dict)
                if (i_episode+1) % 10 == 0:
                    pbar.set_postfix({'episode': '%d' % (num_episodes/10 * i + i_episode+1), 'return': '%.3f' % np.mean(return_list[-10:])})
                pbar.update(1)
    return return_list

def train_off_policy_agent(env, agent, num_episodes, replay_buffer, minimal_size, batch_size):
    return_list = []
    for i in range(10):
        with tqdm(total=int(num_episodes/10), desc='Iteration %d' % i) as pbar:
            for i_episode in range(int(num_episodes/10)):
                episode_return = 0
                state = env.reset()
                done = False
                while not done:
                    action = agent.take_action(state)
                    next_state, reward, done, _ = env.step(action)
                    replay_buffer.add(state, action, reward, next_state, done)
                    state = next_state
                    episode_return += reward
                    if replay_buffer.size() > minimal_size:
                        b_s, b_a, b_r, b_ns, b_d = replay_buffer.sample(batch_size)
                        transition_dict = {'states': b_s, 'actions': b_a, 'next_states': b_ns, 'rewards': b_r, 'dones': b_d}
                        agent.update(transition_dict)
                return_list.append(episode_return)
                if (i_episode+1) % 10 == 0:
                    pbar.set_postfix({'episode': '%d' % (num_episodes/10 * i + i_episode+1), 'return': '%.3f' % np.mean(return_list[-10:])})
                pbar.update(1)
    return return_list


def compute_advantage(gamma, lmbda, td_delta):
    td_delta = td_delta.detach().numpy()
    advantage_list = []
    advantage = 0.0
    for delta in td_delta[::-1]:
        advantage = gamma * lmbda * advantage + delta
        advantage_list.append(advantage)
    advantage_list.reverse()
    return torch.tensor(advantage_list, dtype=torch.float)

DQN 算法
 遇强则强（八）：从Q-table到DQN - 知乎使用Pytorch实现强化学习——DQN算法_dqn pytorch-CSDN博客

https://www.cnblogs.com/xiaohuiduan/p/12993691.html

https://www.cnblogs.com/xiaohuiduan/p/12945449.html

强化学习第五节（DQN）【个人知识分享】_哔哩哔哩_bilibili

CSDN

组会讲解强化学习的DQN算法_哔哩哔哩_bilibili

3-ε-greedy_ReplayBuffer_FixedQ-targets_哔哩哔哩_bilibili

4-代码实战DQN_Agent和Env整体交互_哔哩哔哩_bilibili

DQN基本概念和算法流程（附Pytorch代码） - 知乎

CSDN

DQN 算法

明朝百晓生

关注

0
点赞
踩
0

收藏

觉得还不错? 一键收藏
0
评论
[PyTorch][chapter 64][强化学习-DQN]

Q-learning的核心在于Q表格，通过建立Q表格来为行动提供指引，但这适用于状态和动作空间是离散且维数不高时，当状态和动作空间是高维连续时Q表格将变得十分巨大，对于维护Q表格和查找都是不现实的。在Q Learning 中，我们有个Q table ，记录不同状态下，各个动作的Q 值。Q table 的作用：是我们输入S，通过查表返回能够获得最大Q值的动作A.DQN 跟机器学习的时序差分学习里面的Q-Learning 算法相似。Q-table: 内部维护 Q Tabel。3：DQN 网络模型搭建。
复制链接

扫一扫