update projects

projects/codes/DQN/README.md

@@ -0,0 +1,218 @@
+# DQN
+
+## 原理简介
+
+DQN是Q-leanning算法的优化和延伸，Q-leaning中使用有限的Q表存储值的信息，而DQN中则用神经网络替代Q表存储信息，这样更适用于高维的情况，相关知识基础可参考[datawhale李宏毅笔记-Q学习](https://datawhalechina.github.io/easy-rl/#/chapter6/chapter6)。
+
+论文方面主要可以参考两篇，一篇就是2013年谷歌DeepMind团队的[Playing Atari with Deep Reinforcement Learning](https://www.cs.toronto.edu/~vmnih/docs/dqn.pdf)，一篇是也是他们团队后来在Nature杂志上发表的[Human-level control through deep reinforcement learning](https://web.stanford.edu/class/psych209/Readings/MnihEtAlHassibis15NatureControlDeepRL.pdf)。后者在算法层面增加target q-net，也可以叫做Nature DQN。
+
+Nature DQN使用了两个Q网络，一个当前Q网络𝑄用来选择动作，更新模型参数，另一个目标Q网络𝑄′用于计算目标Q值。目标Q网络的网络参数不需要迭代更新，而是每隔一段时间从当前Q网络𝑄复制过来，即延时更新，这样可以减少目标Q值和当前的Q值相关性。
+
+要注意的是，两个Q网络的结构是一模一样的。这样才可以复制网络参数。Nature DQN和[Playing Atari with Deep Reinforcement Learning](https://www.cs.toronto.edu/~vmnih/docs/dqn.pdf)相比，除了用一个新的相同结构的目标Q网络来计算目标Q值以外，其余部分基本是完全相同的。细节也可参考[强化学习（九）Deep Q-Learning进阶之Nature DQN](https://www.cnblogs.com/pinard/p/9756075.html)。
+
+https://blog.csdn.net/JohnJim0/article/details/109557173)
+
+## 伪代码
+
+<img src="assets/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L0pvaG5KaW0w,size_16,color_FFFFFF,t_70.png" alt="img" style="zoom:50%;" />
+
+## 代码实现
+
+### RL接口
+
+首先是强化学习训练的基本接口，即通用的训练模式：
+```python
+for i_episode in range(MAX_EPISODES):
+	state = env.reset() # reset环境状态
+	for i_step in range(MAX_STEPS):
+    action = agent.choose_action(state) # 根据当前环境state选择action
+    next_state, reward, done, _ = env.step(action) # 更新环境参数
+    agent.memory.push(state, action, reward, next_state, done) # 将state等这些transition存入memory
+    agent.update() # 每步更新网络
+    state = next_state # 跳转到下一个状态
+    if done:
+    	break        
+```
+每个episode加一个MAX_STEPS，也可以使用while not done, 加这个max_steps有时是因为比如gym环境训练目标就是在200个step下达到200的reward，或者是当完成一个episode的步数较多时也可以设置，基本流程跟所有伪代码一致，如下：
+1. agent选择动作
+2. 环境根据agent的动作反馈出next_state和reward
+3. agent进行更新，如有memory就会将transition(包含state，reward，action等)存入memory中
+4. 跳转到下一个状态
+5. 如果done了，就跳出循环，进行下一个episode的训练。
+
+想要实现完整的算法还需要创建Qnet，Replaybuffer等类
+
+### 两个Q网络
+
+上文讲了Nature DQN中有两个Q网络，一个是policy_net，一个是延时更新的target_net，两个网络的结构是一模一样的，如下(见```model.py```)，注意DQN使用的Qnet就是全连接网络即FCH：
+```python
+import torch.nn as nn
+import torch.nn.functional as F
+
+class FCN(nn.Module):
+    def __init__(self, n_states=4, n_actions=18):
+        """ 初始化q网络，为全连接网络
+            n_states: 输入的feature即环境的state数目
+            n_actions: 输出的action总个数
+        """
+        super(FCN, self).__init__()
+        self.fc1 = nn.Linear(n_states, 128) # 输入层
+        self.fc2 = nn.Linear(128, 128) # 隐藏层
+        self.fc3 = nn.Linear(128, n_actions) # 输出层
+        
+    def forward(self, x):
+        # 各层对应的激活函数
+        x = F.relu(self.fc1(x)) 
+        x = F.relu(self.fc2(x))
+        return self.fc3(x)
+```
+输入为n_states，输出为n_actions，包含一个128维度的隐藏层，这里根据需要可增加隐藏层维度和数量，然后一般使用relu激活函数，这里跟深度学习的网路设置是一样的。
+
+### Replay Buffer
+
+然后就是Replay Memory了，其作用主要是是克服经验数据的相关性（correlated data）和非平稳分布（non-stationary distribution）问题，实现如下(见```memory.py```)：
+
+```python
+import random
+import numpy as np
+
+class ReplayBuffer:
+    
+    def __init__(self, capacity):
+        self.capacity = capacity
+        self.buffer = []
+        self.position = 0
+    
+    def push(self, state, action, reward, next_state, done):
+        if len(self.buffer) < self.capacity:
+            self.buffer.append(None)
+        self.buffer[self.position] = (state, action, reward, next_state, done)
+        self.position = (self.position + 1) % self.capacity
+    
+    def sample(self, batch_size):
+        batch = random.sample(self.buffer, batch_size)
+        state, action, reward, next_state, done =  zip(*batch)
+        return state, action, reward, next_state, done
+    
+    def __len__(self):
+        return len(self.buffer)
+```
+
+参数capacity表示buffer的容量，主要包括push和sample两个步骤，push是将transitions放到memory中，sample是从memory随机抽取一些transition。
+
+### Agent类
+
+在```agent.py```中我们定义强化学习算法类，包括```choose_action```(选择动作，使用e-greedy策略时会多一个```predict```函数，下面会将到)和```update```(更新)等函数。
+
+在类中建立两个网络，以及optimizer和memory，
+
+```python
+self.policy_net = MLP(n_states, n_actions,hidden_dim=cfg.hidden_dim).to(self.device)
+self.target_net = MLP(n_states, n_actions,hidden_dim=cfg.hidden_dim).to(self.device)
+for target_param, param in zip(self.target_net.parameters(),self.policy_net.parameters()): # copy params from policy net
+    target_param.data.copy_(param.data)
+self.optimizer = optim.Adam(self.policy_net.parameters(), lr=cfg.lr)
+self.memory = ReplayBuffer(cfg.memory_capacity)
+```
+然后是选择action：
+
+```python
+def choose_action(self, state):
+        '''选择动作
+        '''
+    self.frame_idx += 1
+    if random.random() > self.epsilon(self.frame_idx):
+        action = self.predict(state)
+    else:
+        action = random.randrange(self.n_actions)
+    return action
+```
+
+这里使用e-greedy策略，即设置一个参数epsilon，如果生成的随机数大于epsilon，就根据网络预测的选择action，否则还是随机选择action，这个epsilon是会逐渐减小的，可以使用线性或者指数减小的方式，但不会减小到零，这样在训练稳定时还能保持一定的探索，这部分可以学习探索与利用(exploration and exploition)相关知识。
+
+上面讲到的预测函数其实就是根据state选取q值最大的action，如下：
+
+```python
+def predict(self,state):
+    with torch.no_grad():
+        state = torch.tensor([state], device=self.device, dtype=torch.float32)
+        q_values = self.policy_net(state)
+        action = q_values.max(1)[1].item()
+```
+
+然后是更新函数了：
+
+```python
+def update(self):
+
+        if len(self.memory) < self.batch_size:
+            return
+        # 从memory中随机采样transition
+        state_batch, action_batch, reward_batch, next_state_batch, done_batch = self.memory.sample(
+            self.batch_size)
+        '''转为张量
+        例如tensor([[-4.5543e-02, -2.3910e-01,  1.8344e-02,  2.3158e-01],...,[-1.8615e-02, -2.3921e-01, -1.1791e-02,  2.3400e-01]])'''
+        state_batch = torch.tensor(
+            state_batch, device=self.device, dtype=torch.float)
+        action_batch = torch.tensor(action_batch, device=self.device).unsqueeze(
+            1)  # 例如tensor([[1],...,[0]])
+        reward_batch = torch.tensor(
+            reward_batch, device=self.device, dtype=torch.float)  # tensor([1., 1.,...,1])
+        next_state_batch = torch.tensor(
+            next_state_batch, device=self.device, dtype=torch.float)
+        done_batch = torch.tensor(np.float32(
+            done_batch), device=self.device)
+
+        '''计算当前(s_t,a)对应的Q(s_t, a)'''
+        '''torch.gather:对于a=torch.Tensor([[1,2],[3,4]]),那么a.gather(1,torch.Tensor([[0],[1]]))=torch.Tensor([[1],[3]])'''
+        q_values = self.policy_net(state_batch).gather(
+            dim=1, index=action_batch)  # 等价于self.forward
+        # 计算所有next states的V(s_{t+1})，即通过target_net中选取reward最大的对应states
+        next_q_values = self.target_net(next_state_batch).max(
+            1)[0].detach()  # 比如tensor([ 0.0060, -0.0171,...,])
+        # 计算 expected_q_value
+        # 对于终止状态，此时done_batch[0]=1, 对应的expected_q_value等于reward
+        expected_q_values = reward_batch + \
+            self.gamma * next_q_values * (1-done_batch)
+        # self.loss = F.smooth_l1_loss(q_values,expected_q_values.unsqueeze(1)) # 计算 Huber loss
+        loss = nn.MSELoss()(q_values, expected_q_values.unsqueeze(1))  # 计算 均方误差loss
+        # 优化模型
+        self.optimizer.zero_grad()  # zero_grad清除上一步所有旧的gradients from the last step
+        # loss.backward()使用backpropagation计算loss相对于所有parameters(需要gradients)的微分
+        loss.backward()
+        # for param in self.policy_net.parameters():  # clip防止梯度爆炸
+        #     param.grad.data.clamp_(-1, 1)
+        self.optimizer.step()  # 更新模型
+```
+
+更新遵循伪代码的以下部分：
+
+<img src="assets/image-20210507162813393.png" alt="image-20210507162813393" style="zoom:50%;" />
+
+首先从replay buffer中选取一个batch的数据，计算loss，然后进行minibatch SGD。
+
+然后是保存与加载模型的部分，如下：
+
+```python
+def save(self, path):
+        torch.save(self.target_net.state_dict(), path+'dqn_checkpoint.pth')
+def load(self, path):
+    self.target_net.load_state_dict(torch.load(path+'dqn_checkpoint.pth'))
+    for target_param, param in zip(self.target_net.parameters(), self.policy_net.parameters()):
+        param.data.copy_(target_param.data)
+```
+
+
+
+### 实验结果
+
+训练结果如下：
+
+<img src="assets/train_rewards_curve.png" alt="train_rewards_curve" style="zoom: 67%;" />
+
+<img src="assets/eval_rewards_curve.png" alt="eval_rewards_curve" style="zoom:67%;" />
+
+## 参考
+
+[with torch.no_grad()](https://www.jianshu.com/p/1cea017f5d11)
+

projects/codes/DQN/dqn.py

@@ -0,0 +1,126 @@
+#!/usr/bin/env python
+# coding=utf-8
+'''
+@Author: John
+@Email: johnjim0816@gmail.com
+@Date: 2020-06-12 00:50:49
+@LastEditor: John
+LastEditTime: 2022-07-20 23:57:16
+@Discription: 
+@Environment: python 3.7.7
+'''
+'''off-policy
+'''
+
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+import torch.optim as optim
+import random
+import math
+import numpy as np
+
+class MLP(nn.Module):
+    def __init__(self, n_states,n_actions,hidden_dim=128):
+        """ 初始化q网络，为全连接网络
+            n_states: 输入的特征数即环境的状态维度
+            n_actions: 输出的动作维度
+        """
+        super(MLP, self).__init__()
+        self.fc1 = nn.Linear(n_states, hidden_dim) # 输入层
+        self.fc2 = nn.Linear(hidden_dim,hidden_dim) # 隐藏层
+        self.fc3 = nn.Linear(hidden_dim, n_actions) # 输出层
+        
+    def forward(self, x):
+        # 各层对应的激活函数
+        x = F.relu(self.fc1(x)) 
+        x = F.relu(self.fc2(x))
+        return self.fc3(x)
+
+class ReplayBuffer:
+    def __init__(self, capacity):
+        self.capacity = capacity # 经验回放的容量
+        self.buffer = [] # 缓冲区
+        self.position = 0 
+    
+    def push(self, state, action, reward, next_state, done):
+        ''' 缓冲区是一个队列，容量超出时去掉开始存入的转移(transition)
+        '''
+        if len(self.buffer) < self.capacity:
+            self.buffer.append(None)
+        self.buffer[self.position] = (state, action, reward, next_state, done)
+        self.position = (self.position + 1) % self.capacity 
+    
+    def sample(self, batch_size):
+        batch = random.sample(self.buffer, batch_size) # 随机采出小批量转移
+        state, action, reward, next_state, done =  zip(*batch) # 解压成状态，动作等
+        return state, action, reward, next_state, done
+    
+    def __len__(self):
+        ''' 返回当前存储的量
+        '''
+        return len(self.buffer)
+
+class DQN:
+    def __init__(self, n_states,n_actions,cfg):
+
+        self.n_actions = n_actions  
+        self.device = torch.device(cfg.device)  # cpu or cuda
+        self.gamma = cfg.gamma  # 奖励的折扣因子
+        # e-greedy策略相关参数
+        self.frame_idx = 0  # 用于epsilon的衰减计数
+        self.epsilon = lambda frame_idx: cfg.epsilon_end + \
+            (cfg.epsilon_start - cfg.epsilon_end) * \
+            math.exp(-1. * frame_idx / cfg.epsilon_decay)
+        self.batch_size = cfg.batch_size
+        self.policy_net = MLP(n_states,n_actions).to(self.device)
+        self.target_net = MLP(n_states,n_actions).to(self.device)
+        for target_param, param in zip(self.target_net.parameters(),self.policy_net.parameters()): # 复制参数到目标网路targe_net
+            target_param.data.copy_(param.data)
+        self.optimizer = optim.Adam(self.policy_net.parameters(), lr=cfg.lr) # 优化器
+        self.memory = ReplayBuffer(cfg.memory_capacity) # 经验回放
+
+    def choose_action(self, state):
+        ''' 选择动作
+        '''
+        self.frame_idx += 1
+        if random.random() > self.epsilon(self.frame_idx):
+            with torch.no_grad():
+                state = torch.tensor(state, device=self.device, dtype=torch.float32).unsqueeze(dim=0)
+                q_values = self.policy_net(state)
+                action = q_values.max(1)[1].item() # 选择Q值最大的动作
+        else:
+            action = random.randrange(self.n_actions)
+        return action
+    def update(self):
+        if len(self.memory) < self.batch_size: # 当memory中不满足一个批量时，不更新策略
+            return
+        # 从经验回放中(replay memory)中随机采样一个批量的转移(transition)
+        # print('updating')
+        
+        state_batch, action_batch, reward_batch, next_state_batch, done_batch = self.memory.sample(
+            self.batch_size)
+        state_batch = torch.tensor(np.array(state_batch), device=self.device, dtype=torch.float)
+        action_batch = torch.tensor(action_batch, device=self.device).unsqueeze(1)  
+        reward_batch = torch.tensor(reward_batch, device=self.device, dtype=torch.float)  
+        next_state_batch = torch.tensor(np.array(next_state_batch), device=self.device, dtype=torch.float)
+        done_batch = torch.tensor(np.float32(done_batch), device=self.device)
+        q_values = self.policy_net(state_batch).gather(dim=1, index=action_batch) # 计算当前状态(s_t,a)对应的Q(s_t, a)
+        next_q_values = self.target_net(next_state_batch).max(1)[0].detach() # 计算下一时刻的状态(s_t_,a)对应的Q值
+        # 计算期望的Q值，对于终止状态，此时done_batch[0]=1, 对应的expected_q_value等于reward
+        expected_q_values = reward_batch + self.gamma * next_q_values * (1-done_batch)
+        loss = nn.MSELoss()(q_values, expected_q_values.unsqueeze(1))  # 计算均方根损失
+        # 优化更新模型
+        self.optimizer.zero_grad()  
+        loss.backward()
+        for param in self.policy_net.parameters():  # clip防止梯度爆炸
+            param.grad.data.clamp_(-1, 1)
+        self.optimizer.step() 
+
+    def save(self, path):
+        torch.save(self.target_net.state_dict(), path+'dqn_checkpoint.pth')
+
+    def load(self, path):
+        self.target_net.load_state_dict(torch.load(path+'dqn_checkpoint.pth'))
+        for target_param, param in zip(self.target_net.parameters(), self.policy_net.parameters()):
+            param.data.copy_(target_param.data)

projects/codes/DQN/dqn_cnn.py

@@ -0,0 +1,134 @@
+import torch
+import torch.nn as nn
+import torch.optim as optim
+import torch.autograd as autograd 
+import random
+import math
+class CNN(nn.Module):
+    def __init__(self, input_dim, output_dim):
+        super(CNN, self).__init__()
+        
+        self.input_dim = input_dim
+        self.output_dim = output_dim
+        
+        self.features = nn.Sequential(
+            nn.Conv2d(input_dim[0], 32, kernel_size=8, stride=4),
+            nn.ReLU(),
+            nn.Conv2d(32, 64, kernel_size=4, stride=2),
+            nn.ReLU(),
+            nn.Conv2d(64, 64, kernel_size=3, stride=1),
+            nn.ReLU()
+        )
+        
+        self.fc = nn.Sequential(
+            nn.Linear(self.feature_size(), 512),
+            nn.ReLU(),
+            nn.Linear(512, self.output_dim)
+        )
+        
+    def forward(self, x):
+        x = self.features(x)
+        x = x.view(x.size(0), -1)
+        x = self.fc(x)
+        return x
+    
+    def feature_size(self):
+        return self.features(autograd.Variable(torch.zeros(1, *self.input_dim))).view(1, -1).size(1)
+
+
+    def act(self, state, epsilon):
+        if random.random() > epsilon:
+            state   = Variable(torch.FloatTensor(np.float32(state)).unsqueeze(0), volatile=True)
+            q_value = self.forward(state)
+            action  = q_value.max(1)[1].data[0]
+        else:
+            action = random.randrange(env.action_space.n)
+        return action
+
+class ReplayBuffer:
+    def __init__(self, capacity):
+        self.capacity = capacity # 经验回放的容量
+        self.buffer = [] # 缓冲区
+        self.position = 0 
+    
+    def push(self, state, action, reward, next_state, done):
+        ''' 缓冲区是一个队列，容量超出时去掉开始存入的转移(transition)
+        '''
+        if len(self.buffer) < self.capacity:
+            self.buffer.append(None)
+        self.buffer[self.position] = (state, action, reward, next_state, done)
+        self.position = (self.position + 1) % self.capacity 
+    
+    def sample(self, batch_size):
+        batch = random.sample(self.buffer, batch_size) # 随机采出小批量转移
+        state, action, reward, next_state, done =  zip(*batch) # 解压成状态，动作等
+        return state, action, reward, next_state, done
+    
+    def __len__(self):
+        ''' 返回当前存储的量
+        '''
+        return len(self.buffer)
+
+class DQN:
+    def __init__(self, n_states, n_actions, cfg):
+
+        self.n_actions = n_actions  # 总的动作个数
+        self.device = cfg.device  # 设备，cpu或gpu等
+        self.gamma = cfg.gamma  # 奖励的折扣因子
+        # e-greedy策略相关参数
+        self.frame_idx = 0  # 用于epsilon的衰减计数
+        self.epsilon = lambda frame_idx: cfg.epsilon_end + \
+            (cfg.epsilon_start - cfg.epsilon_end) * \
+            math.exp(-1. * frame_idx / cfg.epsilon_decay)
+        self.batch_size = cfg.batch_size
+        self.policy_net = CNN(n_states, n_actions).to(self.device)
+        self.target_net = CNN(n_states, n_actions).to(self.device)
+        for target_param, param in zip(self.target_net.parameters(),self.policy_net.parameters()): # 复制参数到目标网路targe_net
+            target_param.data.copy_(param.data)
+        self.optimizer = optim.Adam(self.policy_net.parameters(), lr=cfg.lr) # 优化器
+        self.memory = ReplayBuffer(cfg.memory_capacity) # 经验回放
+
+    def choose_action(self, state):
+        ''' 选择动作
+        '''
+        self.frame_idx += 1
+        if random.random() > self.epsilon(self.frame_idx):
+            with torch.no_grad():
+                print(type(state))
+                state = torch.tensor([state], device=self.device, dtype=torch.float32)
+                q_values = self.policy_net(state)
+                action = q_values.max(1)[1].item() # 选择Q值最大的动作
+        else:
+            action = random.randrange(self.n_actions)
+        return action
+    def update(self):
+        if len(self.memory) < self.batch_size: # 当memory中不满足一个批量时，不更新策略
+            return
+        # 从经验回放中(replay memory)中随机采样一个批量的转移(transition)
+        state_batch, action_batch, reward_batch, next_state_batch, done_batch = self.memory.sample(
+            self.batch_size)
+        # 转为张量
+        state_batch = torch.tensor(state_batch, device=self.device, dtype=torch.float)
+        action_batch = torch.tensor(action_batch, device=self.device).unsqueeze(1)  
+        reward_batch = torch.tensor(reward_batch, device=self.device, dtype=torch.float)  
+        next_state_batch = torch.tensor(next_state_batch, device=self.device, dtype=torch.float)
+        done_batch = torch.tensor(np.float32(done_batch), device=self.device)
+        q_values = self.policy_net(state_batch).gather(dim=1, index=action_batch) # 计算当前状态(s_t,a)对应的Q(s_t, a)
+        next_q_values = self.target_net(next_state_batch).max(1)[0].detach() # 计算下一时刻的状态(s_t_,a)对应的Q值
+        # 计算期望的Q值，对于终止状态，此时done_batch[0]=1, 对应的expected_q_value等于reward
+        expected_q_values = reward_batch + self.gamma * next_q_values * (1-done_batch)
+        loss = nn.MSELoss()(q_values, expected_q_values.unsqueeze(1))  # 计算均方根损失
+        # 优化更新模型
+        self.optimizer.zero_grad()  
+        loss.backward()
+        for param in self.policy_net.parameters():  # clip防止梯度爆炸
+            param.grad.data.clamp_(-1, 1)
+        self.optimizer.step() 
+
+    def save(self, path):
+        torch.save(self.target_net.state_dict(), path+'dqn_checkpoint.pth')
+
+    def load(self, path):
+        self.target_net.load_state_dict(torch.load(path+'dqn_checkpoint.pth'))
+        for target_param, param in zip(self.target_net.parameters(), self.policy_net.parameters()):
+            param.data.copy_(target_param.data)

projects/codes/DQN/dqn_cnn2.py

@@ -0,0 +1,142 @@
+import torch
+import torch.nn as nn
+import torch.optim as optim
+import torch.autograd as autograd 
+import random
+import math
+import numpy as np
+class CNN(nn.Module):
+    def __init__(self, n_frames, n_actions):
+        super(CNN,self).__init__()
+        self.n_frames = n_frames
+        self.n_actions = n_actions
+        
+        # Layers
+        self.conv1 = nn.Conv2d(
+            in_channels=n_frames,
+            out_channels=16,
+            kernel_size=8,
+            stride=4,
+            padding=2
+            )
+        self.conv2 = nn.Conv2d(
+            in_channels=16,
+            out_channels=32,
+            kernel_size=4,
+            stride=2,
+            padding=1
+            )
+        self.fc1 = nn.Linear(
+            in_features=3200,
+            out_features=256,
+            )
+        self.fc2 = nn.Linear(
+            in_features=256,
+            out_features=n_actions,
+            )
+        
+        # Activation Functions
+        self.relu = nn.ReLU()
+    
+    def flatten(self, x):
+        batch_size = x.size()[0]
+        x = x.view(batch_size, -1)
+        return x
+    
+    def forward(self, x):
+        
+        # Forward pass
+        x = self.relu(self.conv1(x))  # In: (80, 80, 4)  Out: (20, 20, 16)
+        x = self.relu(self.conv2(x))  # In: (20, 20, 16) Out: (10, 10, 32)
+        x = self.flatten(x)           # In: (10, 10, 32) Out: (3200,)
+        x = self.relu(self.fc1(x))    # In: (3200,)      Out: (256,)
+        x = self.fc2(x)               # In: (256,)       Out: (4,)
+        
+        return x
+
+class ReplayBuffer:
+    def __init__(self, capacity):
+        self.capacity = capacity # 经验回放的容量
+        self.buffer = [] # 缓冲区
+        self.position = 0 
+    
+    def push(self, state, action, reward, next_state, done):
+        ''' 缓冲区是一个队列，容量超出时去掉开始存入的转移(transition)
+        '''
+        if len(self.buffer) < self.capacity:
+            self.buffer.append(None)
+        self.buffer[self.position] = (state, action, reward, next_state, done)
+        self.position = (self.position + 1) % self.capacity 
+    
+    def sample(self, batch_size):
+        batch = random.sample(self.buffer, batch_size) # 随机采出小批量转移
+        state, action, reward, next_state, done =  zip(*batch) # 解压成状态，动作等
+        return state, action, reward, next_state, done
+    
+    def __len__(self):
+        ''' 返回当前存储的量
+        '''
+        return len(self.buffer)
+
+class DQN:
+    def __init__(self, n_states, n_actions, cfg):
+
+        self.n_actions = n_actions  # 总的动作个数
+        self.device = cfg.device  # 设备，cpu或gpu等
+        self.gamma = cfg.gamma  # 奖励的折扣因子
+        # e-greedy策略相关参数
+        self.frame_idx = 0  # 用于epsilon的衰减计数
+        self.epsilon = lambda frame_idx: cfg.epsilon_end + \
+            (cfg.epsilon_start - cfg.epsilon_end) * \
+            math.exp(-1. * frame_idx / cfg.epsilon_decay)
+        self.batch_size = cfg.batch_size
+        self.policy_net = CNN(n_states, n_actions).to(self.device)
+        self.target_net = CNN(n_states, n_actions).to(self.device)
+        for target_param, param in zip(self.target_net.parameters(),self.policy_net.parameters()): # 复制参数到目标网路targe_net
+            target_param.data.copy_(param.data)
+        self.optimizer = optim.Adam(self.policy_net.parameters(), lr=cfg.lr) # 优化器
+        self.memory = ReplayBuffer(cfg.memory_capacity) # 经验回放
+
+    def choose_action(self, state):
+        ''' 选择动作
+        '''
+        self.frame_idx += 1
+        if random.random() > self.epsilon(self.frame_idx):
+            with torch.no_grad():
+                state = torch.tensor([state], device=self.device, dtype=torch.float32)
+                q_values = self.policy_net(state)
+                action = q_values.max(1)[1].item() # 选择Q值最大的动作
+        else:
+            action = random.randrange(self.n_actions)
+        return action
+    def update(self):
+        if len(self.memory) < self.batch_size: # 当memory中不满足一个批量时，不更新策略
+            return
+        # 从经验回放中(replay memory)中随机采样一个批量的转移(transition)
+        state_batch, action_batch, reward_batch, next_state_batch, done_batch = self.memory.sample(
+            self.batch_size)
+        # 转为张量
+        state_batch = torch.tensor(state_batch, device=self.device, dtype=torch.float)
+        action_batch = torch.tensor(action_batch, device=self.device).unsqueeze(1)  
+        reward_batch = torch.tensor(reward_batch, device=self.device, dtype=torch.float)  
+        next_state_batch = torch.tensor(next_state_batch, device=self.device, dtype=torch.float)
+        done_batch = torch.tensor(np.float32(done_batch), device=self.device)
+        q_values = self.policy_net(state_batch).gather(dim=1, index=action_batch) # 计算当前状态(s_t,a)对应的Q(s_t, a)
+        next_q_values = self.target_net(next_state_batch).max(1)[0].detach() # 计算下一时刻的状态(s_t_,a)对应的Q值
+        # 计算期望的Q值，对于终止状态，此时done_batch[0]=1, 对应的expected_q_value等于reward
+        expected_q_values = reward_batch + self.gamma * next_q_values * (1-done_batch)
+        loss = nn.MSELoss()(q_values, expected_q_values.unsqueeze(1))  # 计算均方根损失
+        # 优化更新模型
+        self.optimizer.zero_grad()  
+        loss.backward()
+        for param in self.policy_net.parameters():  # clip防止梯度爆炸
+            param.grad.data.clamp_(-1, 1)
+        self.optimizer.step() 
+
+    def save(self, path):
+        torch.save(self.target_net.state_dict(), path+'dqn_checkpoint.pth')
+
+    def load(self, path):
+        self.target_net.load_state_dict(torch.load(path+'dqn_checkpoint.pth'))
+        for target_param, param in zip(self.target_net.parameters(), self.policy_net.parameters()):
+            param.data.copy_(target_param.data)

projects/codes/DQN/outputs/CartPole-v0/20220713-211653/results/params.json

@@ -0,0 +1,19 @@
+{
+    "algo_name": "DQN",
+    "env_name": "CartPole-v0",
+    "train_eps": 200,
+    "test_eps": 20,
+    "gamma": 0.95,
+    "epsilon_start": 0.95,
+    "epsilon_end": 0.01,
+    "epsilon_decay": 500,
+    "lr": 0.0001,
+    "memory_capacity": 100000,
+    "batch_size": 64,
+    "target_update": 4,
+    "hidden_dim": 256,
+    "deivce": "cpu",
+    "result_path": "C:\\Users\\24438\\Desktop\\rl-tutorials/outputs/CartPole-v0/20220713-211653/results/",
+    "model_path": "C:\\Users\\24438\\Desktop\\rl-tutorials/outputs/CartPole-v0/20220713-211653/models/",
+    "save_fig": true
+}

projects/codes/DQN/task0.py

@@ -0,0 +1,148 @@
+import sys,os
+curr_path = os.path.dirname(os.path.abspath(__file__))  # current path
+parent_path = os.path.dirname(curr_path)  # parent path
+sys.path.append(parent_path)  # add to system path
+import torch.nn as nn
+import torch.nn.functional as F
+
+import gym
+import torch
+import datetime
+import numpy as np
+import argparse
+from common.utils import save_results, make_dir
+from common.utils import plot_rewards,save_args
+from dqn import DQN
+
+def get_args():
+    """ Hyperparameters
+    """
+    curr_time = datetime.datetime.now().strftime("%Y%m%d-%H%M%S")  # Obtain current time
+    parser = argparse.ArgumentParser(description="hyperparameters")      
+    parser.add_argument('--algo_name',default='DQN',type=str,help="name of algorithm")
+    parser.add_argument('--env_name',default='CartPole-v0',type=str,help="name of environment")
+    parser.add_argument('--train_eps',default=200,type=int,help="episodes of training")
+    parser.add_argument('--test_eps',default=20,type=int,help="episodes of testing")
+    parser.add_argument('--gamma',default=0.95,type=float,help="discounted factor")
+    parser.add_argument('--epsilon_start',default=0.95,type=float,help="initial value of epsilon")
+    parser.add_argument('--epsilon_end',default=0.01,type=float,help="final value of epsilon")
+    parser.add_argument('--epsilon_decay',default=500,type=int,help="decay rate of epsilon")
+    parser.add_argument('--lr',default=0.0001,type=float,help="learning rate")
+    parser.add_argument('--memory_capacity',default=100000,type=int,help="memory capacity")
+    parser.add_argument('--batch_size',default=64,type=int)
+    parser.add_argument('--target_update',default=4,type=int)
+    parser.add_argument('--hidden_dim',default=256,type=int)
+    parser.add_argument('--device',default='cpu',type=str,help="cpu or cuda") 
+    parser.add_argument('--result_path',default=curr_path + "/outputs/" + parser.parse_args().env_name + \
+            '/' + curr_time + '/results/' )
+    parser.add_argument('--model_path',default=curr_path + "/outputs/" + parser.parse_args().env_name + \
+            '/' + curr_time + '/models/' ) # path to save models
+    parser.add_argument('--save_fig',default=True,type=bool,help="if save figure or not")           
+    args = parser.parse_args()                          
+    return args
+
+def env_agent_config(cfg,seed=1):
+    ''' 创建环境和智能体
+    '''
+    env = gym.make(cfg.env_name)  # 创建环境
+    n_states = env.observation_space.shape[0]  # 状态维度
+    n_actions = env.action_space.n  # 动作维度
+    print(f"n states: {n_states}, n actions: {n_actions}")
+    agent = DQN(n_states,n_actions, cfg)  # 创建智能体
+    if seed !=0: # 设置随机种子
+        torch.manual_seed(seed)
+        env.seed(seed)
+        np.random.seed(seed)
+    return env, agent
+
+def train(cfg, env, agent):
+    ''' Training
+    '''
+    print('Start training!')
+    print(f'Env:{cfg.env_name}, A{cfg.algo_name}, 设备：{cfg.device}')
+    rewards = []  # 记录所有回合的奖励
+    ma_rewards = []  # 记录所有回合的滑动平均奖励
+    steps = []
+    for i_ep in range(cfg.train_eps):
+        ep_reward = 0  # 记录一回合内的奖励
+        ep_step = 0
+        state = env.reset()  # 重置环境，返回初始状态
+        while True:
+            ep_step += 1
+            action = agent.choose_action(state)  # 选择动作
+            next_state, reward, done, _ = env.step(action)  # 更新环境，返回transition
+            agent.memory.push(state, action, reward,
+                              next_state, done)  # 保存transition
+            state = next_state  # 更新下一个状态
+            agent.update()  # 更新智能体
+            ep_reward += reward  # 累加奖励
+            if done:
+                break
+        if (i_ep + 1) % cfg.target_update == 0:  # 智能体目标网络更新
+            agent.target_net.load_state_dict(agent.policy_net.state_dict())
+        steps.append(ep_step)
+        rewards.append(ep_reward)
+        if ma_rewards:
+            ma_rewards.append(0.9 * ma_rewards[-1] + 0.1 * ep_reward)
+        else:
+            ma_rewards.append(ep_reward)
+        if (i_ep + 1) % 1 == 0:
+            print(f'Episode：{i_ep+1}/{cfg.train_eps}, Reward:{ep_reward:.2f}, Step:{ep_step:.2f} Epislon:{agent.epsilon(agent.frame_idx):.3f}')
+    print('Finish training!')
+    env.close()
+    res_dic = {'rewards':rewards,'ma_rewards':ma_rewards,'steps':steps}
+    return res_dic
+
+
+def test(cfg, env, agent):
+    print('Start testing!')
+    print(f'Env:{cfg.env_name}, A{cfg.algo_name}, 设备：{cfg.device}')
+    ############# 由于测试不需要使用epsilon-greedy策略，所以相应的值设置为0 ###############
+    cfg.epsilon_start = 0.0  # e-greedy策略中初始epsilon
+    cfg.epsilon_end = 0.0  # e-greedy策略中的终止epsilon
+    ################################################################################
+    rewards = []  # 记录所有回合的奖励
+    ma_rewards = []  # 记录所有回合的滑动平均奖励
+    steps = []
+    for i_ep in range(cfg.test_eps):
+        ep_reward = 0  # 记录一回合内的奖励
+        ep_step = 0
+        state = env.reset()  # 重置环境，返回初始状态
+        while True:
+            ep_step+=1
+            action = agent.choose_action(state)  # 选择动作
+            next_state, reward, done, _ = env.step(action)  # 更新环境，返回transition
+            state = next_state  # 更新下一个状态
+            ep_reward += reward  # 累加奖励
+            if done:
+                break
+        steps.append(ep_step)
+        rewards.append(ep_reward)
+        if ma_rewards:
+            ma_rewards.append(ma_rewards[-1] * 0.9 + ep_reward * 0.1)
+        else:
+            ma_rewards.append(ep_reward)
+        print(f'Episode：{i_ep+1}/{cfg.test_eps}, Reward:{ep_reward:.2f}, Step:{ep_step:.2f}')
+    print('Finish testing')
+    env.close()
+    return {'rewards':rewards,'ma_rewards':ma_rewards,'steps':steps}
+
+
+if __name__ == "__main__":
+    cfg = get_args()
+    # 训练
+    env, agent = env_agent_config(cfg)
+    res_dic = train(cfg, env, agent)
+    make_dir(cfg.result_path, cfg.model_path)  
+    save_args(cfg) # save parameters
+    agent.save(path=cfg.model_path)  # save model
+    save_results(res_dic, tag='train',
+                 path=cfg.result_path)  
+    plot_rewards(res_dic['rewards'], res_dic['ma_rewards'], cfg, tag="train")  
+    # 测试
+    env, agent = env_agent_config(cfg)
+    agent.load(path=cfg.model_path)  # 导入模型
+    res_dic = test(cfg, env, agent)
+    save_results(res_dic, tag='test',
+                 path=cfg.result_path)  # 保存结果
+    plot_rewards(res_dic['rewards'], res_dic['ma_rewards'],cfg, tag="test")  # 画出结果