hot update

projects/codes/DQN/dqn.py

@@ -5,7 +5,7 @@
 @Email: johnjim0816@gmail.com
 @Date: 2020-06-12 00:50:49
 @LastEditor: John
-LastEditTime: 2022-08-11 09:52:23
+LastEditTime: 2022-08-18 14:27:18
 @Discription: 
 @Environment: python 3.7.7
 '''
@@ -23,10 +23,10 @@ class DQN:
     def __init__(self,n_actions,model,memory,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.sample_count = 0  # 用于epsilon的衰减计数
+        self.device = torch.device(cfg.device) 
+        self.gamma = cfg.gamma  
+        ## e-greedy parameters
+        self.sample_count = 0  # sample count for epsilon decay
         self.epsilon = cfg.epsilon_start
         self.sample_count = 0  
         self.epsilon_start = cfg.epsilon_start
@@ -35,61 +35,78 @@ class DQN:
         self.batch_size = cfg.batch_size
         self.policy_net = model.to(self.device)
         self.target_net = model.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
+        ## copy parameters from policy net to target net
+        for target_param, param in zip(self.target_net.parameters(),self.policy_net.parameters()): 
             target_param.data.copy_(param.data)
-        self.optimizer = optim.Adam(self.policy_net.parameters(), lr=cfg.lr) # 优化器
-        self.memory = memory # 经验回放
+        # self.target_net.load_state_dict(self.policy_net.state_dict()) # or use this to copy parameters
+        self.optimizer = optim.Adam(self.policy_net.parameters(), lr=cfg.lr) 
+        self.memory = memory 
+        self.update_flag = False 
 
-    def sample(self, state):
-        ''' 选择动作
+    def sample_action(self, state):
+        ''' sample action with e-greedy policy
         '''
         self.sample_count += 1
+        # epsilon must decay(linear,exponential and etc.) for balancing exploration and exploitation
         self.epsilon = self.epsilon_end + (self.epsilon_start - self.epsilon_end) * \
-            math.exp(-1. * self.sample_count / self.epsilon_decay) # epsilon是会递减的，这里选择指数递减
+            math.exp(-1. * self.sample_count / self.epsilon_decay) 
         if random.random() > self.epsilon:
             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
+                action = q_values.max(1)[1].item() # choose action corresponding to the maximum q value
         else:
             action = random.randrange(self.n_actions)
         return action
-    def predict(self,state):
+    def predict_action(self,state):
+        ''' predict action
+        '''
         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
+            action = q_values.max(1)[1].item() # choose action corresponding to the maximum q value
         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
+        if len(self.memory) < self.batch_size: # when transitions in memory donot meet a batch, not update
             return
-        # 从经验回放中(replay memory)中随机采样一个批量的转移(transition)
-        
+        else:
+            if not self.update_flag:
+                print("begin to update!")
+                self.update_flag = True
+        # sample a batch of transitions from replay buffer
         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
+        state_batch = torch.tensor(np.array(state_batch), device=self.device, dtype=torch.float) # shape(batchsize,n_states)
+        action_batch = torch.tensor(action_batch, device=self.device).unsqueeze(1) # shape(batchsize,1)
+        reward_batch = torch.tensor(reward_batch, device=self.device, dtype=torch.float).unsqueeze(1) # shape(batchsize)
+        next_state_batch = torch.tensor(np.array(next_state_batch), device=self.device, dtype=torch.float) # shape(batchsize,n_states)
+        done_batch = torch.tensor(np.float32(done_batch), device=self.device).unsqueeze(1) # shape(batchsize,1)
+        # print(state_batch.shape,action_batch.shape,reward_batch.shape,next_state_batch.shape,done_batch.shape)
+        # compute current Q(s_t,a), it is 'y_j' in pseucodes
+        q_value_batch = self.policy_net(state_batch).gather(dim=1, index=action_batch) # shape(batchsize,1),requires_grad=True
+        # print(q_values.requires_grad)
+        # compute max(Q(s_t+1,A_t+1)) respects to actions A, next_max_q_value comes from another net and is just regarded as constant for q update formula below, thus should detach to requires_grad=False
+        next_max_q_value_batch = self.target_net(next_state_batch).max(1)[0].detach().unsqueeze(1) 
+        # print(q_values.shape,next_q_values.shape)
+        # compute expected q value, for terminal state, done_batch[0]=1, and expected_q_value=rewardcorrespondingly
+        expected_q_value_batch = reward_batch + self.gamma * next_max_q_value_batch* (1-done_batch)
+        # print(expected_q_value_batch.shape,expected_q_value_batch.requires_grad)
+        loss = nn.MSELoss()(q_value_batch, expected_q_value_batch)  # shape same to 
+        # backpropagation
         self.optimizer.zero_grad()  
         loss.backward()
-        for param in self.policy_net.parameters():  # clip防止梯度爆炸
+        # clip to avoid gradient explosion
+        for param in self.policy_net.parameters():  
             param.grad.data.clamp_(-1, 1)
         self.optimizer.step() 
 
-    def save(self, path):
+    def save_model(self, path):
         from pathlib import Path
+        # create path
         Path(path).mkdir(parents=True, exist_ok=True)
-        torch.save(self.target_net.state_dict(), path+'checkpoint.pth')
+        torch.save(self.target_net.state_dict(), f"{path}/checkpoint.pt")
 
-    def load(self, path):
-        self.target_net.load_state_dict(torch.load(path+'checkpoint.pth'))
+    def load_model(self, path):
+        self.target_net.load_state_dict(torch.load(f"{path}/checkpoint.pt"))
         for target_param, param in zip(self.target_net.parameters(), self.policy_net.parameters()):
             param.data.copy_(target_param.data)