update codes

codes/PPO/agent.py

@@ -18,6 +18,7 @@ from PPO.memory import PPOMemory
 class PPO:
     def __init__(self, state_dim, action_dim,cfg):
         self.gamma = cfg.gamma
+        self.continuous = cfg.continuous 
         self.policy_clip = cfg.policy_clip
         self.n_epochs = cfg.n_epochs
         self.gae_lambda = cfg.gae_lambda
@@ -29,13 +30,13 @@ class PPO:
         self.memory = PPOMemory(cfg.batch_size)
         self.loss = 0
 
-    def choose_action(self, state,continuous=False):
+    def choose_action(self, state):
         state = torch.tensor([state], dtype=torch.float).to(self.device)
         dist = self.actor(state)
         value = self.critic(state)
         action = dist.sample()
         probs = torch.squeeze(dist.log_prob(action)).item()
-        if continuous:
+        if self.continuous:
             action = torch.tanh(action)
         else:
             action = torch.squeeze(action).item()

codes/PPO/task0.py

@@ -0,0 +1,67 @@
+import sys,os
+curr_path = os.path.dirname(os.path.abspath(__file__)) # 当前文件所在绝对路径
+parent_path = os.path.dirname(curr_path) # 父路径
+sys.path.append(parent_path) # 添加路径到系统路径
+
+import gym
+import torch
+import datetime
+from common.plot import plot_rewards
+from common.utils import save_results,make_dir
+from PPO.agent import PPO
+from PPO.train import train
+
+curr_time = datetime.datetime.now().strftime("%Y%m%d-%H%M%S")  # 获取当前时间
+
+class PPOConfig:
+    def __init__(self) -> None:
+        self.algo = "DQN"  # 算法名称
+        self.env_name = 'CartPole-v0' # 环境名称
+        self.continuous = False # 环境是否为连续动作
+        self.device = torch.device("cuda" if torch.cuda.is_available() else "cpu")  # 检测GPU
+        self.train_eps = 200 # 训练的回合数
+        self.eval_eps = 20 # 测试的回合数
+        self.batch_size = 5
+        self.gamma=0.99
+        self.n_epochs = 4
+        self.actor_lr = 0.0003
+        self.critic_lr = 0.0003
+        self.gae_lambda=0.95
+        self.policy_clip=0.2
+        self.hidden_dim = 256
+        self.update_fre = 20 # frequency of agent update
+
+class PlotConfig:
+    def __init__(self) -> None:
+        self.algo = "DQN"  # 算法名称
+        self.env_name = 'CartPole-v0' # 环境名称
+        self.device = torch.device("cuda" if torch.cuda.is_available() else "cpu")  # 检测GPU
+        self.result_path = curr_path+"/outputs/" + self.env_name + \
+            '/'+curr_time+'/results/'  # 保存结果的路径
+        self.model_path = curr_path+"/outputs/" + self.env_name + \
+            '/'+curr_time+'/models/'  # 保存模型的路径
+        self.save = True # 是否保存图片
+
+def env_agent_config(cfg,seed=1):
+    env = gym.make(cfg.env_name)  
+    env.seed(seed)
+    state_dim = env.observation_space.shape[0]
+    action_dim = env.action_space.n
+    agent = PPO(state_dim,action_dim,cfg)
+    return env,agent
+
+cfg  = PPOConfig()
+plot_cfg = PlotConfig()
+# 训练
+env,agent = env_agent_config(cfg,seed=1)
+rewards, ma_rewards = train(cfg, env, agent)
+make_dir(plot_cfg.result_path, plot_cfg.model_path) # 创建保存结果和模型路径的文件夹
+agent.save(path=plot_cfg.model_path)
+save_results(rewards, ma_rewards, tag='train', path=plot_cfg.result_path)
+plot_rewards(rewards, ma_rewards, plot_cfg, tag="train")
+# 测试
+env,agent = env_agent_config(cfg,seed=10)
+agent.load(path=plot_cfg.model_path)
+rewards,ma_rewards = eval(cfg,env,agent)
+save_results(rewards,ma_rewards,tag='eval',path=plot_cfg.result_path)
+plot_rewards(rewards,ma_rewards,plot_cfg,tag="eval")

codes/PPO/task1.py

@@ -0,0 +1,68 @@
+import sys,os
+curr_path = os.path.dirname(os.path.abspath(__file__)) # 当前文件所在绝对路径
+parent_path = os.path.dirname(curr_path) # 父路径
+sys.path.append(parent_path) # 添加路径到系统路径
+
+import gym
+import torch
+import datetime
+from common.plot import plot_rewards
+from common.utils import save_results,make_dir
+from PPO.agent import PPO
+from PPO.train import train
+
+curr_time = datetime.datetime.now().strftime("%Y%m%d-%H%M%S")  # 获取当前时间
+
+class PPOConfig:
+    def __init__(self) -> None:
+        self.algo = "PPO"  # 算法名称
+        self.env_name = 'Pendulum-v1' # 环境名称
+        self.continuous = True # 环境是否为连续动作
+        self.device = torch.device("cuda" if torch.cuda.is_available() else "cpu")  # 检测GPU
+        self.train_eps = 200 # 训练的回合数
+        self.eval_eps = 20 # 测试的回合数
+        self.batch_size = 5
+        self.gamma=0.99
+        self.n_epochs = 4
+        self.actor_lr = 0.0003
+        self.critic_lr = 0.0003
+        self.gae_lambda=0.95
+        self.policy_clip=0.2
+        self.hidden_dim = 256
+        self.update_fre = 20 # frequency of agent update
+
+class PlotConfig:
+    def __init__(self) -> None:
+        self.algo = "PPO"  # 算法名称
+        self.env_name = 'Pendulum-v1' # 环境名称
+        self.device = torch.device("cuda" if torch.cuda.is_available() else "cpu")  # 检测GPU
+        self.result_path = curr_path+"/outputs/" + self.env_name + \
+            '/'+curr_time+'/results/'  # 保存结果的路径
+        self.model_path = curr_path+"/outputs/" + self.env_name + \
+            '/'+curr_time+'/models/'  # 保存模型的路径
+        self.save = True # 是否保存图片
+        
+def env_agent_config(cfg,seed=1):
+    env = gym.make(cfg.env_name)  
+    env.seed(seed)
+    state_dim = env.observation_space.shape[0]
+    action_dim = env.action_space.shape[0]
+    agent = PPO(state_dim,action_dim,cfg)
+    return env,agent
+
+
+cfg  = PPOConfig()
+plot_cfg = PlotConfig()
+# 训练
+env,agent = env_agent_config(cfg,seed=1)
+rewards, ma_rewards = train(cfg, env, agent)
+make_dir(plot_cfg.result_path, plot_cfg.model_path) # 创建保存结果和模型路径的文件夹
+agent.save(path=plot_cfg.model_path)
+save_results(rewards, ma_rewards, tag='train', path=plot_cfg.result_path)
+plot_rewards(rewards, ma_rewards, plot_cfg, tag="train")
+# 测试
+env,agent = env_agent_config(cfg,seed=10)
+agent.load(path=plot_cfg.model_path)
+rewards,ma_rewards = eval(cfg,env,agent)
+save_results(rewards,ma_rewards,tag='eval',path=plot_cfg.result_path)
+plot_rewards(rewards,ma_rewards,plot_cfg,tag="eval")

codes/PPO/task1_train.py

@@ -1,132 +0,0 @@
-#!/usr/bin/env python
-# coding=utf-8
-'''
-Author: John
-Email: johnjim0816@gmail.com
-Date: 2021-03-22 16:18:10
-LastEditor: John
-LastEditTime: 2021-09-26 22:05:00
-Discription: 
-Environment: 
-'''
-import sys,os
-curr_path = os.path.dirname(os.path.abspath(__file__)) # 当前文件所在绝对路径
-parent_path = os.path.dirname(curr_path) # 父路径
-sys.path.append(parent_path) # 添加路径到系统路径
-
-import gym
-import torch
-import datetime
-from PPO.agent import PPO
-from common.plot import plot_rewards
-from common.utils import save_results,make_dir
-
-curr_time = datetime.datetime.now().strftime("%Y%m%d-%H%M%S") # obtain current time
-
-class PPOConfig:
-    def __init__(self) -> None:
-        self.algo = "PPO"  # 算法名称
-        self.env_name = 'Pendulum-v1' # 环境名称
-        self.continuous = True # 环境是否为连续动作
-        self.device = torch.device("cuda" if torch.cuda.is_available() else "cpu")  # 检测GPU
-        self.train_eps = 200 # 训练的回合数
-        self.eval_eps = 20 # 测试的回合数
-        self.batch_size = 5
-        self.gamma=0.99
-        self.n_epochs = 4
-        self.actor_lr = 0.0003
-        self.critic_lr = 0.0003
-        self.gae_lambda=0.95
-        self.policy_clip=0.2
-        self.hidden_dim = 256
-        self.update_fre = 20 # frequency of agent update
-
-class PlotConfig:
-    def __init__(self) -> None:
-        self.algo = "PPO"  # 算法名称
-        self.env_name = 'Pendulum-v1' # 环境名称
-        self.device = torch.device("cuda" if torch.cuda.is_available() else "cpu")  # 检测GPU
-        self.result_path = curr_path+"/outputs/" + self.env_name + \
-            '/'+curr_time+'/results/'  # 保存结果的路径
-        self.model_path = curr_path+"/outputs/" + self.env_name + \
-            '/'+curr_time+'/models/'  # 保存模型的路径
-        self.save = True # 是否保存图片
-        
-def env_agent_config(cfg,seed=1):
-    env = gym.make(cfg.env_name)  
-    env.seed(seed)
-    state_dim = env.observation_space.shape[0]
-    action_dim = env.action_space.shape[0]
-    agent = PPO(state_dim,action_dim,cfg)
-    return env,agent
-            
-def train(cfg,env,agent):
-    print('开始训练！')
-    print(f'环境：{cfg.env_name}, 算法：{cfg.algo}, 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 = 0
-    for i_ep in range(cfg.train_eps):
-        state = env.reset()
-        done = False
-        ep_reward = 0
-        while not done:
-            action, prob, val = agent.choose_action(state,continuous=cfg.continuous)
-            print(action)
-            state_, reward, done, _ = env.step(action)
-            steps += 1
-            ep_reward += reward
-            agent.memory.push(state, action, prob, val, reward, done)
-            if steps % cfg.update_fre == 0:
-                agent.update()
-            state = state_
-        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)%10 == 0: 
-            print(f"回合：{i_ep+1}/{cfg.train_eps}，奖励：{ep_reward:.2f}")
-    print('完成训练！')
-    return rewards,ma_rewards
-
-def eval(cfg,env,agent):
-    print('开始测试!')
-    print(f'环境：{cfg.env_name}, 算法：{cfg.algo}, 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
-    for i_ep in range(cfg.eval_eps):
-        state = env.reset()
-        done = False
-        ep_reward = 0
-        while not done:
-            action, prob, val = agent.choose_action(state,continuous=False)
-            state_, reward, done, _ = env.step(action)
-            ep_reward += reward
-            state = state_
-        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)
-        print('回合：{}/{}, 奖励：{}'.format(i_ep+1, cfg.eval_eps, ep_reward))
-    print('完成训练！')
-    return rewards,ma_rewards
-    
-if __name__ == '__main__':
-    cfg  = PPOConfig()
-    plot_cfg = PlotConfig()
-    # 训练
-    env,agent = env_agent_config(cfg,seed=1)
-    rewards, ma_rewards = train(cfg, env, agent)
-    make_dir(plot_cfg.result_path, plot_cfg.model_path) # 创建保存结果和模型路径的文件夹
-    agent.save(path=plot_cfg.model_path)
-    save_results(rewards, ma_rewards, tag='train', path=plot_cfg.result_path)
-    plot_rewards(rewards, ma_rewards, plot_cfg, tag="train")
-    # 测试
-    env,agent = env_agent_config(cfg,seed=10)
-    agent.load(path=plot_cfg.model_path)
-    rewards,ma_rewards = eval(cfg,env,agent)
-    save_results(rewards,ma_rewards,tag='eval',path=plot_cfg.result_path)
-    plot_rewards(rewards,ma_rewards,plot_cfg,tag="eval")

codes/PPO/train.py

@@ -1,65 +1,3 @@
-#!/usr/bin/env python
-# coding=utf-8
-'''
-Author: John
-Email: johnjim0816@gmail.com
-Date: 2021-03-22 16:18:10
-LastEditor: John
-LastEditTime: 2021-09-26 22:05:00
-Discription: 
-Environment: 
-'''
-import sys,os
-curr_path = os.path.dirname(os.path.abspath(__file__)) # 当前文件所在绝对路径
-parent_path = os.path.dirname(curr_path) # 父路径
-sys.path.append(parent_path) # 添加路径到系统路径
-
-import gym
-import torch
-import datetime
-from PPO.agent import PPO
-from common.plot import plot_rewards
-from common.utils import save_results,make_dir
-
-curr_time = datetime.datetime.now().strftime("%Y%m%d-%H%M%S") # obtain current time
-
-class PPOConfig:
-    def __init__(self) -> None:
-        self.algo = "DQN"  # 算法名称
-        self.env_name = 'CartPole-v0' # 环境名称
-        self.continuous = False # 环境是否为连续动作
-        self.device = torch.device("cuda" if torch.cuda.is_available() else "cpu")  # 检测GPU
-        self.train_eps = 200 # 训练的回合数
-        self.eval_eps = 20 # 测试的回合数
-        self.batch_size = 5
-        self.gamma=0.99
-        self.n_epochs = 4
-        self.actor_lr = 0.0003
-        self.critic_lr = 0.0003
-        self.gae_lambda=0.95
-        self.policy_clip=0.2
-        self.hidden_dim = 256
-        self.update_fre = 20 # frequency of agent update
-
-class PlotConfig:
-    def __init__(self) -> None:
-        self.algo = "DQN"  # 算法名称
-        self.env_name = 'CartPole-v0' # 环境名称
-        self.device = torch.device("cuda" if torch.cuda.is_available() else "cpu")  # 检测GPU
-        self.result_path = curr_path+"/outputs/" + self.env_name + \
-            '/'+curr_time+'/results/'  # 保存结果的路径
-        self.model_path = curr_path+"/outputs/" + self.env_name + \
-            '/'+curr_time+'/models/'  # 保存模型的路径
-        self.save = True # 是否保存图片
-        
-def env_agent_config(cfg,seed=1):
-    env = gym.make(cfg.env_name)  
-    env.seed(seed)
-    state_dim = env.observation_space.shape[0]
-    action_dim = env.action_space.n
-    agent = PPO(state_dim,action_dim,cfg)
-    return env,agent
-            
 def train(cfg,env,agent):
     print('开始训练！')
     print(f'环境：{cfg.env_name}, 算法：{cfg.algo}, 设备：{cfg.device}')
@@ -71,7 +9,7 @@ def train(cfg,env,agent):
         done = False
         ep_reward = 0
         while not done:
-            action, prob, val = agent.choose_action(state,continuous=cfg.continuous)
+            action, prob, val = agent.choose_action(state)
             state_, reward, done, _ = env.step(action)
             steps += 1
             ep_reward += reward
@@ -99,7 +37,7 @@ def eval(cfg,env,agent):
         done = False
         ep_reward = 0
         while not done:
-            action, prob, val = agent.choose_action(state,cfg.continuous)
+            action, prob, val = agent.choose_action(state)
             state_, reward, done, _ = env.step(action)
             ep_reward += reward
             state = state_
@@ -112,8 +50,60 @@ def eval(cfg,env,agent):
         print('回合：{}/{}, 奖励：{}'.format(i_ep+1, cfg.eval_eps, ep_reward))
     print('完成训练！')
     return rewards,ma_rewards
-    
+
 if __name__ == '__main__':
+    import sys,os
+    curr_path = os.path.dirname(os.path.abspath(__file__)) # 当前文件所在绝对路径
+    parent_path = os.path.dirname(curr_path) # 父路径
+    sys.path.append(parent_path) # 添加路径到系统路径
+
+    import gym
+    import torch
+    import datetime
+    from common.plot import plot_rewards
+    from common.utils import save_results,make_dir
+    from PPO.agent import PPO
+    from PPO.train import train
+
+    curr_time = datetime.datetime.now().strftime("%Y%m%d-%H%M%S")  # 获取当前时间
+
+    class PPOConfig:
+        def __init__(self) -> None:
+            self.algo = "DQN"  # 算法名称
+            self.env_name = 'CartPole-v0' # 环境名称
+            self.continuous = False # 环境是否为连续动作
+            self.device = torch.device("cuda" if torch.cuda.is_available() else "cpu")  # 检测GPU
+            self.train_eps = 200 # 训练的回合数
+            self.eval_eps = 20 # 测试的回合数
+            self.batch_size = 5
+            self.gamma=0.99
+            self.n_epochs = 4
+            self.actor_lr = 0.0003
+            self.critic_lr = 0.0003
+            self.gae_lambda=0.95
+            self.policy_clip=0.2
+            self.hidden_dim = 256
+            self.update_fre = 20 # frequency of agent update
+
+    class PlotConfig:
+        def __init__(self) -> None:
+            self.algo = "DQN"  # 算法名称
+            self.env_name = 'CartPole-v0' # 环境名称
+            self.device = torch.device("cuda" if torch.cuda.is_available() else "cpu")  # 检测GPU
+            self.result_path = curr_path+"/outputs/" + self.env_name + \
+                '/'+curr_time+'/results/'  # 保存结果的路径
+            self.model_path = curr_path+"/outputs/" + self.env_name + \
+                '/'+curr_time+'/models/'  # 保存模型的路径
+            self.save = True # 是否保存图片
+
+    def env_agent_config(cfg,seed=1):
+        env = gym.make(cfg.env_name)  
+        env.seed(seed)
+        state_dim = env.observation_space.shape[0]
+        action_dim = env.action_space.n
+        agent = PPO(state_dim,action_dim,cfg)
+        return env,agent
+
     cfg  = PPOConfig()
     plot_cfg = PlotConfig()
     # 训练
@@ -128,4 +118,4 @@ if __name__ == '__main__':
     agent.load(path=plot_cfg.model_path)
     rewards,ma_rewards = eval(cfg,env,agent)
     save_results(rewards,ma_rewards,tag='eval',path=plot_cfg.result_path)
-    plot_rewards(rewards,ma_rewards,plot_cfg,tag="eval")
+    plot_rewards(rewards,ma_rewards,plot_cfg,tag="eval")

codes/README.md

@@ -1,6 +1,3 @@
-
-[Eng](https://github.com/JohnJim0816/reinforcement-learning-tutorials/blob/master/README_en.md)|[中文](https://github.com/JohnJim0816/reinforcement-learning-tutorials/blob/master/README.md)
-
 ## 写在前面
 
 本项目用于学习RL基础算法，尽量做到: **注释详细**，**结构清晰**。
@@ -12,7 +9,7 @@
 * ```plot.py``` 利用matplotlib或seaborn绘制rewards图，包括滑动平均的reward，结果保存在result文件夹中
 * ```env.py``` 用于构建强化学习环境，也可以重新自定义环境，比如给action加noise
 * ```agent.py``` RL核心算法，比如dqn等，主要包含update和choose_action两个方法，
-* ```main.py``` 运行主函数
+* ```train.py``` 保存用于训练和测试的函数
 
 其中```model.py```,```memory.py```,```plot.py``` 由于不同算法都会用到，所以放入```common```文件夹中。
 
@@ -22,8 +19,8 @@ python 3.7、pytorch 1.6.0-1.8.1、gym 0.17.0-0.19.0
 
 ## 使用说明
 
-运行带有```train```的py文件或ipynb文件进行训练，如果前面带有```task```如```task0_train.py```，表示对task0任务训练，
+直接运行带有```train```的py文件或ipynb文件会进行训练默认的任务；  
-类似的带有```eval```即为测试。
+也可以运行带有```task```的py文件训练不同的任务
 
 ## 内容导航
 

codes/SAC/task0_train.py

@@ -10,10 +10,9 @@ Discription:
 Environment: 
 '''
 import sys,os
-curr_path = os.path.dirname(__file__)
+curr_path = os.path.dirname(os.path.abspath(__file__)) # 当前文件所在绝对路径
-parent_path = os.path.dirname(curr_path)
+parent_path = os.path.dirname(curr_path) # 父路径
-sys.path.append(parent_path)  # add current terminal path to sys.path
+sys.path.append(parent_path) # 添加路径到系统路径
-
 
 import gym
 import torch
@@ -24,7 +23,7 @@ from SAC.agent import SAC
 from common.utils import save_results, make_dir
 from common.plot import plot_rewards
 
-curr_time = datetime.datetime.now().strftime("%Y%m%d-%H%M%S") # obtain current time
+curr_time = datetime.datetime.now().strftime("%Y%m%d-%H%M%S")  # 获取当前时间
 
 class SACConfig:
     def __init__(self) -> None:
@@ -48,6 +47,14 @@ class SACConfig:
         self.hidden_dim = 256
         self.batch_size  = 128
         self.device=torch.device("cuda" if torch.cuda.is_available() else "cpu")
+class PlotConfig(SACConfig):
+	def __init__(self) -> None:
+		super().__init__()
+		self.result_path = curr_path+"/outputs/" + self.env_name + \
+            '/'+curr_time+'/results/'  # 保存结果的路径
+		self.model_path = curr_path+"/outputs/" + self.env_name + \
+            '/'+curr_time+'/models/'  # 保存模型的路径
+		self.save = True # 是否保存图片
 
 def env_agent_config(cfg,seed=1):
     env = NormalizedActions(gym.make(cfg.env_name))
@@ -58,13 +65,13 @@ def env_agent_config(cfg,seed=1):
     return env,agent
 
 def train(cfg,env,agent):
-    print('Start to train !')
+    print('开始训练!')
-    print(f'Env: {cfg.env_name}, Algorithm: {cfg.algo}, Device: {cfg.device}')
+    print(f'环境：{cfg.env_name}, 算法：{cfg.algo}, 设备：{cfg.device}')
-    rewards  = []
+    rewards = [] # 记录所有回合的奖励
-    ma_rewards = [] # moveing average reward
+    ma_rewards = []  # 记录所有回合的滑动平均奖励
     for i_ep in range(cfg.train_eps):
-        state = env.reset()
+        ep_reward = 0 # 记录一回合内的奖励
-        ep_reward = 0
+        state = env.reset() # 重置环境，返回初始状态
         for i_step in range(cfg.train_steps):
             action = agent.policy_net.get_action(state)
             next_state, reward, done, _ = env.step(action)
@@ -111,21 +118,20 @@ def eval(cfg,env,agent):
 
 if __name__ == "__main__":
     cfg=SACConfig()
-    
+    plot_cfg = PlotConfig()
     # train
     env,agent = env_agent_config(cfg,seed=1)
     rewards, ma_rewards = train(cfg, env, agent)
-    make_dir(cfg.result_path, cfg.model_path)
+    make_dir(plot_cfg.result_path, plot_cfg.model_path)
-    agent.save(path=cfg.model_path)
+    agent.save(path=plot_cfg.model_path)
-    save_results(rewards, ma_rewards, tag='train', path=cfg.result_path)
+    save_results(rewards, ma_rewards, tag='train', path=plot_cfg.result_path)
-    plot_rewards(rewards, ma_rewards, tag="train",
+    plot_rewards(rewards, ma_rewards, plot_cfg, tag="train")
-                 algo=cfg.algo, path=cfg.result_path)
     # eval
     env,agent = env_agent_config(cfg,seed=10)
-    agent.load(path=cfg.model_path)
+    agent.load(path=plot_cfg.model_path)
     rewards,ma_rewards = eval(cfg,env,agent)
-    save_results(rewards,ma_rewards,tag='eval',path=cfg.result_path)
+    save_results(rewards,ma_rewards,tag='eval',path=plot_cfg.result_path)
-    plot_rewards(rewards,ma_rewards,tag="eval",env=cfg.env,algo = cfg.algo,path=cfg.result_path)
+    plot_rewards(rewards,ma_rewards,plot_cfg,tag="eval")
 
 
 

codes/TD3/task1_train.py

@@ -1,41 +1,47 @@
 import sys,os
-curr_path = os.path.dirname(__file__)
+curr_path = os.path.dirname(os.path.abspath(__file__)) # 当前文件所在绝对路径
-parent_path=os.path.dirname(curr_path) 
+parent_path = os.path.dirname(curr_path) # 父路径
-sys.path.append(parent_path) # add current terminal path to sys.path
+sys.path.append(parent_path) # 添加路径到系统路径
 
 import torch
 import gym
 import numpy as np
 import datetime
 
-
 from TD3.agent import TD3
 from common.plot import plot_rewards
 from common.utils import save_results,make_dir
 
-curr_time = datetime.datetime.now().strftime("%Y%m%d-%H%M%S") # obtain current time
+curr_time = datetime.datetime.now().strftime("%Y%m%d-%H%M%S")  # 获取当前时间
 	
 
 class TD3Config:
 	def __init__(self) -> None:
-		self.algo = 'TD3'
+		self.algo = 'TD3' # 算法名称
-		self.env = 'Pendulum-v0'
+		self.env_name = 'Pendulum-v1' # 环境名称
-		self.seed = 0
+		self.device = torch.device("cuda" if torch.cuda.is_available() else "cpu")  # 检测GPU
-		self.result_path = curr_path+"/results/" +self.env+'/'+curr_time+'/results/'  # path to save results
+		self.train_eps = 600 # 训练的回合数
-		self.model_path = curr_path+"/results/" +self.env+'/'+curr_time+'/models/'  # path to save models
 		self.start_timestep = 25e3 # Time steps initial random policy is used
-		self.start_ep = 50 # Episodes initial random policy is used
+		self.epsilon_start = 50 # Episodes initial random policy is used
 		self.eval_freq = 10 # How often (episodes) we evaluate
-		self.train_eps = 600
 		self.max_timestep = 100000 # Max time steps to run environment
 		self.expl_noise = 0.1 # Std of Gaussian exploration noise
 		self.batch_size = 256 # Batch size for both actor and critic
 		self.gamma = 0.9 # gamma factor
-		self.lr = 0.0005 # Target network update rate 
+		self.lr = 0.0005 # 学习率
 		self.policy_noise = 0.2 # Noise added to target policy during critic update
 		self.noise_clip = 0.3  # Range to clip target policy noise
 		self.policy_freq = 2 # Frequency of delayed policy updates
-		self.device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+class PlotConfig(TD3Config):
+	def __init__(self) -> None:
+		super().__init__()
+		self.result_path = curr_path+"/outputs/" + self.env_name + \
+            '/'+curr_time+'/results/'  # 保存结果的路径
+		self.model_path = curr_path+"/outputs/" + self.env_name + \
+            '/'+curr_time+'/models/'  # 保存模型的路径
+		self.save = True # 是否保存图片
+		
+    
 
 # Runs policy for X episodes and returns average reward
 # A fixed seed is used for the eval environment
@@ -57,8 +63,10 @@ def eval(env,agent, seed, eval_episodes=10):
 	return avg_reward
 
 def train(cfg,env,agent):
-	rewards = []
+	print('开始训练!')
-	ma_rewards = [] # moveing average reward
+	print(f'环境：{cfg.env_name}, 算法：{cfg.algo}, 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
 	for i_ep in range(int(cfg.train_eps)):
 		ep_reward = 0
 		ep_timesteps = 0
@@ -66,7 +74,7 @@ def train(cfg,env,agent):
 		while not done:
 			ep_timesteps += 1
 			# Select action randomly or according to policy
-			if i_ep < cfg.start_ep:
+			if i_ep < cfg.epsilon_start:
 				action = env.action_space.sample()
 			else:
 				action = (
@@ -81,32 +89,34 @@ def train(cfg,env,agent):
 			state = next_state
 			ep_reward += reward
 			# Train agent after collecting sufficient data
-			if i_ep+1 >= cfg.start_ep:
+			if i_ep+1 >= cfg.epsilon_start:
 				agent.update()
-		print(f"Episode:{i_ep+1}/{cfg.train_eps}, Step:{ep_timesteps}, Reward:{ep_reward:.3f}")
+		if (i_ep+1)%10 == 0: 
+			print('回合：{}/{}, 奖励：{:.2f}'.format(i_ep+1, cfg.train_eps, ep_reward))
 		rewards.append(ep_reward)
-		# 计算滑动窗口的reward
 		if ma_rewards:
 			ma_rewards.append(0.9*ma_rewards[-1]+0.1*ep_reward)
 		else:
-			ma_rewards.append(ep_reward) 	
+			ma_rewards.append(ep_reward) 
+	print('完成训练！')	
 	return rewards, ma_rewards
 
 
 if __name__ == "__main__":
 	cfg  = TD3Config()
-	env = gym.make(cfg.env)
+	plot_cfg = PlotConfig()
-	env.seed(cfg.seed) # Set seeds
+	env = gym.make(cfg.env_name)
-	torch.manual_seed(cfg.seed)
+	env.seed(1) # 随机种子
-	np.random.seed(cfg.seed)
+	torch.manual_seed(1)
+	np.random.seed(1)
 	state_dim = env.observation_space.shape[0]
 	action_dim = env.action_space.shape[0] 
 	max_action = float(env.action_space.high[0])
 	agent = TD3(state_dim,action_dim,max_action,cfg)
 	rewards,ma_rewards = train(cfg,env,agent)
-	make_dir(cfg.result_path,cfg.model_path)
+	make_dir(plot_cfg.result_path,plot_cfg.model_path)
-	agent.save(path=cfg.model_path)
+	agent.save(path=plot_cfg.model_path)
-	save_results(rewards,ma_rewards,tag='train',path=cfg.result_path)
+	save_results(rewards,ma_rewards,tag='train',path=plot_cfg.result_path)
-	plot_rewards(rewards,ma_rewards,tag="train",env=cfg.env,algo = cfg.algo,path=cfg.result_path)
+	plot_rewards(rewards,ma_rewards,plot_cfg,tag="train")