update

codes/DDPG/agent.py

@@ -5,7 +5,7 @@
 @Email: johnjim0816@gmail.com
 @Date: 2020-06-09 20:25:52
 @LastEditor: John
-LastEditTime: 2021-05-04 14:50:17
+LastEditTime: 2021-09-16 00:55:30
 @Discription: 
 @Environment: python 3.7.7
 '''
@@ -26,7 +26,7 @@ class DDPG:
         self.target_critic = Critic(state_dim, action_dim, cfg.hidden_dim).to(cfg.device)
         self.target_actor = Actor(state_dim, action_dim, cfg.hidden_dim).to(cfg.device)
 
-        # copy parameters to target net
+        # 复制参数到目标网络
         for target_param, param in zip(self.target_critic.parameters(), self.critic.parameters()):
             target_param.data.copy_(param.data)
         for target_param, param in zip(self.target_actor.parameters(), self.actor.parameters()):
@@ -37,7 +37,7 @@ class DDPG:
         self.actor_optimizer = optim.Adam(self.actor.parameters(), lr=cfg.actor_lr)
         self.memory = ReplayBuffer(cfg.memory_capacity)
         self.batch_size = cfg.batch_size
-        self.soft_tau = cfg.soft_tau
+        self.soft_tau = cfg.soft_tau # 软更新参数
         self.gamma = cfg.gamma
 
     def choose_action(self, state):
@@ -46,11 +46,11 @@ class DDPG:
         return action.detach().cpu().numpy()[0, 0]
 
     def update(self):
-        if len(self.memory) < self.batch_size:
+        if len(self.memory) < self.batch_size: # 当 memory 中不满足一个批量时，不更新策略
             return
-        state, action, reward, next_state, done = self.memory.sample(
-            self.batch_size)
-        # convert variables to Tensor
+        # 从经验回放中(replay memory)中随机采样一个批量的转移(transition)
+        state, action, reward, next_state, done = self.memory.sample(self.batch_size)
+        # 转变为张量
         state = torch.FloatTensor(state).to(self.device)
         next_state = torch.FloatTensor(next_state).to(self.device)
         action = torch.FloatTensor(action).to(self.device)
@@ -70,10 +70,10 @@ class DDPG:
         self.actor_optimizer.zero_grad()
         policy_loss.backward()
         self.actor_optimizer.step()
-
         self.critic_optimizer.zero_grad()
         value_loss.backward()
         self.critic_optimizer.step()
+        # 软更新
         for target_param, param in zip(self.target_critic.parameters(), self.critic.parameters()):
             target_param.data.copy_(
                 target_param.data * (1.0 - self.soft_tau) +

codes/DDPG/env.py

@@ -5,7 +5,7 @@
 @Email: johnjim0816@gmail.com
 @Date: 2020-06-10 15:28:30
 @LastEditor: John
-LastEditTime: 2021-03-19 19:56:46
+LastEditTime: 2021-09-16 00:52:30
 @Discription: 
 @Environment: python 3.7.7
 '''
@@ -32,12 +32,12 @@ class NormalizedActions(gym.ActionWrapper):
         return action
 
 class OUNoise(object):
-    '''Ornstein–Uhlenbeck
+    '''Ornstein–Uhlenbeck噪声
     '''
     def __init__(self, action_space, mu=0.0, theta=0.15, max_sigma=0.3, min_sigma=0.3, decay_period=100000):
-        self.mu           = mu
-        self.theta        = theta
-        self.sigma        = max_sigma
+        self.mu           = mu # OU噪声的参数
+        self.theta        = theta # OU噪声的参数
+        self.sigma        = max_sigma # OU噪声的参数
         self.max_sigma    = max_sigma
         self.min_sigma    = min_sigma
         self.decay_period = decay_period
@@ -45,17 +45,14 @@ class OUNoise(object):
         self.low          = action_space.low
         self.high         = action_space.high
         self.reset()
-        
     def reset(self):
         self.obs = np.ones(self.action_dim) * self.mu
-        
     def evolve_obs(self):
         x  = self.obs
         dx = self.theta * (self.mu - x) + self.sigma * np.random.randn(self.action_dim)
         self.obs = x + dx
         return self.obs
-    
     def get_action(self, action, t=0):
         ou_obs = self.evolve_obs()
-        self.sigma = self.max_sigma - (self.max_sigma - self.min_sigma) * min(1.0, t / self.decay_period)
-        return np.clip(action + ou_obs, self.low, self.high)
+        self.sigma = self.max_sigma - (self.max_sigma - self.min_sigma) * min(1.0, t / self.decay_period) # sigma会逐渐衰减
+        return np.clip(action + ou_obs, self.low, self.high) # 动作加上噪声后进行剪切

codes/DDPG/task0_train.py

@@ -5,14 +5,14 @@
 @Email: johnjim0816@gmail.com
 @Date: 2020-06-11 20:58:21
 @LastEditor: John
-LastEditTime: 2021-05-04 14:49:45
+LastEditTime: 2021-09-16 01:31:33
 @Discription: 
 @Environment: python 3.7.7
 '''
 import sys,os
-curr_path = os.path.dirname(__file__)
-parent_path = os.path.dirname(curr_path)
-sys.path.append(parent_path)  # add current terminal path to sys.path
+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) # 添加父路径到系统路径sys.path
 
 import datetime
 import gym
@@ -21,49 +21,45 @@ import torch
 from DDPG.env import NormalizedActions, OUNoise
 from DDPG.agent import DDPG
 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
+from common.plot import plot_rewards, plot_rewards_cn
 
+curr_time = datetime.datetime.now().strftime("%Y%m%d-%H%M%S")  # 获取当前时间
 
 class DDPGConfig:
     def __init__(self):
-        self.algo = 'DDPG'
-        self.env = 'Pendulum-v0' # env name
+        self.algo = 'DDPG' # 算法名称
+        self.env = 'Pendulum-v0' # 环境名称
         self.result_path = curr_path+"/outputs/" + self.env + \
-            '/'+curr_time+'/results/'  # path to save results
+            '/'+curr_time+'/results/'  # 保存结果的路径
         self.model_path = curr_path+"/outputs/" + self.env + \
-            '/'+curr_time+'/models/'  # path to save results
-        self.gamma = 0.99
-        self.critic_lr = 1e-3
-        self.actor_lr = 1e-4
-        self.memory_capacity = 10000
+            '/'+curr_time+'/models/'  # 保存模型的路径
+        self.train_eps = 300 # 训练的回合数
+        self.eval_eps = 50 # 测试的回合数
+        self.gamma = 0.99 # 折扣因子
+        self.critic_lr = 1e-3 # 评论家网络的学习率
+        self.actor_lr = 1e-4 # 演员网络的学习率
+        self.memory_capacity = 8000 
         self.batch_size = 128
-        self.train_eps = 300
-        self.eval_eps = 50
-        self.eval_steps = 200
-        self.target_update = 4
-        self.hidden_dim = 30
-        self.soft_tau = 1e-2
-        self.device = torch.device(
-            "cuda" if torch.cuda.is_available() else "cpu")
+        self.target_update = 2
+        self.hidden_dim = 256
+        self.soft_tau = 1e-2 # 软更新参数
+        self.device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
 
 def env_agent_config(cfg,seed=1):
     env = NormalizedActions(gym.make(cfg.env)) 
-    env.seed(seed)
+    env.seed(seed) # 随机种子
     state_dim = env.observation_space.shape[0]
     action_dim = env.action_space.shape[0]
     agent = DDPG(state_dim,action_dim,cfg)
     return env,agent
 
 def train(cfg, env, agent):
-    print('Start to train ! ')
-    print(f'Env:{cfg.env}, Algorithm:{cfg.algo}, Device:{cfg.device}')
-    ou_noise = OUNoise(env.action_space)  # action noise
-    rewards = []
-    ma_rewards = []  # moving average rewards
-    for i_episode in range(cfg.train_eps):
+    print('开始训练！')
+    print(f'环境：{cfg.env}，算法：{cfg.algo}，设备：{cfg.device}')
+    ou_noise = OUNoise(env.action_space)  # 动作噪声
+    rewards = [] # 记录奖励
+    ma_rewards = []  # 记录滑动平均奖励
+    for i_ep in range(cfg.train_eps):
         state = env.reset()
         ou_noise.reset()
         done = False
@@ -72,29 +68,29 @@ def train(cfg, env, agent):
         while not done:
             i_step += 1
             action = agent.choose_action(state)
-            action = ou_noise.get_action(
-                action, i_step)  # 即paper中的random process
+            action = ou_noise.get_action(action, i_step) 
             next_state, reward, done, _ = env.step(action)
             ep_reward += reward
             agent.memory.push(state, action, reward, next_state, done)
             agent.update()
             state = next_state
-        print('Episode:{}/{}, Reward:{}'.format(i_episode+1, cfg.train_eps, ep_reward))
+        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)
         if ma_rewards:
             ma_rewards.append(0.9*ma_rewards[-1]+0.1*ep_reward)
         else:
             ma_rewards.append(ep_reward)
-    print('Complete training！')
+    print('完成训练！')
     return rewards, ma_rewards
 
 def eval(cfg, env, agent):
-    print('Start to Eval ! ')
-    print(f'Env:{cfg.env}, Algorithm:{cfg.algo}, Device:{cfg.device}')
-    rewards = []
-    ma_rewards = []  # moving average rewards
-    for i_episode in range(cfg.eval_eps):
-        state = env.reset()
+    print('开始测试！')
+    print(f'环境：{cfg.env}, 算法：{cfg.algo}, 设备：{cfg.device}')
+    rewards = [] # 记录奖励
+    ma_rewards = []  # 记录滑动平均奖励
+    for i_ep in range(cfg.eval_eps):
+        state = env.reset() 
         done = False
         ep_reward = 0
         i_step = 0
@@ -104,32 +100,29 @@ def eval(cfg, env, agent):
             next_state, reward, done, _ = env.step(action)
             ep_reward += reward
             state = next_state
-        print('Episode:{}/{}, Reward:{}'.format(i_episode+1, cfg.train_eps, ep_reward))
+        print('回合：{}/{}, 奖励：{}'.format(i_ep+1, cfg.train_eps, ep_reward))
         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('Complete Eval！')
+    print('完成测试！')
     return rewards, ma_rewards
 
 
 if __name__ == "__main__":
     cfg = DDPGConfig()
-
-    # train
+    # 训练
     env,agent = env_agent_config(cfg,seed=1)
     rewards, ma_rewards = train(cfg, env, agent)
     make_dir(cfg.result_path, cfg.model_path)
     agent.save(path=cfg.model_path)
     save_results(rewards, ma_rewards, tag='train', path=cfg.result_path)
-    plot_rewards(rewards, ma_rewards, tag="train",
-                 algo=cfg.algo, path=cfg.result_path)
-    
-    # eval
+    plot_rewards_cn(rewards, ma_rewards, tag="train", env = cfg.env, algo=cfg.algo, path=cfg.result_path)
+    # 测试
     env,agent = env_agent_config(cfg,seed=10)
     agent.load(path=cfg.model_path)
     rewards,ma_rewards = eval(cfg,env,agent)
-    save_results(rewards,ma_rewards,tag='eval',path=cfg.result_path)
-    plot_rewards(rewards,ma_rewards,tag="eval",env=cfg.env,algo = cfg.algo,path=cfg.result_path)
+    save_results(rewards,ma_rewards,tag = 'eval',path = cfg.result_path)
+    plot_rewards_cn(rewards,ma_rewards,tag = "eval",env = cfg.env,algo = cfg.algo,path=cfg.result_path)