update rainbowdqn

codes/DDPG/ddpg.py

@@ -39,11 +39,11 @@ class ReplayBuffer:
         '''
         return len(self.buffer)
 class Actor(nn.Module):
-    def __init__(self, state_dim, action_dim, hidden_dim, init_w=3e-3):
+    def __init__(self, n_states, n_actions, hidden_dim, init_w=3e-3):
         super(Actor, self).__init__()  
-        self.linear1 = nn.Linear(state_dim, hidden_dim)
+        self.linear1 = nn.Linear(n_states, hidden_dim)
         self.linear2 = nn.Linear(hidden_dim, hidden_dim)
-        self.linear3 = nn.Linear(hidden_dim, action_dim)
+        self.linear3 = nn.Linear(hidden_dim, n_actions)
         
         self.linear3.weight.data.uniform_(-init_w, init_w)
         self.linear3.bias.data.uniform_(-init_w, init_w)
@@ -54,10 +54,10 @@ class Actor(nn.Module):
         x = torch.tanh(self.linear3(x))
         return x
 class Critic(nn.Module):
-    def __init__(self, state_dim, action_dim, hidden_dim, init_w=3e-3):
+    def __init__(self, n_states, n_actions, hidden_dim, init_w=3e-3):
         super(Critic, self).__init__()
         
-        self.linear1 = nn.Linear(state_dim + action_dim, hidden_dim)
+        self.linear1 = nn.Linear(n_states + n_actions, hidden_dim)
         self.linear2 = nn.Linear(hidden_dim, hidden_dim)
         self.linear3 = nn.Linear(hidden_dim, 1)
         # 随机初始化为较小的值
@@ -72,12 +72,12 @@ class Critic(nn.Module):
         x = self.linear3(x)
         return x
 class DDPG:
-    def __init__(self, state_dim, action_dim, cfg):
+    def __init__(self, n_states, n_actions, cfg):
         self.device = cfg.device
-        self.critic = Critic(state_dim, action_dim, cfg.hidden_dim).to(cfg.device)
-        self.actor = Actor(state_dim, action_dim, cfg.hidden_dim).to(cfg.device)
-        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)
+        self.critic = Critic(n_states, n_actions, cfg.hidden_dim).to(cfg.device)
+        self.actor = Actor(n_states, n_actions, cfg.hidden_dim).to(cfg.device)
+        self.target_critic = Critic(n_states, n_actions, cfg.hidden_dim).to(cfg.device)
+        self.target_actor = Actor(n_states, n_actions, cfg.hidden_dim).to(cfg.device)
 
         # 复制参数到目标网络
         for target_param, param in zip(self.target_critic.parameters(), self.critic.parameters()):

codes/DDPG/env.py

@@ -39,15 +39,15 @@ class OUNoise(object):
         self.max_sigma    = max_sigma
         self.min_sigma    = min_sigma
         self.decay_period = decay_period
-        self.action_dim   = action_space.shape[0]
+        self.n_actions   = action_space.shape[0]
         self.low          = action_space.low
         self.high         = action_space.high
         self.reset()
     def reset(self):
-        self.obs = np.ones(self.action_dim) * self.mu
+        self.obs = np.ones(self.n_actions) * self.mu
     def evolve_obs(self):
         x  = self.obs
-        dx = self.theta * (self.mu - x) + self.sigma * np.random.randn(self.action_dim)
+        dx = self.theta * (self.mu - x) + self.sigma * np.random.randn(self.n_actions)
         self.obs = x + dx
         return self.obs
     def get_action(self, action, t=0):

codes/DDPG/task0.py

@@ -5,7 +5,7 @@
 @Email: johnjim0816@gmail.com
 @Date: 2020-06-11 20:58:21
 @LastEditor: John
-LastEditTime: 2021-09-16 01:31:33
+LastEditTime: 2022-02-10 06:23:27
 @Discription: 
 @Environment: python 3.7.7
 '''
@@ -18,23 +18,29 @@ import datetime
 import gym
 import torch
 
-from DDPG.env import NormalizedActions
-from DDPG.agent import DDPG
+from env import NormalizedActions,OUNoise
+from ddpg import DDPG
 from DDPG.train import train,test
 from common.utils import save_results,make_dir
 from common.utils import plot_rewards
 
 curr_time = datetime.datetime.now().strftime("%Y%m%d-%H%M%S")  # 获取当前时间
-algo_name = 'DDPG'  # 算法名称
-env_name = 'Pendulum-v1'  # 环境名称，gym新版本（约0.21.0之后）中Pendulum-v0改为Pendulum-v1
+class Config:
+    '''超参数
+    '''
 
-class DDPGConfig:
     def __init__(self):
-        self.algo_name = algo_name # 算法名称
-        self.env_name = env_name # 环境名称
-        self.device = torch.device("cuda" if torch.cuda.is_available() else "cpu") # 检测GPU
+        ################################## 环境超参数 ###################################
+        self.algo_name = 'DDPG'  # 算法名称
+        self.env_name = 'Pendulum-v1'  # 环境名称，gym新版本（约0.21.0之后）中Pendulum-v0改为Pendulum-v1
+        self.device = torch.device(
+            "cuda" if torch.cuda.is_available() else "cpu")  # 检测GPUgjgjlkhfsf风刀霜的撒发十
+        self.seed = 10 # 随机种子，置0则不设置随机种子
         self.train_eps = 300 # 训练的回合数
         self.test_eps = 50 # 测试的回合数
+        ################################################################################
+        
+        ################################## 算法超参数 ###################################
         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
@@ -43,39 +49,92 @@ class DDPGConfig:
         self.target_update = 2 # 目标网络的更新频率
         self.hidden_dim = 256 # 网络隐藏层维度
         self.soft_tau = 1e-2 # 软更新参数
+        ################################################################################
         
-class PlotConfig:
-    def __init__(self) -> None:
-        self.algo_name = algo_name  # 算法名称
-        self.env_name = env_name # 环境名称
-        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
+        ################################# 保存结果相关参数 ################################
+        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
-        self.device = torch.device("cuda" if torch.cuda.is_available() else "cpu")  # 检测GPU
+        ################################################################################
 
 def env_agent_config(cfg,seed=1):
     env = NormalizedActions(gym.make(cfg.env_name)) # 装饰action噪声
     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)
+    n_states = env.observation_space.shape[0]
+    n_actions = env.action_space.shape[0]
+    agent = DDPG(n_states,n_actions,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}')
+    ou_noise = OUNoise(env.action_space)  # 动作噪声
+    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.train_eps):
+        state = env.reset()
+        ou_noise.reset()
+        done = False
+        ep_reward = 0
+        i_step = 0
+        while not done:
+            i_step += 1
+            action = agent.choose_action(state)
+            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
+        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('完成训练！')
+    return rewards, ma_rewards
 
-cfg = DDPGConfig()
-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)
-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")  # 画出结果
-# 测试
-env,agent = env_agent_config(cfg,seed=10)
-agent.load(path=plot_cfg.model_path)
-rewards,ma_rewards = test(plot_cfg,env,agent)
-save_results(rewards,ma_rewards,tag = 'test',path = cfg.result_path)
-plot_rewards(rewards, ma_rewards, plot_cfg, tag="test")  # 画出结果
+def test(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.test_eps):
+        state = env.reset() 
+        done = False
+        ep_reward = 0
+        i_step = 0
+        while not done:
+            i_step += 1
+            action = agent.choose_action(state)
+            next_state, reward, done, _ = env.step(action)
+            ep_reward += reward
+            state = next_state
+        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(f"回合：{i_ep+1}/{cfg.test_eps}，奖励：{ep_reward:.1f}")
+    print('完成测试！')
+    return rewards, ma_rewards
+if __name__ == "__main__":
+    cfg = Config()
+    # 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, cfg, tag="train")  # 画出结果
+    # 测试
+    env,agent = env_agent_config(cfg,seed=10)
+    agent.load(path=cfg.model_path)
+    rewards,ma_rewards = test(cfg,env,agent)
+    save_results(rewards,ma_rewards,tag = 'test',path = cfg.result_path)
+    plot_rewards(rewards, ma_rewards, cfg, tag="test")  # 画出结果
 

codes/DDPG/train.py

@@ -1,64 +0,0 @@
-import sys
-import 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
-
-from DDPG.env import OUNoise
-
-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}')
-    ou_noise = OUNoise(env.action_space)  # 动作噪声
-    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.train_eps):
-        state = env.reset()
-        ou_noise.reset()
-        done = False
-        ep_reward = 0
-        i_step = 0
-        while not done:
-            i_step += 1
-            action = agent.choose_action(state)
-            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
-        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('完成训练！')
-    return rewards, ma_rewards
-
-def test(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.test_eps):
-        state = env.reset() 
-        done = False
-        ep_reward = 0
-        i_step = 0
-        while not done:
-            i_step += 1
-            action = agent.choose_action(state)
-            next_state, reward, done, _ = env.step(action)
-            ep_reward += reward
-            state = next_state
-        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(f"回合：{i_ep+1}/{cfg.test_eps}，奖励：{ep_reward:.1f}")
-    print('完成测试！')
-    return rewards, ma_rewards