add DDPG

codes/DDPG/README.md

@@ -0,0 +1,5 @@
+# DDPG
+
+## 伪代码
+
+![image-20210320151900695](assets/image-20210320151900695.png)

codes/DDPG/agent.py

@@ -0,0 +1,93 @@
+#!/usr/bin/env python
+# coding=utf-8
+'''
+@Author: John
+@Email: johnjim0816@gmail.com
+@Date: 2020-06-09 20:25:52
+@LastEditor: John
+LastEditTime: 2021-03-17 20:43:25
+@Discription: 
+@Environment: python 3.7.7
+'''
+import numpy as np
+import torch
+import torch.nn as nn
+import torch.optim as optim
+
+from common.model import Actor, Critic
+from common.memory import ReplayBuffer
+
+
+class DDPG:
+    def __init__(self, n_states, n_actions, cfg):
+        self.device = 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)
+
+        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()):
+            target_param.data.copy_(param.data)
+
+        self.critic_optimizer = optim.Adam(
+            self.critic.parameters(),  lr=cfg.critic_lr)
+        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.gamma = cfg.gamma
+
+    def choose_action(self, state):
+        state = torch.FloatTensor(state).unsqueeze(0).to(self.device)
+        action = self.actor(state)
+        # torch.detach()用于切断反向传播
+        return action.detach().cpu().numpy()[0, 0]
+
+    def update(self):
+        if len(self.memory) < self.batch_size:
+            return
+        state, action, reward, next_state, done = self.memory.sample(
+            self.batch_size)
+        # 将所有变量转为张量
+        state = torch.FloatTensor(state).to(self.device)
+        next_state = torch.FloatTensor(next_state).to(self.device)
+        action = torch.FloatTensor(action).to(self.device)
+        reward = torch.FloatTensor(reward).unsqueeze(1).to(self.device)
+        done = torch.FloatTensor(np.float32(done)).unsqueeze(1).to(self.device)
+        # 注意critic将(s_t,a)作为输入
+        policy_loss = self.critic(state, self.actor(state))
+        
+        policy_loss = -policy_loss.mean()
+
+        next_action = self.target_actor(next_state)
+        target_value = self.target_critic(next_state, next_action.detach())
+        expected_value = reward + (1.0 - done) * self.gamma * target_value
+        expected_value = torch.clamp(expected_value, -np.inf, np.inf)
+
+        value = self.critic(state, action)
+        value_loss = nn.MSELoss()(value, expected_value.detach())
+        
+        self.actor_optimizer.zero_grad()
+        policy_loss.backward()
+        self.actor_optimizer.step()
+
+        self.critic_optimizer.zero_grad()
+        value_loss.backward()
+        self.critic_optimizer.step()
+        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) +
+                param.data * self.soft_tau
+            )
+        for target_param, param in zip(self.target_actor.parameters(), self.actor.parameters()):
+            target_param.data.copy_(
+                target_param.data * (1.0 - self.soft_tau) +
+                param.data * self.soft_tau
+            )
+    def save(self,path):
+        torch.save(self.target_net.state_dict(), path+'DDPG_checkpoint.pth')
+
+    def load(self,path):
+        self.actor.load_state_dict(torch.load(path+'DDPG_checkpoint.pth')) 

codes/DDPG/env.py

@@ -0,0 +1,61 @@
+#!/usr/bin/env python
+# coding=utf-8
+'''
+@Author: John
+@Email: johnjim0816@gmail.com
+@Date: 2020-06-10 15:28:30
+@LastEditor: John
+LastEditTime: 2021-03-19 19:56:46
+@Discription: 
+@Environment: python 3.7.7
+'''
+import gym
+import numpy as np
+
+class NormalizedActions(gym.ActionWrapper):
+    ''' 将action范围重定在[0.1]之间
+    '''
+    def action(self, action):
+        
+        low_bound   = self.action_space.low
+        upper_bound = self.action_space.high
+        action = low_bound + (action + 1.0) * 0.5 * (upper_bound - low_bound)
+        action = np.clip(action, low_bound, upper_bound)
+        
+        return action
+
+    def reverse_action(self, action):
+        low_bound   = self.action_space.low
+        upper_bound = self.action_space.high
+        action = 2 * (action - low_bound) / (upper_bound - low_bound) - 1
+        action = np.clip(action, low_bound, upper_bound)
+        return action
+
+class OUNoise(object):
+    '''Ornstein–Uhlenbeck
+    '''
+    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.max_sigma    = max_sigma
+        self.min_sigma    = min_sigma
+        self.decay_period = decay_period
+        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.n_actions) * self.mu
+        
+    def evolve_obs(self):
+        x  = self.obs
+        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):
+        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)

codes/DDPG/main.py

@@ -0,0 +1,92 @@
+#!/usr/bin/env python
+# coding=utf-8
+'''
+@Author: John
+@Email: johnjim0816@gmail.com
+@Date: 2020-06-11 20:58:21
+@LastEditor: John
+LastEditTime: 2021-03-19 19:57:00
+@Discription: 
+@Environment: python 3.7.7
+'''
+import sys,os
+sys.path.append(os.getcwd()) # 添加当前终端路径
+import torch
+import gym
+import numpy as np
+import datetime
+from DDPG.agent import DDPG
+from DDPG.env import NormalizedActions,OUNoise
+from common.plot import plot_rewards
+from common.utils import save_results
+
+SEQUENCE = datetime.datetime.now().strftime("%Y%m%d-%H%M%S") # 获取当前时间
+SAVED_MODEL_PATH = os.path.split(os.path.abspath(__file__))[0]+"/saved_model/"+SEQUENCE+'/' # 生成保存的模型路径
+if not os.path.exists(os.path.split(os.path.abspath(__file__))[0]+"/saved_model/"): # 检测是否存在文件夹
+    os.mkdir(os.path.split(os.path.abspath(__file__))[0]+"/saved_model/")
+if not os.path.exists(SAVED_MODEL_PATH): # 检测是否存在文件夹
+    os.mkdir(SAVED_MODEL_PATH)
+RESULT_PATH = os.path.split(os.path.abspath(__file__))[0]+"/results/"+SEQUENCE+'/' # 存储reward的路径
+if not os.path.exists(os.path.split(os.path.abspath(__file__))[0]+"/results/"): # 检测是否存在文件夹
+    os.mkdir(os.path.split(os.path.abspath(__file__))[0]+"/results/")
+if not os.path.exists(RESULT_PATH): # 检测是否存在文件夹
+    os.mkdir(RESULT_PATH)
+
+class DDPGConfig:
+    def __init__(self):
+        self.gamma = 0.99
+        self.critic_lr = 1e-3  
+        self.actor_lr = 1e-4 
+        self.memory_capacity = 10000
+        self.batch_size = 128
+        self.train_eps =300
+        self.train_steps = 200
+        self.eval_eps = 200
+        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")
+def train(cfg,env,agent):
+    print('Start to train ! ')
+    ou_noise = OUNoise(env.action_space) # action noise
+    rewards = []
+    ma_rewards = [] # moving average rewards
+    ep_steps = []
+    for i_episode in range(cfg.train_eps):
+        state = env.reset()
+        ou_noise.reset()
+        ep_reward = 0
+        for i_step in range(cfg.train_steps):
+            action = agent.choose_action(state)
+            action = ou_noise.get_action(
+                action, i_step)  # 即paper中的random process
+            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 done:
+                break
+        print('Episode:{}/{}, Reward:{}, Steps:{}, Done:{}'.format(i_episode+1,cfg.train_eps,ep_reward,i_step+1,done))
+        ep_steps.append(i_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)
+    print('Complete training！')
+    return rewards,ma_rewards
+
+if __name__ == "__main__":
+    cfg = DDPGConfig()
+    env = NormalizedActions(gym.make("Pendulum-v0"))
+    env.seed(1) # 设置env随机种子
+    n_states = env.observation_space.shape[0]
+    n_actions = env.action_space.shape[0]
+    agent = DDPG(n_states,n_actions,cfg)
+    rewards,ma_rewards = train(cfg,env,agent)
+    agent.save(path=SAVED_MODEL_PATH)
+    save_results(rewards,ma_rewards,tag='train',path=RESULT_PATH)
+    plot_rewards(rewards,ma_rewards,tag="train",algo = cfg.algo,path=RESULT_PATH)
+