ddpg代码pytorch实现

以下是使用PyTorch实现DDPG算法的代码示例：

import torch
import torch.nn as nn
import torch.optim as optim
import torch.nn.functional as F
import numpy as np
import gym

# 定义Actor网络
class Actor(nn.Module):
    def __init__(self, state_dim, action_dim, max_action):
        super(Actor, self).__init__()
        self.layer1 = nn.Linear(state_dim, 400)
        self.layer2 = nn.Linear(400, 300)
        self.layer3 = nn.Linear(300, action_dim)
        self.max_action = max_action

    def forward(self, x):
        x = F.relu(self.layer1(x))
        x = F.relu(self.layer2(x))
        x = self.max_action * torch.tanh(self.layer3(x))
        return x

# 定义Critic网络
class Critic(nn.Module):
    def __init__(self, state_dim, action_dim):
        super(Critic, self).__init__()
        self.layer1 = nn.Linear(state_dim + action_dim,400)
        self.layer2 = nn.Linear(400, 300)
        self.layer3 = nn.Linear(300, 1)

    def forward(self, x, u):
        xu = torch.cat([x, u], 1)
        x = F.relu(self.layer1(xu))
        x = F.relu(self.layer2(x))
        x = self.layer3(x)
        return x

# 定义DDPG算法
class DDPG(object):
    def __init__(self, state_dim, action_dim, max_action):
        self.actor = Actor(state_dim, action_dim, max_action).to(device)
        self.actor_target = Actor(state_dim, action_dim, max_action).to(device)
        self.actor_target.load_state_dict(self.actor.state_dict())
        self.actor_optimizer = optim.Adam(self.actor.parameters(), lr=1e-4)

        self.critic = Critic(state_dim, action_dim).to(device)
        self.critic_target = Critic(state_dim, action_dim).to(device)
        self.critic_target.load_state_dict(self.critic.state_dict())
        self.critic_optimizer = optim.Adam(self.critic.parameters(), lr=1e-3)

        self.max_action = max_action

    def select_action(self, state):
        state = torch.FloatTensor(state.reshape(1, -1)).to(device)
        return self.actor(state).cpu().data.numpy().flatten()

    def train(self, replay_buffer, iterations, batch_size=64, discount=0.99, tau=0.001):
        for it in range(iterations):
            # 从缓存中随机采样一批数据
            batch_states, batch_next_states, batch_actions, batch_rewards, batch_dones = replay_buffer.sample(batch_size)
            state = torch.FloatTensor(batch_states).to(device)
            next_state = torch.FloatTensor(batch_next_states).to(device)
            action = torch.FloatTensor(batch_actions).to(device)
            reward = torch.FloatTensor(batch_rewards).to(device)
            done = torch.FloatTensor(batch_dones).to(device)

            # 计算Q值
            Q = self.critic(state, action)
            next_action = self.actor_target(next_state)
            next_Q = self.critic_target(next_state, next_action.detach())
            target_Q = reward + (1 - done) * discount * next_Q

            # 计算Critic损失并更新网络
            critic_loss = F.mse_loss(Q, target_Q.detach())
            self.critic_optimizer.zero_grad()
            critic_loss.backward()
            self.critic_optimizer.step()

            # 计算Actor损失并更新网络
            actor_loss = -self.critic(state, self.actor(state)).mean()
            self.actor_optimizer.zero_grad()
            actor_loss.backward()
            self.actor_optimizer.step()

            # 更新目标网络
            for param, target_param in zip(self.critic.parameters(), self.critic_target.parameters()):
                target_param.data.copy_(tau * param.data + (1 - tau) * target_param.data)

            for param, target_param in zip(self.actor.parameters(), self.actor_target.parameters()):
                target_param.data.copy_(tau * param.data + (1 - tau) * target_param.data)

# 定义经验回放缓存
class ReplayBuffer(object):
    def __init__(self, max_size=1000000):
        self.storage = []
        self.max_size = max_size
        self.ptr = 0

    def add(self, state, next_state, action, reward, done):
        data = (state, next_state, action, reward, done)
        if len(self.storage) == self.max_size:
            self.storage[int(self.ptr)] = data
            self.ptr = (self.ptr + 1) % self.max_size
        else:
            self.storage.append(data)

    def sample(self, batch_size):
        ind = np.random.randint(0, len(self.storage), size=batch_size)
        batch_states, batch_next_states, batch_actions, batch_rewards, batch_dones = [], [], [], [], []
        for i in ind:
            state, next_state, action, reward, done = self.storage[i]
            batch_states.append(np.array(state, copy=False))
            batch_next_states.append(np.array(next_state, copy=False))
            batch_actions.append(np.array(action, copy=False))
            batch_rewards.append(np.array(reward, copy=False))
            batch_dones.append(np.array(done, copy=False))
        return np.array(batch_states), np.array(batch_next_states), np.array(batch_actions), np.array(batch_rewards).reshape(-1, 1), np.array(batch_dones).reshape(-1, 1)

# 定义超参数
env_name = "MountainCarContinuous-v0"
seed = 0
start_timesteps = 1e4
max_timesteps = 5e5
expl_noise = 0.1
batch_size = 64
discount = 0.99
tau = 0.001
policy_noise = 0.2
noise_clip = 0.5
policy_freq = 2

# 创建环境
env = gym.make(env_name)
state_dim = env.observation_space.shape[0]
action_dim = env.action_space.shape[0]
max_action = float(env.action_space.high[0])

# 设置随机种子
env.seed(seed)
torch.manual_seed(seed)
np.random.seed(seed)

# 创建DDPG对象
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
ddpg = DDPG(state_dim, action_dim, max_action)

# 创建经验回放缓存
replay_buffer = ReplayBuffer()

# 初始化状态和分数
state, done, score = env.reset(), False, 0

# 开始训练
for t in range(int(max_timesteps)):
    if t < start_timesteps:
        action = env.action_space.sample()
    else:
        action = (
            ddpg.select_action(np.array(state))
            + np.random.normal(0, max_action * expl_noise, size=action_dim)
        ).clip(-max_action, max_action)

    next_state, reward, done, _ = env.step(action)
    replay_buffer.add(state, next_state, action, reward, done)

    state = next_state
    score += reward

    if done:
        state, done, score = env.reset(), False, 0

    if t >= start_timesteps:
        ddpg.train(replay_buffer, batch_size, discount, tau)

    if t % 10000 == 0:
        print("Timestep: {} Score: {}".format(t, score))

# 关闭环境
env.close()