GCN DDPG Python代码

以下是GCN-DDPG的Python代码示例：

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

# Define the actor and critic networks
class Actor(nn.Module):
    def __init__(self, state_dim, action_dim, hidden_dim):
        super(Actor, self).__init__()

        self.fc1 = nn.Linear(state_dim, hidden_dim)
        self.fc2 = nn.Linear(hidden_dim, hidden_dim)
        self.fc3 = nn.Linear(hidden_dim, action_dim)

    def forward(self, state):
        x = F.relu(self.fc1(state))
        x = F.relu(self.fc2(x))
        x = torch.tanh(self.fc3(x))
        return x

class Critic(nn.Module):
    def __init__(self, state_dim, action_dim, hidden_dim):
        super(Critic, self).__init__()

        self.fc1 = nn.Linear(state_dim + action_dim, hidden_dim)
        self.fc2 = nn.Linear(hidden_dim, hidden_dim)
        self.fc3 = nn.Linear(hidden_dim, 1)

    def forward(self, state, action):
        x = torch.cat([state, action], dim=1)
        x = F.relu(self.fc1(x))
        x = F.relu(self.fc2(x))
        x = self.fc3(x)
        return x

# Define the GCN layer
class GCNLayer(nn.Module):
    def __init__(self, input_dim, output_dim):
        super(GCNLayer, self).__init__()

        self.fc = nn.Linear(input_dim, output_dim)

    def forward(self, adj, features):
        x = torch.spmm(adj, features)
        x = self.fc(x)
        x = F.relu(x)
        return x

# Define the GCN-DDPG agent
class GCN_DDPG_Agent:
    def __init__(self, state_dim, action_dim, hidden_dim, gcn_hidden_dim, replay_buffer_size, batch_size, gamma, tau, lr):

        self.device = torch.device("cuda" if torch.cuda.is_available() else "cpu")

        # Define the actor and critic networks
        self.actor = Actor(state_dim, action_dim, hidden_dim).to(self.device)
        self.target_actor = Actor(state_dim, action_dim, hidden_dim).to(self.device)
        self.critic = Critic(state_dim, action_dim, hidden_dim).to(self.device)
        self.target_critic = Critic(state_dim, action_dim, hidden_dim).to(self.device)

        # Define the GCN layers
        self.gcn1 = GCNLayer(state_dim, gcn_hidden_dim).to(self.device)
        self.gcn2 = GCNLayer(gcn_hidden_dim, gcn_hidden_dim).to(self.device)

        # Initialize the target networks with the same parameters as the online networks
        self.target_actor.load_state_dict(self.actor.state_dict())
        self.target_critic.load_state_dict(self.critic.state_dict())

        # Define the optimizer
        self.actor_optimizer = optim.Adam(self.actor.parameters(), lr=lr)
        self.critic_optimizer = optim.Adam(self.critic.parameters(), lr=lr)

        # Define the replay buffer
        self.replay_buffer_size = replay_buffer_size
        self.batch_size = batch_size
        self.replay_buffer = []

        # Define the hyperparameters
        self.gamma = gamma
        self.tau = tau

    def get_action(self, state, adj):
        # Convert the state to a tensor and pass it through the GCN layers
        state = torch.FloatTensor(state).unsqueeze(0).to(self.device)
        adj = torch.FloatTensor(adj).to(self.device)
        x = self.gcn1(adj, state)
        x = self.gcn2(adj, x)

        # Pass the GCN output through the actor network to get the action
        action = self.actor(x).cpu().data.numpy().flatten()
        return action

    def add_to_replay_buffer(self, state, adj, action, reward, next_state, next_adj, done):
        # Add the transition to the replay buffer
        self.replay_buffer.append((state, adj, action, reward, next_state, next_adj, done))

        # If the replay buffer size exceeds the maximum size, remove the oldest transition
        if len(self.replay_buffer) > self.replay_buffer_size:
            self.replay_buffer.pop(0)

    def train(self):
        # If the replay buffer size is smaller than the batch size, do not train
        if len(self.replay_buffer) < self.batch_size:
            return

        # Sample a batch of transitions from the replay buffer
        batch = random.sample(self.replay_buffer, self.batch_size)
        state_batch = torch.FloatTensor([transition[0] for transition in batch]).to(self.device)
        adj_batch = torch.FloatTensor([transition[1] for transition in batch]).to(self.device)
        action_batch = torch.FloatTensor([transition[2] for transition in batch]).to(self.device)
        reward_batch = torch.FloatTensor([transition[3] for transition in batch]).unsqueeze(1).to(self.device)
        next_state_batch = torch.FloatTensor([transition[4] for transition in batch]).to(self.device)
        next_adj_batch = torch.FloatTensor([transition[5] for transition in batch]).to(self.device)
        done_batch = torch.FloatTensor([transition[6] for transition in batch]).unsqueeze(1).to(self.device)

        # Compute the target Q value
        with torch.no_grad():
            next_x = self.gcn1(next_adj_batch, next_state_batch)
            next_x = self.gcn2(next_adj_batch, next_x)
            next_action = self.target_actor(next_x)
            target_q = self.target_critic(next_x, next_action)
            target_q = reward_batch + (self.gamma * target_q * (1 - done_batch))

        # Compute the critic loss
        q = self.critic(state_batch, action_batch)
        critic_loss = F.mse_loss(q, target_q)

        # Optimize the critic
        self.critic_optimizer.zero_grad()
        critic_loss.backward()
        self.critic_optimizer.step()

        # Compute the actor loss
        x = self.gcn1(adj_batch, state_batch)
        x = self.gcn2(adj_batch, x)
        actor_loss = -self.critic(x, self.actor(x)).mean()

        # Optimize the actor
        self.actor_optimizer.zero_grad()
        actor_loss.backward()
        self.actor_optimizer.step()

        # Update the target networks
        for param, target_param in zip(self.critic.parameters(), self.target_critic.parameters()):
            target_param.data.copy_(self.tau * param.data + (1 - self.tau) * target_param.data)

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

# Define the environment and the agent
env = gym.make('Pendulum-v0')
state_dim = env.observation_space.shape[0]
action_dim = env.action_space.shape[0]
hidden_dim = 256
gcn_hidden_dim = 64
replay_buffer_size = 1000000
batch_size = 256
gamma = 0.99
tau = 0.001
lr = 0.001
agent = GCN_DDPG_Agent(state_dim, action_dim, hidden_dim, gcn_hidden_dim, replay_buffer_size, batch_size, gamma, tau, lr)

# Train the agent
max_episodes = 1000
max_steps_per_episode = 1000
for episode in range(max_episodes):
    state = env.reset()
    adj = np.eye(state_dim)
    episode_reward = 0
    for step in range(max_steps_per_episode):
        action = agent.get_action(state, adj)
        next_state, reward, done, _ = env.step(action)
        next_adj = np.eye(state_dim)
        episode_reward += reward
        agent.add_to_replay_buffer(state, adj, action, reward, next_state, next_adj, done)
        agent.train()
        state = next_state
        adj = next_adj
        if done:
            break
    print("Episode: {}, Episode Reward: {}, Steps: {}".format(episode, episode_reward, step+1))
env.close()

请注意，此代码实现中使用的GCN层是基于稀疏矩阵的，因此需要将邻接矩阵作为稀疏矩阵传递，而不是密集矩阵。请确保在使用此代码时理解这一点。