【深度强化学习】 PPO 模型解析，附Pytorch完整代码

import torch
import torch.nn as nn
import torch.nn.functional as F
import torch.optim as optim
from torch.distributions import Categorical

class PolicyNetwork(nn.Module):
    def __init__(self, num_inputs, num_actions):
        super(PolicyNetwork, self).__init__()
        self.fc1 = nn.Linear(num_inputs, 64)
        self.fc2 = nn.Linear(64, num_actions)

    def forward(self, x):
        x = F.relu(self.fc1(x))
        action_scores = self.fc2(x)
        return action_scores

    def get_action_probs(self, state):
        state = torch.from_numpy(state).float().unsqueeze(0)
        action_scores = self.forward(state)
        action_probs = F.softmax(action_scores, dim=1).data.squeeze()
        return action_probs

    def get_action(self, state):
        action_probs = self.get_action_probs(state)
        dist = Categorical(action_probs)
        action = dist.sample()
        return action.item(), action_probs

class ValueNetwork(nn.Module):
    def __init__(self, num_inputs):
        super(ValueNetwork, self).__init__()
        self.fc1 = nn.Linear(num_inputs, 64)
        self.fc2 = nn.Linear(64, 1)

    def forward(self, x):
        x = F.relu(self.fc1(x))
        state_values = self.fc2(x)
        return state_values

    def get_value(self, state):
        state = torch.from_numpy(state).float().unsqueeze(0)
        state_value = self.forward(state).data.squeeze().item()
        return state_value

def compute_advantages(batch_size, gamma, lambd, values, rewards, masks, next_values):
    advantages = torch.zeros(batch_size)
    adv_t = 0
    for t in reversed(range(len(rewards))):
        adv_t = rewards[t] + gamma * adv_t * masks[t]
        advantages[t] = adv_t - values[t]
    advantages = (advantages - advantages.mean()) / (advantages.std() + 1e-8)
    return advantages

def train_ppo(env, policy_net, value_net, optimizer_policy, optimizer_value, gamma=0.99, lambd=0.95, num_episodes=1000, batch_size=64):
    for i_episode in range(num_episodes):
        state = env.reset()
        log_probs = []
        values = []
        rewards = []
        masks = []

        for t in range(100):
            action, log_prob = policy_net.get_action_probs(state)
            next_state, reward, done, _ = env.step(action)
            log_probs.append(log_prob)
            values.append(value