main_ddpg.py

import gym
import numpy as np
import random
from agent_ddpg import DDPGAgent
import os
import torch
import time
import pygame
# 初始化环境
env = gym.make(id='Pendulum-v1', render_mode="human")
STATE_DIM = env.observation_space.shape[0]
ACTION_DIM = env.action_space.shape[0]

agent =DDPGAgent(STATE_DIM, ACTION_DIM)

#超参数
NUM_EPISODE = 100
NUM_STEP = 200
EPSILON_START = 1
EPSILON_END = 0.02
EPSILON_DECAY = 10000

REWARD_BUFFER = np.empty(shape=NUM_EPISODE)
for episode_i in range(NUM_EPISODE):
    state, others = env.reset()
    episode_reward = 0

    for step_i in range(NUM_STEP):
        epsilon = np.interp(x=episode_i*NUM_STEP+step_i, xp=[0,EPSILON_DECAY], fp=[EPSILON_START,EPSILON_END])
        random_sample = random.random()
        if random_sample <=epsilon:
            action = np.random.uniform(low=-2, high=2,size=ACTION_DIM)
        else:
            action = agent.get_action(state)
        next_state, reward, done, truncation, info = env.step(action)

        agent.replay_buffer.add_memo(state, action, reward, next_state,done)

        state = next_state
        episode_reward += reward

        agent.update()   #每个时间步更新
        if done:
            break
    REWARD_BUFFER[episode_i] = episode_reward
    print(f"Episode: {episode_i+1}, Reward: {round(episode_reward,2)}")

#模型保存地址
current_path = os.path.dirname(os.path.realpath(__file__))
model = current_path + '/models/'
timestamp = time.strftime("%Y%m%d%H%M%S")

#保存模型
torch.save(agent.actor.state_dict(), model + f"ddpg_actor_{timestamp}.path")
torch.save(agent.critic.state_dict(), model + f"ddpg_critic_{timestamp}.path")

env.close()