bffeefe23c0c8f372e337a925e8c28b97556e5fc,Reinforcement_learning_TUT/8_Actor_Critic_Advantage/AC_continue_Pendulum.py,,,#,120

Before Change

actor = Actor(n_features=env.observation_space.shape[0], action_range=[env.action_space.low[0], env.action_space.high[0]], lr=0.001)
critic = Critic(n_features=env.observation_space.shape[0], lr=0.002)

with tf.Session() as sess:
    if OUTPUT_GRAPH:
        tf.summary.FileWriter("logs/", sess.graph)

    actor.sess, critic.sess = sess, sess    // define the tf session
    tf.global_variables_initializer().run()

    for i_episode in range(3000):
        observation = env.reset()
        t = 0
        ep_rs = []
        while True:
            // if RENDER:
            env.render()
            action, mu, sigma = actor.choose_action(observation)

            observation_, reward, done, info = env.step(action)
            reward /= 10
            TD_target = reward + GAMMA * critic.evaluate(observation_)    // r + gamma * V_next
            TD_eval = critic.evaluate(observation)    // V_now
            TD_error = TD_target - TD_eval

            actor.update(s=observation, a=action, adv=TD_error)
            critic.update(s=observation, target=TD_target)

            observation = observation_
            t += 1
            // print(reward)
            ep_rs.append(reward)
            if t > EPISODE_TIME_THRESHOLD:
                ep_rs_sum = sum(ep_rs)
                if "running_reward" not in globals():
                    running_reward = ep_rs_sum
                else:
                    running_reward = running_reward * 0.9 + ep_rs_sum * 0.1
                if running_reward > DISPLAY_REWARD_THRESHOLD: RENDER = True  // rendering
                print("episode:", i_episode, "  reward:", int(running_reward))
                break

After Change

env = gym.make("Pendulum-v0")
env.seed(1)  // reproducible

sess = tf.Session()

actor = Actor(sess, n_features=env.observation_space.shape[0], action_range=[env.action_space.low[0], env.action_space.high[0]], lr=LR_A)
critic = Critic(sess, n_features=env.observation_space.shape[0], lr=LR_C)

Instances