Use TD3 to Play Pendulum-v1¶

TensorFlow version

In [1]:
%matplotlib inline

import sys
import logging
import itertools

import numpy as np
np.random.seed(0)
import pandas as pd
import gym
import matplotlib.pyplot as plt
import tensorflow.compat.v2 as tf
tf.random.set_seed(0)
from tensorflow import keras
from tensorflow import nn
from tensorflow import optimizers
from tensorflow import losses
from tensorflow.keras import layers
from tensorflow.keras import models

logging.basicConfig(level=logging.INFO,
        format='%(asctime)s [%(levelname)s] %(message)s',
        stream=sys.stdout, datefmt='%H:%M:%S')
In [2]:
env = gym.make('Pendulum-v1')
for key in vars(env.spec):
    logging.info('%s: %s', key, vars(env.spec)[key])
for key in vars(env.unwrapped):
    logging.info('%s: %s', key, vars(env.unwrapped)[key])

00:22:00 [INFO] id: Pendulum-v1
00:22:00 [INFO] entry_point: gym.envs.classic_control:PendulumEnv
00:22:00 [INFO] reward_threshold: None
00:22:00 [INFO] nondeterministic: False
00:22:00 [INFO] max_episode_steps: 200
00:22:00 [INFO] order_enforce: True
00:22:00 [INFO] _kwargs: {}
00:22:00 [INFO] _env_name: Pendulum
00:22:00 [INFO] max_speed: 8
00:22:00 [INFO] max_torque: 2.0
00:22:00 [INFO] dt: 0.05
00:22:00 [INFO] g: 10.0
00:22:00 [INFO] m: 1.0
00:22:00 [INFO] l: 1.0
00:22:00 [INFO] viewer: None
00:22:00 [INFO] action_space: Box([-2.], [2.], (1,), float32)
00:22:00 [INFO] observation_space: Box([-1. -1. -8.], [1. 1. 8.], (3,), float32)
00:22:00 [INFO] np_random: RandomState(MT19937)
00:22:00 [INFO] spec: EnvSpec(Pendulum-v1)
In [3]:
class DQNReplayer:
    def __init__(self, capacity):
        self.memory = pd.DataFrame(index=range(capacity),
                columns=['state', 'action', 'reward', 'next_state', 'terminated'])
        self.i = 0
        self.count = 0
        self.capacity = capacity

    def store(self, *args):
        self.memory.loc[self.i] = np.asarray(args, dtype=object)
        self.i = (self.i + 1) % self.capacity
        self.count = min(self.count + 1, self.capacity)

    def sample(self, size):
        indices = np.random.choice(self.count, size=size)
        return (np.stack(self.memory.loc[indices, field]) for field in
                self.memory.columns)
In [4]:
class OrnsteinUhlenbeckProcess:
    def __init__(self, x0):
        self.x = x0

    def __call__(self, mu=0., sigma=1., theta=.15, dt=.01):
        n = np.random.normal(size=self.x.shape)
        self.x += (theta * (mu - self.x) * dt + sigma * np.sqrt(dt) * n)
        return self.x
In [5]:
class TD3Agent:
    def __init__(self, env):
        state_dim = env.observation_space.shape[0]
        self.action_dim = env.action_space.shape[0]
        self.action_low = env.action_space.low
        self.action_high = env.action_space.high
        self.gamma = 0.99

        self.replayer = DQNReplayer(20000)

        self.actor_evaluate_net = self.build_net(
                input_size=state_dim, hidden_sizes=[32, 64],
                output_size=self.action_dim, output_activation=nn.tanh)
        self.actor_target_net = models.clone_model(self.actor_evaluate_net)
        self.actor_target_net.set_weights(self.actor_evaluate_net.get_weights())

        self.critic0_evaluate_net = self.build_net(
                input_size=state_dim+self.action_dim, hidden_sizes=[64, 128])
        self.critic0_target_net = models.clone_model(self.critic0_evaluate_net)
        self.critic0_target_net.set_weights(self.critic0_evaluate_net.get_weights())

        self.critic1_evaluate_net = self.build_net(
                input_size=state_dim+self.action_dim, hidden_sizes=[64, 128])
        self.critic1_target_net = models.clone_model(self.critic1_evaluate_net)
        self.critic1_target_net.set_weights(self.critic1_evaluate_net.get_weights())

    def build_net(self, input_size=None, hidden_sizes=None, output_size=1,
                activation=nn.relu, output_activation=None,
                loss=losses.mse, learning_rate=0.001):
        model = keras.Sequential()
        for layer, hidden_size in enumerate(hidden_sizes):
            kwargs = {'input_shape' : (input_size,)} if layer == 0 else {}
            model.add(layers.Dense(units=hidden_size,
                    activation=activation, **kwargs))
        model.add(layers.Dense(units=output_size,
                activation=output_activation))
        optimizer = optimizers.Adam(learning_rate)
        model.compile(optimizer=optimizer, loss=loss)
        return model

    def reset(self, mode=None):
        self.mode = mode
        if self.mode == 'train':
            self.trajectory = []
            self.noise = OrnsteinUhlenbeckProcess(np.zeros((self.action_dim,)))

    def step(self, observation, reward, terminated):
        if self.mode == 'train' and self.replayer.count < 3000:
            action = np.random.uniform(self.action_low, self.action_high)
        else:
            action = self.actor_evaluate_net.predict(observation[np.newaxis],
                    verbose=0)[0]
        if self.mode == 'train':
            # noisy action
            noise = self.noise(sigma=0.1)
            action = (action + noise).clip(self.action_low, self.action_high)

            self.trajectory += [observation, reward, terminated, action]
            if len(self.trajectory) >= 8:
                state, _, _, act, next_state, reward, terminated, _ = \
                        self.trajectory[-8:]
                self.replayer.store(state, act, reward, next_state, terminated)

            if self.replayer.count >= 3000:
                self.learn()
        return action

    def close(self):
        pass

    def update_net(self, target_net, evaluate_net, learning_rate=0.005):
        average_weights = [(1. - learning_rate) * t + learning_rate * e for t, e
                in zip(target_net.get_weights(), evaluate_net.get_weights())]
        target_net.set_weights(average_weights)

    def learn(self):
        # replay
        states, actions, rewards, next_states, terminateds = \
                self.replayer.sample(64)
        state_tensor = tf.convert_to_tensor(states, dtype=tf.float32)

        # update critic
        next_actions = self.actor_target_net.predict(next_states, verbose=0)
        next_noises = np.random.normal(0, 0.2, size=next_actions.shape)
        next_actions = (next_actions + next_noises).clip(self.action_low,
                self.action_high)
        state_actions = np.hstack([states, actions])
        next_state_actions = np.hstack([next_states, next_actions])
        next_q0s = self.critic0_target_net.predict(next_state_actions,
                verbose=0)[:, 0]
        next_q1s = self.critic1_target_net.predict(next_state_actions,
                verbose=0)[:, 0]
        next_qs = np.minimum(next_q0s, next_q1s)
        targets = rewards + (1. - terminateds) * self.gamma * next_qs
        self.critic0_evaluate_net.fit(state_actions, targets[:, np.newaxis],
                verbose=0)
        self.critic1_evaluate_net.fit(state_actions, targets[:, np.newaxis],
                verbose=0)

        # update actor
        with tf.GradientTape() as tape:
            action_tensor = self.actor_evaluate_net(state_tensor)
            state_action_tensor = tf.concat([state_tensor, action_tensor], axis=1)
            q_tensor = self.critic0_evaluate_net(state_action_tensor)
            loss_tensor = -tf.reduce_mean(q_tensor)
        grad_tensors = tape.gradient(loss_tensor,
                self.actor_evaluate_net.variables)
        self.actor_evaluate_net.optimizer.apply_gradients(zip(
                grad_tensors, self.actor_evaluate_net.variables))

        self.update_net(self.critic0_target_net, self.critic0_evaluate_net)
        self.update_net(self.critic1_target_net, self.critic1_evaluate_net)
        self.update_net(self.actor_target_net, self.actor_evaluate_net)


agent = TD3Agent(env)
In [6]:
def play_episode(env, agent, seed=None, mode=None, render=False):
    observation, _ = env.reset(seed=seed)
    reward, terminated, truncated = 0., False, False
    agent.reset(mode=mode)
    episode_reward, elapsed_steps = 0., 0
    while True:
        action = agent.step(observation, reward, terminated)
        if render:
            env.render()
        if terminated or truncated:
            break
        observation, reward, terminated, truncated, _ = env.step(action)
        episode_reward += reward
        elapsed_steps += 1
    agent.close()
    return episode_reward, elapsed_steps


logging.info('==== train ====')
episode_rewards = []
for episode in itertools.count():
    episode_reward, elapsed_steps = play_episode(env, agent, seed=episode,
            mode='train')
    episode_rewards.append(episode_reward)
    logging.info('train episode %d: reward = %.2f, steps = %d',
            episode, episode_reward, elapsed_steps)
    if np.mean(episode_rewards[-10:]) > -150:
        break
plt.plot(episode_rewards)


logging.info('==== test ====')
episode_rewards = []
for episode in range(100):
    episode_reward, elapsed_steps = play_episode(env, agent)
    episode_rewards.append(episode_reward)
    logging.info('test episode %d: reward = %.2f, steps = %d',
            episode, episode_reward, elapsed_steps)
logging.info('average episode reward = %.2f ± %.2f',
        np.mean(episode_rewards), np.std(episode_rewards))

00:22:51 [INFO] ==== train ====
00:22:51 [INFO] train episode 0: reward = -1744.13, steps = 200
00:22:51 [INFO] train episode 1: reward = -1025.25, steps = 200
00:22:52 [INFO] train episode 2: reward = -1590.20, steps = 200
00:22:52 [INFO] train episode 3: reward = -1137.77, steps = 200
00:22:52 [INFO] train episode 4: reward = -1675.82, steps = 200
00:22:52 [INFO] train episode 5: reward = -1632.97, steps = 200
00:22:52 [INFO] train episode 6: reward = -753.85, steps = 200
00:22:52 [INFO] train episode 7: reward = -1833.66, steps = 200
00:22:52 [INFO] train episode 8: reward = -936.49, steps = 200
00:22:52 [INFO] train episode 9: reward = -1622.68, steps = 200
00:22:52 [INFO] train episode 10: reward = -1307.43, steps = 200
00:22:52 [INFO] train episode 11: reward = -908.99, steps = 200
00:22:52 [INFO] train episode 12: reward = -1504.19, steps = 200
00:22:53 [INFO] train episode 13: reward = -1003.41, steps = 200
00:22:53 [INFO] train episode 14: reward = -921.67, steps = 200
00:24:27 [INFO] train episode 15: reward = -1156.67, steps = 200
00:26:07 [INFO] train episode 16: reward = -1467.30, steps = 200
00:27:30 [INFO] train episode 17: reward = -1550.39, steps = 200
00:28:55 [INFO] train episode 18: reward = -1696.63, steps = 200
00:30:14 [INFO] train episode 19: reward = -1086.24, steps = 200
00:31:49 [INFO] train episode 20: reward = -1313.67, steps = 200
00:33:18 [INFO] train episode 21: reward = -1413.73, steps = 200
00:34:43 [INFO] train episode 22: reward = -1518.39, steps = 200
00:36:18 [INFO] train episode 23: reward = -1183.81, steps = 200
00:38:12 [INFO] train episode 24: reward = -991.98, steps = 200
00:40:13 [INFO] train episode 25: reward = -1113.87, steps = 200
00:42:31 [INFO] train episode 26: reward = -1015.62, steps = 200
00:44:54 [INFO] train episode 27: reward = -752.92, steps = 200
00:47:12 [INFO] train episode 28: reward = -625.19, steps = 200
00:49:13 [INFO] train episode 29: reward = -991.54, steps = 200
00:51:29 [INFO] train episode 30: reward = -376.51, steps = 200
00:53:38 [INFO] train episode 31: reward = -497.96, steps = 200
00:55:53 [INFO] train episode 32: reward = -126.30, steps = 200
00:58:03 [INFO] train episode 33: reward = -242.28, steps = 200
01:00:18 [INFO] train episode 34: reward = -481.37, steps = 200
01:02:46 [INFO] train episode 35: reward = -497.32, steps = 200
01:05:10 [INFO] train episode 36: reward = -606.38, steps = 200
01:07:26 [INFO] train episode 37: reward = -126.52, steps = 200
01:09:43 [INFO] train episode 38: reward = -126.26, steps = 200
01:12:19 [INFO] train episode 39: reward = -241.38, steps = 200
01:15:23 [INFO] train episode 40: reward = -652.64, steps = 200
01:18:22 [INFO] train episode 41: reward = -790.80, steps = 200
01:21:21 [INFO] train episode 42: reward = -250.58, steps = 200
01:24:24 [INFO] train episode 43: reward = -1.79, steps = 200
01:27:23 [INFO] train episode 44: reward = -254.04, steps = 200
01:30:22 [INFO] train episode 45: reward = -121.56, steps = 200
01:33:22 [INFO] train episode 46: reward = -239.31, steps = 200
01:36:20 [INFO] train episode 47: reward = -120.61, steps = 200
01:39:24 [INFO] train episode 48: reward = -126.17, steps = 200
01:42:28 [INFO] train episode 49: reward = -125.57, steps = 200
01:45:45 [INFO] train episode 50: reward = -484.86, steps = 200
01:49:07 [INFO] train episode 51: reward = -122.38, steps = 200
01:52:26 [INFO] train episode 52: reward = -238.84, steps = 200
01:55:33 [INFO] train episode 53: reward = -245.61, steps = 200
01:58:32 [INFO] train episode 54: reward = -247.65, steps = 200
02:01:29 [INFO] train episode 55: reward = -486.61, steps = 200
02:04:27 [INFO] train episode 56: reward = -125.16, steps = 200
02:07:29 [INFO] train episode 57: reward = -243.55, steps = 200
02:10:27 [INFO] train episode 58: reward = -119.64, steps = 200
02:13:24 [INFO] train episode 59: reward = -237.30, steps = 200
02:16:22 [INFO] train episode 60: reward = -123.70, steps = 200
02:19:16 [INFO] train episode 61: reward = -240.52, steps = 200
02:22:15 [INFO] train episode 62: reward = -238.56, steps = 200
02:25:12 [INFO] train episode 63: reward = -121.25, steps = 200
02:28:12 [INFO] train episode 64: reward = -240.12, steps = 200
02:31:09 [INFO] train episode 65: reward = -122.46, steps = 200
02:34:13 [INFO] train episode 66: reward = -125.17, steps = 200
02:37:11 [INFO] train episode 67: reward = -365.42, steps = 200
02:40:07 [INFO] train episode 68: reward = -475.81, steps = 200
02:43:04 [INFO] train episode 69: reward = -239.04, steps = 200
02:45:58 [INFO] train episode 70: reward = -120.79, steps = 200
02:48:54 [INFO] train episode 71: reward = -124.61, steps = 200
02:51:51 [INFO] train episode 72: reward = -468.28, steps = 200
02:54:47 [INFO] train episode 73: reward = -126.16, steps = 200
02:57:43 [INFO] train episode 74: reward = -234.84, steps = 200
03:00:44 [INFO] train episode 75: reward = -125.65, steps = 200
03:03:40 [INFO] train episode 76: reward = -357.30, steps = 200
03:06:37 [INFO] train episode 77: reward = -123.65, steps = 200
03:09:35 [INFO] train episode 78: reward = -120.43, steps = 200
03:12:31 [INFO] train episode 79: reward = -124.02, steps = 200
03:15:26 [INFO] train episode 80: reward = -505.90, steps = 200
03:18:22 [INFO] train episode 81: reward = -126.72, steps = 200
03:21:17 [INFO] train episode 82: reward = -596.69, steps = 200
03:24:13 [INFO] train episode 83: reward = -124.80, steps = 200
03:27:11 [INFO] train episode 84: reward = -121.46, steps = 200
03:30:06 [INFO] train episode 85: reward = -121.98, steps = 200
03:33:04 [INFO] train episode 86: reward = -593.72, steps = 200
03:36:00 [INFO] train episode 87: reward = -240.93, steps = 200
03:38:58 [INFO] train episode 88: reward = -247.89, steps = 200
03:41:53 [INFO] train episode 89: reward = -121.46, steps = 200
03:44:50 [INFO] train episode 90: reward = -123.04, steps = 200
03:47:46 [INFO] train episode 91: reward = -120.72, steps = 200
03:50:42 [INFO] train episode 92: reward = -236.45, steps = 200
03:53:39 [INFO] train episode 93: reward = -125.30, steps = 200
03:56:35 [INFO] train episode 94: reward = -126.42, steps = 200
03:59:30 [INFO] train episode 95: reward = -506.16, steps = 200
04:02:25 [INFO] train episode 96: reward = -122.54, steps = 200
04:05:20 [INFO] train episode 97: reward = -360.87, steps = 200
04:08:19 [INFO] train episode 98: reward = -121.46, steps = 200
04:11:16 [INFO] train episode 99: reward = -471.62, steps = 200
04:14:13 [INFO] train episode 100: reward = -363.90, steps = 200
04:17:09 [INFO] train episode 101: reward = -238.32, steps = 200
04:20:04 [INFO] train episode 102: reward = -598.70, steps = 200
04:23:01 [INFO] train episode 103: reward = -494.53, steps = 200
04:25:58 [INFO] train episode 104: reward = -492.42, steps = 200
04:28:58 [INFO] train episode 105: reward = -124.28, steps = 200
04:31:52 [INFO] train episode 106: reward = -244.48, steps = 200
04:34:48 [INFO] train episode 107: reward = -356.00, steps = 200
04:37:44 [INFO] train episode 108: reward = -119.03, steps = 200
04:40:40 [INFO] train episode 109: reward = -489.83, steps = 200
04:43:39 [INFO] train episode 110: reward = -124.21, steps = 200
04:46:35 [INFO] train episode 111: reward = -125.19, steps = 200
04:49:30 [INFO] train episode 112: reward = -354.99, steps = 200
04:52:27 [INFO] train episode 113: reward = -121.10, steps = 200
04:55:23 [INFO] train episode 114: reward = -119.28, steps = 200
04:58:05 [INFO] train episode 115: reward = -468.05, steps = 200
05:00:41 [INFO] train episode 116: reward = -367.44, steps = 200
05:03:16 [INFO] train episode 117: reward = -124.05, steps = 200
05:05:51 [INFO] train episode 118: reward = -237.68, steps = 200
05:08:28 [INFO] train episode 119: reward = -123.57, steps = 200
05:11:02 [INFO] train episode 120: reward = -363.97, steps = 200
05:13:37 [INFO] train episode 121: reward = -120.68, steps = 200
05:16:13 [INFO] train episode 122: reward = -365.84, steps = 200
05:18:47 [INFO] train episode 123: reward = -497.26, steps = 200
05:21:18 [INFO] train episode 124: reward = -123.05, steps = 200
05:23:51 [INFO] train episode 125: reward = -483.80, steps = 200
05:26:23 [INFO] train episode 126: reward = -125.38, steps = 200
05:28:57 [INFO] train episode 127: reward = -125.54, steps = 200
05:31:28 [INFO] train episode 128: reward = -121.31, steps = 200
05:33:58 [INFO] train episode 129: reward = -349.95, steps = 200
05:36:27 [INFO] train episode 130: reward = -1.79, steps = 200
05:38:59 [INFO] train episode 131: reward = -360.27, steps = 200
05:41:31 [INFO] train episode 132: reward = -127.78, steps = 200
05:44:02 [INFO] train episode 133: reward = -184.57, steps = 200
05:46:33 [INFO] train episode 134: reward = -492.26, steps = 200
05:49:04 [INFO] train episode 135: reward = -522.55, steps = 200
05:51:36 [INFO] train episode 136: reward = -241.31, steps = 200
05:54:08 [INFO] train episode 137: reward = -475.77, steps = 200
05:56:38 [INFO] train episode 138: reward = -242.52, steps = 200
05:59:10 [INFO] train episode 139: reward = -483.96, steps = 200
06:01:42 [INFO] train episode 140: reward = -238.91, steps = 200
06:04:13 [INFO] train episode 141: reward = -355.65, steps = 200
06:06:43 [INFO] train episode 142: reward = -525.41, steps = 200
06:09:16 [INFO] train episode 143: reward = -122.80, steps = 200
06:11:57 [INFO] train episode 144: reward = -119.83, steps = 200
06:14:27 [INFO] train episode 145: reward = -121.93, steps = 200
06:16:57 [INFO] train episode 146: reward = -129.13, steps = 200
06:19:23 [INFO] train episode 147: reward = -366.97, steps = 200
06:21:37 [INFO] train episode 148: reward = -127.54, steps = 200
06:23:50 [INFO] train episode 149: reward = -1677.82, steps = 200
06:26:04 [INFO] train episode 150: reward = -1708.18, steps = 200
06:28:19 [INFO] train episode 151: reward = -359.16, steps = 200
06:30:33 [INFO] train episode 152: reward = -3.74, steps = 200
06:32:38 [INFO] train episode 153: reward = -237.41, steps = 200
06:34:35 [INFO] train episode 154: reward = -370.23, steps = 200
06:36:34 [INFO] train episode 155: reward = -366.13, steps = 200
06:38:27 [INFO] train episode 156: reward = -243.46, steps = 200
06:40:19 [INFO] train episode 157: reward = -125.16, steps = 200
06:42:12 [INFO] train episode 158: reward = -237.18, steps = 200
06:44:05 [INFO] train episode 159: reward = -124.80, steps = 200
06:45:57 [INFO] train episode 160: reward = -351.52, steps = 200
06:47:50 [INFO] train episode 161: reward = -244.16, steps = 200
06:49:43 [INFO] train episode 162: reward = -596.23, steps = 200
06:51:30 [INFO] train episode 163: reward = -246.55, steps = 200
06:53:16 [INFO] train episode 164: reward = -123.55, steps = 200
06:55:01 [INFO] train episode 165: reward = -122.38, steps = 200
06:56:42 [INFO] train episode 166: reward = -243.91, steps = 200
06:58:23 [INFO] train episode 167: reward = -596.74, steps = 200
07:00:08 [INFO] train episode 168: reward = -243.37, steps = 200
07:01:50 [INFO] train episode 169: reward = -123.00, steps = 200
07:03:34 [INFO] train episode 170: reward = -591.54, steps = 200
07:05:16 [INFO] train episode 171: reward = -123.34, steps = 200
07:06:57 [INFO] train episode 172: reward = -122.74, steps = 200
07:08:51 [INFO] train episode 173: reward = -570.28, steps = 200
07:10:37 [INFO] train episode 174: reward = -237.73, steps = 200
07:12:21 [INFO] train episode 175: reward = -473.73, steps = 200
07:14:03 [INFO] train episode 176: reward = -125.72, steps = 200
07:15:48 [INFO] train episode 177: reward = -499.51, steps = 200
07:17:33 [INFO] train episode 178: reward = -122.84, steps = 200
07:19:18 [INFO] train episode 179: reward = -354.69, steps = 200
07:21:00 [INFO] train episode 180: reward = -611.68, steps = 200
07:22:42 [INFO] train episode 181: reward = -242.68, steps = 200
07:24:24 [INFO] train episode 182: reward = -237.96, steps = 200
07:26:10 [INFO] train episode 183: reward = -244.58, steps = 200
07:27:55 [INFO] train episode 184: reward = -356.48, steps = 200
07:29:37 [INFO] train episode 185: reward = -122.68, steps = 200
07:31:18 [INFO] train episode 186: reward = -121.29, steps = 200
07:33:02 [INFO] train episode 187: reward = -2.76, steps = 200
07:34:45 [INFO] train episode 188: reward = -356.27, steps = 200
07:36:29 [INFO] train episode 189: reward = -4.24, steps = 200
07:38:12 [INFO] train episode 190: reward = -238.95, steps = 200
07:39:55 [INFO] train episode 191: reward = -468.01, steps = 200
07:41:38 [INFO] train episode 192: reward = -118.63, steps = 200
07:43:21 [INFO] train episode 193: reward = -564.73, steps = 200
07:45:05 [INFO] train episode 194: reward = -125.75, steps = 200
07:46:48 [INFO] train episode 195: reward = -361.16, steps = 200
07:48:30 [INFO] train episode 196: reward = -240.85, steps = 200
07:50:04 [INFO] train episode 197: reward = -477.32, steps = 200
07:51:37 [INFO] train episode 198: reward = -364.28, steps = 200
07:53:10 [INFO] train episode 199: reward = -127.64, steps = 200
07:54:44 [INFO] train episode 200: reward = -490.00, steps = 200
07:56:16 [INFO] train episode 201: reward = -121.42, steps = 200
07:57:43 [INFO] train episode 202: reward = -127.72, steps = 200
07:59:07 [INFO] train episode 203: reward = -120.75, steps = 200
08:00:24 [INFO] train episode 204: reward = -246.98, steps = 200
08:01:40 [INFO] train episode 205: reward = -127.79, steps = 200
08:02:57 [INFO] train episode 206: reward = -128.53, steps = 200
08:04:15 [INFO] train episode 207: reward = -122.72, steps = 200
08:05:31 [INFO] train episode 208: reward = -121.22, steps = 200
08:06:48 [INFO] train episode 209: reward = -128.93, steps = 200
08:08:05 [INFO] train episode 210: reward = -250.15, steps = 200
08:08:05 [INFO] ==== test ====
08:08:17 [INFO] test episode 0: reward = -121.06, steps = 200
08:08:30 [INFO] test episode 1: reward = -121.41, steps = 200
08:08:43 [INFO] test episode 2: reward = -4.54, steps = 200
08:08:56 [INFO] test episode 3: reward = -606.56, steps = 200
08:09:09 [INFO] test episode 4: reward = -126.56, steps = 200
08:09:22 [INFO] test episode 5: reward = -126.88, steps = 200
08:09:35 [INFO] test episode 6: reward = -119.16, steps = 200
08:09:48 [INFO] test episode 7: reward = -486.39, steps = 200
08:10:00 [INFO] test episode 8: reward = -501.57, steps = 200
08:10:13 [INFO] test episode 9: reward = -119.34, steps = 200
08:10:26 [INFO] test episode 10: reward = -126.63, steps = 200
08:10:39 [INFO] test episode 11: reward = -0.48, steps = 200
08:10:52 [INFO] test episode 12: reward = -485.65, steps = 200
08:11:05 [INFO] test episode 13: reward = -358.65, steps = 200
08:11:18 [INFO] test episode 14: reward = -125.33, steps = 200
08:11:31 [INFO] test episode 15: reward = -573.09, steps = 200
08:11:44 [INFO] test episode 16: reward = -239.49, steps = 200
08:11:57 [INFO] test episode 17: reward = -492.63, steps = 200
08:12:10 [INFO] test episode 18: reward = -128.90, steps = 200
08:12:23 [INFO] test episode 19: reward = -232.10, steps = 200
08:12:36 [INFO] test episode 20: reward = -362.95, steps = 200
08:12:48 [INFO] test episode 21: reward = -122.09, steps = 200
08:13:02 [INFO] test episode 22: reward = -125.79, steps = 200
08:13:14 [INFO] test episode 23: reward = -362.22, steps = 200
08:13:27 [INFO] test episode 24: reward = -584.94, steps = 200
08:13:41 [INFO] test episode 25: reward = -123.96, steps = 200
08:13:53 [INFO] test episode 26: reward = -127.59, steps = 200
08:14:06 [INFO] test episode 27: reward = -240.20, steps = 200
08:14:19 [INFO] test episode 28: reward = -232.60, steps = 200
08:14:32 [INFO] test episode 29: reward = -126.65, steps = 200
08:14:45 [INFO] test episode 30: reward = -555.29, steps = 200
08:14:58 [INFO] test episode 31: reward = -127.88, steps = 200
08:15:11 [INFO] test episode 32: reward = -242.91, steps = 200
08:15:24 [INFO] test episode 33: reward = -601.64, steps = 200
08:15:37 [INFO] test episode 34: reward = -126.87, steps = 200
08:15:50 [INFO] test episode 35: reward = -492.83, steps = 200
08:16:02 [INFO] test episode 36: reward = -471.13, steps = 200
08:16:15 [INFO] test episode 37: reward = -540.78, steps = 200
08:16:28 [INFO] test episode 38: reward = -127.28, steps = 200
08:16:41 [INFO] test episode 39: reward = -120.67, steps = 200
08:16:54 [INFO] test episode 40: reward = -358.97, steps = 200
08:17:06 [INFO] test episode 41: reward = -124.68, steps = 200
08:17:20 [INFO] test episode 42: reward = -506.01, steps = 200
08:17:32 [INFO] test episode 43: reward = -126.74, steps = 200
08:17:45 [INFO] test episode 44: reward = -350.33, steps = 200
08:17:58 [INFO] test episode 45: reward = -238.96, steps = 200
08:18:11 [INFO] test episode 46: reward = -122.70, steps = 200
08:18:24 [INFO] test episode 47: reward = -122.26, steps = 200
08:18:37 [INFO] test episode 48: reward = -358.71, steps = 200
08:18:49 [INFO] test episode 49: reward = -121.65, steps = 200
08:19:02 [INFO] test episode 50: reward = -478.52, steps = 200
08:19:16 [INFO] test episode 51: reward = -127.19, steps = 200
08:19:28 [INFO] test episode 52: reward = -0.54, steps = 200
08:19:41 [INFO] test episode 53: reward = -240.66, steps = 200
08:19:54 [INFO] test episode 54: reward = -356.13, steps = 200
08:20:07 [INFO] test episode 55: reward = -358.22, steps = 200
08:20:20 [INFO] test episode 56: reward = -119.34, steps = 200
08:20:33 [INFO] test episode 57: reward = -123.71, steps = 200
08:20:47 [INFO] test episode 58: reward = -122.33, steps = 200
08:21:00 [INFO] test episode 59: reward = -593.70, steps = 200
08:21:12 [INFO] test episode 60: reward = -0.54, steps = 200
08:21:26 [INFO] test episode 61: reward = -122.09, steps = 200
08:21:39 [INFO] test episode 62: reward = -121.40, steps = 200
08:21:53 [INFO] test episode 63: reward = -361.02, steps = 200
08:22:07 [INFO] test episode 64: reward = -122.36, steps = 200
08:22:19 [INFO] test episode 65: reward = -238.50, steps = 200
08:22:32 [INFO] test episode 66: reward = -245.23, steps = 200
08:22:45 [INFO] test episode 67: reward = -618.04, steps = 200
08:22:58 [INFO] test episode 68: reward = -122.65, steps = 200
08:23:11 [INFO] test episode 69: reward = -123.31, steps = 200
08:23:24 [INFO] test episode 70: reward = -246.25, steps = 200
08:23:37 [INFO] test episode 71: reward = -124.21, steps = 200
08:23:50 [INFO] test episode 72: reward = -237.99, steps = 200
08:24:03 [INFO] test episode 73: reward = -245.20, steps = 200
08:24:16 [INFO] test episode 74: reward = -125.24, steps = 200
08:24:29 [INFO] test episode 75: reward = -118.42, steps = 200
08:24:42 [INFO] test episode 76: reward = -615.05, steps = 200
08:24:54 [INFO] test episode 77: reward = -123.28, steps = 200
08:25:07 [INFO] test episode 78: reward = -128.32, steps = 200
08:25:20 [INFO] test episode 79: reward = -122.68, steps = 200
08:25:33 [INFO] test episode 80: reward = -351.11, steps = 200
08:25:46 [INFO] test episode 81: reward = -235.15, steps = 200
08:25:59 [INFO] test episode 82: reward = -4.49, steps = 200
08:26:12 [INFO] test episode 83: reward = -127.92, steps = 200
08:26:25 [INFO] test episode 84: reward = -232.36, steps = 200
08:26:38 [INFO] test episode 85: reward = -0.42, steps = 200
08:26:50 [INFO] test episode 86: reward = -358.81, steps = 200
08:27:04 [INFO] test episode 87: reward = -361.15, steps = 200
08:27:16 [INFO] test episode 88: reward = -235.18, steps = 200
08:27:29 [INFO] test episode 89: reward = -125.39, steps = 200
08:27:42 [INFO] test episode 90: reward = -353.71, steps = 200
08:27:55 [INFO] test episode 91: reward = -238.33, steps = 200
08:28:08 [INFO] test episode 92: reward = -121.59, steps = 200
08:28:21 [INFO] test episode 93: reward = -498.14, steps = 200
08:28:34 [INFO] test episode 94: reward = -244.71, steps = 200
08:28:46 [INFO] test episode 95: reward = -123.82, steps = 200
08:28:59 [INFO] test episode 96: reward = -1.30, steps = 200
08:29:13 [INFO] test episode 97: reward = -241.89, steps = 200
08:29:26 [INFO] test episode 98: reward = -127.86, steps = 200
08:29:38 [INFO] test episode 99: reward = -126.27, steps = 200
08:29:38 [INFO] average episode reward = -241.35 ± 168.17
In [7]:
env.close()