Use Soft Actor-Critic to Play LunarLander-v2

TensorFlow version

%matplotlib inline

import sys
import logging
import itertools

import numpy as np
np.random.seed(0)
import scipy.special
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 nn
from tensorflow import losses
from tensorflow import optimizers
from tensorflow import keras
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')

env = gym.make('LunarLander-v2')
for key in vars(env):
    logging.info('%s: %s', key, vars(env)[key])
for key in vars(env.spec):
    logging.info('%s: %s', key, vars(env.spec)[key])

08:26:06 [INFO] env: <LunarLander<LunarLander-v2>>
08:26:06 [INFO] action_space: Discrete(4)
08:26:06 [INFO] observation_space: Box(-inf, inf, (8,), float32)
08:26:06 [INFO] reward_range: (-inf, inf)
08:26:06 [INFO] metadata: {'render.modes': ['human', 'rgb_array'], 'video.frames_per_second': 50}
08:26:06 [INFO] _max_episode_steps: 1000
08:26:06 [INFO] _elapsed_steps: None
08:26:06 [INFO] id: LunarLander-v2
08:26:06 [INFO] entry_point: gym.envs.box2d:LunarLander
08:26:06 [INFO] reward_threshold: 200
08:26:06 [INFO] nondeterministic: False
08:26:06 [INFO] max_episode_steps: 1000
08:26:06 [INFO] _kwargs: {}
08:26:06 [INFO] _env_name: LunarLander

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)

class SACAgent:
    def __init__(self, env):
        self.action_n = env.action_space.n
        self.gamma = 0.99

        self.replayer = DQNReplayer(100000)

        self.alpha = 0.02

        # create actor
        def sac_loss(y_true, y_pred):
            """ y_true is Q(*, action_n), y_pred is pi(*, action_n) """
            qs = self.alpha * tf.math.xlogy(y_pred, y_pred) - y_pred * y_true
            return tf.reduce_sum(qs, axis=-1)
        self.actor_net = self.build_net(
                hidden_sizes=[256, 256],
                output_size=self.action_n, output_activation=nn.softmax,
                loss=sac_loss)

        # create Q critic
        self.q0_net = self.build_net(
                hidden_sizes=[256, 256],
                output_size=self.action_n)
        self.q1_net = self.build_net(
                hidden_sizes=[256, 256],
                output_size=self.action_n)

        # create V critic
        self.v_evaluate_net = self.build_net(
                hidden_sizes=[256, 256])
        self.v_target_net = models.clone_model(self.v_evaluate_net)

    def build_net(self, hidden_sizes, output_size=1,
                activation=nn.relu, output_activation=None,
                loss=losses.mse, learning_rate=0.0003):
        model = keras.Sequential()
        for hidden_size in hidden_sizes:
            model.add(layers.Dense(units=hidden_size,
                    activation=activation))
        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 = []

    def step(self, observation, reward, terminated):
        probs = self.actor_net.predict(observation[np.newaxis], verbose=0)[0]
        action = np.random.choice(self.action_n, p=probs)
        if self.mode == 'train':
            self.trajectory += [observation, reward, terminated, action]
            if len(self.trajectory) >= 8:
                state, _, _, action, next_state, reward, terminated, _ = \
                        self.trajectory[-8:]
                self.replayer.store(state, action, reward, next_state, terminated)
            if self.replayer.count >= 500:
                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):
        states, actions, rewards, next_states, terminateds = \
                self.replayer.sample(128)

        # update actor
        q0s = self.q0_net.predict(states, verbose=0)
        q1s = self.q1_net.predict(states, verbose=0)
        self.actor_net.fit(states, q0s, verbose=0)

        # update V critic
        q01s = np.minimum(q0s, q1s)
        pis = self.actor_net.predict(states, verbose=0)
        entropic_q01s = pis * q01s - self.alpha * \
                scipy.special.xlogy(pis, pis)
        v_targets = entropic_q01s.sum(axis=-1)
        self.v_evaluate_net.fit(states, v_targets, verbose=0)

        # update Q critic
        next_vs = self.v_target_net.predict(next_states, verbose=0)
        q_targets = rewards[:, np.newaxis] + \
                self.gamma * (1. - terminateds[:, np.newaxis]) * next_vs
        np.put_along_axis(q0s, actions.reshape(-1, 1), q_targets, -1)
        np.put_along_axis(q1s, actions.reshape(-1, 1), q_targets, -1)
        self.q0_net.fit(states, q0s, verbose=0)
        self.q1_net.fit(states, q1s, verbose=0)

        # update v network
        self.update_net(self.v_target_net, self.v_evaluate_net)


agent = SACAgent(env)

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:]) > 250:
        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))

08:26:07 [INFO] ==== train ====
08:26:18 [INFO] train episode 0: reward = -74.72, steps = 113
08:26:29 [INFO] train episode 1: reward = -122.94, steps = 112
08:26:38 [INFO] train episode 2: reward = -365.03, steps = 101
08:26:46 [INFO] train episode 3: reward = -326.18, steps = 90
08:26:53 [INFO] train episode 4: reward = -104.81, steps = 80
08:27:53 [INFO] train episode 5: reward = -198.31, steps = 72
08:28:43 [INFO] train episode 6: reward = -501.15, steps = 61
08:29:41 [INFO] train episode 7: reward = -151.48, steps = 73
08:30:43 [INFO] train episode 8: reward = -115.98, steps = 75
08:31:32 [INFO] train episode 9: reward = -306.37, steps = 61
08:32:38 [INFO] train episode 10: reward = -421.95, steps = 82
08:34:11 [INFO] train episode 11: reward = -381.04, steps = 116
08:35:36 [INFO] train episode 12: reward = -250.08, steps = 106
08:36:50 [INFO] train episode 13: reward = -301.79, steps = 91
08:38:21 [INFO] train episode 14: reward = -81.56, steps = 113
08:40:20 [INFO] train episode 15: reward = -301.11, steps = 148
08:43:04 [INFO] train episode 16: reward = -172.73, steps = 203
08:45:02 [INFO] train episode 17: reward = -225.58, steps = 146
08:48:06 [INFO] train episode 18: reward = -173.76, steps = 230
08:49:40 [INFO] train episode 19: reward = -162.77, steps = 117
08:51:21 [INFO] train episode 20: reward = -313.96, steps = 125
08:54:14 [INFO] train episode 21: reward = -345.04, steps = 202
08:59:11 [INFO] train episode 22: reward = -244.46, steps = 363
09:00:26 [INFO] train episode 23: reward = -237.82, steps = 92
09:02:11 [INFO] train episode 24: reward = -391.98, steps = 128
09:04:28 [INFO] train episode 25: reward = -223.44, steps = 169
09:05:34 [INFO] train episode 26: reward = -250.25, steps = 80
09:07:40 [INFO] train episode 27: reward = -225.68, steps = 154
09:10:00 [INFO] train episode 28: reward = -234.29, steps = 173
09:11:59 [INFO] train episode 29: reward = -231.30, steps = 145
09:13:34 [INFO] train episode 30: reward = -148.45, steps = 117
09:15:32 [INFO] train episode 31: reward = -200.62, steps = 144
09:16:56 [INFO] train episode 32: reward = -60.58, steps = 103
09:18:33 [INFO] train episode 33: reward = -175.80, steps = 118
09:20:49 [INFO] train episode 34: reward = -269.26, steps = 167
09:22:00 [INFO] train episode 35: reward = -125.53, steps = 87
09:23:27 [INFO] train episode 36: reward = -180.59, steps = 106
09:24:45 [INFO] train episode 37: reward = -129.33, steps = 95
09:25:47 [INFO] train episode 38: reward = -163.53, steps = 76
09:29:43 [INFO] train episode 39: reward = -157.50, steps = 289
09:35:22 [INFO] train episode 40: reward = -217.52, steps = 418
09:37:37 [INFO] train episode 41: reward = -386.43, steps = 166
09:38:57 [INFO] train episode 42: reward = -191.00, steps = 99
09:40:28 [INFO] train episode 43: reward = -170.17, steps = 111
09:42:40 [INFO] train episode 44: reward = -277.10, steps = 161
09:47:24 [INFO] train episode 45: reward = -290.33, steps = 349
09:49:40 [INFO] train episode 46: reward = -161.16, steps = 167
09:52:56 [INFO] train episode 47: reward = -55.62, steps = 241
09:55:13 [INFO] train episode 48: reward = -33.28, steps = 170
09:58:01 [INFO] train episode 49: reward = -54.79, steps = 206
09:59:41 [INFO] train episode 50: reward = -15.06, steps = 123
10:01:35 [INFO] train episode 51: reward = -257.93, steps = 141
10:04:50 [INFO] train episode 52: reward = -52.59, steps = 240
10:07:46 [INFO] train episode 53: reward = -68.54, steps = 216
10:18:20 [INFO] train episode 54: reward = -222.73, steps = 781
10:31:54 [INFO] train episode 55: reward = -118.68, steps = 1000
10:45:27 [INFO] train episode 56: reward = -114.53, steps = 1000
10:48:13 [INFO] train episode 57: reward = -31.13, steps = 204
11:01:45 [INFO] train episode 58: reward = -136.05, steps = 1000
11:15:22 [INFO] train episode 59: reward = -108.62, steps = 1000
11:19:43 [INFO] train episode 60: reward = -115.93, steps = 321
11:33:17 [INFO] train episode 61: reward = -78.74, steps = 1000
11:41:32 [INFO] train episode 62: reward = -229.38, steps = 609
11:51:17 [INFO] train episode 63: reward = -108.99, steps = 717
11:55:01 [INFO] train episode 64: reward = -53.12, steps = 274
12:08:12 [INFO] train episode 65: reward = -427.70, steps = 963
12:13:35 [INFO] train episode 66: reward = -90.73, steps = 394
12:27:15 [INFO] train episode 67: reward = -121.45, steps = 1000
12:38:28 [INFO] train episode 68: reward = -159.10, steps = 820
12:51:24 [INFO] train episode 69: reward = -136.81, steps = 946
12:59:12 [INFO] train episode 70: reward = -81.67, steps = 570
13:01:03 [INFO] train episode 71: reward = 11.44, steps = 135
13:11:06 [INFO] train episode 72: reward = -76.24, steps = 1000
13:13:45 [INFO] train episode 73: reward = 11.55, steps = 279
13:23:15 [INFO] train episode 74: reward = -46.13, steps = 1000
13:32:45 [INFO] train episode 75: reward = -82.33, steps = 1000
13:42:15 [INFO] train episode 76: reward = -34.55, steps = 1000
13:51:45 [INFO] train episode 77: reward = -57.85, steps = 1000
14:01:17 [INFO] train episode 78: reward = -40.40, steps = 1000
14:10:48 [INFO] train episode 79: reward = -58.30, steps = 1000
14:20:19 [INFO] train episode 80: reward = -33.73, steps = 1000
14:29:53 [INFO] train episode 81: reward = -65.05, steps = 1000
14:39:24 [INFO] train episode 82: reward = -104.65, steps = 1000
14:48:56 [INFO] train episode 83: reward = -48.76, steps = 1000
14:58:29 [INFO] train episode 84: reward = -49.37, steps = 1000
15:08:01 [INFO] train episode 85: reward = -55.60, steps = 1000
15:17:32 [INFO] train episode 86: reward = -36.14, steps = 1000
15:27:06 [INFO] train episode 87: reward = -48.96, steps = 1000
15:36:36 [INFO] train episode 88: reward = -38.01, steps = 1000
15:46:08 [INFO] train episode 89: reward = -30.15, steps = 1000
15:55:38 [INFO] train episode 90: reward = -73.12, steps = 1000
16:05:10 [INFO] train episode 91: reward = -14.05, steps = 1000
16:14:43 [INFO] train episode 92: reward = -61.70, steps = 1000
16:24:16 [INFO] train episode 93: reward = -26.32, steps = 1000
16:33:47 [INFO] train episode 94: reward = -41.12, steps = 1000
16:43:20 [INFO] train episode 95: reward = -36.86, steps = 1000
16:52:50 [INFO] train episode 96: reward = -63.19, steps = 1000
17:02:22 [INFO] train episode 97: reward = -49.44, steps = 1000
17:11:55 [INFO] train episode 98: reward = -34.78, steps = 1000
17:21:26 [INFO] train episode 99: reward = -17.34, steps = 1000
17:31:01 [INFO] train episode 100: reward = -56.42, steps = 1000
17:40:34 [INFO] train episode 101: reward = -47.08, steps = 1000
17:50:06 [INFO] train episode 102: reward = -21.59, steps = 1000
17:59:39 [INFO] train episode 103: reward = -88.11, steps = 1000
18:09:14 [INFO] train episode 104: reward = -21.69, steps = 1000
18:18:47 [INFO] train episode 105: reward = -40.15, steps = 1000
18:28:24 [INFO] train episode 106: reward = -50.33, steps = 1000
18:37:57 [INFO] train episode 107: reward = -43.57, steps = 1000
18:47:33 [INFO] train episode 108: reward = -47.32, steps = 1000
18:57:09 [INFO] train episode 109: reward = -50.09, steps = 1000
19:06:47 [INFO] train episode 110: reward = -32.77, steps = 1000
19:16:25 [INFO] train episode 111: reward = -42.50, steps = 1000
19:26:07 [INFO] train episode 112: reward = -29.16, steps = 1000
19:35:43 [INFO] train episode 113: reward = -43.81, steps = 1000
19:45:21 [INFO] train episode 114: reward = -44.79, steps = 1000
19:55:00 [INFO] train episode 115: reward = -50.32, steps = 1000
20:04:42 [INFO] train episode 116: reward = -36.38, steps = 1000
20:12:59 [INFO] train episode 117: reward = -52.33, steps = 1000
20:21:13 [INFO] train episode 118: reward = -23.33, steps = 1000
20:29:29 [INFO] train episode 119: reward = -14.81, steps = 1000
20:36:11 [INFO] train episode 120: reward = -45.71, steps = 1000
20:42:36 [INFO] train episode 121: reward = -18.99, steps = 1000
20:48:58 [INFO] train episode 122: reward = -24.43, steps = 1000
20:55:42 [INFO] train episode 123: reward = -31.85, steps = 1000
21:02:42 [INFO] train episode 124: reward = -22.17, steps = 1000
21:09:08 [INFO] train episode 125: reward = -28.34, steps = 1000
21:15:45 [INFO] train episode 126: reward = -9.23, steps = 1000
21:22:31 [INFO] train episode 127: reward = -13.24, steps = 1000
21:29:19 [INFO] train episode 128: reward = -30.17, steps = 1000
21:36:22 [INFO] train episode 129: reward = -25.92, steps = 1000
21:43:10 [INFO] train episode 130: reward = -303.78, steps = 921
21:49:41 [INFO] train episode 131: reward = -4.38, steps = 1000
21:56:12 [INFO] train episode 132: reward = -19.49, steps = 1000
22:02:36 [INFO] train episode 133: reward = -68.60, steps = 1000
22:09:00 [INFO] train episode 134: reward = -14.10, steps = 1000
22:15:44 [INFO] train episode 135: reward = -49.63, steps = 1000
22:22:36 [INFO] train episode 136: reward = -44.75, steps = 1000
22:35:52 [INFO] train episode 137: reward = -9.70, steps = 1000
22:46:22 [INFO] train episode 138: reward = -45.48, steps = 1000
22:57:56 [INFO] train episode 139: reward = -33.41, steps = 1000
23:04:27 [INFO] train episode 140: reward = -42.26, steps = 1000
23:11:10 [INFO] train episode 141: reward = -36.48, steps = 1000
23:18:18 [INFO] train episode 142: reward = -31.02, steps = 1000
23:25:37 [INFO] train episode 143: reward = -74.55, steps = 1000
23:32:44 [INFO] train episode 144: reward = -58.16, steps = 1000
23:39:50 [INFO] train episode 145: reward = -81.09, steps = 1000
23:46:55 [INFO] train episode 146: reward = -11.47, steps = 1000
23:54:00 [INFO] train episode 147: reward = -50.28, steps = 1000
07:41:10 [INFO] train episode 148: reward = -80.80, steps = 1000
07:49:21 [INFO] train episode 149: reward = -29.60, steps = 1000
07:57:27 [INFO] train episode 150: reward = -40.62, steps = 1000
08:05:33 [INFO] train episode 151: reward = -44.47, steps = 1000
08:13:40 [INFO] train episode 152: reward = -62.70, steps = 1000
08:17:32 [INFO] train episode 153: reward = 260.86, steps = 476
08:25:38 [INFO] train episode 154: reward = -34.27, steps = 1000
08:33:47 [INFO] train episode 155: reward = -29.38, steps = 1000
08:41:57 [INFO] train episode 156: reward = -57.31, steps = 1000
08:50:06 [INFO] train episode 157: reward = -29.60, steps = 1000
08:58:15 [INFO] train episode 158: reward = -41.89, steps = 1000
09:06:20 [INFO] train episode 159: reward = -32.09, steps = 1000
09:14:26 [INFO] train episode 160: reward = -26.52, steps = 1000
09:22:34 [INFO] train episode 161: reward = -67.42, steps = 1000
09:30:38 [INFO] train episode 162: reward = -20.92, steps = 1000
09:38:45 [INFO] train episode 163: reward = -67.53, steps = 1000
09:46:50 [INFO] train episode 164: reward = -49.77, steps = 1000
09:54:53 [INFO] train episode 165: reward = -43.56, steps = 1000
10:03:02 [INFO] train episode 166: reward = -9.79, steps = 1000
10:11:04 [INFO] train episode 167: reward = 6.04, steps = 1000
10:19:08 [INFO] train episode 168: reward = -31.60, steps = 1000
10:27:16 [INFO] train episode 169: reward = -68.71, steps = 1000
10:35:20 [INFO] train episode 170: reward = -1.14, steps = 1000
10:43:25 [INFO] train episode 171: reward = -51.36, steps = 1000
10:51:32 [INFO] train episode 172: reward = -54.58, steps = 1000
10:59:41 [INFO] train episode 173: reward = -59.57, steps = 1000
11:07:51 [INFO] train episode 174: reward = -55.77, steps = 1000
11:13:30 [INFO] train episode 175: reward = 136.31, steps = 695
11:21:39 [INFO] train episode 176: reward = 76.23, steps = 1000
11:29:47 [INFO] train episode 177: reward = -38.80, steps = 1000
11:37:55 [INFO] train episode 178: reward = -65.98, steps = 1000
11:46:04 [INFO] train episode 179: reward = 8.52, steps = 1000
11:54:10 [INFO] train episode 180: reward = -32.85, steps = 1000
12:02:16 [INFO] train episode 181: reward = 130.46, steps = 993
12:07:45 [INFO] train episode 182: reward = -135.03, steps = 674
12:09:20 [INFO] train episode 183: reward = 46.50, steps = 194
12:16:21 [INFO] train episode 184: reward = 126.24, steps = 864
12:22:29 [INFO] train episode 185: reward = 132.08, steps = 752
12:28:50 [INFO] train episode 186: reward = 232.98, steps = 785
12:34:51 [INFO] train episode 187: reward = 186.61, steps = 744
12:40:20 [INFO] train episode 188: reward = 158.82, steps = 675
12:46:06 [INFO] train episode 189: reward = 161.32, steps = 708
12:54:13 [INFO] train episode 190: reward = -24.72, steps = 1000
13:00:43 [INFO] train episode 191: reward = 177.27, steps = 799
13:02:10 [INFO] train episode 192: reward = 67.46, steps = 176
13:10:24 [INFO] train episode 193: reward = -29.08, steps = 1000
13:18:31 [INFO] train episode 194: reward = 0.13, steps = 1000
13:24:09 [INFO] train episode 195: reward = 169.79, steps = 690
13:27:48 [INFO] train episode 196: reward = 244.96, steps = 448
13:31:28 [INFO] train episode 197: reward = -50.39, steps = 450
13:33:00 [INFO] train episode 198: reward = 3.41, steps = 188
13:34:21 [INFO] train episode 199: reward = -14.05, steps = 165
13:42:28 [INFO] train episode 200: reward = -24.38, steps = 1000
13:48:14 [INFO] train episode 201: reward = 219.05, steps = 711
13:51:24 [INFO] train episode 202: reward = 0.68, steps = 389
13:56:50 [INFO] train episode 203: reward = 194.04, steps = 669
13:58:33 [INFO] train episode 204: reward = 4.17, steps = 211
14:06:42 [INFO] train episode 205: reward = -47.56, steps = 1000
14:13:36 [INFO] train episode 206: reward = 178.83, steps = 848
14:18:47 [INFO] train episode 207: reward = 190.24, steps = 641
14:22:44 [INFO] train episode 208: reward = 239.04, steps = 485
14:25:44 [INFO] train episode 209: reward = 253.79, steps = 374
14:27:06 [INFO] train episode 210: reward = -16.37, steps = 167
14:28:03 [INFO] train episode 211: reward = -41.71, steps = 117
14:31:41 [INFO] train episode 212: reward = 225.23, steps = 453
14:32:56 [INFO] train episode 213: reward = -13.32, steps = 153
14:34:32 [INFO] train episode 214: reward = -19.53, steps = 199
14:36:03 [INFO] train episode 215: reward = 74.19, steps = 188
14:37:00 [INFO] train episode 216: reward = -42.01, steps = 117
14:37:41 [INFO] train episode 217: reward = -75.08, steps = 83
14:40:21 [INFO] train episode 218: reward = -24.57, steps = 330
14:48:28 [INFO] train episode 219: reward = 21.95, steps = 1000
14:50:23 [INFO] train episode 220: reward = 5.00, steps = 237
14:54:10 [INFO] train episode 221: reward = 215.37, steps = 465
15:00:00 [INFO] train episode 222: reward = 162.49, steps = 714
15:05:04 [INFO] train episode 223: reward = 166.78, steps = 624
15:06:34 [INFO] train episode 224: reward = 39.70, steps = 183
15:12:04 [INFO] train episode 225: reward = 209.77, steps = 678
15:15:17 [INFO] train episode 226: reward = 205.89, steps = 396
15:18:59 [INFO] train episode 227: reward = 231.10, steps = 457
15:20:27 [INFO] train episode 228: reward = 26.82, steps = 179
15:24:12 [INFO] train episode 229: reward = 239.76, steps = 462
15:28:13 [INFO] train episode 230: reward = 230.77, steps = 488
15:31:36 [INFO] train episode 231: reward = 263.25, steps = 416
15:34:23 [INFO] train episode 232: reward = -56.34, steps = 343
15:36:37 [INFO] train episode 233: reward = -35.99, steps = 275
15:39:23 [INFO] train episode 234: reward = -44.81, steps = 339
15:41:48 [INFO] train episode 235: reward = -46.11, steps = 295
15:44:37 [INFO] train episode 236: reward = 227.03, steps = 345
15:47:52 [INFO] train episode 237: reward = 219.44, steps = 398
15:51:13 [INFO] train episode 238: reward = 206.12, steps = 411
15:53:50 [INFO] train episode 239: reward = 203.33, steps = 322
15:54:50 [INFO] train episode 240: reward = -54.80, steps = 121
15:57:42 [INFO] train episode 241: reward = 223.04, steps = 351
16:01:13 [INFO] train episode 242: reward = 241.18, steps = 431
16:05:05 [INFO] train episode 243: reward = 181.11, steps = 474
16:08:04 [INFO] train episode 244: reward = 223.16, steps = 366
16:10:44 [INFO] train episode 245: reward = 234.13, steps = 327
16:13:40 [INFO] train episode 246: reward = 17.85, steps = 360
16:16:41 [INFO] train episode 247: reward = 259.35, steps = 372
16:19:19 [INFO] train episode 248: reward = 208.94, steps = 324
16:20:38 [INFO] train episode 249: reward = 59.40, steps = 157
16:23:10 [INFO] train episode 250: reward = 231.79, steps = 311
16:26:01 [INFO] train episode 251: reward = 263.69, steps = 350
16:29:18 [INFO] train episode 252: reward = 216.53, steps = 402
16:32:28 [INFO] train episode 253: reward = 240.86, steps = 390
16:35:36 [INFO] train episode 254: reward = 251.86, steps = 384
16:38:02 [INFO] train episode 255: reward = 228.69, steps = 300
16:42:17 [INFO] train episode 256: reward = 267.35, steps = 525
16:46:41 [INFO] train episode 257: reward = -51.68, steps = 544
16:50:12 [INFO] train episode 258: reward = 242.94, steps = 437
16:53:04 [INFO] train episode 259: reward = 274.39, steps = 356
16:55:36 [INFO] train episode 260: reward = 268.78, steps = 313
16:58:13 [INFO] train episode 261: reward = -55.59, steps = 323
17:00:48 [INFO] train episode 262: reward = 219.12, steps = 318
17:03:14 [INFO] train episode 263: reward = 266.82, steps = 304
17:05:58 [INFO] train episode 264: reward = 273.98, steps = 338
17:10:38 [INFO] train episode 265: reward = -149.89, steps = 579
17:13:58 [INFO] train episode 266: reward = 241.03, steps = 416
17:17:10 [INFO] train episode 267: reward = 227.13, steps = 400
17:20:11 [INFO] train episode 268: reward = 220.56, steps = 376
17:22:42 [INFO] train episode 269: reward = 236.69, steps = 309
17:25:07 [INFO] train episode 270: reward = 237.90, steps = 301
17:28:36 [INFO] train episode 271: reward = 246.74, steps = 434
17:32:04 [INFO] train episode 272: reward = -157.47, steps = 431
17:35:02 [INFO] train episode 273: reward = 249.70, steps = 366
17:36:09 [INFO] train episode 274: reward = 48.31, steps = 138
17:38:49 [INFO] train episode 275: reward = 225.81, steps = 332
17:40:47 [INFO] train episode 276: reward = -18.32, steps = 245
17:43:15 [INFO] train episode 277: reward = 285.75, steps = 304
17:46:25 [INFO] train episode 278: reward = 221.22, steps = 392
17:49:31 [INFO] train episode 279: reward = 230.84, steps = 384
17:50:29 [INFO] train episode 280: reward = 47.23, steps = 119
17:55:23 [INFO] train episode 281: reward = 276.33, steps = 607
17:57:35 [INFO] train episode 282: reward = 284.50, steps = 273
18:01:19 [INFO] train episode 283: reward = 129.49, steps = 459
18:02:35 [INFO] train episode 284: reward = 69.78, steps = 157
18:05:10 [INFO] train episode 285: reward = 234.56, steps = 319
18:08:36 [INFO] train episode 286: reward = 193.87, steps = 425
18:11:01 [INFO] train episode 287: reward = 220.68, steps = 300
18:14:03 [INFO] train episode 288: reward = 211.85, steps = 378
18:17:43 [INFO] train episode 289: reward = 178.74, steps = 454
18:18:47 [INFO] train episode 290: reward = 16.04, steps = 131
18:21:19 [INFO] train episode 291: reward = 244.04, steps = 309
18:24:33 [INFO] train episode 292: reward = 211.80, steps = 403
18:27:37 [INFO] train episode 293: reward = 234.26, steps = 379
18:28:43 [INFO] train episode 294: reward = 4.04, steps = 136
18:32:03 [INFO] train episode 295: reward = -202.65, steps = 413
18:36:41 [INFO] train episode 296: reward = -67.90, steps = 580
18:38:34 [INFO] train episode 297: reward = 13.09, steps = 232
18:40:55 [INFO] train episode 298: reward = 244.17, steps = 293
18:43:24 [INFO] train episode 299: reward = 266.41, steps = 307
18:45:34 [INFO] train episode 300: reward = 287.59, steps = 268
18:48:08 [INFO] train episode 301: reward = 182.31, steps = 319
18:53:06 [INFO] train episode 302: reward = -84.44, steps = 619
18:55:57 [INFO] train episode 303: reward = 221.66, steps = 355
18:58:55 [INFO] train episode 304: reward = 237.39, steps = 369
19:01:21 [INFO] train episode 305: reward = 238.81, steps = 302
19:04:13 [INFO] train episode 306: reward = 216.63, steps = 358
19:06:40 [INFO] train episode 307: reward = 248.08, steps = 305
19:09:14 [INFO] train episode 308: reward = 186.99, steps = 319
19:12:07 [INFO] train episode 309: reward = 214.30, steps = 357
19:14:46 [INFO] train episode 310: reward = 256.59, steps = 326
19:18:28 [INFO] train episode 311: reward = -188.49, steps = 459
19:21:27 [INFO] train episode 312: reward = 235.55, steps = 368
19:23:29 [INFO] train episode 313: reward = 17.29, steps = 251
19:26:01 [INFO] train episode 314: reward = 243.72, steps = 313
19:28:19 [INFO] train episode 315: reward = 225.16, steps = 284
19:30:48 [INFO] train episode 316: reward = 260.91, steps = 309
19:33:44 [INFO] train episode 317: reward = 214.20, steps = 362
19:36:11 [INFO] train episode 318: reward = 241.11, steps = 303
19:38:37 [INFO] train episode 319: reward = 251.71, steps = 300
19:41:09 [INFO] train episode 320: reward = 232.01, steps = 314
19:45:06 [INFO] train episode 321: reward = 145.96, steps = 487
19:47:13 [INFO] train episode 322: reward = -40.42, steps = 261
19:49:25 [INFO] train episode 323: reward = 264.88, steps = 270
19:50:23 [INFO] train episode 324: reward = -120.74, steps = 118
19:52:34 [INFO] train episode 325: reward = 46.12, steps = 269
19:54:47 [INFO] train episode 326: reward = -42.49, steps = 273
19:56:51 [INFO] train episode 327: reward = 231.45, steps = 254
19:59:06 [INFO] train episode 328: reward = 251.18, steps = 277
20:01:28 [INFO] train episode 329: reward = 236.58, steps = 291
20:03:05 [INFO] train episode 330: reward = 38.96, steps = 201
20:05:18 [INFO] train episode 331: reward = 256.06, steps = 276
20:08:16 [INFO] train episode 332: reward = 265.92, steps = 357
20:10:34 [INFO] train episode 333: reward = 246.93, steps = 269
20:13:15 [INFO] train episode 334: reward = 223.24, steps = 306
20:15:45 [INFO] train episode 335: reward = 256.67, steps = 286
20:17:57 [INFO] train episode 336: reward = 7.83, steps = 263
20:21:03 [INFO] train episode 337: reward = 288.15, steps = 365
20:23:06 [INFO] train episode 338: reward = 5.72, steps = 252
20:25:38 [INFO] train episode 339: reward = 212.55, steps = 303
20:28:25 [INFO] train episode 340: reward = 242.82, steps = 337
20:31:54 [INFO] train episode 341: reward = 272.67, steps = 419
20:34:06 [INFO] train episode 342: reward = -33.61, steps = 262
20:37:04 [INFO] train episode 343: reward = 268.03, steps = 337
20:40:48 [INFO] train episode 344: reward = 237.82, steps = 431
20:43:30 [INFO] train episode 345: reward = 257.01, steps = 330
20:45:50 [INFO] train episode 346: reward = 262.86, steps = 271
20:49:02 [INFO] train episode 347: reward = 297.80, steps = 274
20:52:22 [INFO] train episode 348: reward = 228.71, steps = 265
20:56:33 [INFO] train episode 349: reward = 250.48, steps = 322
21:01:47 [INFO] train episode 350: reward = 270.99, steps = 492
21:04:31 [INFO] train episode 351: reward = 8.77, steps = 301
21:06:55 [INFO] train episode 352: reward = 276.75, steps = 266
21:11:44 [INFO] train episode 353: reward = 267.92, steps = 355
21:15:47 [INFO] train episode 354: reward = -27.88, steps = 305
21:20:38 [INFO] train episode 355: reward = 254.92, steps = 352
21:25:59 [INFO] train episode 356: reward = 228.31, steps = 397
21:29:36 [INFO] train episode 357: reward = 237.62, steps = 283
21:33:10 [INFO] train episode 358: reward = 261.00, steps = 290
21:38:22 [INFO] train episode 359: reward = 272.50, steps = 390
21:41:42 [INFO] train episode 360: reward = 259.38, steps = 268
21:45:08 [INFO] train episode 361: reward = 257.31, steps = 278
21:49:10 [INFO] train episode 362: reward = 259.91, steps = 318
21:52:47 [INFO] train episode 363: reward = 11.26, steps = 280
21:56:25 [INFO] train episode 364: reward = 220.58, steps = 271
22:00:50 [INFO] train episode 365: reward = 230.59, steps = 334
22:06:01 [INFO] train episode 366: reward = 239.75, steps = 406
22:09:19 [INFO] train episode 367: reward = 287.08, steps = 251
22:13:13 [INFO] train episode 368: reward = 280.66, steps = 288
22:16:47 [INFO] train episode 369: reward = 255.24, steps = 270
22:22:32 [INFO] train episode 370: reward = 240.68, steps = 418
22:26:01 [INFO] train episode 371: reward = 278.65, steps = 253
22:30:23 [INFO] train episode 372: reward = 284.44, steps = 325
22:35:41 [INFO] train episode 373: reward = 212.39, steps = 388
22:35:41 [INFO] ==== test ====
22:36:05 [INFO] test episode 0: reward = -32.11, steps = 289
22:36:28 [INFO] test episode 1: reward = -32.23, steps = 255
22:36:56 [INFO] test episode 2: reward = 278.20, steps = 312
22:37:21 [INFO] test episode 3: reward = 265.34, steps = 287
22:37:43 [INFO] test episode 4: reward = 247.70, steps = 253
22:38:09 [INFO] test episode 5: reward = 211.36, steps = 277
22:38:34 [INFO] test episode 6: reward = 226.89, steps = 272
22:38:58 [INFO] test episode 7: reward = 281.83, steps = 263
22:39:30 [INFO] test episode 8: reward = 216.34, steps = 356
22:39:55 [INFO] test episode 9: reward = -27.73, steps = 276
22:40:20 [INFO] test episode 10: reward = 219.64, steps = 263
22:40:54 [INFO] test episode 11: reward = 284.99, steps = 375
22:41:28 [INFO] test episode 12: reward = 216.88, steps = 353
22:41:55 [INFO] test episode 13: reward = 264.33, steps = 307
22:42:21 [INFO] test episode 14: reward = 252.44, steps = 290
22:42:54 [INFO] test episode 15: reward = -32.08, steps = 357
22:43:21 [INFO] test episode 16: reward = 287.76, steps = 292
22:43:50 [INFO] test episode 17: reward = 288.85, steps = 335
22:44:14 [INFO] test episode 18: reward = 245.75, steps = 280
22:44:38 [INFO] test episode 19: reward = 252.16, steps = 259
22:45:04 [INFO] test episode 20: reward = 235.86, steps = 285
22:45:29 [INFO] test episode 21: reward = -41.28, steps = 282
22:46:02 [INFO] test episode 22: reward = -32.77, steps = 354
22:46:26 [INFO] test episode 23: reward = 251.89, steps = 281
22:46:51 [INFO] test episode 24: reward = 245.60, steps = 286
22:47:27 [INFO] test episode 25: reward = 233.73, steps = 378
22:47:48 [INFO] test episode 26: reward = -51.75, steps = 230
22:48:19 [INFO] test episode 27: reward = 251.16, steps = 341
22:48:51 [INFO] test episode 28: reward = 1.56, steps = 340
22:49:21 [INFO] test episode 29: reward = 285.28, steps = 338
22:49:54 [INFO] test episode 30: reward = 267.00, steps = 364
22:50:25 [INFO] test episode 31: reward = 269.83, steps = 337
22:50:54 [INFO] test episode 32: reward = 204.35, steps = 325
22:52:27 [INFO] test episode 33: reward = 83.91, steps = 1000
22:52:50 [INFO] test episode 34: reward = 259.93, steps = 260
22:53:31 [INFO] test episode 35: reward = 277.19, steps = 461
22:54:02 [INFO] test episode 36: reward = 224.20, steps = 333
22:54:43 [INFO] test episode 37: reward = 195.79, steps = 498
22:55:10 [INFO] test episode 38: reward = 286.14, steps = 311
22:55:45 [INFO] test episode 39: reward = 6.13, steps = 370
22:56:23 [INFO] test episode 40: reward = 211.74, steps = 443
22:56:51 [INFO] test episode 41: reward = 282.40, steps = 310
22:57:19 [INFO] test episode 42: reward = 220.18, steps = 327
22:57:49 [INFO] test episode 43: reward = 266.89, steps = 319
22:58:11 [INFO] test episode 44: reward = -12.16, steps = 231
22:58:34 [INFO] test episode 45: reward = -20.85, steps = 272
22:58:56 [INFO] test episode 46: reward = 241.66, steps = 251
22:59:19 [INFO] test episode 47: reward = 239.20, steps = 268
22:59:45 [INFO] test episode 48: reward = 260.02, steps = 294
23:00:11 [INFO] test episode 49: reward = -26.97, steps = 279
23:00:37 [INFO] test episode 50: reward = 269.95, steps = 301
23:01:09 [INFO] test episode 51: reward = 263.39, steps = 351
23:01:32 [INFO] test episode 52: reward = 236.69, steps = 251
23:02:00 [INFO] test episode 53: reward = 287.57, steps = 334
23:02:27 [INFO] test episode 54: reward = 275.76, steps = 300
23:02:53 [INFO] test episode 55: reward = 247.01, steps = 263
23:03:18 [INFO] test episode 56: reward = 243.62, steps = 251
23:03:44 [INFO] test episode 57: reward = 226.51, steps = 302
23:04:12 [INFO] test episode 58: reward = 279.56, steps = 280
23:04:43 [INFO] test episode 59: reward = 274.73, steps = 311
23:05:07 [INFO] test episode 60: reward = 264.30, steps = 253
23:05:33 [INFO] test episode 61: reward = 242.01, steps = 263
23:06:00 [INFO] test episode 62: reward = 263.38, steps = 257
23:06:23 [INFO] test episode 63: reward = 228.83, steps = 274
23:06:49 [INFO] test episode 64: reward = 243.65, steps = 269
23:07:24 [INFO] test episode 65: reward = 299.48, steps = 342
23:07:48 [INFO] test episode 66: reward = 237.23, steps = 263
23:08:14 [INFO] test episode 67: reward = 237.08, steps = 270
23:08:49 [INFO] test episode 68: reward = 285.57, steps = 384
23:09:16 [INFO] test episode 69: reward = -3.69, steps = 287
23:09:48 [INFO] test episode 70: reward = 284.44, steps = 355
23:10:17 [INFO] test episode 71: reward = 230.55, steps = 343
23:10:42 [INFO] test episode 72: reward = 219.30, steps = 273
23:11:08 [INFO] test episode 73: reward = 276.11, steps = 288
23:11:33 [INFO] test episode 74: reward = 258.88, steps = 291
23:11:58 [INFO] test episode 75: reward = 2.95, steps = 330
23:12:19 [INFO] test episode 76: reward = 242.20, steps = 264
23:12:42 [INFO] test episode 77: reward = 221.27, steps = 289
23:13:17 [INFO] test episode 78: reward = 228.65, steps = 436
23:13:43 [INFO] test episode 79: reward = 279.87, steps = 324
23:14:10 [INFO] test episode 80: reward = 288.13, steps = 338
23:14:37 [INFO] test episode 81: reward = 271.24, steps = 354
23:15:04 [INFO] test episode 82: reward = 265.60, steps = 342
23:15:25 [INFO] test episode 83: reward = 208.81, steps = 265
23:15:46 [INFO] test episode 84: reward = 217.22, steps = 274
23:16:08 [INFO] test episode 85: reward = 252.02, steps = 282
23:16:28 [INFO] test episode 86: reward = -30.86, steps = 256
23:16:51 [INFO] test episode 87: reward = -31.61, steps = 281
23:17:22 [INFO] test episode 88: reward = 288.17, steps = 398
23:18:18 [INFO] test episode 89: reward = 154.86, steps = 693
23:18:43 [INFO] test episode 90: reward = 261.03, steps = 309
23:19:10 [INFO] test episode 91: reward = 271.89, steps = 320
23:19:30 [INFO] test episode 92: reward = 264.77, steps = 249
23:19:50 [INFO] test episode 93: reward = -32.41, steps = 249
23:20:15 [INFO] test episode 94: reward = 289.10, steps = 301
23:20:38 [INFO] test episode 95: reward = 203.40, steps = 301
23:21:03 [INFO] test episode 96: reward = -21.69, steps = 312
23:21:25 [INFO] test episode 97: reward = 234.05, steps = 264
23:22:15 [INFO] test episode 98: reward = 160.78, steps = 620
23:22:38 [INFO] test episode 99: reward = -13.88, steps = 280
23:22:38 [INFO] average episode reward = 196.80 ± 111.02

env.close()