
%matplotlib inline

import copy
import logging
import itertools
import sys

import numpy as np
np.random.seed(0)
import pandas as pd
import gym
from gym.wrappers.atari_preprocessing import AtariPreprocessing
from gym.wrappers.frame_stack import FrameStack
import matplotlib.pyplot as plt
import torch
torch.manual_seed(0)
from torch import nn
from torch import optim

logging.basicConfig(level=logging.INFO,
        format='%(asctime)s [%(levelname)s] %(message)s',
        stream=sys.stdout, datefmt='%H:%M:%S')


env = gym.make('PongNoFrameskip-v4')
env = FrameStack(AtariPreprocessing(env), num_stack=4)
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])

11:37:24 [INFO] env: <AtariPreprocessing<TimeLimit<AtariEnv<PongNoFrameskip-v4>>>>
11:37:24 [INFO] action_space: Discrete(6)
11:37:24 [INFO] observation_space: : Box([[[0 0 0 ... 0 0 0]
  [0 0 0 ... 0 0 0]
  [0 0 0 ... 0 0 0]
  ...
  [0 0 0 ... 0 0 0]
  [0 0 0 ... 0 0 0]
  [0 0 0 ... 0 0 0]]

 [[0 0 0 ... 0 0 0]
  [0 0 0 ... 0 0 0]
  [0 0 0 ... 0 0 0]
  ...
  [0 0 0 ... 0 0 0]
  [0 0 0 ... 0 0 0]
  [0 0 0 ... 0 0 0]]

 [[0 0 0 ... 0 0 0]
  [0 0 0 ... 0 0 0]
  [0 0 0 ... 0 0 0]
  ...
  [0 0 0 ... 0 0 0]
  [0 0 0 ... 0 0 0]
  [0 0 0 ... 0 0 0]]

 [[0 0 0 ... 0 0 0]
  [0 0 0 ... 0 0 0]
  [0 0 0 ... 0 0 0]
  ...
  [0 0 0 ... 0 0 0]
  [0 0 0 ... 0 0 0]
  [0 0 0 ... 0 0 0]]], [[[255 255 255 ... 255 255 255]
  [255 255 255 ... 255 255 255]
  [255 255 255 ... 255 255 255]
  ...
  [255 255 255 ... 255 255 255]
  [255 255 255 ... 255 255 255]
  [255 255 255 ... 255 255 255]]

 [[255 255 255 ... 255 255 255]
  [255 255 255 ... 255 255 255]
  [255 255 255 ... 255 255 255]
  ...
  [255 255 255 ... 255 255 255]
  [255 255 255 ... 255 255 255]
  [255 255 255 ... 255 255 255]]

 [[255 255 255 ... 255 255 255]
  [255 255 255 ... 255 255 255]
  [255 255 255 ... 255 255 255]
  ...
  [255 255 255 ... 255 255 255]
  [255 255 255 ... 255 255 255]
  [255 255 255 ... 255 255 255]]

 [[255 255 255 ... 255 255 255]
  [255 255 255 ... 255 255 255]
  [255 255 255 ... 255 255 255]
  ...
  [255 255 255 ... 255 255 255]
  [255 255 255 ... 255 255 255]
  [255 255 255 ... 255 255 255]]], (4, 84, 84), uint8)
11:37:24 [INFO] reward_range: (-inf, inf)
11:37:24 [INFO] metadata: {'render.modes': ['human', 'rgb_array']}
11:37:24 [INFO] num_stack: 4
11:37:24 [INFO] lz4_compress: False
11:37:24 [INFO] frames: deque([], maxlen=4)
11:37:24 [INFO] id: PongNoFrameskip-v4
11:37:24 [INFO] entry_point: gym.envs.atari:AtariEnv
11:37:24 [INFO] reward_threshold: None
11:37:24 [INFO] nondeterministic: False
11:37:24 [INFO] max_episode_steps: 400000
11:37:24 [INFO] _kwargs: {'game': 'pong', 'obs_type': 'image', 'frameskip': 1}
11:37:24 [INFO] _env_name: PongNoFrameskip


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 CategoricalDQNAgent:
    def __init__(self, env):
        self.action_n = env.action_space.n
        self.gamma = 0.99
        self.epsilon = 1.  # exploration

        self.replayer = DQNReplayer(capacity=100000)

        self.atom_count = 51
        self.atom_min = -10.
        self.atom_max = 10.
        self.atom_difference = (self.atom_max - self.atom_min) \
                / (self.atom_count - 1)
        self.atom_tensor = torch.linspace(self.atom_min, self.atom_max,
                self.atom_count)

        self.evaluate_net = nn.Sequential(
                nn.Conv2d(4, 32, kernel_size=8, stride=4), nn.ReLU(),
                nn.Conv2d(32, 64, kernel_size=4, stride=2), nn.ReLU(),
                nn.Conv2d(64, 64, kernel_size=3, stride=1), nn.ReLU(),
                nn.Flatten(),
                nn.Linear(3136, 512), nn.ReLU(inplace=True),
                nn.Linear(512, self.action_n * self.atom_count))
        self.target_net = copy.deepcopy(self.evaluate_net)
        self.optimizer = optim.Adam(self.evaluate_net.parameters(), lr=0.0001)

    def reset(self, mode=None):
        self.mode = mode
        if mode == 'train':
            self.trajectory = []

    def step(self, observation, reward, terminated):
        state_tensor = torch.as_tensor(observation,
                dtype=torch.float).unsqueeze(0)
        logit_tensor = self.evaluate_net(state_tensor).view(-1, self.action_n,
                self.atom_count)
        prob_tensor = logit_tensor.softmax(dim=-1)
        q_component_tensor = prob_tensor * self.atom_tensor
        q_tensor = q_component_tensor.mean(2)
        action_tensor = q_tensor.argmax(dim=1)
        actions = action_tensor.detach().numpy()
        action = actions[0]
        if self.mode == 'train':
            if np.random.rand() < self.epsilon:
                action = np.random.randint(0, self.action_n)
            
            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 >= 1024 and self.replayer.count % 10 == 0:
                self.learn()
        return action

    def close(self):
        pass

    def update_net(self, target_net, evaluate_net, learning_rate=0.005):
        for target_param, evaluate_param in zip(
                target_net.parameters(), evaluate_net.parameters()):
            target_param.data.copy_(learning_rate * evaluate_param.data
                    + (1 - learning_rate) * target_param.data)

    def learn(self):
        # replay
        batch_size = 32
        states, actions, rewards, next_states, terminateds = \
                self.replayer.sample(batch_size)
        state_tensor = torch.as_tensor(states, dtype=torch.float)
        reward_tensor = torch.as_tensor(rewards, dtype=torch.float)
        terminated_tensor = torch.as_tensor(terminateds, dtype=torch.float)
        next_state_tensor = torch.as_tensor(next_states, dtype=torch.float)

        # compute target
        next_logit_tensor = self.target_net(next_state_tensor).view(-1,
                self.action_n, self.atom_count)
        next_prob_tensor = next_logit_tensor.softmax(dim=-1)
        next_q_tensor = (next_prob_tensor * self.atom_tensor).sum(2)
        next_action_tensor = next_q_tensor.argmax(dim=1)
        next_actions = next_action_tensor.detach().numpy()
        next_dist_tensor = next_prob_tensor[np.arange(batch_size),
                next_actions, :].unsqueeze(1)
        
        # project
        target_tensor = reward_tensor.reshape(batch_size, 1) + self.gamma \
                * self.atom_tensor.repeat(batch_size, 1) \
                * (1. - terminated_tensor).reshape(-1, 1)
        clipped_target_tensor = target_tensor.clamp(self.atom_min,
                self.atom_max)
        projection_tensor = (1. - (clipped_target_tensor.unsqueeze(1)
                - self.atom_tensor.view(1, -1, 1)).abs()
                / self.atom_difference).clamp(0, 1)
        projected_tensor = (projection_tensor * next_dist_tensor).sum(-1)

        logit_tensor = self.evaluate_net(state_tensor).view(-1, self.action_n,
                self.atom_count)
        all_q_prob_tensor = logit_tensor.softmax(dim=-1)
        q_prob_tensor = all_q_prob_tensor[range(batch_size), actions, :]

        cross_entropy_tensor = -torch.xlogy(projected_tensor, q_prob_tensor
                + 1e-8).sum(1)
        loss_tensor = cross_entropy_tensor.mean()
        self.optimizer.zero_grad()
        loss_tensor.backward()
        self.optimizer.step()

        self.update_net(self.target_net, self.evaluate_net)

        self.epsilon = max(self.epsilon - 1e-5, 0.05)


agent = CategoricalDQNAgent(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, 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[-5:]) > 16.:
        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))

11:37:24 [INFO] ==== train ====
11:37:47 [INFO] train episode 0: reward = -19.00, steps = 1095
11:38:20 [INFO] train episode 1: reward = -20.00, steps = 945
11:38:53 [INFO] train episode 2: reward = -20.00, steps = 917
11:39:25 [INFO] train episode 3: reward = -21.00, steps = 879
11:39:56 [INFO] train episode 4: reward = -21.00, steps = 863
11:40:28 [INFO] train episode 5: reward = -20.00, steps = 837
11:41:05 [INFO] train episode 6: reward = -20.00, steps = 925
11:41:41 [INFO] train episode 7: reward = -20.00, steps = 966
11:42:12 [INFO] train episode 8: reward = -21.00, steps = 785
11:42:41 [INFO] train episode 9: reward = -21.00, steps = 757
11:43:16 [INFO] train episode 10: reward = -19.00, steps = 919
11:43:53 [INFO] train episode 11: reward = -20.00, steps = 960
11:44:24 [INFO] train episode 12: reward = -21.00, steps = 761
11:44:55 [INFO] train episode 13: reward = -21.00, steps = 816
11:45:25 [INFO] train episode 14: reward = -21.00, steps = 762
11:46:02 [INFO] train episode 15: reward = -20.00, steps = 943
11:46:36 [INFO] train episode 16: reward = -20.00, steps = 887
11:47:15 [INFO] train episode 17: reward = -21.00, steps = 996
11:47:55 [INFO] train episode 18: reward = -20.00, steps = 1020
11:48:28 [INFO] train episode 19: reward = -21.00, steps = 852
11:49:11 [INFO] train episode 20: reward = -20.00, steps = 1098
11:49:53 [INFO] train episode 21: reward = -19.00, steps = 1006
11:50:29 [INFO] train episode 22: reward = -20.00, steps = 882
11:51:03 [INFO] train episode 23: reward = -20.00, steps = 863
11:51:36 [INFO] train episode 24: reward = -21.00, steps = 837
11:52:14 [INFO] train episode 25: reward = -20.00, steps = 1004
11:52:51 [INFO] train episode 26: reward = -20.00, steps = 939
11:53:39 [INFO] train episode 27: reward = -19.00, steps = 1249
11:54:15 [INFO] train episode 28: reward = -20.00, steps = 928
11:54:58 [INFO] train episode 29: reward = -20.00, steps = 1051
11:55:28 [INFO] train episode 30: reward = -21.00, steps = 757
11:56:06 [INFO] train episode 31: reward = -21.00, steps = 938
11:56:43 [INFO] train episode 32: reward = -19.00, steps = 938
11:57:23 [INFO] train episode 33: reward = -20.00, steps = 1008
11:57:57 [INFO] train episode 34: reward = -21.00, steps = 847
11:58:34 [INFO] train episode 35: reward = -20.00, steps = 895
11:59:08 [INFO] train episode 36: reward = -21.00, steps = 846
11:59:43 [INFO] train episode 37: reward = -21.00, steps = 878
12:00:26 [INFO] train episode 38: reward = -21.00, steps = 1060
12:01:03 [INFO] train episode 39: reward = -20.00, steps = 922
12:01:39 [INFO] train episode 40: reward = -21.00, steps = 910
12:02:16 [INFO] train episode 41: reward = -20.00, steps = 917
12:02:49 [INFO] train episode 42: reward = -21.00, steps = 849
12:03:21 [INFO] train episode 43: reward = -21.00, steps = 791
12:03:54 [INFO] train episode 44: reward = -20.00, steps = 840
12:04:33 [INFO] train episode 45: reward = -21.00, steps = 973
12:05:04 [INFO] train episode 46: reward = -21.00, steps = 790
12:05:39 [INFO] train episode 47: reward = -21.00, steps = 883
12:06:15 [INFO] train episode 48: reward = -21.00, steps = 905
12:07:03 [INFO] train episode 49: reward = -18.00, steps = 1186
12:07:43 [INFO] train episode 50: reward = -21.00, steps = 1032
12:08:25 [INFO] train episode 51: reward = -20.00, steps = 1020
12:08:57 [INFO] train episode 52: reward = -21.00, steps = 817
12:09:32 [INFO] train episode 53: reward = -21.00, steps = 866
12:10:15 [INFO] train episode 54: reward = -19.00, steps = 1070
12:10:53 [INFO] train episode 55: reward = -19.00, steps = 933
12:11:32 [INFO] train episode 56: reward = -19.00, steps = 971
12:12:11 [INFO] train episode 57: reward = -20.00, steps = 983
12:12:44 [INFO] train episode 58: reward = -21.00, steps = 804
12:13:17 [INFO] train episode 59: reward = -21.00, steps = 823
12:13:52 [INFO] train episode 60: reward = -20.00, steps = 864
12:14:33 [INFO] train episode 61: reward = -20.00, steps = 1020
12:15:09 [INFO] train episode 62: reward = -21.00, steps = 898
12:15:51 [INFO] train episode 63: reward = -20.00, steps = 1037
12:16:30 [INFO] train episode 64: reward = -19.00, steps = 974
12:17:04 [INFO] train episode 65: reward = -21.00, steps = 824
12:17:40 [INFO] train episode 66: reward = -21.00, steps = 867
12:18:18 [INFO] train episode 67: reward = -20.00, steps = 930
12:19:03 [INFO] train episode 68: reward = -20.00, steps = 1112
12:19:38 [INFO] train episode 69: reward = -21.00, steps = 848
12:20:11 [INFO] train episode 70: reward = -21.00, steps = 806
12:20:44 [INFO] train episode 71: reward = -21.00, steps = 808
12:21:24 [INFO] train episode 72: reward = -20.00, steps = 956
12:21:59 [INFO] train episode 73: reward = -21.00, steps = 848
12:22:41 [INFO] train episode 74: reward = -20.00, steps = 1042
12:23:26 [INFO] train episode 75: reward = -18.00, steps = 1099
12:24:00 [INFO] train episode 76: reward = -21.00, steps = 847
12:24:39 [INFO] train episode 77: reward = -20.00, steps = 958
12:25:27 [INFO] train episode 78: reward = -19.00, steps = 1016
12:26:15 [INFO] train episode 79: reward = -19.00, steps = 1025
12:26:57 [INFO] train episode 80: reward = -20.00, steps = 914
12:27:37 [INFO] train episode 81: reward = -21.00, steps = 848
12:28:19 [INFO] train episode 82: reward = -21.00, steps = 884
12:29:04 [INFO] train episode 83: reward = -21.00, steps = 946
12:30:00 [INFO] train episode 84: reward = -20.00, steps = 1163
12:30:44 [INFO] train episode 85: reward = -20.00, steps = 948
12:31:30 [INFO] train episode 86: reward = -21.00, steps = 957
12:32:11 [INFO] train episode 87: reward = -21.00, steps = 876
12:33:04 [INFO] train episode 88: reward = -17.00, steps = 1112
12:33:52 [INFO] train episode 89: reward = -20.00, steps = 1020
12:34:37 [INFO] train episode 90: reward = -21.00, steps = 947
12:35:20 [INFO] train episode 91: reward = -21.00, steps = 899
12:36:04 [INFO] train episode 92: reward = -19.00, steps = 930
12:36:46 [INFO] train episode 93: reward = -21.00, steps = 880
12:37:23 [INFO] train episode 94: reward = -21.00, steps = 792
12:38:09 [INFO] train episode 95: reward = -20.00, steps = 975
12:38:48 [INFO] train episode 96: reward = -21.00, steps = 824
12:39:32 [INFO] train episode 97: reward = -20.00, steps = 918
12:40:17 [INFO] train episode 98: reward = -20.00, steps = 956
12:41:09 [INFO] train episode 99: reward = -21.00, steps = 1101
12:41:46 [INFO] train episode 100: reward = -21.00, steps = 777
12:42:25 [INFO] train episode 101: reward = -21.00, steps = 826
12:43:06 [INFO] train episode 102: reward = -21.00, steps = 865
12:43:52 [INFO] train episode 103: reward = -21.00, steps = 961
12:44:42 [INFO] train episode 104: reward = -20.00, steps = 1041
12:45:26 [INFO] train episode 105: reward = -20.00, steps = 929
12:46:16 [INFO] train episode 106: reward = -20.00, steps = 1038
12:47:28 [INFO] train episode 107: reward = -21.00, steps = 847
12:51:20 [INFO] train episode 108: reward = -20.00, steps = 918
12:55:25 [INFO] train episode 109: reward = -21.00, steps = 968
12:59:56 [INFO] train episode 110: reward = -19.00, steps = 1066
13:03:52 [INFO] train episode 111: reward = -20.00, steps = 926
13:07:26 [INFO] train episode 112: reward = -20.00, steps = 842
13:11:47 [INFO] train episode 113: reward = -21.00, steps = 1028
13:15:52 [INFO] train episode 114: reward = -21.00, steps = 959
13:19:13 [INFO] train episode 115: reward = -21.00, steps = 786
13:24:31 [INFO] train episode 116: reward = -18.00, steps = 1238
13:28:22 [INFO] train episode 117: reward = -21.00, steps = 909
13:32:46 [INFO] train episode 118: reward = -21.00, steps = 1033
13:36:50 [INFO] train episode 119: reward = -20.00, steps = 959
13:41:40 [INFO] train episode 120: reward = -19.00, steps = 1133
13:45:28 [INFO] train episode 121: reward = -20.00, steps = 857
13:48:57 [INFO] train episode 122: reward = -21.00, steps = 805
13:53:25 [INFO] train episode 123: reward = -20.00, steps = 1042
13:58:36 [INFO] train episode 124: reward = -21.00, steps = 1210
14:03:15 [INFO] train episode 125: reward = -20.00, steps = 1078
14:08:46 [INFO] train episode 126: reward = -20.00, steps = 1290
14:14:59 [INFO] train episode 127: reward = -19.00, steps = 1454
14:19:40 [INFO] train episode 128: reward = -21.00, steps = 1101
14:25:15 [INFO] train episode 129: reward = -20.00, steps = 1291
14:33:24 [INFO] train episode 130: reward = -15.00, steps = 1903
14:40:21 [INFO] train episode 131: reward = -14.00, steps = 1628
14:45:46 [INFO] train episode 132: reward = -19.00, steps = 1264
14:51:24 [INFO] train episode 133: reward = -18.00, steps = 1314
14:56:42 [INFO] train episode 134: reward = -19.00, steps = 1233
15:02:00 [INFO] train episode 135: reward = -20.00, steps = 1231
15:09:38 [INFO] train episode 136: reward = -18.00, steps = 1775
15:16:36 [INFO] train episode 137: reward = -18.00, steps = 1618
15:22:21 [INFO] train episode 138: reward = -17.00, steps = 1318
15:28:06 [INFO] train episode 139: reward = -18.00, steps = 1351
15:36:12 [INFO] train episode 140: reward = -18.00, steps = 1897
15:41:54 [INFO] train episode 141: reward = -21.00, steps = 1340
15:49:30 [INFO] train episode 142: reward = -15.00, steps = 1778
15:54:32 [INFO] train episode 143: reward = -21.00, steps = 1175
16:01:08 [INFO] train episode 144: reward = -18.00, steps = 1537
16:06:48 [INFO] train episode 145: reward = -19.00, steps = 1317
16:14:13 [INFO] train episode 146: reward = -17.00, steps = 1724
16:20:01 [INFO] train episode 147: reward = -19.00, steps = 1340
16:26:59 [INFO] train episode 148: reward = -15.00, steps = 1596
16:35:04 [INFO] train episode 149: reward = -14.00, steps = 1855
16:43:51 [INFO] train episode 150: reward = -13.00, steps = 2009
16:49:06 [INFO] train episode 151: reward = -20.00, steps = 1196
16:55:58 [INFO] train episode 152: reward = -18.00, steps = 1555
17:06:21 [INFO] train episode 153: reward = -17.00, steps = 1619
17:16:03 [INFO] train episode 154: reward = -14.00, steps = 2129
17:28:38 [INFO] train episode 155: reward = -7.00, steps = 2760
17:39:39 [INFO] train episode 156: reward = -8.00, steps = 2394
17:51:37 [INFO] train episode 157: reward = -6.00, steps = 2600
18:03:12 [INFO] train episode 158: reward = -8.00, steps = 2499
18:08:18 [INFO] train episode 159: reward = -18.00, steps = 1097
18:16:56 [INFO] train episode 160: reward = -16.00, steps = 1860
18:26:40 [INFO] train episode 161: reward = -15.00, steps = 2073
18:36:32 [INFO] train episode 162: reward = -9.00, steps = 2118
18:45:22 [INFO] train episode 163: reward = -14.00, steps = 1901
18:57:19 [INFO] train episode 164: reward = -7.00, steps = 2458
19:08:58 [INFO] train episode 165: reward = -6.00, steps = 2374
19:20:28 [INFO] train episode 166: reward = -8.00, steps = 2336
19:27:05 [INFO] train episode 167: reward = -19.00, steps = 1323
19:40:36 [INFO] train episode 168: reward = -2.00, steps = 2728
19:54:10 [INFO] train episode 169: reward = -4.00, steps = 2725
20:07:06 [INFO] train episode 170: reward = -4.00, steps = 2590
20:21:43 [INFO] train episode 171: reward = -2.00, steps = 2909
20:29:59 [INFO] train episode 172: reward = -14.00, steps = 1647
20:43:15 [INFO] train episode 173: reward = -5.00, steps = 2642
20:53:09 [INFO] train episode 174: reward = -11.00, steps = 1961
21:06:29 [INFO] train episode 175: reward = -5.00, steps = 2636
21:16:24 [INFO] train episode 176: reward = -12.00, steps = 1956
21:28:36 [INFO] train episode 177: reward = -8.00, steps = 2395
21:41:56 [INFO] train episode 178: reward = -7.00, steps = 2623
21:51:14 [INFO] train episode 179: reward = -16.00, steps = 1804
22:05:11 [INFO] train episode 180: reward = -3.00, steps = 2865
22:15:25 [INFO] train episode 181: reward = -10.00, steps = 2101
22:26:28 [INFO] train episode 182: reward = -11.00, steps = 2254
22:35:37 [INFO] train episode 183: reward = -14.00, steps = 1831
22:48:30 [INFO] train episode 184: reward = -8.00, steps = 2513
23:02:31 [INFO] train episode 185: reward = -7.00, steps = 2565
23:17:12 [INFO] train episode 186: reward = -8.00, steps = 2996
23:25:48 [INFO] train episode 187: reward = -16.00, steps = 1749
23:36:40 [INFO] train episode 188: reward = -12.00, steps = 2241
23:47:09 [INFO] train episode 189: reward = -12.00, steps = 2066
00:02:27 [INFO] train episode 190: reward = -3.00, steps = 3220
00:16:31 [INFO] train episode 191: reward = -3.00, steps = 2959
00:29:26 [INFO] train episode 192: reward = -3.00, steps = 2711
00:37:48 [INFO] train episode 193: reward = -12.00, steps = 1766
00:47:19 [INFO] train episode 194: reward = -12.00, steps = 2017
00:57:44 [INFO] train episode 195: reward = -11.00, steps = 2209
01:08:05 [INFO] train episode 196: reward = -13.00, steps = 2190
01:21:10 [INFO] train episode 197: reward = -2.00, steps = 2753
01:34:42 [INFO] train episode 198: reward = -5.00, steps = 2868
01:45:16 [INFO] train episode 199: reward = -11.00, steps = 2234
01:58:27 [INFO] train episode 200: reward = -7.00, steps = 2778
02:13:35 [INFO] train episode 201: reward = -3.00, steps = 3187
02:29:25 [INFO] train episode 202: reward = 1.00, steps = 3316
02:41:58 [INFO] train episode 203: reward = -4.00, steps = 2640
02:55:32 [INFO] train episode 204: reward = -2.00, steps = 2856
03:05:44 [INFO] train episode 205: reward = -7.00, steps = 2150
03:20:04 [INFO] train episode 206: reward = -3.00, steps = 3035
03:29:48 [INFO] train episode 207: reward = -12.00, steps = 2037
03:43:40 [INFO] train episode 208: reward = 6.00, steps = 2929
03:57:17 [INFO] train episode 209: reward = 3.00, steps = 2874
04:10:48 [INFO] train episode 210: reward = 6.00, steps = 2859
04:24:09 [INFO] train episode 211: reward = 3.00, steps = 2802
04:36:43 [INFO] train episode 212: reward = 11.00, steps = 2642
04:48:37 [INFO] train episode 213: reward = 8.00, steps = 2515
05:01:34 [INFO] train episode 214: reward = -3.00, steps = 2707
05:14:26 [INFO] train episode 215: reward = -2.00, steps = 2700
05:26:14 [INFO] train episode 216: reward = 11.00, steps = 2473
05:39:36 [INFO] train episode 217: reward = -4.00, steps = 2822
05:52:36 [INFO] train episode 218: reward = 10.00, steps = 2710
06:07:37 [INFO] train episode 219: reward = -2.00, steps = 3138
06:21:51 [INFO] train episode 220: reward = 8.00, steps = 2950
06:34:26 [INFO] train episode 221: reward = 10.00, steps = 2624
06:47:50 [INFO] train episode 222: reward = -4.00, steps = 2809
07:02:28 [INFO] train episode 223: reward = -1.00, steps = 3059
07:15:04 [INFO] train episode 224: reward = 9.00, steps = 2643
07:26:37 [INFO] train episode 225: reward = 17.00, steps = 2397
07:42:03 [INFO] train episode 226: reward = 1.00, steps = 3215
07:55:24 [INFO] train episode 227: reward = -1.00, steps = 2777
08:04:59 [INFO] train episode 228: reward = 16.00, steps = 1999
08:19:30 [INFO] train episode 229: reward = 4.00, steps = 2903
08:31:44 [INFO] train episode 230: reward = 11.00, steps = 2504
08:43:04 [INFO] train episode 231: reward = 14.00, steps = 2336
08:56:32 [INFO] train episode 232: reward = 6.00, steps = 2758
09:10:14 [INFO] train episode 233: reward = -3.00, steps = 2809
09:22:52 [INFO] train episode 234: reward = 11.00, steps = 2570
09:36:38 [INFO] train episode 235: reward = 5.00, steps = 2916
09:47:02 [INFO] train episode 236: reward = 14.00, steps = 2421
10:00:14 [INFO] train episode 237: reward = -1.00, steps = 3071
10:10:57 [INFO] train episode 238: reward = 10.00, steps = 2501
10:20:26 [INFO] train episode 239: reward = 13.00, steps = 2208
10:34:08 [INFO] train episode 240: reward = 2.00, steps = 2966
10:48:15 [INFO] train episode 241: reward = 6.00, steps = 2906
10:58:47 [INFO] train episode 242: reward = 15.00, steps = 2167
11:10:59 [INFO] train episode 243: reward = 10.00, steps = 2514
11:23:40 [INFO] train episode 244: reward = 7.00, steps = 2556
11:36:24 [INFO] train episode 245: reward = 3.00, steps = 2603
11:47:40 [INFO] train episode 246: reward = 11.00, steps = 2317
12:01:45 [INFO] train episode 247: reward = 6.00, steps = 2814
12:13:31 [INFO] train episode 248: reward = 13.00, steps = 2377
12:26:06 [INFO] train episode 249: reward = 10.00, steps = 2570
12:37:33 [INFO] train episode 250: reward = 11.00, steps = 2341
12:48:24 [INFO] train episode 251: reward = 9.00, steps = 2231
13:03:09 [INFO] train episode 252: reward = 6.00, steps = 3022
13:14:27 [INFO] train episode 253: reward = 16.00, steps = 2314
13:25:18 [INFO] train episode 254: reward = 14.00, steps = 2190
13:35:46 [INFO] train episode 255: reward = 13.00, steps = 2137
13:46:51 [INFO] train episode 256: reward = 14.00, steps = 2284
13:56:18 [INFO] train episode 257: reward = 16.00, steps = 1944
14:06:26 [INFO] train episode 258: reward = 11.00, steps = 2067
14:18:27 [INFO] train episode 259: reward = 7.00, steps = 2425
14:27:05 [INFO] train episode 260: reward = 16.00, steps = 2118
14:33:36 [INFO] train episode 261: reward = 19.00, steps = 1703
14:40:48 [INFO] train episode 262: reward = 16.00, steps = 1886
14:48:27 [INFO] train episode 263: reward = 15.00, steps = 2011
14:57:13 [INFO] train episode 264: reward = 13.00, steps = 2281
15:06:12 [INFO] train episode 265: reward = 12.00, steps = 2357
15:14:12 [INFO] train episode 266: reward = 12.00, steps = 2108
15:22:09 [INFO] train episode 267: reward = 14.00, steps = 2090
15:30:04 [INFO] train episode 268: reward = 15.00, steps = 2081
15:36:42 [INFO] train episode 269: reward = 20.00, steps = 1738
15:45:25 [INFO] train episode 270: reward = 10.00, steps = 2294
15:53:22 [INFO] train episode 271: reward = 14.00, steps = 2089
16:03:27 [INFO] train episode 272: reward = -1.00, steps = 2638
16:11:53 [INFO] train episode 273: reward = 11.00, steps = 2211
16:18:55 [INFO] train episode 274: reward = 17.00, steps = 1843
16:26:12 [INFO] train episode 275: reward = 18.00, steps = 1895
16:32:20 [INFO] train episode 276: reward = 20.00, steps = 1599
16:40:22 [INFO] train episode 277: reward = 16.00, steps = 2103
16:40:22 [INFO] ==== test ====
16:41:02 [INFO] test episode 0: reward = 14.00, steps = 1946
16:41:44 [INFO] test episode 1: reward = 14.00, steps = 2076
16:42:26 [INFO] test episode 2: reward = 14.00, steps = 2072
16:43:06 [INFO] test episode 3: reward = 14.00, steps = 1952
16:43:48 [INFO] test episode 4: reward = 14.00, steps = 2076
16:44:22 [INFO] test episode 5: reward = 20.00, steps = 1688
16:45:04 [INFO] test episode 6: reward = 14.00, steps = 2074
16:45:43 [INFO] test episode 7: reward = 14.00, steps = 1949
16:46:18 [INFO] test episode 8: reward = 20.00, steps = 1689
16:47:01 [INFO] test episode 9: reward = 14.00, steps = 2076
16:47:35 [INFO] test episode 10: reward = 20.00, steps = 1692
16:48:17 [INFO] test episode 11: reward = 14.00, steps = 2072
16:48:57 [INFO] test episode 12: reward = 14.00, steps = 1948
16:49:37 [INFO] test episode 13: reward = 14.00, steps = 1953
16:50:12 [INFO] test episode 14: reward = 20.00, steps = 1693
16:50:46 [INFO] test episode 15: reward = 20.00, steps = 1673
16:51:21 [INFO] test episode 16: reward = 20.00, steps = 1694
16:52:00 [INFO] test episode 17: reward = 14.00, steps = 1948
16:52:35 [INFO] test episode 18: reward = 20.00, steps = 1690
16:53:10 [INFO] test episode 19: reward = 20.00, steps = 1691
16:53:44 [INFO] test episode 20: reward = 20.00, steps = 1669
16:54:23 [INFO] test episode 21: reward = 14.00, steps = 1950
16:54:59 [INFO] test episode 22: reward = 20.00, steps = 1691
16:55:39 [INFO] test episode 23: reward = 14.00, steps = 1947
16:56:19 [INFO] test episode 24: reward = 14.00, steps = 1953
16:57:01 [INFO] test episode 25: reward = 14.00, steps = 2077
16:57:43 [INFO] test episode 26: reward = 14.00, steps = 2072
16:58:22 [INFO] test episode 27: reward = 14.00, steps = 1947
16:59:05 [INFO] test episode 28: reward = 14.00, steps = 2077
16:59:39 [INFO] test episode 29: reward = 20.00, steps = 1693
17:00:13 [INFO] test episode 30: reward = 20.00, steps = 1692
17:00:47 [INFO] test episode 31: reward = 20.00, steps = 1693
17:01:27 [INFO] test episode 32: reward = 14.00, steps = 1952
17:02:07 [INFO] test episode 33: reward = 14.00, steps = 1948
17:02:47 [INFO] test episode 34: reward = 14.00, steps = 1953
17:03:21 [INFO] test episode 35: reward = 20.00, steps = 1687
17:03:55 [INFO] test episode 36: reward = 20.00, steps = 1688
17:04:35 [INFO] test episode 37: reward = 14.00, steps = 1951
17:05:09 [INFO] test episode 38: reward = 20.00, steps = 1687
17:05:49 [INFO] test episode 39: reward = 14.00, steps = 1948
17:06:29 [INFO] test episode 40: reward = 14.00, steps = 1949
17:07:08 [INFO] test episode 41: reward = 14.00, steps = 1950
17:07:43 [INFO] test episode 42: reward = 20.00, steps = 1692
17:08:17 [INFO] test episode 43: reward = 20.00, steps = 1671
17:08:57 [INFO] test episode 44: reward = 14.00, steps = 1949
17:09:31 [INFO] test episode 45: reward = 20.00, steps = 1673
17:10:05 [INFO] test episode 46: reward = 20.00, steps = 1670
17:10:39 [INFO] test episode 47: reward = 20.00, steps = 1669
17:11:14 [INFO] test episode 48: reward = 20.00, steps = 1690
17:11:55 [INFO] test episode 49: reward = 14.00, steps = 2076
17:12:30 [INFO] test episode 50: reward = 20.00, steps = 1688
17:13:12 [INFO] test episode 51: reward = 14.00, steps = 2075
17:13:54 [INFO] test episode 52: reward = 14.00, steps = 2072
17:14:29 [INFO] test episode 53: reward = 20.00, steps = 1687
17:15:11 [INFO] test episode 54: reward = 14.00, steps = 2072
17:15:45 [INFO] test episode 55: reward = 20.00, steps = 1669
17:16:18 [INFO] test episode 56: reward = 20.00, steps = 1673
17:16:52 [INFO] test episode 57: reward = 20.00, steps = 1668
17:17:26 [INFO] test episode 58: reward = 20.00, steps = 1687
17:18:01 [INFO] test episode 59: reward = 20.00, steps = 1672
17:18:34 [INFO] test episode 60: reward = 20.00, steps = 1667
17:19:09 [INFO] test episode 61: reward = 20.00, steps = 1669
17:19:53 [INFO] test episode 62: reward = 14.00, steps = 2076
17:20:37 [INFO] test episode 63: reward = 14.00, steps = 2073
17:21:13 [INFO] test episode 64: reward = 20.00, steps = 1667
17:21:47 [INFO] test episode 65: reward = 20.00, steps = 1672
17:22:29 [INFO] test episode 66: reward = 14.00, steps = 2074
17:23:09 [INFO] test episode 67: reward = 14.00, steps = 1947
17:23:43 [INFO] test episode 68: reward = 20.00, steps = 1692
17:24:23 [INFO] test episode 69: reward = 14.00, steps = 1953
17:25:02 [INFO] test episode 70: reward = 14.00, steps = 1949
17:25:36 [INFO] test episode 71: reward = 20.00, steps = 1694
17:26:12 [INFO] test episode 72: reward = 20.00, steps = 1688
17:26:51 [INFO] test episode 73: reward = 14.00, steps = 1950
17:27:25 [INFO] test episode 74: reward = 20.00, steps = 1673
17:28:07 [INFO] test episode 75: reward = 14.00, steps = 2073
17:28:41 [INFO] test episode 76: reward = 20.00, steps = 1694
17:29:22 [INFO] test episode 77: reward = 14.00, steps = 1951
17:29:56 [INFO] test episode 78: reward = 20.00, steps = 1671
17:30:35 [INFO] test episode 79: reward = 14.00, steps = 1948
17:31:10 [INFO] test episode 80: reward = 20.00, steps = 1690
17:31:52 [INFO] test episode 81: reward = 14.00, steps = 2078
17:32:31 [INFO] test episode 82: reward = 14.00, steps = 1951
17:33:06 [INFO] test episode 83: reward = 20.00, steps = 1689
17:33:40 [INFO] test episode 84: reward = 20.00, steps = 1689
17:34:23 [INFO] test episode 85: reward = 14.00, steps = 2074
17:34:57 [INFO] test episode 86: reward = 20.00, steps = 1687
17:35:39 [INFO] test episode 87: reward = 14.00, steps = 2074
17:36:13 [INFO] test episode 88: reward = 20.00, steps = 1694
17:36:47 [INFO] test episode 89: reward = 20.00, steps = 1691
17:37:22 [INFO] test episode 90: reward = 20.00, steps = 1694
17:38:02 [INFO] test episode 91: reward = 14.00, steps = 1947
17:38:36 [INFO] test episode 92: reward = 20.00, steps = 1673
17:39:15 [INFO] test episode 93: reward = 14.00, steps = 1951
17:39:55 [INFO] test episode 94: reward = 14.00, steps = 1950
17:40:34 [INFO] test episode 95: reward = 14.00, steps = 1952
17:41:17 [INFO] test episode 96: reward = 14.00, steps = 2077
17:41:57 [INFO] test episode 97: reward = 14.00, steps = 1953
17:42:31 [INFO] test episode 98: reward = 20.00, steps = 1673
17:43:05 [INFO] test episode 99: reward = 20.00, steps = 1692
17:43:05 [INFO] average episode reward = 16.94 ± 3.00

Use Categorical DQN to Play Pong-v4¶