vwxyzjn commited on
Commit
b2a8a54
1 Parent(s): 7efa4df

pushing model

Browse files
.gitattributes CHANGED
@@ -32,3 +32,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
32
  *.zip filter=lfs diff=lfs merge=lfs -text
33
  *.zst filter=lfs diff=lfs merge=lfs -text
34
  *tfevents* filter=lfs diff=lfs merge=lfs -text
 
 
32
  *.zip filter=lfs diff=lfs merge=lfs -text
33
  *.zst filter=lfs diff=lfs merge=lfs -text
34
  *tfevents* filter=lfs diff=lfs merge=lfs -text
35
+ dqn_atari.cleanrl_model filter=lfs diff=lfs merge=lfs -text
README.md ADDED
@@ -0,0 +1,66 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ tags:
3
+ - PongNoFrameskip-v4
4
+ - deep-reinforcement-learning
5
+ - reinforcement-learning
6
+ - custom-implementation
7
+ library_name: cleanrl
8
+ model-index:
9
+ - name: DQN
10
+ results:
11
+ - task:
12
+ type: reinforcement-learning
13
+ name: reinforcement-learning
14
+ dataset:
15
+ name: PongNoFrameskip-v4
16
+ type: PongNoFrameskip-v4
17
+ metrics:
18
+ - type: mean_reward
19
+ value: 17.80 +/- 1.83
20
+ name: mean_reward
21
+ verified: false
22
+ ---
23
+
24
+ # (CleanRL) **DQN** Agent Playing **PongNoFrameskip-v4**
25
+
26
+ This is a trained model of a DQN agent playing PongNoFrameskip-v4.
27
+ The model was trained by using [CleanRL](https://github.com/vwxyzjn/cleanrl) and the most up-to-date training code can be
28
+ found [here](https://github.com/vwxyzjn/cleanrl/blob/master/cleanrl/dqn_atari.py).
29
+
30
+ ## Command to reproduce the training
31
+
32
+ ```bash
33
+ curl -OL https://huggingface.co/cleanrl/PongNoFrameskip-v4-dqn_atari-seed1/raw/main/dqn.py
34
+ curl -OL https://huggingface.co/cleanrl/PongNoFrameskip-v4-dqn_atari-seed1/raw/main/pyproject.toml
35
+ curl -OL https://huggingface.co/cleanrl/PongNoFrameskip-v4-dqn_atari-seed1/raw/main/poetry.lock
36
+ poetry install --all-extras
37
+ python dqn_atari.py --track --capture-video --save-model --upload-model --hf-entity cleanrl --env-id PongNoFrameskip-v4 --seed 1
38
+ ```
39
+
40
+ # Hyperparameters
41
+ ```python
42
+ {'batch_size': 32,
43
+ 'buffer_size': 1000000,
44
+ 'capture_video': True,
45
+ 'cuda': True,
46
+ 'end_e': 0.01,
47
+ 'env_id': 'PongNoFrameskip-v4',
48
+ 'exp_name': 'dqn_atari',
49
+ 'exploration_fraction': 0.1,
50
+ 'gamma': 0.99,
51
+ 'hf_entity': 'cleanrl',
52
+ 'learning_rate': 0.0001,
53
+ 'learning_starts': 80000,
54
+ 'save_model': True,
55
+ 'seed': 1,
56
+ 'start_e': 1,
57
+ 'target_network_frequency': 1000,
58
+ 'torch_deterministic': True,
59
+ 'total_timesteps': 10000000,
60
+ 'track': True,
61
+ 'train_frequency': 4,
62
+ 'upload_model': True,
63
+ 'wandb_entity': None,
64
+ 'wandb_project_name': 'cleanRL'}
65
+ ```
66
+
dqn_atari.cleanrl_model ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1ffa4e32dc1be3ec0465d9db898229bca4bb61fb9c35c9d9b55332c51e1d4735
3
+ size 6751815
dqn_atari.py ADDED
@@ -0,0 +1,264 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # docs and experiment results can be found at https://docs.cleanrl.dev/rl-algorithms/dqn/#dqn_ataripy
2
+ import argparse
3
+ import os
4
+ import random
5
+ import time
6
+ from distutils.util import strtobool
7
+
8
+ import gym
9
+ import numpy as np
10
+ import torch
11
+ import torch.nn as nn
12
+ import torch.nn.functional as F
13
+ import torch.optim as optim
14
+ from stable_baselines3.common.atari_wrappers import (
15
+ ClipRewardEnv,
16
+ EpisodicLifeEnv,
17
+ FireResetEnv,
18
+ MaxAndSkipEnv,
19
+ NoopResetEnv,
20
+ )
21
+ from stable_baselines3.common.buffers import ReplayBuffer
22
+ from torch.utils.tensorboard import SummaryWriter
23
+
24
+
25
+ def parse_args():
26
+ # fmt: off
27
+ parser = argparse.ArgumentParser()
28
+ parser.add_argument("--exp-name", type=str, default=os.path.basename(__file__).rstrip(".py"),
29
+ help="the name of this experiment")
30
+ parser.add_argument("--seed", type=int, default=1,
31
+ help="seed of the experiment")
32
+ parser.add_argument("--torch-deterministic", type=lambda x: bool(strtobool(x)), default=True, nargs="?", const=True,
33
+ help="if toggled, `torch.backends.cudnn.deterministic=False`")
34
+ parser.add_argument("--cuda", type=lambda x: bool(strtobool(x)), default=True, nargs="?", const=True,
35
+ help="if toggled, cuda will be enabled by default")
36
+ parser.add_argument("--track", type=lambda x: bool(strtobool(x)), default=False, nargs="?", const=True,
37
+ help="if toggled, this experiment will be tracked with Weights and Biases")
38
+ parser.add_argument("--wandb-project-name", type=str, default="cleanRL",
39
+ help="the wandb's project name")
40
+ parser.add_argument("--wandb-entity", type=str, default=None,
41
+ help="the entity (team) of wandb's project")
42
+ parser.add_argument("--capture-video", type=lambda x: bool(strtobool(x)), default=False, nargs="?", const=True,
43
+ help="whether to capture videos of the agent performances (check out `videos` folder)")
44
+ parser.add_argument("--save-model", type=lambda x: bool(strtobool(x)), default=False, nargs="?", const=True,
45
+ help="whether to save model into the `runs/{run_name}` folder")
46
+ parser.add_argument("--upload-model", type=lambda x: bool(strtobool(x)), default=False, nargs="?", const=True,
47
+ help="whether to upload the saved model to huggingface")
48
+ parser.add_argument("--hf-entity", type=str, default="",
49
+ help="the user or org name of the model repository from the Hugging Face Hub")
50
+
51
+ # Algorithm specific arguments
52
+ parser.add_argument("--env-id", type=str, default="BreakoutNoFrameskip-v4",
53
+ help="the id of the environment")
54
+ parser.add_argument("--total-timesteps", type=int, default=10000000,
55
+ help="total timesteps of the experiments")
56
+ parser.add_argument("--learning-rate", type=float, default=1e-4,
57
+ help="the learning rate of the optimizer")
58
+ parser.add_argument("--buffer-size", type=int, default=1000000,
59
+ help="the replay memory buffer size")
60
+ parser.add_argument("--gamma", type=float, default=0.99,
61
+ help="the discount factor gamma")
62
+ parser.add_argument("--target-network-frequency", type=int, default=1000,
63
+ help="the timesteps it takes to update the target network")
64
+ parser.add_argument("--batch-size", type=int, default=32,
65
+ help="the batch size of sample from the reply memory")
66
+ parser.add_argument("--start-e", type=float, default=1,
67
+ help="the starting epsilon for exploration")
68
+ parser.add_argument("--end-e", type=float, default=0.01,
69
+ help="the ending epsilon for exploration")
70
+ parser.add_argument("--exploration-fraction", type=float, default=0.10,
71
+ help="the fraction of `total-timesteps` it takes from start-e to go end-e")
72
+ parser.add_argument("--learning-starts", type=int, default=80000,
73
+ help="timestep to start learning")
74
+ parser.add_argument("--train-frequency", type=int, default=4,
75
+ help="the frequency of training")
76
+ args = parser.parse_args()
77
+ # fmt: on
78
+ return args
79
+
80
+
81
+ def make_env(env_id, seed, idx, capture_video, run_name):
82
+ def thunk():
83
+ env = gym.make(env_id)
84
+ env = gym.wrappers.RecordEpisodeStatistics(env)
85
+ if capture_video:
86
+ if idx == 0:
87
+ env = gym.wrappers.RecordVideo(env, f"videos/{run_name}")
88
+ env = NoopResetEnv(env, noop_max=30)
89
+ env = MaxAndSkipEnv(env, skip=4)
90
+ env = EpisodicLifeEnv(env)
91
+ if "FIRE" in env.unwrapped.get_action_meanings():
92
+ env = FireResetEnv(env)
93
+ env = ClipRewardEnv(env)
94
+ env = gym.wrappers.ResizeObservation(env, (84, 84))
95
+ env = gym.wrappers.GrayScaleObservation(env)
96
+ env = gym.wrappers.FrameStack(env, 4)
97
+ env.seed(seed)
98
+ env.action_space.seed(seed)
99
+ env.observation_space.seed(seed)
100
+ return env
101
+
102
+ return thunk
103
+
104
+
105
+ # ALGO LOGIC: initialize agent here:
106
+ class QNetwork(nn.Module):
107
+ def __init__(self, env):
108
+ super().__init__()
109
+ self.network = nn.Sequential(
110
+ nn.Conv2d(4, 32, 8, stride=4),
111
+ nn.ReLU(),
112
+ nn.Conv2d(32, 64, 4, stride=2),
113
+ nn.ReLU(),
114
+ nn.Conv2d(64, 64, 3, stride=1),
115
+ nn.ReLU(),
116
+ nn.Flatten(),
117
+ nn.Linear(3136, 512),
118
+ nn.ReLU(),
119
+ nn.Linear(512, env.single_action_space.n),
120
+ )
121
+
122
+ def forward(self, x):
123
+ return self.network(x / 255.0)
124
+
125
+
126
+ def linear_schedule(start_e: float, end_e: float, duration: int, t: int):
127
+ slope = (end_e - start_e) / duration
128
+ return max(slope * t + start_e, end_e)
129
+
130
+
131
+ if __name__ == "__main__":
132
+ args = parse_args()
133
+ run_name = f"{args.env_id}__{args.exp_name}__{args.seed}__{int(time.time())}"
134
+ if args.track:
135
+ import wandb
136
+
137
+ wandb.init(
138
+ project=args.wandb_project_name,
139
+ entity=args.wandb_entity,
140
+ sync_tensorboard=True,
141
+ config=vars(args),
142
+ name=run_name,
143
+ monitor_gym=True,
144
+ save_code=True,
145
+ )
146
+ writer = SummaryWriter(f"runs/{run_name}")
147
+ writer.add_text(
148
+ "hyperparameters",
149
+ "|param|value|\n|-|-|\n%s" % ("\n".join([f"|{key}|{value}|" for key, value in vars(args).items()])),
150
+ )
151
+
152
+ # TRY NOT TO MODIFY: seeding
153
+ random.seed(args.seed)
154
+ np.random.seed(args.seed)
155
+ torch.manual_seed(args.seed)
156
+ torch.backends.cudnn.deterministic = args.torch_deterministic
157
+
158
+ device = torch.device("cuda" if torch.cuda.is_available() and args.cuda else "cpu")
159
+
160
+ # env setup
161
+ envs = gym.vector.SyncVectorEnv([make_env(args.env_id, args.seed, 0, args.capture_video, run_name)])
162
+ assert isinstance(envs.single_action_space, gym.spaces.Discrete), "only discrete action space is supported"
163
+
164
+ q_network = QNetwork(envs).to(device)
165
+ optimizer = optim.Adam(q_network.parameters(), lr=args.learning_rate)
166
+ target_network = QNetwork(envs).to(device)
167
+ target_network.load_state_dict(q_network.state_dict())
168
+
169
+ rb = ReplayBuffer(
170
+ args.buffer_size,
171
+ envs.single_observation_space,
172
+ envs.single_action_space,
173
+ device,
174
+ optimize_memory_usage=True,
175
+ handle_timeout_termination=True,
176
+ )
177
+ start_time = time.time()
178
+
179
+ # TRY NOT TO MODIFY: start the game
180
+ obs = envs.reset()
181
+ for global_step in range(args.total_timesteps):
182
+ # ALGO LOGIC: put action logic here
183
+ epsilon = linear_schedule(args.start_e, args.end_e, args.exploration_fraction * args.total_timesteps, global_step)
184
+ if random.random() < epsilon:
185
+ actions = np.array([envs.single_action_space.sample() for _ in range(envs.num_envs)])
186
+ else:
187
+ q_values = q_network(torch.Tensor(obs).to(device))
188
+ actions = torch.argmax(q_values, dim=1).cpu().numpy()
189
+
190
+ # TRY NOT TO MODIFY: execute the game and log data.
191
+ next_obs, rewards, dones, infos = envs.step(actions)
192
+
193
+ # TRY NOT TO MODIFY: record rewards for plotting purposes
194
+ for info in infos:
195
+ if "episode" in info.keys():
196
+ print(f"global_step={global_step}, episodic_return={info['episode']['r']}")
197
+ writer.add_scalar("charts/episodic_return", info["episode"]["r"], global_step)
198
+ writer.add_scalar("charts/episodic_length", info["episode"]["l"], global_step)
199
+ writer.add_scalar("charts/epsilon", epsilon, global_step)
200
+ break
201
+
202
+ # TRY NOT TO MODIFY: save data to reply buffer; handle `terminal_observation`
203
+ real_next_obs = next_obs.copy()
204
+ for idx, d in enumerate(dones):
205
+ if d:
206
+ real_next_obs[idx] = infos[idx]["terminal_observation"]
207
+ rb.add(obs, real_next_obs, actions, rewards, dones, infos)
208
+
209
+ # TRY NOT TO MODIFY: CRUCIAL step easy to overlook
210
+ obs = next_obs
211
+
212
+ # ALGO LOGIC: training.
213
+ if global_step > args.learning_starts:
214
+ if global_step % args.train_frequency == 0:
215
+ data = rb.sample(args.batch_size)
216
+ with torch.no_grad():
217
+ target_max, _ = target_network(data.next_observations).max(dim=1)
218
+ td_target = data.rewards.flatten() + args.gamma * target_max * (1 - data.dones.flatten())
219
+ old_val = q_network(data.observations).gather(1, data.actions).squeeze()
220
+ loss = F.mse_loss(td_target, old_val)
221
+
222
+ if global_step % 100 == 0:
223
+ writer.add_scalar("losses/td_loss", loss, global_step)
224
+ writer.add_scalar("losses/q_values", old_val.mean().item(), global_step)
225
+ print("SPS:", int(global_step / (time.time() - start_time)))
226
+ writer.add_scalar("charts/SPS", int(global_step / (time.time() - start_time)), global_step)
227
+
228
+ # optimize the model
229
+ optimizer.zero_grad()
230
+ loss.backward()
231
+ optimizer.step()
232
+
233
+ # update the target network
234
+ if global_step % args.target_network_frequency == 0:
235
+ target_network.load_state_dict(q_network.state_dict())
236
+
237
+ if args.save_model:
238
+ model_path = f"runs/{run_name}/{args.exp_name}.cleanrl_model"
239
+ torch.save(q_network.state_dict(), model_path)
240
+ print(f"model saved to {model_path}")
241
+ from cleanrl_utils.evals.dqn_eval import evaluate
242
+
243
+ episodic_returns = evaluate(
244
+ model_path,
245
+ make_env,
246
+ args.env_id,
247
+ eval_episodes=10,
248
+ run_name=f"{run_name}-eval",
249
+ Model=QNetwork,
250
+ device=device,
251
+ epsilon=0.05,
252
+ )
253
+ for idx, episodic_return in enumerate(episodic_returns):
254
+ writer.add_scalar("eval/episodic_return", episodic_return, idx)
255
+
256
+ if args.upload_model:
257
+ from cleanrl_utils.huggingface import push_to_hub
258
+
259
+ repo_name = f"{args.env_id}-{args.exp_name}-seed{args.seed}"
260
+ repo_id = f"{args.hf_entity}/{repo_name}" if args.hf_entity else repo_name
261
+ push_to_hub(args, episodic_returns, repo_id, "DQN", f"runs/{run_name}", f"videos/{run_name}-eval")
262
+
263
+ envs.close()
264
+ writer.close()
events.out.tfevents.1671340144.pop-os.1595202.0 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ecc2823a75bf8db7e62dacbf6b1708f6c9b5e9c3a864b23597a580bccb94eb18
3
+ size 17036419
poetry.lock ADDED
The diff for this file is too large to render. See raw diff
 
pyproject.toml ADDED
@@ -0,0 +1,158 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ [tool.poetry]
2
+ name = "cleanrl-test"
3
+ version = "1.1.0"
4
+ description = "High-quality single file implementation of Deep Reinforcement Learning algorithms with research-friendly features"
5
+ authors = ["Costa Huang <costa.huang@outlook.com>"]
6
+ packages = [
7
+ { include = "cleanrl" },
8
+ { include = "cleanrl_utils" },
9
+ ]
10
+ keywords = ["reinforcement", "machine", "learning", "research"]
11
+ license="MIT"
12
+ readme = "README.md"
13
+
14
+ [tool.poetry.dependencies]
15
+ python = ">=3.7.1,<3.10"
16
+ tensorboard = "^2.10.0"
17
+ wandb = "^0.13.6"
18
+ gym = "0.23.1"
19
+ torch = ">=1.12.1"
20
+ stable-baselines3 = "1.2.0"
21
+ gymnasium = "^0.26.3"
22
+ moviepy = "^1.0.3"
23
+ pygame = "2.1.0"
24
+ huggingface-hub = "^0.11.1"
25
+
26
+ ale-py = {version = "0.7.4", optional = true}
27
+ AutoROM = {extras = ["accept-rom-license"], version = "^0.4.2"}
28
+ opencv-python = {version = "^4.6.0.66", optional = true}
29
+ pybullet = {version = "3.1.8", optional = true}
30
+ procgen = {version = "^0.10.7", optional = true}
31
+ pytest = {version = "^7.1.3", optional = true}
32
+ mujoco = {version = "^2.2", optional = true}
33
+ imageio = {version = "^2.14.1", optional = true}
34
+ free-mujoco-py = {version = "^2.1.6", optional = true}
35
+ mkdocs-material = {version = "^8.4.3", optional = true}
36
+ markdown-include = {version = "^0.7.0", optional = true}
37
+ jax = {version = "^0.3.17", optional = true}
38
+ jaxlib = {version = "^0.3.15", optional = true}
39
+ flax = {version = "^0.6.0", optional = true}
40
+ optuna = {version = "^3.0.1", optional = true}
41
+ optuna-dashboard = {version = "^0.7.2", optional = true}
42
+ rich = {version = "<12.0", optional = true}
43
+ envpool = {version = "^0.6.4", optional = true}
44
+ PettingZoo = {version = "1.18.1", optional = true}
45
+ SuperSuit = {version = "3.4.0", optional = true}
46
+ multi-agent-ale-py = {version = "0.1.11", optional = true}
47
+ boto3 = {version = "^1.24.70", optional = true}
48
+ awscli = {version = "^1.25.71", optional = true}
49
+ shimmy = {version = "^0.1.0", optional = true}
50
+ dm-control = {version = "^1.0.8", optional = true}
51
+
52
+ [tool.poetry.group.dev.dependencies]
53
+ pre-commit = "^2.20.0"
54
+
55
+ [tool.poetry.group.atari]
56
+ optional = true
57
+ [tool.poetry.group.atari.dependencies]
58
+ ale-py = "0.7.4"
59
+ AutoROM = {extras = ["accept-rom-license"], version = "^0.4.2"}
60
+ opencv-python = "^4.6.0.66"
61
+
62
+ [tool.poetry.group.pybullet]
63
+ optional = true
64
+ [tool.poetry.group.pybullet.dependencies]
65
+ pybullet = "3.1.8"
66
+
67
+ [tool.poetry.group.procgen]
68
+ optional = true
69
+ [tool.poetry.group.procgen.dependencies]
70
+ procgen = "^0.10.7"
71
+
72
+ [tool.poetry.group.pytest]
73
+ optional = true
74
+ [tool.poetry.group.pytest.dependencies]
75
+ pytest = "^7.1.3"
76
+
77
+ [tool.poetry.group.mujoco]
78
+ optional = true
79
+ [tool.poetry.group.mujoco.dependencies]
80
+ mujoco = "^2.2"
81
+ imageio = "^2.14.1"
82
+
83
+ [tool.poetry.group.mujoco_py]
84
+ optional = true
85
+ [tool.poetry.group.mujoco_py.dependencies]
86
+ free-mujoco-py = "^2.1.6"
87
+
88
+ [tool.poetry.group.docs]
89
+ optional = true
90
+ [tool.poetry.group.docs.dependencies]
91
+ mkdocs-material = "^8.4.3"
92
+ markdown-include = "^0.7.0"
93
+
94
+ [tool.poetry.group.jax]
95
+ optional = true
96
+ [tool.poetry.group.jax.dependencies]
97
+ jax = "^0.3.17"
98
+ jaxlib = "^0.3.15"
99
+ flax = "^0.6.0"
100
+
101
+ [tool.poetry.group.optuna]
102
+ optional = true
103
+ [tool.poetry.group.optuna.dependencies]
104
+ optuna = "^3.0.1"
105
+ optuna-dashboard = "^0.7.2"
106
+ rich = "<12.0"
107
+
108
+ [tool.poetry.group.envpool]
109
+ optional = true
110
+ [tool.poetry.group.envpool.dependencies]
111
+ envpool = "^0.6.4"
112
+
113
+ [tool.poetry.group.pettingzoo]
114
+ optional = true
115
+ [tool.poetry.group.pettingzoo.dependencies]
116
+ PettingZoo = "1.18.1"
117
+ SuperSuit = "3.4.0"
118
+ multi-agent-ale-py = "0.1.11"
119
+
120
+ [tool.poetry.group.cloud]
121
+ optional = true
122
+ [tool.poetry.group.cloud.dependencies]
123
+ boto3 = "^1.24.70"
124
+ awscli = "^1.25.71"
125
+
126
+ [tool.poetry.group.isaacgym]
127
+ optional = true
128
+ [tool.poetry.group.isaacgym.dependencies]
129
+ isaacgymenvs = {git = "https://github.com/vwxyzjn/IsaacGymEnvs.git", rev = "poetry"}
130
+ isaacgym = {path = "cleanrl/ppo_continuous_action_isaacgym/isaacgym", develop = true}
131
+
132
+ [tool.poetry.group.dm_control]
133
+ optional = true
134
+ [tool.poetry.group.dm_control.dependencies]
135
+ shimmy = "^0.1.0"
136
+ dm-control = "^1.0.8"
137
+ mujoco = "^2.2"
138
+
139
+ [build-system]
140
+ requires = ["poetry-core"]
141
+ build-backend = "poetry.core.masonry.api"
142
+
143
+ [tool.poetry.extras]
144
+ atari = ["ale-py", "AutoROM", "opencv-python"]
145
+ pybullet = ["pybullet"]
146
+ procgen = ["procgen"]
147
+ plot = ["pandas", "seaborn"]
148
+ spyder = ["spyder"]
149
+ pytest = ["pytest"]
150
+ mujoco = ["mujoco", "imageio"]
151
+ mujoco_py = ["free-mujoco-py"]
152
+ jax = ["jax", "jaxlib", "flax"]
153
+ docs = ["mkdocs-material", "markdown-include"]
154
+ envpool = ["envpool"]
155
+ optuna = ["optuna", "optuna-dashboard", "rich"]
156
+ pettingzoo = ["PettingZoo", "SuperSuit", "multi-agent-ale-py"]
157
+ cloud = ["boto3", "awscli"]
158
+ dm_control = ["shimmy", "dm-control", "mujoco"]
replay.mp4 ADDED
Binary file (321 kB). View file
 
videos/PongNoFrameskip-v4__dqn_atari__1__1671340143-eval/rl-video-episode-0.mp4 ADDED
Binary file (378 kB). View file
 
videos/PongNoFrameskip-v4__dqn_atari__1__1671340143-eval/rl-video-episode-1.mp4 ADDED
Binary file (359 kB). View file
 
videos/PongNoFrameskip-v4__dqn_atari__1__1671340143-eval/rl-video-episode-8.mp4 ADDED
Binary file (321 kB). View file