Add atari ppo example (#523)

I needed a policy gradient baseline myself and it has been requested several times (#497, #374, #440). I used https://github.com/vwxyzjn/cleanrl/blob/master/cleanrl/ppo_atari.py as a reference for hyper-parameters.

Note that using lr=2.5e-4 will result in "Invalid Value" error for 2 games. The fix is to reduce the learning rate. That's why I set the default lr to 1e-4. See discussion in https://github.com/DLR-RM/rl-baselines3-zoo/issues/156.

examples/atari/README.md

@@ -95,3 +95,17 @@ One epoch here is equal to 100,000 env step, 100 epochs stand for 10M.
 | MsPacmanNoFrameskip-v4      | 3101        | ![](results/rainbow/MsPacman_rew.png)     | `python3 atari_rainbow.py --task "MsPacmanNoFrameskip-v4"`  |
 | SeaquestNoFrameskip-v4      | 2126       | ![](results/rainbow/Seaquest_rew.png)     | `python3 atari_rainbow.py --task "SeaquestNoFrameskip-v4"`  |
 | SpaceInvadersNoFrameskip-v4 | 1794.5       | ![](results/rainbow/SpaceInvaders_rew.png) | `python3 atari_rainbow.py --task "SpaceInvadersNoFrameskip-v4"`  |
+
+# PPO (single run)
+
+One epoch here is equal to 100,000 env step, 100 epochs stand for 10M.
+
+| task                        | best reward | reward curve                          | parameters                                                   |
+| --------------------------- | ----------- | ------------------------------------- | ------------------------------------------------------------ |
+| PongNoFrameskip-v4          | 20        | ![](results/ppo/Pong_rew.png)         | `python3 atari_ppo.py --task "PongNoFrameskip-v4"` |
+| BreakoutNoFrameskip-v4      | 442.1        | ![](results/ppo/Breakout_rew.png)     | `python3 atari_ppo.py --task "BreakoutNoFrameskip-v4"` |
+| EnduroNoFrameskip-v4        | 1386.4       | ![](results/ppo/Enduro_rew.png)       | `python3 atari_ppo.py --task "EnduroNoFrameskip-v4"`  |
+| QbertNoFrameskip-v4         | 19585      | ![](results/ppo/Qbert_rew.png)        | `python3 atari_ppo.py --task "QbertNoFrameskip-v4"`  |
+| MsPacmanNoFrameskip-v4      | 2319       | ![](results/ppo/MsPacman_rew.png)     | `python3 atari_ppo.py --task "MsPacmanNoFrameskip-v4"`  |
+| SeaquestNoFrameskip-v4      | 1764       | ![](results/ppo/Seaquest_rew.png)     | `python3 atari_ppo.py --task "SeaquestNoFrameskip-v4"`  |
+| SpaceInvadersNoFrameskip-v4 | 1184       | ![](results/ppo/SpaceInvaders_rew.png) | `python3 atari_ppo.py --task "SpaceInvadersNoFrameskip-v4"`  |

examples/atari/atari_network.py

@@ -22,6 +22,7 @@ class DQN(nn.Module):
         action_shape: Sequence[int],
         device: Union[str, int, torch.device] = "cpu",
         features_only: bool = False,
+        output_dim: Optional[int] = None,
     ) -> None:
         super().__init__()
         self.device = device
@@ -39,6 +40,12 @@ class DQN(nn.Module):
                 nn.Linear(512, np.prod(action_shape))
             )
             self.output_dim = np.prod(action_shape)
+        elif output_dim is not None:
+            self.net = nn.Sequential(
+                self.net, nn.Linear(self.output_dim, output_dim),
+                nn.ReLU(inplace=True)
+            )
+            self.output_dim = output_dim
 
     def forward(
         self,

examples/atari/atari_ppo.py

@@ -0,0 +1,297 @@
+import argparse
+import os
+import pprint
+
+import numpy as np
+import torch
+from atari_network import DQN
+from atari_wrapper import wrap_deepmind
+from torch.optim.lr_scheduler import LambdaLR
+from torch.utils.tensorboard import SummaryWriter
+
+from tianshou.data import Collector, VectorReplayBuffer
+from tianshou.env import ShmemVectorEnv
+from tianshou.policy import ICMPolicy, PPOPolicy
+from tianshou.trainer import onpolicy_trainer
+from tianshou.utils import TensorboardLogger, WandbLogger
+from tianshou.utils.net.common import ActorCritic
+from tianshou.utils.net.discrete import Actor, Critic, IntrinsicCuriosityModule
+
+
+def get_args():
+    parser = argparse.ArgumentParser()
+    parser.add_argument('--task', type=str, default='PongNoFrameskip-v4')
+    parser.add_argument('--seed', type=int, default=4213)
+    parser.add_argument('--scale-obs', type=int, default=0)
+    parser.add_argument('--buffer-size', type=int, default=100000)
+    parser.add_argument('--lr', type=float, default=1e-4)
+    parser.add_argument('--gamma', type=float, default=0.99)
+    parser.add_argument('--epoch', type=int, default=100)
+    parser.add_argument('--step-per-epoch', type=int, default=100000)
+    parser.add_argument('--step-per-collect', type=int, default=1000)
+    parser.add_argument('--repeat-per-collect', type=int, default=4)
+    parser.add_argument('--batch-size', type=int, default=256)
+    parser.add_argument('--hidden-size', type=int, default=512)
+    parser.add_argument('--training-num', type=int, default=10)
+    parser.add_argument('--test-num', type=int, default=10)
+    parser.add_argument('--rew-norm', type=int, default=False)
+    parser.add_argument('--vf-coef', type=float, default=0.5)
+    parser.add_argument('--ent-coef', type=float, default=0.01)
+    parser.add_argument('--gae-lambda', type=float, default=0.95)
+    parser.add_argument('--lr-decay', type=int, default=True)
+    parser.add_argument('--max-grad-norm', type=float, default=0.5)
+    parser.add_argument('--eps-clip', type=float, default=0.2)
+    parser.add_argument('--dual-clip', type=float, default=None)
+    parser.add_argument('--value-clip', type=int, default=0)
+    parser.add_argument('--norm-adv', type=int, default=1)
+    parser.add_argument('--recompute-adv', type=int, default=0)
+    parser.add_argument('--logdir', type=str, default='log')
+    parser.add_argument('--render', type=float, default=0.)
+    parser.add_argument(
+        '--device', type=str, default='cuda' if torch.cuda.is_available() else 'cpu'
+    )
+    parser.add_argument('--frames-stack', type=int, default=4)
+    parser.add_argument('--resume-path', type=str, default=None)
+    parser.add_argument('--resume-id', type=str, default=None)
+    parser.add_argument(
+        '--logger',
+        type=str,
+        default="tensorboard",
+        choices=["tensorboard", "wandb"],
+    )
+    parser.add_argument(
+        '--watch',
+        default=False,
+        action='store_true',
+        help='watch the play of pre-trained policy only'
+    )
+    parser.add_argument('--save-buffer-name', type=str, default=None)
+    parser.add_argument(
+        '--icm-lr-scale',
+        type=float,
+        default=0.,
+        help='use intrinsic curiosity module with this lr scale'
+    )
+    parser.add_argument(
+        '--icm-reward-scale',
+        type=float,
+        default=0.01,
+        help='scaling factor for intrinsic curiosity reward'
+    )
+    parser.add_argument(
+        '--icm-forward-loss-weight',
+        type=float,
+        default=0.2,
+        help='weight for the forward model loss in ICM'
+    )
+    return parser.parse_args()
+
+
+def make_atari_env(args):
+    return wrap_deepmind(
+        args.task, frame_stack=args.frames_stack, scale=args.scale_obs
+    )
+
+
+def make_atari_env_watch(args):
+    return wrap_deepmind(
+        args.task,
+        frame_stack=args.frames_stack,
+        episode_life=False,
+        clip_rewards=False,
+        scale=args.scale_obs
+    )
+
+
+def test_ppo(args=get_args()):
+    env = make_atari_env(args)
+    args.state_shape = env.observation_space.shape or env.observation_space.n
+    args.action_shape = env.action_space.shape or env.action_space.n
+    # should be N_FRAMES x H x W
+    print("Observations shape:", args.state_shape)
+    print("Actions shape:", args.action_shape)
+    # make environments
+    train_envs = ShmemVectorEnv(
+        [lambda: make_atari_env(args) for _ in range(args.training_num)]
+    )
+    test_envs = ShmemVectorEnv(
+        [lambda: make_atari_env_watch(args) for _ in range(args.test_num)]
+    )
+    # seed
+    np.random.seed(args.seed)
+    torch.manual_seed(args.seed)
+    train_envs.seed(args.seed)
+    test_envs.seed(args.seed)
+    # define model
+    net = DQN(
+        *args.state_shape,
+        args.action_shape,
+        device=args.device,
+        features_only=True,
+        output_dim=args.hidden_size
+    )
+    actor = Actor(net, args.action_shape, device=args.device, softmax_output=False)
+    critic = Critic(net, device=args.device)
+    optim = torch.optim.Adam(ActorCritic(actor, critic).parameters(), lr=args.lr)
+
+    lr_scheduler = None
+    if args.lr_decay:
+        # decay learning rate to 0 linearly
+        max_update_num = np.ceil(
+            args.step_per_epoch / args.step_per_collect
+        ) * args.epoch
+
+        lr_scheduler = LambdaLR(
+            optim, lr_lambda=lambda epoch: 1 - epoch / max_update_num
+        )
+
+    # define policy
+    def dist(p):
+        return torch.distributions.Categorical(logits=p)
+
+    policy = PPOPolicy(
+        actor,
+        critic,
+        optim,
+        dist,
+        discount_factor=args.gamma,
+        gae_lambda=args.gae_lambda,
+        max_grad_norm=args.max_grad_norm,
+        vf_coef=args.vf_coef,
+        ent_coef=args.ent_coef,
+        reward_normalization=args.rew_norm,
+        action_scaling=False,
+        lr_scheduler=lr_scheduler,
+        action_space=env.action_space,
+        eps_clip=args.eps_clip,
+        value_clip=args.value_clip,
+        dual_clip=args.dual_clip,
+        advantage_normalization=args.norm_adv,
+        recompute_advantage=args.recompute_adv
+    ).to(args.device)
+    if args.icm_lr_scale > 0:
+        feature_net = DQN(
+            *args.state_shape, args.action_shape, args.device, features_only=True
+        )
+        action_dim = np.prod(args.action_shape)
+        feature_dim = feature_net.output_dim
+        icm_net = IntrinsicCuriosityModule(
+            feature_net.net,
+            feature_dim,
+            action_dim,
+            hidden_sizes=args.hidden_sizes,
+            device=args.device
+        )
+        icm_optim = torch.optim.Adam(icm_net.parameters(), lr=args.lr)
+        policy = ICMPolicy(
+            policy, icm_net, icm_optim, args.icm_lr_scale, args.icm_reward_scale,
+            args.icm_forward_loss_weight
+        ).to(args.device)
+    # load a previous policy
+    if args.resume_path:
+        policy.load_state_dict(torch.load(args.resume_path, map_location=args.device))
+        print("Loaded agent from: ", args.resume_path)
+    # replay buffer: `save_last_obs` and `stack_num` can be removed together
+    # when you have enough RAM
+    buffer = VectorReplayBuffer(
+        args.buffer_size,
+        buffer_num=len(train_envs),
+        ignore_obs_next=True,
+        save_only_last_obs=True,
+        stack_num=args.frames_stack
+    )
+    # collector
+    train_collector = Collector(policy, train_envs, buffer, exploration_noise=True)
+    test_collector = Collector(policy, test_envs, exploration_noise=True)
+    # log
+    log_name = 'ppo_icm' if args.icm_lr_scale > 0 else 'ppo'
+    log_path = os.path.join(args.logdir, args.task, log_name)
+    if args.logger == "tensorboard":
+        writer = SummaryWriter(log_path)
+        writer.add_text("args", str(args))
+        logger = TensorboardLogger(writer)
+    else:
+        logger = WandbLogger(
+            save_interval=1,
+            project=args.task,
+            name=log_name,
+            run_id=args.resume_id,
+            config=args,
+        )
+
+    def save_fn(policy):
+        torch.save(policy.state_dict(), os.path.join(log_path, 'policy.pth'))
+
+    def stop_fn(mean_rewards):
+        if env.spec.reward_threshold:
+            return mean_rewards >= env.spec.reward_threshold
+        elif 'Pong' in args.task:
+            return mean_rewards >= 20
+        else:
+            return False
+
+    def save_checkpoint_fn(epoch, env_step, gradient_step):
+        # see also: https://pytorch.org/tutorials/beginner/saving_loading_models.html
+        ckpt_path = os.path.join(log_path, 'checkpoint.pth')
+        torch.save({'model': policy.state_dict()}, ckpt_path)
+        return ckpt_path
+
+    # watch agent's performance
+    def watch():
+        print("Setup test envs ...")
+        policy.eval()
+        test_envs.seed(args.seed)
+        if args.save_buffer_name:
+            print(f"Generate buffer with size {args.buffer_size}")
+            buffer = VectorReplayBuffer(
+                args.buffer_size,
+                buffer_num=len(test_envs),
+                ignore_obs_next=True,
+                save_only_last_obs=True,
+                stack_num=args.frames_stack
+            )
+            collector = Collector(policy, test_envs, buffer, exploration_noise=True)
+            result = collector.collect(n_step=args.buffer_size)
+            print(f"Save buffer into {args.save_buffer_name}")
+            # Unfortunately, pickle will cause oom with 1M buffer size
+            buffer.save_hdf5(args.save_buffer_name)
+        else:
+            print("Testing agent ...")
+            test_collector.reset()
+            result = test_collector.collect(
+                n_episode=args.test_num, render=args.render
+            )
+        rew = result["rews"].mean()
+        print(f'Mean reward (over {result["n/ep"]} episodes): {rew}')
+
+    if args.watch:
+        watch()
+        exit(0)
+
+    # test train_collector and start filling replay buffer
+    train_collector.collect(n_step=args.batch_size * args.training_num)
+    # trainer
+    result = onpolicy_trainer(
+        policy,
+        train_collector,
+        test_collector,
+        args.epoch,
+        args.step_per_epoch,
+        args.repeat_per_collect,
+        args.test_num,
+        args.batch_size,
+        step_per_collect=args.step_per_collect,
+        stop_fn=stop_fn,
+        save_fn=save_fn,
+        logger=logger,
+        test_in_train=False,
+        resume_from_log=args.resume_id is not None,
+        save_checkpoint_fn=save_checkpoint_fn,
+    )
+
+    pprint.pprint(result)
+    watch()
+
+
+if __name__ == '__main__':
+    test_ppo(get_args())