import argparse import os import pprint import gymnasium as gym import numpy as np import torch from torch.utils.tensorboard import SummaryWriter from tianshou.data import Collector, PrioritizedVectorReplayBuffer, VectorReplayBuffer from tianshou.env import DummyVectorEnv from tianshou.policy import DQNPolicy, ICMPolicy from tianshou.trainer import OffpolicyTrainer from tianshou.utils import TensorboardLogger from tianshou.utils.net.common import MLP, Net from tianshou.utils.net.discrete import IntrinsicCuriosityModule def get_args(): parser = argparse.ArgumentParser() parser.add_argument('--task', type=str, default='CartPole-v0') parser.add_argument('--reward-threshold', type=float, default=None) parser.add_argument('--seed', type=int, default=1626) parser.add_argument('--eps-test', type=float, default=0.05) parser.add_argument('--eps-train', type=float, default=0.1) parser.add_argument('--buffer-size', type=int, default=20000) parser.add_argument('--lr', type=float, default=1e-3) parser.add_argument('--gamma', type=float, default=0.9) parser.add_argument('--n-step', type=int, default=3) parser.add_argument('--target-update-freq', type=int, default=320) parser.add_argument('--epoch', type=int, default=20) parser.add_argument('--step-per-epoch', type=int, default=10000) parser.add_argument('--step-per-collect', type=int, default=10) parser.add_argument('--update-per-step', type=float, default=0.1) parser.add_argument('--batch-size', type=int, default=64) parser.add_argument( '--hidden-sizes', type=int, nargs='*', default=[128, 128, 128, 128] ) parser.add_argument('--training-num', type=int, default=10) parser.add_argument('--test-num', type=int, default=100) parser.add_argument('--logdir', type=str, default='log') parser.add_argument('--render', type=float, default=0.) parser.add_argument('--prioritized-replay', action="store_true", default=False) parser.add_argument('--alpha', type=float, default=0.6) parser.add_argument('--beta', type=float, default=0.4) parser.add_argument( '--device', type=str, default='cuda' if torch.cuda.is_available() else 'cpu' ) parser.add_argument( '--lr-scale', type=float, default=1., help='use intrinsic curiosity module with this lr scale' ) parser.add_argument( '--reward-scale', type=float, default=0.01, help='scaling factor for intrinsic curiosity reward' ) parser.add_argument( '--forward-loss-weight', type=float, default=0.2, help='weight for the forward model loss in ICM' ) args = parser.parse_known_args()[0] return args def test_dqn_icm(args=get_args()): env = gym.make(args.task) args.state_shape = env.observation_space.shape or env.observation_space.n args.action_shape = env.action_space.shape or env.action_space.n if args.reward_threshold is None: default_reward_threshold = {"CartPole-v0": 195} args.reward_threshold = default_reward_threshold.get( args.task, env.spec.reward_threshold ) # train_envs = gym.make(args.task) # you can also use tianshou.env.SubprocVectorEnv train_envs = DummyVectorEnv( [lambda: gym.make(args.task) for _ in range(args.training_num)] ) # test_envs = gym.make(args.task) test_envs = DummyVectorEnv( [lambda: gym.make(args.task) 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) # Q_param = V_param = {"hidden_sizes": [128]} # model net = Net( args.state_shape, args.action_shape, hidden_sizes=args.hidden_sizes, device=args.device, # dueling=(Q_param, V_param), ).to(args.device) optim = torch.optim.Adam(net.parameters(), lr=args.lr) policy = DQNPolicy( net, optim, args.gamma, args.n_step, target_update_freq=args.target_update_freq, ) feature_dim = args.hidden_sizes[-1] feature_net = MLP( np.prod(args.state_shape), output_dim=feature_dim, hidden_sizes=args.hidden_sizes[:-1], device=args.device ) action_dim = np.prod(args.action_shape) icm_net = IntrinsicCuriosityModule( feature_net, feature_dim, action_dim, hidden_sizes=args.hidden_sizes[-1:], device=args.device ).to(args.device) icm_optim = torch.optim.Adam(icm_net.parameters(), lr=args.lr) policy = ICMPolicy( policy, icm_net, icm_optim, args.lr_scale, args.reward_scale, args.forward_loss_weight ) # buffer if args.prioritized_replay: buf = PrioritizedVectorReplayBuffer( args.buffer_size, buffer_num=len(train_envs), alpha=args.alpha, beta=args.beta, ) else: buf = VectorReplayBuffer(args.buffer_size, buffer_num=len(train_envs)) # collector train_collector = Collector(policy, train_envs, buf, exploration_noise=True) test_collector = Collector(policy, test_envs, exploration_noise=True) # policy.set_eps(1) train_collector.collect(n_step=args.batch_size * args.training_num) # log log_path = os.path.join(args.logdir, args.task, 'dqn_icm') writer = SummaryWriter(log_path) logger = TensorboardLogger(writer) def save_best_fn(policy): torch.save(policy.state_dict(), os.path.join(log_path, 'policy.pth')) def stop_fn(mean_rewards): return mean_rewards >= args.reward_threshold def train_fn(epoch, env_step): # eps annnealing, just a demo if env_step <= 10000: policy.set_eps(args.eps_train) elif env_step <= 50000: eps = args.eps_train - (env_step - 10000) / \ 40000 * (0.9 * args.eps_train) policy.set_eps(eps) else: policy.set_eps(0.1 * args.eps_train) def test_fn(epoch, env_step): policy.set_eps(args.eps_test) # trainer result = OffpolicyTrainer( policy=policy, train_collector=train_collector, test_collector=test_collector, max_epoch=args.epoch, step_per_epoch=args.step_per_epoch, step_per_collect=args.step_per_collect, episode_per_test=args.test_num, batch_size=args.batch_size, update_per_step=args.update_per_step, train_fn=train_fn, test_fn=test_fn, stop_fn=stop_fn, save_best_fn=save_best_fn, logger=logger, ).run() assert stop_fn(result['best_reward']) if __name__ == '__main__': pprint.pprint(result) # Let's watch its performance! env = gym.make(args.task) policy.eval() policy.set_eps(args.eps_test) collector = Collector(policy, env) result = collector.collect(n_episode=1, render=args.render) rews, lens = result["rews"], result["lens"] print(f"Final reward: {rews.mean()}, length: {lens.mean()}") if __name__ == '__main__': test_dqn_icm(get_args())