Tianshou/test/discrete/test_sac.py

import argparse
import os
import pprint

import gym
import numpy as np
import torch
from torch.utils.tensorboard import SummaryWriter

from tianshou.data import Collector, VectorReplayBuffer
from tianshou.env import DummyVectorEnv
from tianshou.policy import DiscreteSACPolicy
from tianshou.trainer import offpolicy_trainer
from tianshou.utils import TensorboardLogger
from tianshou.utils.net.common import Net
from tianshou.utils.net.discrete import Actor, Critic


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=1)
    parser.add_argument('--buffer-size', type=int, default=20000)
    parser.add_argument('--actor-lr', type=float, default=1e-4)
    parser.add_argument('--critic-lr', type=float, default=1e-3)
    parser.add_argument('--alpha-lr', type=float, default=3e-4)
    parser.add_argument('--gamma', type=float, default=0.95)
    parser.add_argument('--tau', type=float, default=0.005)
    parser.add_argument('--alpha', type=float, default=0.05)
    parser.add_argument('--auto-alpha', action="store_true", default=False)
    parser.add_argument('--epoch', type=int, default=5)
    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=[64, 64])
    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=0.0)
    parser.add_argument('--rew-norm', action="store_true", default=False)
    parser.add_argument('--n-step', type=int, default=3)
    parser.add_argument(
        '--device', type=str, default='cuda' if torch.cuda.is_available() else 'cpu'
    )
    args = parser.parse_known_args()[0]
    return args


def test_discrete_sac(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": 170}  # lower the goal
        args.reward_threshold = default_reward_threshold.get(
            args.task, env.spec.reward_threshold
        )

    train_envs = DummyVectorEnv(
        [lambda: gym.make(args.task) for _ in range(args.training_num)]
    )
    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)
    # model
    net = Net(args.state_shape, hidden_sizes=args.hidden_sizes, device=args.device)
    actor = Actor(net, args.action_shape, softmax_output=False,
                  device=args.device).to(args.device)
    actor_optim = torch.optim.Adam(actor.parameters(), lr=args.actor_lr)
    net_c1 = Net(args.state_shape, hidden_sizes=args.hidden_sizes, device=args.device)
    critic1 = Critic(net_c1, last_size=args.action_shape,
                     device=args.device).to(args.device)
    critic1_optim = torch.optim.Adam(critic1.parameters(), lr=args.critic_lr)
    net_c2 = Net(args.state_shape, hidden_sizes=args.hidden_sizes, device=args.device)
    critic2 = Critic(net_c2, last_size=args.action_shape,
                     device=args.device).to(args.device)
    critic2_optim = torch.optim.Adam(critic2.parameters(), lr=args.critic_lr)

    # better not to use auto alpha in CartPole
    if args.auto_alpha:
        target_entropy = 0.98 * np.log(np.prod(args.action_shape))
        log_alpha = torch.zeros(1, requires_grad=True, device=args.device)
        alpha_optim = torch.optim.Adam([log_alpha], lr=args.alpha_lr)
        args.alpha = (target_entropy, log_alpha, alpha_optim)

    policy = DiscreteSACPolicy(
        actor,
        actor_optim,
        critic1,
        critic1_optim,
        critic2,
        critic2_optim,
        args.tau,
        args.gamma,
        args.alpha,
        estimation_step=args.n_step,
        reward_normalization=args.rew_norm
    )
    # collector
    train_collector = Collector(
        policy, train_envs, VectorReplayBuffer(args.buffer_size, len(train_envs))
    )
    test_collector = Collector(policy, test_envs)
    # train_collector.collect(n_step=args.buffer_size)
    # log
    log_path = os.path.join(args.logdir, args.task, 'discrete_sac')
    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

    # trainer
    result = offpolicy_trainer(
        policy,
        train_collector,
        test_collector,
        args.epoch,
        args.step_per_epoch,
        args.step_per_collect,
        args.test_num,
        args.batch_size,
        stop_fn=stop_fn,
        save_best_fn=save_best_fn,
        logger=logger,
        update_per_step=args.update_per_step,
        test_in_train=False
    )
    assert stop_fn(result['best_reward'])

    if __name__ == '__main__':
        pprint.pprint(result)
        # Let's watch its performance!
        env = gym.make(args.task)
        policy.eval()
        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_discrete_sac()