DI-engine/dizoo/classic_control/cartpole/entry/cartpole_ppo_main.py

import os
import gym
from tensorboardX import SummaryWriter
from easydict import EasyDict

from ding.config import compile_config
from ding.worker import BaseLearner, SampleSerialCollector, InteractionSerialEvaluator
from ding.envs import BaseEnvManager, DingEnvWrapper
from ding.policy import PPOPolicy
from ding.model import VAC
from ding.utils import set_pkg_seed, deep_merge_dicts
from dizoo.classic_control.cartpole.config.cartpole_ppo_config import cartpole_ppo_config


def wrapped_cartpole_env():
    return DingEnvWrapper(
        gym.make('CartPole-v0'),
        EasyDict(env_wrapper='default'),
    )


def main(cfg, seed=0, max_iterations=int(1e10)):
    cfg = compile_config(
        cfg, BaseEnvManager, PPOPolicy, BaseLearner, SampleSerialCollector, InteractionSerialEvaluator, save_cfg=True
    )
    collector_env_num, evaluator_env_num = cfg.env.collector_env_num, cfg.env.evaluator_env_num
    collector_env = BaseEnvManager(env_fn=[wrapped_cartpole_env for _ in range(collector_env_num)], cfg=cfg.env.manager)
    evaluator_env = BaseEnvManager(env_fn=[wrapped_cartpole_env for _ in range(evaluator_env_num)], cfg=cfg.env.manager)

    collector_env.seed(seed)
    evaluator_env.seed(seed, dynamic_seed=False)
    set_pkg_seed(seed, use_cuda=cfg.policy.cuda)

    model = VAC(**cfg.policy.model)
    policy = PPOPolicy(cfg.policy, model=model)
    tb_logger = SummaryWriter(os.path.join('./{}/log/'.format(cfg.exp_name), 'serial'))
    learner = BaseLearner(cfg.policy.learn.learner, policy.learn_mode, tb_logger, exp_name=cfg.exp_name)
    collector = SampleSerialCollector(
        cfg.policy.collect.collector, collector_env, policy.collect_mode, tb_logger, exp_name=cfg.exp_name
    )
    evaluator = InteractionSerialEvaluator(
        cfg.policy.eval.evaluator, evaluator_env, policy.eval_mode, tb_logger, exp_name=cfg.exp_name
    )

    for _ in range(max_iterations):
        if evaluator.should_eval(learner.train_iter):
            stop, reward = evaluator.eval(learner.save_checkpoint, learner.train_iter, collector.envstep)
            if stop:
                break
        new_data = collector.collect(train_iter=learner.train_iter)
        learner.train(new_data, collector.envstep)


if __name__ == "__main__":
    main(cartpole_ppo_config)