ding.example.collect_demo_data

ding.example.collect_demo_data

Full Source Code

../ding/example/collect_demo_data.py

1import gym 2from ditk import logging 3import torch 4from ding.model import ContinuousQAC 5from ding.policy import SACPolicy 6from ding.envs import DingEnvWrapper, BaseEnvManagerV2 7from ding.data import offline_data_save_type 8from ding.config import compile_config 9from ding.framework import task 10from ding.framework.context import OnlineRLContext 11from ding.framework.middleware import StepCollector, offline_data_saver 12from ding.utils import set_pkg_seed 13from dizoo.classic_control.pendulum.envs.pendulum_env import PendulumEnv 14from dizoo.classic_control.pendulum.config.pendulum_sac_data_generation_config import main_config, create_config 15 16 17def main(): 18 logging.getLogger().setLevel(logging.INFO) 19 cfg = compile_config(main_config, create_cfg=create_config, auto=True, evaluator=None) 20 with task.start(async_mode=False, ctx=OnlineRLContext()): 21 collector_env = BaseEnvManagerV2(env_fn=[lambda: PendulumEnv(cfg.env) for _ in range(10)], cfg=cfg.env.manager) 22 23 set_pkg_seed(cfg.seed, use_cuda=cfg.policy.cuda) 24 25 model = ContinuousQAC(**cfg.policy.model) 26 policy = SACPolicy(cfg.policy, model=model, enable_field=['collect']) 27 state_dict = torch.load(cfg.policy.collect.state_dict_path, map_location='cpu') 28 policy.collect_mode.load_state_dict(state_dict) 29 30 task.use(StepCollector(cfg, policy.collect_mode, collector_env)) 31 task.use(offline_data_saver(cfg.policy.collect.save_path, data_type='hdf5')) 32 task.run(max_step=1) 33 34 35if __name__ == "__main__": 36 main()