PaddlePaddle / PARL

A high-performance distributed training framework for Reinforcement Learning
https://parl.readthedocs.io/
Apache License 2.0
3.22k stars 816 forks source link

examples/PPO/atari_config.py及examples/PPO/mujoco_config.py 里未见‘num_updates’参数定义 #958

Open xhdavid opened 1 year ago

xhdavid commented 1 year ago

examples/PPO/atari_config.py 里未见‘num_updates’参数定义,而在trian.py及agent.py里多次被调用。

ShuaibinLi commented 1 year ago

num_updates 仅与训练步数相关(仅用在定义learn的次数及对应的lr更新),需要根据训练的总步数以及batch_size计算,严格意义不属于需要自主定义的参数。