neardws / Game-Theoretic-Deep-Reinforcement-Learning

Code of Paper "Joint Task Offloading and Resource Optimization in NOMA-based Vehicular Edge Computing: A Game-Theoretic DRL Approach", JSA 2022.
https://www.sciencedirect.com/science/article/abs/pii/S138376212200265X
GNU General Public License v3.0
222 stars 47 forks source link

请教一下作者关于mad4pg的实验问题 #23

Closed Simple11024 closed 1 year ago

Simple11024 commented 1 year ago

Describe the bug A clear and concise description of what the bug is.

To Reproduce Steps to reproduce the behavior:

  1. Go to '...'
  2. ...
  3. See error

Expected behavior A clear and concise description of what you expected to happen.

Screenshots If applicable, add screenshots to help explain your problem.

Runtime Environment (please complete the following information):

Additional context Add any other context about the problem here.

Simple11024 commented 1 year ago

作者大大,您好,我是一名在读研究生,有幸读到您的论文。我最近在学习您的代码。您的代码整洁美观,可读性极高;论文写得也非常好,让我收获很多。遇到一些问题,希望得到您的指点,感激不尽。 1.environment_old和environment的主要区别 我尝试使用mad4pg调用environment也能跑 您能指点一下吗 2.我是在win10的虚拟机上运行的,纯cpu跑,跑一轮madd4g的时间是6h,请问如果用gpu训练的话,速度能达到多少(如果速度提升大我想试试用gpu跑跑试试) 3.我的mad4pg跑5h就会不正常终止,导致cumulative reward 只有500个数据,无法画出完整的图(您论文中完整的应该为4000个数据是正常的)但是我看到您的Draw/draw_result.ipynb中示例图中的cumulative reward也只有700个数据,但是在figure/draw_cumulative_reward.ipynb中mad4pg的cumulative reward拥有4000个数据 我想请您指点一下

neardws commented 1 year ago
  1. environment是最新的版本,用这个就行
  2. 使用GPU的话(NVIDIA 3090),一轮(4000 iterations)好像是半天吧,记不太清楚了
  3. draw_result.ipynb 里的数据并不是对应算法收敛图,因为所有原始数据其实都是保存在 /用户名/acme/ 文件夹下的。至于跑 5h 就终止,你需要看看它是否迭代到了你设置的iterations,如果没有,它有什么报错信息吗?
Simple11024 commented 1 year ago

感谢您的回复,我的问题都已经解决了,感谢!