search

starry-sky6688 / MADDPG

Pytorch implementation of the MARL algorithm, MADDPG, which correspondings to the paper "Multi-Agent Actor-Critic for Mixed Cooperative-Competitive Environments".
516 stars 80 forks source link

epsilon 更新的问题 #9

Closed shmily326 closed 2 years ago

shmily326 commented 2 years ago

https://github.com/starry-sky6688/MADDPG/blob/2c5a93fb2680daf93e0e5c75298574f34038cc01/runner.py#L62 是否应该改为 self.epsilon = max(0.05, self.epsilon - 0.0000005) 呢？

starry-sky6688 commented 2 years ago

是的，谢谢你的提醒，已经改过来了

不过之前的实验结果应该还是没问题的，因为默认的noise和epsilon的初始值一样，然后下降速度也一样，导致我当时没发现这个问题