sjtu-marl / malib

A parallel framework for population-based multi-agent reinforcement learning.
https://malib.io
MIT License
499 stars 60 forks source link

For SMAC Qimx/MADDPG config #34

Closed Weiyuhong-1998 closed 2 years ago

Weiyuhong-1998 commented 2 years ago

It's really a nice work and according to your paper, running Qmix/MADDPG is really fast. But we didn't find a config about Qmix/MADDPG algorithm, We don't know how to run your Qmix/MADDPG program,so can you give us a config for the Qmix/MADDPG algorithm? WeChat Work Screenshot_20211227124527

zhuerfei commented 2 years ago

Have you solved this problem bro? I have some questions of this framework, shall we talk about this.

KornbergFresnel commented 2 years ago

The repo has been refactored, thus I closed this issue