parachutel
commented
2 years ago This repo is not maintained. Please redirect to https://github.com/sisl/DICG.
Core modules/algorithms are in /dicg/torch, other directories are experiment runners (/exp_runners), environments (/envs) and RL utilities (/garage).
Core files:
/dicg/torch/algos/centralized_ma_ppo.py: An implementation of centralized training PPO/dicg/torch/modules/dicg_base.py: The base class of DICG/dicg/torch/baselines: Some baselines (aka value functions) for PPO- Other modules in
/dicg/torch/modules: GCN, self-attention, MLP policy module, LSTM policy module, etc. /dicg/torch/policies: Containing all the policies including DICG-CE, DEC and CENT, (DICG-DE is a combination ofdecpolicy anddicg_criticinbaselines) and their MLP/LSTM variants.
Launch experiments following the commands in REAME.
minglou1984
你好，最近注意到你的工作《Deep Implicit Coordination Graphs for Multi-agent Reinforcement Learning》。实验结果在多智能体追捕环境中效果非常好，所以研读了你的论文，并且也想把你的开源代码移植在我的电脑上。大致看了你的代码，应该是你的多个项目的集成。你的开源代码的结构我很难理清，你能大致说明各文件目录的含义吗？非常期待你能给我解答！