halleanwoo / agma

Source code of the paper "AGMA: Agreement Generation Multi-agent Reinforcement Learning"

AGMA

This repo is the source code of the paper "AGMA: Agreement Generation Multi-agent Reinforcement Learning". It is constructed based on the SMAC and PyMARL.

Installation

Clone this repo and set up SMAC:

cd Sub_AVG
bash install_sc2.sh

The installation is same with PyMARL. For quick start, you can install the necessary packages directly by (not recommended):

pip install -r requirements.txt

If needed, add directory that contains StarCraftII to the system's path ('xxx' is your local file path):

export PATH="$PATH:/xxx/AGMA/3rdparty/StarCraftII"

Run Sub-AVG

We provide run_AGMA.sh , run_QMIX.sh, run_VDN.sh, run_QTRAN.sh and run_IQL.sh for quick start.

Results

The results save path is: "./results/sacred".