TAPE: Topology-based Multi-Agent Policy Gradient

This is the code repository for AAAI 2024 paper TAPE: Leveraging Agent Topology for Cooperative Multi-Agent Policy Gradient.

This repo is built upon SMAC, DOP and PAC.

To run experiments:

First install SMAC, then

for stochastic TAPE,

cd stochastic  
bash runalgo.sh  

for deterministic TAPE,

cd deterministic  
bash runalgo.sh

To kill running processes:

pkill -u ($youruser) python
pkill -u ($youruser) Main_Thread

To run with other graph models:

You can use networkx package to generate graphs and use the adjcency matrix for the agent topology

import networkx as nx
import matplotlib.pyplot as plt
import numpy as np

# Create a Watts-Strogatz small-world network
n = 20
k = 4 
p = 0.1 
G = nx.watts_strogatz_graph(n, k, p)

# plot
nx.draw(G, with_labels=True)
plt.show()

A = nx.adjacency_matrix(G)
for i in range(n):
    A[i,i]=1
print(A.todense())

Citation

Please cite

@article{lou2023tape,
  title={TAPE: Leveraging Agent Topology for Cooperative Multi-Agent Policy Gradient},
  author={Lou, Xingzhou and Zhang, Junge and Norman, Timothy J and Huang, Kaiqi and Du, Yali},
  journal={arXiv preprint arXiv:2312.15667},
  year={2023}
}