Open Federated Graph Learning (OpenFGL)

OpenFGL is a comprehensive, user-friendly algorithm library, complemented by an integrated evaluation platform, designed specifically for researchers in the field of federated graph learning (FGL).

Paper • Highlights • Installation • Docs • Citation

Highlights

• 2 FGL Scenarios: Graph-FL and Subgraph-FL
• 10+ FGL Algorithms
• 34 FGL Datasets
• 12 GNN Models
• 5 Downstream Tasks
• Comprehensive FGL Data Property Analysis

Get Started

import openfgl.config as config


from openfgl.flcore.trainer import FGLTrainer

args = config.args

args.root = "your_data_root"


args.dataset = ["Cora"]
args.simulation_mode = "subgraph_fl_louvain"
args.num_clients = 10


if True:
    args.fl_algorithm = "fedavg"
    args.model = ["gcn"]
else:
    args.fl_algorithm = "fedproto"
    args.model = ["gcn", "gat", "sgc", "mlp", "graphsage"] # choose multiple gnn models for model heterogeneity setting.

args.metrics = ["accuracy"]



trainer = FGLTrainer(args)

trainer.train()

Citation

Please cite our paper (and the respective papers of the methods used) if you use this code in your own work:

@misc{li2024openfglcomprehensivebenchmarksfederated,
      title={OpenFGL: A Comprehensive Benchmarks for Federated Graph Learning}, 
      author={Xunkai Li and Yinlin Zhu and Boyang Pang and Guochen Yan and Yeyu Yan and Zening Li and Zhengyu Wu and Wentao Zhang and Rong-Hua Li and Guoren Wang},
      year={2024},
      eprint={2408.16288},
      archivePrefix={arXiv},
      primaryClass={cs.LG},
      url={https://arxiv.org/abs/2408.16288}, 
}