The fedhgn's intro from cynricfu

fedhgn's Introduction

FedHGN

This repository provides a reference implementation of FedHGN as described in the following paper.

FedHGN: A Federated Framework for Heterogeneous Graph Neural Networks
Xinyu Fu, Irwin King
International Joint Conference on Artificial Intelligence, 2023

Available at arXiv:2305.09729.

Dependencies

Python 3.9
PyTorch 1.13.1
DGL 0.9.1
scikit-learn 1.1.2
SciPy 1.9.0
PyYAML 6.0
tqdm 4.64.1

Datasets

The raw data are obtained from DGL:

The datasets above are preprocessed by prepare_data.ipynb.

Usage

usage: main.py [-h] --dataset DATASET [--split-strategy SPLIT_STRATEGY] [--framework FRAMEWORK] [--ablation ABLATION]
               [--model MODEL] [--num-clients NUM_CLIENTS] [--gpu GPU] [--random-seed RANDOM_SEED]
               [--config-path CONFIG_PATH]

optional arguments:
  -h, --help            show this help message and exit
  --dataset DATASET, -d DATASET
                        name of dataset
  --split-strategy SPLIT_STRATEGY, -s SPLIT_STRATEGY
                        dataset splitting strategy, either "edges" or "etypes"
  --framework FRAMEWORK, -f FRAMEWORK
                        FedHGN/FedAvg/FedProx/Local/Central
  --ablation ABLATION, -a ABLATION
                        B/C/B+C
  --num-clients NUM_CLIENTS, -c NUM_CLIENTS
                        number of clients, 3/5/10
  --gpu GPU, -g GPU     which gpu to use, specify -1 to use CPU
  --random-seed RANDOM_SEED
                        random seed
  --config-path CONFIG_PATH
                        path to config file

For example, to run FedHGN on the AIFB dataset with the random edges splitting strategy with 5 clients using GPU:

python main.py -d AIFB -s edges -f FedHGN -c 5 -g 0

Citing

If you find FedHGN useful in your research, please cite the following paper:

@inproceedings{fu2023fedhgn,
  author       = {Xinyu Fu and
                  Irwin King},
  title        = {FedHGN: {A} Federated Framework for Heterogeneous Graph Neural Networks},
  booktitle    = {{IJCAI}},
  pages        = {3705--3713},
  publisher    = {ijcai.org},
  year         = {2023}
}

Recommend Projects