Cross-Node Federated Graph Neural Network for Spatio-Temporal Data Modeling

This repository is the official PyTorch implementation of "Cross-Node Federated Graph Neural Network for Spatio-Temporal Modeling". [arXiv]

Setup

Environment

conda create -n fedgnn "python<3.8"
conda activate fedgnn
bash install.sh

Data

Download data.tar.bz2. Then extract it to the root directory of the repository:

tar -xjf data.tar.bz2

Experiments

Main Experiments

submission_exps/exp_main.sh contains all commands used for experiments in Table 2 and Table 3.

Inductive Learning on Unseen Nodes

Run python submission_exps/exp_inductive.py to print all commands for Table 4.

Ablation Study: Effect of Alternating Training of Node-Level and Spatial Models

Run python submission_exps/exp_at.py to print all commands for Figure 2.

Ablation Study: Effect of Client Rounds and Server Rounds

Run python submission_exps/exp_crsr.py to print all commands for Figure 3.