KDD2024_FedGTP

Official PyTorch implementation of "FedGTP: Exploiting Inter-Client Spatial Dependency in Federated Graph-based Traffic Prediction".

Setup

Environment

PyTorch 1.13.1 and conducted on Intel(R) Xeon(R) Gold 6230R CPU @ 2.10GHz and four NVIDIA A100-PCIE-40GB GPUs with CUDA 11.6.

conda create -n fedgtp "python=3.11"
conda activate fedgtp
bash install.sh

Run

bash run.sh

Default Hyperparameter Configuration

Hyperparameters	Values
hidden feature dimension $F$	64
number of hidden layers $L$	2
embedding dimension $d$	2
polynomial coefficient $K$	4
learning rate $\eta$	0.003
batch size	64
number of global epochs $R_g$	200
number of local epochs $R_l$	2
validation ratio	0.2
testing ratio	0.2