Official PyTorch implementation of "FedGTP: Exploiting Inter-Client Spatial Dependency in Federated Graph-based Traffic Prediction".
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
bash run.sh
Hyperparameters | Values |
---|---|
hidden feature dimension |
64 |
number of hidden layers |
2 |
embedding dimension |
2 |
polynomial coefficient |
4 |
learning rate |
0.003 |
batch size | 64 |
number of global epochs |
200 |
number of local epochs |
2 |
validation ratio | 0.2 |
testing ratio | 0.2 |