We recommend creating separate conda environments for data generation and code execution.

1. Python Version: 3.8.18
2. Module
   • pytorch 2.0.0
   • torch-geometric 2.0.4
   • ogb 1.3.6
   • networkx 3.1
   • metis-python 0.1, Link

cd federated/data/generator
python3 split_disjoint.py

1. Python Version: 3.10.14
2. Module
   • pytorch 2.1.1
   • torch-geometric 2.5.3
   • networkx 3.2.1
   • numpy 1.26.3
   • scipy 1.14.0
   • pyyaml 6.0.1

All variables necessary for execution, such as paths and hyperparameters, are stored in YAML files in config directory

Pretrain

python3 main.py [pretrain config file path]

Run CUFL

python3 main.py [config file path]

Here is an example

python3 main.py --config config/fedprox_pretrain.yaml
python3 main.py --config config/cufl_cora.yaml

GNN hidden dim : 128