This project is a federated learning model implemented using PyTorch and PyTorch Geometric. It uses a Graph Convolutional Network (GCN) to process graph data and employs two federated learning algorithms: FedAvg and FedProx.
main.py: The main program entry point, which includes code for parsing command-line arguments, training the model, and evaluating the results.model.py: Contains the definition of the Graph Convolutional Network (GCN) model.data_loader.py: Contains the data loading function, which can load either the WikiCS or Cora datasets.client_server.py: Contains the definitions of the Client and Server classes, which simulate the federated learning environment.train_eval.py: Contains functions for training and evaluating the model.metrics.py: Contains functions for calculating accuracy, precision, recall, and F1 score.
- Install dependencies:
pip install torch torch_geometric - Run the main program:
python main.py --dataset WikiCS --optimizer Adam --clients 10 --fed_algo fedavg --lr 0.01 --local_epoch 5 --communication_rounds 10
--dataset: The name of the dataset, either 'WikiCS' or 'Cora'. Default is 'WikiCS'.--optimizer: The type of optimizer, either 'Adam' or 'SGD'. Default is 'Adam'.--clients: The number of clients, must be a positive integer. Default is 10.--fed_algo: The federated learning algorithm, either 'fedavg' or 'fedprox'. Default is 'fedavg'.--lr: The learning rate, must be a positive float. Default is 0.01.--local_epoch: The number of local training epochs per communication round, must be a positive integer. Default is 5.--communication_rounds: The number of communication rounds, must be a positive integer. Default is 10.
- Please ensure that your Python version is 3.6 or higher.
- Please ensure that you have installed all necessary dependencies.