import flwr as fl from dnn_model import create_model import argparse from utils import load_config def get_on_fit_config_fn(batch_size, local_epochs): ''' Returns a functions which returns the server config file, to be used by the clients ''' def fit_config(server_round): config = { "batch_size": batch_size, "current_round": server_round, "local_epochs": local_epochs, } return config return fit_config def parse_arguments(): '''Parser ''' parser = argparse.ArgumentParser(description='FL server') parser.add_argument('--config', type=str, default='conf/server/config.json', help='Path to the config file: conf/server/config.json') parser.add_argument('--server_address', type=str, help='Address of server') parser.add_argument('--num_rounds', type=int, help='Number of rounds') parser.add_argument('--num_clients', type=int, help='Number of clients') parser.add_argument('--batch_size', type=int, help='Batch size') parser.add_argument('--local_epochs', type=int, help='Number of local epochs') parser.add_argument('--input_shape', type=int, help='Model input shape') parser.add_argument('--num_classes', type=int, help='Number of classes') parser.add_argument('--loss', type=str, help='Loss function of the model') return parser.parse_args() def update_config(args, conf): '''Update config file via cli ''' if args.server_address is not None: conf['server_address'] = args.server_address if args.num_rounds is not None: conf['num_rounds'] = args.num_rounds if args.num_clients is not None: conf['num_clients'] = args.num_clients if args.batch_size is not None: conf['batch_size'] = args.batch_size if args.local_epochs is not None: conf['local_epochs'] = args.local_epochs if args.input_shape is not None: conf['input_shape'] = args.input_shape if args.num_classes is not None: conf['num_classes'] = args.num_classes if args.loss is not None: conf['loss'] = args.loss return conf def main(): # parser args = parse_arguments() # load config conf = load_config(args.config) conf = update_config(args, conf) # model architecture model = create_model(conf['input_shape'], conf['num_classes']) # compile model model.compile(loss='sparse_categorical_crossentropy', metrics=['accuracy']) # get model weights weights = model.get_weights() # Serialize ndarrays to "Parameters" parameters = fl.common.ndarrays_to_parameters(weights) # Server's strategy strategy = fl.server.strategy.FedAvg( min_available_clients=conf["num_clients"], on_fit_config_fn=get_on_fit_config_fn(conf["batch_size"], conf['local_epochs']), initial_parameters=parameters, ) # Start server fl.server.start_server( server_address=conf['server_address'], config=fl.server.ServerConfig(num_rounds=conf['num_rounds']), strategy=strategy, ) if __name__ == "__main__": main()