Ejemplo n.º 1
0
CLIENT_ID = 1

if __name__ == "__main__":
    # CLIENT_ID = int(sys.argv[1])

    dataset_path = os.path.join(os.path.abspath("../"), "cifa10_demo",
                                "cifa10", "train_dataset_dir",
                                "train_dataset_{}".format(CLIENT_ID))

    dataset = torch.load(dataset_path)

    client = FLClient()
    pfl_models = client.get_remote_pfl_models()

    for pfl_model in pfl_models:
        optimizer = torch.optim.SGD(pfl_model.get_model().parameters(),
                                    lr=0.001,
                                    momentum=0.9)
        train_strategy = TrainStrategy(optimizer=optimizer,
                                       batch_size=32,
                                       loss_function=LossStrategy.NLL_LOSS)
        pfl_model.set_train_strategy(train_strategy)

    TrainerController(work_mode=WorkModeStrategy.WORKMODE_STANDALONE,
                      models=pfl_models,
                      data=dataset,
                      client_id=CLIENT_ID,
                      curve=True,
                      concurrent_num=3).start()
Ejemplo n.º 2
0
    train_ids = ['4_10', '4_11', '5_10', '5_11', '7_10', '7_11']
    # test_ids = ['5', '21', '15', '30']

    train_set = ISPRS_dataset(train_ids, cache=CACHE)

    client = FLClient()
    gfl_models = client.get_remote_gfl_models()

    for gfl_model in gfl_models:
        optimizer = torch.optim.SGD(gfl_model.get_model().parameters(),
                                    lr=0.01,
                                    momentum=0.9,
                                    weight_decay=0.0005)
        scheduler = torch.optim.lr_scheduler.MultiStepLR(optimizer,
                                                         [25, 35, 45],
                                                         gamma=0.1)
        train_strategy = TrainStrategy(optimizer=optimizer,
                                       scheduler=scheduler,
                                       batch_size=10,
                                       loss_function=CrossEntropy2d,
                                       test_function=test,
                                       accuracy_function=accuracy)
        gfl_model.set_train_strategy(train_strategy)

    TrainerController(work_mode=WorkModeStrategy.WORKMODE_STANDALONE,
                      models=gfl_models,
                      data=train_set,
                      client_id=CLIENT_ID,
                      curve=True,
                      local_epoch=3,
                      concurrent_num=3).start()