fed_train.py 24.6 KB
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608
import copy
import argparse
import time
import math
import numpy as np
import os
from collections import OrderedDict
import torch
import torch.optim as optim
import torch.nn as nn

import model
import utils
import dataset

# for google colab reload
import importlib
importlib.reload(model)
importlib.reload(utils)
importlib.reload(dataset)


## paramter

# shared
criterion = nn.CrossEntropyLoss()
C = 0.1
#

# prox
mu = 0.001
#

# time weight
twa_exp = 1.1
#

# dynamic weight
H = 0.5
P = 0.1
G = 0.1
R = 0.1
alpha, beta, gamma = 40.0/100.0, 40.0/100.0, 20.0/100.0
#

## end


def add_args(parser):
    parser.add_argument('--packet_num', type=int, default=1,
                        help='packet number used in training, 1 ~ 3')
    parser.add_argument('--fold_num', type=int, default=0, 
                        help='5-fold, 0 ~ 4')
    parser.add_argument('--batch_size', type=int, default=128,
                        help='input batch size for training')
    parser.add_argument('--lr', type=float, default=0.001,
                        help='learning rate')
    parser.add_argument('--n_nets', type=int, default=100,
                        help='number of workers in a distributed cluster')
    parser.add_argument('--comm_type', type=str, default='fedprox', 
                            help='type of communication, [fedavg, fedprox, fedtwa, feddw, edge]')  
    parser.add_argument('--comm_round', type=int, default=50, 
                            help='how many round of communications we shoud use')
    parser.add_argument('--weight_save_path', type=str, default='./weights', 
                            help='model weight save path')
    args = parser.parse_args(args=[])
    return args


def test_model(fed_model, args, testloader, device):
    fed_model.to(device)
    fed_model.eval()

    cnt = 0
    step_acc = 0.0
    with torch.no_grad():
        packet_state = torch.zeros(args.batch_size, model.STATE_DIM).to(device)
        for i, (inputs, labels) in enumerate(testloader):
            inputs, labels = inputs.to(device), labels.to(device)
            
            outputs, packet_state = fed_model(inputs, packet_state)
            packet_state = torch.autograd.Variable(packet_state, requires_grad=False)

            _, preds = torch.max(outputs, 1)

            cnt += inputs.shape[0]
            corr_sum = torch.sum(preds == labels.data)
            step_acc += corr_sum.double()
            if i % 200 == 0:
                print('test [%4d/%4d] acc: %.3f' % (i, len(testloader), (step_acc / cnt).item()))
                # break
    fed_accuracy = (step_acc / cnt).item()
    print('test acc', fed_accuracy)
    fed_model.to('cpu')
    fed_model.train()
    torch.save(fed_model.state_dict(), os.path.join(args.weight_save_path, '%s_%d_%.4f.pth' % (args.comm_type, cr, fed_accuracy)))


def start_fedavg(fed_model, args,
                          train_data_set,
                          data_idx_map,
                          net_data_count,
                          testloader,
                          edges,
                          device):
    print("start fed avg")

    num_edge = int(max(C * args.n_nets, 1))
    total_data_count = 0
    for _, data_count in net_data_count.items():
        total_data_count += data_count
    print("total data: %d" % total_data_count)

    for cr in range(1, args.comm_round + 1):
        print("Communication round : %d" % (cr))

        np.random.seed(cr)  # make sure for each comparison, select the same clients each round
        selected_edge = np.random.choice(args.n_nets, num_edge, replace=False)
        print("selected edge", selected_edge)

        for edge_progress, edge_index in enumerate(selected_edge):
            train_data_set.set_idx_map(data_idx_map[edge_index])
            sampler = dataset.BatchIntervalSampler(len(train_data_set), args.batch_size)
            train_loader = torch.utils.data.DataLoader(train_data_set, batch_size=args.batch_size, sampler=sampler,
                                                    shuffle=False, num_workers=2, drop_last=True)
            print("[%2d/%2d] edge: %d, data len: %d" % (edge_progress, len(selected_edge), edge_index, len(train_data_set)))

            edges[edge_index] = copy.deepcopy(fed_model)
            edges[edge_index].to(device)
            edges[edge_index].train()
            edge_opt = optim.Adam(params=edges[edge_index].parameters(), lr=args.lr)

            # train
            packet_state = torch.zeros(args.batch_size, model.STATE_DIM).to(device)
            for data_idx, (inputs, labels) in enumerate(train_loader):
                inputs, labels = inputs.to(device), labels.to(device)

                edge_pred, packet_state = edges[edge_index](inputs, packet_state)
                packet_state = torch.autograd.Variable(packet_state, requires_grad=False)

                edge_opt.zero_grad()

                edge_loss = criterion(edge_pred, labels)
                edge_loss.backward()

                edge_opt.step()
                edge_loss = edge_loss.item()
                if data_idx % 100 == 0:
                    print('[%4d] loss: %.3f' % (data_idx, edge_loss))
                    # break
            edges[edge_index].to('cpu')

        # cal weight using fed avg
        update_state = OrderedDict()
        for k, edge in enumerate(edges):
            local_state = edge.state_dict()
            for key in fed_model.state_dict().keys():
                if k == 0:
                    update_state[key] = local_state[key] * (net_data_count[k] / total_data_count)
                else:
                    update_state[key] += local_state[key] * (net_data_count[k] / total_data_count)
        
        fed_model.load_state_dict(update_state)
        if cr % 10 == 0:
            test_model(fed_model, args, testloader, device)


def start_fedprox(fed_model, args,
                          train_data_set,
                          data_idx_map,
                          testloader,
                          device):
    print("start fed prox")

    num_edge = int(max(C * args.n_nets, 1))
    fed_model.to(device)

    for cr in range(1, args.comm_round + 1):
        print("Communication round : %d" % (cr))
        edge_weight_dict = {}
        fed_weight_dict = {}
        for fed_name, fed_param in fed_model.named_parameters():
          edge_weight_dict[fed_name] = []
          fed_weight_dict[fed_name] = fed_param

        np.random.seed(cr)  # make sure for each comparison, select the same clients each round
        selected_edge = np.random.choice(args.n_nets, num_edge, replace=False)
        print("selected edge", selected_edge)

        for edge_progress, edge_index in enumerate(selected_edge):
            train_data_set.set_idx_map(data_idx_map[edge_index])
            sampler = dataset.BatchIntervalSampler(len(train_data_set), args.batch_size)
            train_loader = torch.utils.data.DataLoader(train_data_set, batch_size=args.batch_size, sampler=sampler,
                                                    shuffle=False, num_workers=2, drop_last=True)
            print("[%2d/%2d] edge: %d, data len: %d" % (edge_progress, len(selected_edge), edge_index, len(train_data_set)))

            edge_model = copy.deepcopy(fed_model)
            edge_model.to(device)
            edge_model.train()
            edge_opt = optim.Adam(params=edge_model.parameters(),lr=args.lr)
            # train
            packet_state = torch.zeros(args.batch_size, model.STATE_DIM).to(device)
            for data_idx, (inputs, labels) in enumerate(train_loader):
                inputs, labels = inputs.to(device), labels.to(device)

                edge_pred, packet_state = edge_model(inputs, packet_state)
                packet_state = torch.autograd.Variable(packet_state, requires_grad=False)

                edge_opt.zero_grad()

                edge_loss = criterion(edge_pred, labels)
                # prox term
                fed_prox_reg = 0.0
                for edge_name, edge_param in edge_model.named_parameters():
                    fed_prox_reg += ((mu / 2) * torch.norm((fed_weight_dict[edge_name] - edge_param))**2)
                edge_loss += fed_prox_reg
                
                edge_loss.backward()

                edge_opt.step()
                edge_loss = edge_loss.item()
                if data_idx % 100 == 0:
                    print('[%4d] loss: %.3f' % (data_idx, edge_loss))
                    # break

            edge_model.to('cpu')
            # save edge weight
            for edge_name, edge_param in edge_model.named_parameters():
                edge_weight_dict[edge_name].append(edge_param)

        fed_model.to('cpu')
        # cal weight, / number of edge
        for fed_name, fed_param in fed_model.named_parameters():
            fed_param.data.copy_( sum(weight / num_edge for weight in edge_weight_dict[fed_name]) )
        fed_model.to(device)

        if cr % 10 == 0:
            test_model(fed_model, args, testloader, device)
            fed_model.to(device)


def start_fedtwa(fed_model, args,
                          train_data_set,
                          data_idx_map,
                          net_data_count,
                          testloader,
                          edges,
                          device):
    # TEFL, without asynchronous model update
    print("start fed temporally weighted aggregation")

    time_stamp = [0 for worker in range(args.n_nets)]
    num_edge = int(max(C * args.n_nets, 1))
    total_data_count = 0
    for _, data_count in net_data_count.items():
        total_data_count += data_count
    print("total data: %d" % total_data_count)
    
    for cr in range(1, args.comm_round + 1):
        print("Communication round : %d" % (cr))
        
        np.random.seed(cr)  # make sure for each comparison, select the same clients each round
        selected_edge = np.random.choice(args.n_nets, num_edge, replace=False)
        print("selected edge", selected_edge)

        for edge_progress, edge_index in enumerate(selected_edge):
            time_stamp[edge_index] = cr
            train_data_set.set_idx_map(data_idx_map[edge_index])
            sampler = dataset.BatchIntervalSampler(len(train_data_set), args.batch_size)
            train_loader = torch.utils.data.DataLoader(train_data_set, batch_size=args.batch_size, sampler=sampler,
                                                    shuffle=False, num_workers=2, drop_last=True)
            print("[%2d/%2d] edge: %d, data len: %d" % (edge_progress, len(selected_edge), edge_index, len(train_data_set)))

            edges[edge_index] = copy.deepcopy(fed_model)
            edges[edge_index].to(device)
            edges[edge_index].train()
            edge_opt = optim.Adam(params=edges[edge_index].parameters(), lr=args.lr)

            # train
            packet_state = torch.zeros(args.batch_size, model.STATE_DIM).to(device)
            for data_idx, (inputs, labels) in enumerate(train_loader):
                inputs, labels = inputs.float().to(device), labels.long().to(device)

                edge_pred, packet_state = edges[edge_index](inputs, packet_state)
                packet_state = torch.autograd.Variable(packet_state, requires_grad=False)

                edge_opt.zero_grad()

                edge_loss = criterion(edge_pred, labels)
                edge_loss.backward()

                edge_opt.step()
                edge_loss = edge_loss.item()
                if data_idx % 100 == 0:
                    print('[%4d] loss: %.3f' % (data_idx, edge_loss))
                    # break
            edges[edge_index].to('cpu')

        # cal weight using time stamp
        # in paper, cr - time_stamp[k] used, but error is high
        update_state = OrderedDict()
        for k, edge in enumerate(edges):
            local_state = edge.state_dict()
            for key in fed_model.state_dict().keys():
                if k == 0:
                    update_state[key] = local_state[key] * (net_data_count[k] / total_data_count) * math.pow(twa_exp, -(cr -2 - time_stamp[k]))
                else:
                    update_state[key] += local_state[key] * (net_data_count[k] / total_data_count) * math.pow(twa_exp, -(cr -2 - time_stamp[k]))

        fed_model.load_state_dict(update_state)
        if cr % 10 == 0:
            test_model(fed_model, args, testloader, device)


def start_feddw(fed_model, args,
                          train_data_set,
                          data_idx_map,
                          net_data_count,
                          testloader,
                          local_test_loader,
                          edges,
                          device):
    print("start fed Node-aware Dynamic Weighting")

    worker_selected_frequency = [0 for worker in range(args.n_nets)]
    num_edge = int(max(G * args.n_nets, 1))
    
    # cal data weight for selecting participants
    total_data_count = 0
    for _, data_count in net_data_count.items():
        total_data_count += data_count
    print("total data: %d" % total_data_count)

    total_data_weight = 0.0
    net_weight_dict = {}
    for net_key, data_count in net_data_count.items():
        net_data_count[net_key] = data_count / total_data_count
        net_weight_dict[net_key] = total_data_count / data_count
        total_data_weight += net_weight_dict[net_key]
    
    for net_key, data_count in net_weight_dict.items():
        net_weight_dict[net_key] = net_weight_dict[net_key] / total_data_weight
    # end

    worker_local_accuracy = [0 for worker in range(args.n_nets)]
    
    for cr in range(1, args.comm_round + 1):
        print("Communication round : %d" % (cr))

        # select participants
        candidates = []
        sum_frequency = sum(worker_selected_frequency)
        if sum_frequency == 0:
            sum_frequency = 1
        for worker_index in range(args.n_nets):
          candidates.append((H * worker_selected_frequency[worker_index] / sum_frequency + (1 - H) * net_weight_dict[worker_index], worker_index))
        candidates = sorted(candidates)[:int(R * args.n_nets)]
        candidates = [temp[1] for temp in candidates]

        np.random.seed(cr)
        selected_edge = np.random.choice(candidates, num_edge, replace=False)
        # end select

        # weighted frequency
        avg_selected_frequency = sum(worker_selected_frequency) / len(worker_selected_frequency)
        weighted_frequency = [P * (avg_selected_frequency - worker_frequency) for worker_frequency in worker_selected_frequency]
        frequency_prime = min(weighted_frequency)
        weighted_frequency = [frequency + frequency_prime + 1 for frequency in weighted_frequency]
        # end weigthed

        print("selected edge", selected_edge)
        for edge_progress, edge_index in enumerate(selected_edge):
            worker_selected_frequency[edge_index] += 1
            train_data_set.set_idx_map(data_idx_map[edge_index])
            sampler = dataset.BatchIntervalSampler(len(train_data_set), args.batch_size)
            train_loader = torch.utils.data.DataLoader(train_data_set, batch_size=args.batch_size, sampler=sampler,
                                                    shuffle=False, num_workers=2, drop_last=True)
            print("[%2d/%2d] edge: %d, data len: %d" % (edge_progress, len(selected_edge), edge_index, len(train_data_set)))

            edges[edge_index] = copy.deepcopy(fed_model)
            edges[edge_index].to(device)
            edges[edge_index].train()
            edge_opt = optim.Adam(params=edges[edge_index].parameters(), lr=args.lr)

            # train
            packet_state = torch.zeros(args.batch_size, model.STATE_DIM).to(device)
            for data_idx, (inputs, labels) in enumerate(train_loader):
                inputs, labels = inputs.float().to(device), labels.long().to(device)

                edge_pred, packet_state = edges[edge_index](inputs, packet_state)
                packet_state = torch.autograd.Variable(packet_state, requires_grad=False)

                edge_opt.zero_grad()

                edge_loss = criterion(edge_pred, labels)
                edge_loss.backward()

                edge_opt.step()
                edge_loss = edge_loss.item()
                if data_idx % 100 == 0:
                    print('[%4d] loss: %.3f' % (data_idx, edge_loss))
                    # break

            # get edge accuracy using subset of testset
            edges[edge_index].eval()
            print("[%2d/%2d] edge: %d, cal local accuracy" % (edge_progress, len(selected_edge), edge_index))
            cnt = 0
            step_acc = 0.0
            with torch.no_grad():
                packet_state = torch.zeros(args.batch_size, model.STATE_DIM).to(device)
                for inputs, labels in local_test_loader:
                  inputs, labels = inputs.float().to(device), labels.long().to(device)

                  edge_pred, packet_state = edges[edge_index](inputs, packet_state)
                  packet_state = torch.autograd.Variable(packet_state, requires_grad=False)

                  _, preds = torch.max(edge_pred, 1)

                  loss = criterion(edge_pred, labels)
                  cnt += inputs.shape[0]

                  corr_sum = torch.sum(preds == labels.data)
                  step_acc += corr_sum.double()
                  # break

            worker_local_accuracy[edge_index] = (step_acc / cnt).item()
            print('edge local accuracy', worker_local_accuracy[edge_index])
            edges[edge_index].to('cpu')

        # cal weight dynamically
        sum_accuracy = sum(worker_local_accuracy)
        sum_weighted_frequency = sum(weighted_frequency)
        update_state = OrderedDict()
        for k, edge in enumerate(edges):
            local_state = edge.state_dict()
            for key in fed_model.state_dict().keys():
                if k == 0:
                    update_state[key] = local_state[key] \
                    * (net_data_count[k] * alpha \
                    + worker_local_accuracy[k] / sum_accuracy * beta \
                    + weighted_frequency[k] / sum_weighted_frequency * gamma)
                else:
                    update_state[key] += local_state[key] \
                    * (net_data_count[k] * alpha \
                    + worker_local_accuracy[k] / sum_accuracy * beta \
                    + weighted_frequency[k] / sum_weighted_frequency * gamma)

        fed_model.load_state_dict(update_state)
        if cr % 10 == 0:
          test_model(fed_model, args, testloader, device)


def start_only_edge(args,
                          train_data_set,
                          data_idx_map,
                          testloader,
                          edges,
                          device):
    print("start only edge")
    total_epoch = int(args.comm_round * C)

    for cr in range(1, total_epoch + 1):
        print("Edge round : %d" % (cr))
        edge_accuracy_list = []

        for edge_index, edge_model in enumerate(edges):
            train_data_set.set_idx_map(data_idx_map[edge_index])
            sampler = dataset.BatchIntervalSampler(len(train_data_set), args.batch_size)
            train_loader = torch.utils.data.DataLoader(train_data_set, batch_size=args.batch_size, sampler=sampler,
                                                    shuffle=False, num_workers=2, drop_last=True)
            print("edge[%2d/%2d] data len: %d" % (edge_index, len(edges), len(train_data_set)))

            edge_model.to(device)
            edge_model.train()
            edge_opt = optim.Adam(params=edge_model.parameters(),lr=args.lr)

            # train
            packet_state = torch.zeros(args.batch_size, model.STATE_DIM).to(device)
            for data_idx, (inputs, labels) in enumerate(train_loader):
                inputs, labels = inputs.float().to(device), labels.long().to(device)

                edge_pred, packet_state = edge_model(inputs, packet_state)
                packet_state = torch.autograd.Variable(packet_state, requires_grad=False)

                edge_opt.zero_grad()

                edge_loss = criterion(edge_pred, labels)
                edge_loss.backward()

                edge_opt.step()
                edge_loss = edge_loss.item()
                if data_idx % 100 == 0:
                    print('[%4d] loss: %.3f' % (data_idx, edge_loss))
                    # break

            # test
            # if cr < 4:
            #     continue
            edge_model.eval()
            total_loss = 0.0
            cnt = 0
            step_acc = 0.0
            with torch.no_grad():
                packet_state = torch.zeros(args.batch_size, model.STATE_DIM).to(device)
                for i, (inputs, labels) in enumerate(testloader):
                    inputs, labels = inputs.float().to(device), labels.long().to(device)
                    
                    outputs, packet_state = edge_model(inputs, packet_state)
                    packet_state = torch.autograd.Variable(packet_state, requires_grad=False)

                    _, preds = torch.max(outputs, 1)

                    loss = criterion(outputs, labels)
                    cnt += inputs.shape[0]

                    corr_sum = torch.sum(preds == labels.data)
                    step_acc += corr_sum.double()
                    running_loss = loss.item() * inputs.shape[0]
                    total_loss += running_loss
                    if i % 200 == 0:
                      print('test [%4d] loss: %.3f' % (i, loss.item()))
                      # break
            edge_accuracy = (step_acc / cnt).item()
            edge_accuracy_list.append(edge_accuracy)
            print("edge[%2d/%2d] acc: %.4f" % (edge_index, len(edges), edge_accuracy))
            edge_model.to('cpu')

        # if cr < 4:
        #     continue
        edge_accuracy_avg = sum(edge_accuracy_list) / len(edge_accuracy_list)
        torch.save(edges[0].state_dict(), os.path.join(weight_path, 'edge_%d_%.4f.pth' % (cr, edge_accuracy_avg)))



def start_train():
    device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
    print('device:', device)

    args = add_args(argparse.ArgumentParser())

    # make weight folder
    os.makedirs(args.weight_save_path, exist_ok=True)

    # for reproductivity
    seed = 0
    np.random.seed(seed)
    torch.manual_seed(seed)

    print("Loading data...")
    train_data_set, data_idx_map, net_data_count, test_data_set = dataset.GetCanDataset(args.n_nets, args.fold_num, args.packet_num, "./dataset/Mixed_dataset.csv", "./dataset/Mixed_dataset_1.txt")

    sampler = dataset.BatchIntervalSampler(len(test_data_set), args.batch_size)
    testloader = torch.utils.data.DataLoader(test_data_set, batch_size=args.batch_size, sampler=sampler,
                                            shuffle=False, num_workers=2, drop_last=True)

    fed_model = model.OneNet(args.packet_num)
    edges = [model.OneNet(args.packet_num) for _ in range(args.n_nets)]

    if args.comm_type == "fedavg":
        start_fedavg(fed_model, args,
                            train_data_set,
                            data_idx_map,
                            net_data_count,
                            testloader,
                            edges,
                            device)
    elif args.comm_type == "fedprox":
        start_fedprox(fed_model, args,
                            train_data_set,
                            data_idx_map,
                            testloader,
                            device)
    elif args.comm_type == "fedtwa":
        start_fedtwa(fed_model, args,
                            train_data_set,
                            data_idx_map,
                            net_data_count,
                            testloader,
                            edges,
                            device)
    elif args.comm_type == "feddw":
        local_test_set = copy.deepcopy(test_data_set)
        # in paper, mnist train 60,000 / test 10,000 / 1,000 - 10%
        # CAN train ~ 1,400,000 / test 300,000 / for speed 15,000 - 5%
        local_test_idx = [idx for idx in range(0, len(local_test_set) // 20)]
        local_test_set.set_idx_map(local_test_idx)
        sampler = dataset.BatchIntervalSampler(len(local_test_set), args.batch_size)
        local_test_loader = torch.utils.data.DataLoader(local_test_set, batch_size=args.batch_size, sampler=sampler,
                                                shuffle=False, num_workers=2, drop_last=True)

        start_feddw(fed_model, args,
                            train_data_set,
                            data_idx_map,
                            net_data_count,
                            testloader,
                            local_test_loader,
                            edges,
                            device)
    elif args.comm_type == "edge":
        start_only_edge(args,
                            train_data_set,
                            data_idx_map,
                            testloader,
                            edges,
                            device)

if __name__ == "__main__":
    start_train()