-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathmain.py
165 lines (137 loc) · 7.42 KB
/
main.py
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
#!/usr/bin/env python
from FLAlgorithms.servers.myserver import ServerMain
from FLAlgorithms.trainmodel.models import *
import argparse
#from utils.data_loader import *
from utils.model_utils import read_data, load_data
from utils.data_utils import split_data
import torch
import warnings
warnings.filterwarnings("ignore")
import os
import sys
sys.stdout.flush()
torch.manual_seed(1)
# torch.backends.cudnn.enabled = True
# torch.backends.cudnn.benchmark = True
def main(dataset, class_num, split_method, split_para, split_num, algorithm, modelname, loss,
batch_size, learning_rate, num_glob_iters, layer, fea_percent, fea_dim, is_interpolated,
local_epochs, optimizer, client_num, personal_learning_rate, device, beta,mode, seed):
post_fix_str = '{}_{}_{}_loss_{}_epoch_{}_{}_client_{}_split_{}_{}'.format(algorithm, modelname, dataset,loss, local_epochs,num_glob_iters, split_num, split_method, split_para)
model_path = []
train_data, test_data = split_data(dataset, class_num, split_num, split_method, split_para)
data = load_data(train_data, test_data) #for testing
if dataset == 'Cifar10':
inputdim = 32*32
if modelname == 'VGG':
model = VGG16(class_num).to(device), modelname
elif modelname == 'RESNET':
model = ResNet18(class_num).to(device), modelname
elif modelname == 'VIT':
model = VisionTransformer(class_num).to(device), modelname
elif modelname == 'MOBNET':
model = MobileNetV2(class_num).to(device), modelname
elif dataset == 'UrbanSound':
model = AudioClassifier(class_num).to(device), modelname
elif dataset == 'UCI_HAR':
model = HARNet(class_num).to(device), modelname
print(model[0])
print('#######', device, '#######' )
pytorch_total_params = sum(p.numel() for p in model[0].parameters())
print('number of parameters:', pytorch_total_params)
for i in range(1): #repeat times
if mode == 'personalisation': # for personlisaton test
model_path = os.path.join("models", 'test')
#checkpoint_path = os.path.join(model_path, "server_FedAvg_MOBNET_Cifar10_loss_CE_epoch_10_100_client_1000_split_quantity_10.0.pt")
#checkpoint_path = os.path.join(model_path, "server_FedAvg_MOBNET_Cifar10_loss_CE_epoch_10_100_client_10_split_quantity_10.0.pt")
checkpoint_path = os.path.join(model_path, "server_FLea_MOBNET_Cifar10_loss_CE_CE_DeC_KL_epoch_10_100_client_100_split_quantity_3.0.pt")
checkpoint= torch.load(checkpoint_path)
print('Load model checkpoint from name succuessfully!')
model = checkpoint, modelname
server = ServerMain(dataset, data, algorithm, model, batch_size, learning_rate, num_glob_iters, layer, fea_percent, fea_dim, is_interpolated,
local_epochs, optimizer, client_num, i, device, personal_learning_rate,
output_dim=class_num, post_fix_str=post_fix_str, loss=loss, beta=beta, seed=seed)
del data
del train_data, test_data
server.personal_train()
else: #Training and test
server = ServerMain(dataset, data, algorithm, model, batch_size, learning_rate, num_glob_iters, layer, fea_percent, fea_dim, is_interpolated,
local_epochs, optimizer, client_num, i, device, personal_learning_rate,
output_dim=class_num, post_fix_str=post_fix_str, loss=loss, beta=beta, seed=seed)
del data
del train_data, test_data
server.train()
def run():
parser = argparse.ArgumentParser()
parser.add_argument("--dataset", type=str, default="CIFAR10")
parser.add_argument("--class_num", type=int, default=10)
parser.add_argument("--split_method", type=str, default='quantity') #distribution
parser.add_argument("--split_para", type=float, default=5.0) #for split it is #preesnt class, for lda, it is \alpha
parser.add_argument("--split_num", type=int, default=10)
parser.add_argument("--client_num", type=int, default=10) # total clients, should be <= split_num
parser.add_argument("--algorithm", type=str, default="FedAvg")
parser.add_argument("--loss", type=str, default="CE")
parser.add_argument("--layer", type=int, default=11)
parser.add_argument("--fea_percent", type=float, default=0.1)
parser.add_argument("--fea_dim", type=int, default=64)
parser.add_argument("--is_interpolated", type=bool, default=False)
parser.add_argument("--batch_size", type=int, default=64)
parser.add_argument("--learning_rate", type=float, default=0.001) #not used
parser.add_argument("--personal_learning_rate", type=float, default=0.001)
parser.add_argument("--num_global_iters", type=int, default=500)
parser.add_argument("--local_epochs", type=int, default=10)
parser.add_argument("--optimizer", type=str, default="Adam")
parser.add_argument("--beta", type=float, default=1.0)
parser.add_argument("--seed", type=int, default=0, help="seed for client selection")
parser.add_argument("--mode", type=str, default="training")
parser.add_argument("--modelname", type=str, default="MOBNET")
args = parser.parse_args()
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
print("=" * 80)
print("Summary of training process:")
print("Dataset : {}".format(args.dataset))
print("Batch size : {}".format(args.batch_size))
print("Learing rate : {}".format(args.personal_learning_rate))
print("Number of total clients: {}".format(args.split_num))
print("Split method : {}".format(args.split_method))
print("Split parameter : {}".format(args.split_para))
print("Clients per round : {}".format(args.client_num))
print("Number of global rounds: {}".format(args.num_global_iters))
print("Number of local rounds : {}".format(args.local_epochs))
print("Feature from layer : {}".format(args.layer))
print("Feature percentage : {}".format(args.fea_percent))
print("Is interplolated : {}".format(args.is_interpolated))
print("Local training loss : {}".format(args.loss))
print("Loss of beta : {}".format(args.beta))
print("Algorithm : {}".format(args.algorithm))
print("Modelname : {}".format(args.modelname))
print("Mode : {}".format(args.mode))
print("Seed : {}".format(args.seed))
print("=" * 80)
return main(
dataset=args.dataset,
class_num=args.class_num,
split_method=args.split_method,
split_para=args.split_para,
split_num=args.split_num,
algorithm=args.algorithm,
modelname=args.modelname,
loss=args.loss,
batch_size=args.batch_size,
learning_rate=args.learning_rate,
num_glob_iters=args.num_global_iters,
layer=args.layer,
fea_percent=args.fea_percent,
fea_dim=args.fea_dim,
is_interpolated=args.is_interpolated,
local_epochs=args.local_epochs,
optimizer=args.optimizer,
client_num=args.client_num,
personal_learning_rate=args.personal_learning_rate,
device=device,
beta=args.beta,
mode=args.mode,
seed=args.seed
)
if __name__ == "__main__":
run()