-
Notifications
You must be signed in to change notification settings - Fork 3
/
Copy pathFeSViBS.py
181 lines (147 loc) · 8.74 KB
/
FeSViBS.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
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
import os
import numpy as np
import models
import random
from dataset import skinCancer, bloodmnisit, isic2019
from utils import weight_dec_global, weight_vec
import argparse
import torch as torch
from torch import nn
def fesvibs(
dataset_name, lr, batch_size, Epochs, input_size, num_workers, save_every_epochs,
model_name, pretrained, opt_name, seed, base_dir, root_dir, csv_file_path, num_clients, DP,
epsilon, delta, resnet_dropout, initial_block, final_block, fesvibs_arg, local_round
):
torch.manual_seed(seed)
random.seed(seed)
np.random.seed(seed)
if fesvibs_arg:
method_flag = 'FeSViBS'
else:
method_flag = 'SViBS'
if torch.cuda.is_available():
device = 'cuda'
else:
device = 'cpu'
if DP:
std = np.sqrt(2 * np.math.log(1.25/delta)) / epsilon
mean=0
dir_name = f"{model_name}_{lr}lr_{dataset_name}_{num_clients}Clients_{initial_block}to{final_block}Blocks_{batch_size}Batch__{epsilon,delta}DP_{method_flag}"
else:
mean = 0
std = 0
dir_name = f"{model_name}_{lr}lr_{dataset_name}_{num_clients}Clients_{initial_block}to{final_block}Blocks_{batch_size}Batch_{method_flag}"
save_dir = f'{dir_name}'
os.mkdir(save_dir)
print(f"Logging to: {dir_name}")
print('Getting the Dataset and Dataloader!')
if dataset_name == 'HAM':
num_classes = 7
_, _, traindataset, testdataset = skinCancer(input_size= input_size, batch_size = batch_size, base_dir= base_dir, num_workers=num_workers)
num_channels = 3
elif dataset_name == 'bloodmnist':
num_classes = 8
_, _, traindataset, testdataset = bloodmnisit(input_size= input_size, batch_size = batch_size, download= True, num_workers=num_workers)
num_channels = 3
elif dataset_name == 'isic2019':
num_classes = 8
DATALOADERS, _, _, _, _, test_loader = isic2019(input_size= input_size, batch_size = batch_size, root_dir=root_dir, csv_file_path=csv_file_path, num_workers=num_workers)
num_channels = 3
criterion = nn.CrossEntropyLoss()
fesvibs_network = models.FeSVBiS(
ViT_name= model_name, num_classes= num_classes,
num_clients = num_clients, in_channels = num_channels,
ViT_pretrained= pretrained,
initial_block= initial_block, final_block= final_block,
resnet_dropout= resnet_dropout, DP=DP, mean= mean, std= std
).to(device)
Split = models.SplitFeSViBS(
num_clients=num_clients, device = device, network = fesvibs_network,
criterion = criterion, base_dir=save_dir,
initial_block= initial_block, final_block= final_block,
)
if dataset_name != 'isic2019':
print('Distribute Images Among Clients')
Split.distribute_images(dataset_name=dataset_name, train_data= traindataset,test_data= testdataset ,batch_size = batch_size)
else:
Split.CLIENTS_DATALOADERS = DATALOADERS
Split.testloader = test_loader
Split.set_optimizer(opt_name, lr = lr)
Split.init_logs()
print('Start Training! \n')
for r in range(Epochs):
print(f"Round {r+1} / {Epochs}")
agg_weights = None
for client_i in range(num_clients):
weight_dict = Split.train_round(client_i)
if client_i == 0:
agg_weights = weight_dict
else:
agg_weights['blocks'] += weight_dict['blocks']
agg_weights['cls'] += weight_dict['cls']
agg_weights['pos_embed'] += weight_dict['pos_embed']
agg_weights['blocks'] /= num_clients
agg_weights['cls'] /= num_clients
agg_weights['pos_embed'] /= num_clients
Split.network.vit.blocks = weight_dec_global(
Split.network.vit.blocks,
agg_weights['blocks'].to(device)
)
Split.network.vit.cls_token.data = agg_weights['cls'].to(device) + 0.0
Split.network.vit.pos_embed.data = agg_weights['pos_embed'].to(device) + 0.0
if fesvibs_arg and ((r+1) % local_round == 0 and r!= 0):
print('========================== \t \t Federation \t \t ==========================')
tails_weights = []
head_weights = []
for head, tail in zip(Split.network.resnet50_clients, Split.network.mlp_clients_tail):
head_weights.append(weight_vec(head).detach().cpu())
tails_weights.append(weight_vec(tail).detach().cpu())
mean_avg_tail = torch.mean(torch.stack(tails_weights), axis = 0)
mean_avg_head = torch.mean(torch.stack(head_weights), axis = 0)
for i in range(num_clients):
Split.network.mlp_clients_tail[i] = weight_dec_global(Split.network.mlp_clients_tail[i],
mean_avg_tail.to(device))
Split.network.resnet50_clients[i] = weight_dec_global(Split.network.resnet50_clients[i],
mean_avg_head.to(device))
for client_i in range(num_clients):
Split.eval_round(client_i)
print('---------')
if (r+1) % save_every_epochs == 0 and r != 0:
Split.save_pickles(save_dir)
print('============================================')
if __name__ == "__main__":
parser = argparse.ArgumentParser(description='Run Centralized Experiments')
parser.add_argument('--dataset_name', type=str, choices=['HAM', 'bloodmnist', 'isic2019'], help='Dataset Name')
parser.add_argument('--input_size', type=int, default= 224, help='Input size --> (input_size, input_size), default : 224')
parser.add_argument('--local_round', type=int, default= 2, help='Local round before federation in FeSViBS, default : 2')
parser.add_argument('--num_workers', type=int, default= 8, help='Number of workers for dataloaders, default : 8')
parser.add_argument('--initial_block', type=int, default= 1, help='Initial Block, default : 1')
parser.add_argument('--final_block', type=int, default= 6, help='Final Block, default : 6')
parser.add_argument('--num_clients', type=int, default= 6, help='Number of Clients, default : 6')
parser.add_argument('--model_name', type=str, default= 'vit_base_r50_s16_224', help='Model name from timm library, default: vit_base_r50_s16_224')
parser.add_argument('--pretrained', type=bool, default= False, help='Pretrained weights flag, default: False')
parser.add_argument('--fesvibs_arg', type=bool, default= False, help='Flag to indicate whether SViBS or FeSViBS, default: False')
parser.add_argument('--batch_size', type=int, default= 32, help='Batch size, default : 32')
parser.add_argument('--Epochs', type=int, default= 200, help='Number of Epochs, default : 200')
parser.add_argument('--opt_name', type=str, choices=['Adam'], default = 'Adam', help='Optimizer name, only ADAM optimizer is available')
parser.add_argument('--lr', type=float, default= 1e-4, help='Learning rate, default : 1e-4')
parser.add_argument('--save_every_epochs', type=int, default= 10, help='Save metrics every this number of epochs, default: 10')
parser.add_argument('--seed', type=int, default= 105, help='Seed, default: 105')
parser.add_argument('--base_dir', type=str, default= None, help='')
parser.add_argument('--root_dir', type=str, default= None, help='')
parser.add_argument('--csv_file_path', type=str, default=None, help='')
parser.add_argument('--DP', type=bool, default= False, help='Differential Privacy , default: False')
parser.add_argument('--epsilon', type=float, default= 0, help='Epsilon Value for differential privacy')
parser.add_argument('--delta', type=float, default= 0.00001, help='Delta Value for differential privacy')
parser.add_argument('--resnet_dropout', type=float, default= 0.5, help='ResNet Dropout, Default: 0.5')
args = parser.parse_args()
fesvibs(
dataset_name = args.dataset_name, input_size= args.input_size,
num_workers= args.num_workers, model_name= args.model_name,
pretrained= args.pretrained, batch_size= args.batch_size,
Epochs= args.Epochs, opt_name= args.opt_name, lr= args.lr,
save_every_epochs= args.save_every_epochs, seed= args.seed,
base_dir= args.base_dir, root_dir= args.root_dir, csv_file_path= args.csv_file_path, num_clients = args.num_clients,
DP = args.DP, epsilon = args.epsilon, delta = args.delta, initial_block= args.initial_block, final_block=args.final_block,
resnet_dropout = args.resnet_dropout, fesvibs_arg = args.fesvibs_arg, local_round = args.local_round
)