-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathutils.py
185 lines (150 loc) · 7.26 KB
/
utils.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
182
183
184
185
import torch
import logging
import random
from models import *
from partition_data import *
import torchvision.models as models
def init_nets_proto(args,n_parties):
nets = {net_i: None for net_i in range(n_parties)}
if args.dataset in ["cifar10"]:
n_classes = 10
elif args.dataset in ["cifar100"]:
n_classes = 100
elif args.dataset in ["femnist_by_writer"]:
n_classes = 62
for net_i in range(n_parties):
if args.dataset in ("cifar10", "cifar100"):
net = model_cifar_proto(args, n_classes ,256)
elif args.dataset == "femnist_by_writer":
net = SimpleFemnist()
nets[net_i] = net
return nets
def init_nets(args,n_parties):
nets = {net_i: None for net_i in range(n_parties)}
if args.dataset in ["cifar10","mnist","fmnist","SVHN"]:
n_classes = 10
elif args.dataset in ["cifar100"]:
n_classes = 100
elif args.dataset in ["femnist_by_writer"]:
n_classes = 62
for net_i in range(n_parties):
if args.dataset in ("cifar10", "cifar100"):
if args.dataset=="cifar100":
net=Wide_ResNet(num_classes=n_classes)
else:
net = model_cifar(args, n_classes)
elif args.dataset in ["mnist","fmnist"]:
args.model = "resnet18"
net = model_cifar(args, n_classes)
elif args.dataset in ["SVHN"]:
args.model = "resnet18"
net=model_cifar(args, n_classes)
nets[net_i] = net
return nets
def init_dataloader(args, net_dataidx_map=None):
print("starting init dataloader")
train_dl_local_set = []
train_ds_local_set = []
if args.dataset in ["femnist_by_writer","synthetic","shakespeare"]:
for i in range(args.n_parties):
train_dl_local, test_dl_local, train_ds_local, test_ds_local = get_dataloader(args, identity=i)
train_dl_local_set.append(train_dl_local)
train_ds_local_set.append(train_ds_local)
print(len(train_dl_local_set))
elif args.dataset in ["fmnist","mnist","cifar10","cifar10l","cifar100","tiny-imagenet","SVHN"]:
for i in range(args.n_parties):
if net_dataidx_map==None:
dataidxs=None
else:
dataidxs = net_dataidx_map[i]
train_dl_local, test_dl_local, train_ds_local, test_ds_local = get_dataloader(args,dataidxs=dataidxs)
train_dl_local_set.append(train_dl_local)
train_ds_local_set.append(train_ds_local)
print(len(train_dl_local_set))
elif args.dataset in ["PACS"]:
train_X,train_Y,test_X,test_Y=build_PACS()
for i in range(args.n_parties):
train_ds=PACS_s3(client_id=i,X=train_X,Y=train_Y)
train_dl_local = DataLoader(dataset=train_ds, batch_size=args.train_batchsize, drop_last=True, shuffle=True,
prefetch_factor=4, persistent_workers=True, num_workers=4)
train_dl_local_set.append(train_dl_local)
train_ds_local_set.append(train_ds)
test_ds_local=PACS_s3(client_id=-1,X=test_X,Y=test_Y)
test_dl_local = DataLoader(dataset=test_ds_local, batch_size=args.train_batchsize, drop_last=True, shuffle=True,
prefetch_factor=4, persistent_workers=True, num_workers=4)
elif args.dataset in ["officehome"]:
train_X, train_Y, test_X, test_Y = build_Officehome()
for i in range(args.n_parties):
train_ds = Officehome_s3(client_id=i, X=train_X, Y=train_Y)
train_dl_local = DataLoader(dataset=train_ds, batch_size=args.train_batchsize, drop_last=True, shuffle=True,
prefetch_factor=4, persistent_workers=True, num_workers=4)
train_dl_local_set.append(train_dl_local)
train_ds_local_set.append(train_ds)
test_ds_local = Officehome_s3(client_id=-1, X=test_X, Y=test_Y)
test_dl_local = DataLoader(dataset=test_ds_local, batch_size=args.train_batchsize, drop_last=True, shuffle=True,
prefetch_factor=4, persistent_workers=True, num_workers=4)
elif args.dataset in ["ISIC"]:
for i in range(args.n_parties):
train_ds = ISIC2019(identity=i)
train_dl_local = DataLoader(dataset=train_ds, batch_size=args.train_batchsize, drop_last=True, shuffle=True,
prefetch_factor=4, persistent_workers=True, num_workers=4)
train_dl_local_set.append(train_dl_local)
train_ds_local_set.append(train_ds)
test_ds_local = ISIC2019(identity=-1, train=False)
test_dl_local = DataLoader(dataset=test_ds_local, batch_size=args.train_batchsize, drop_last=True, shuffle=True,
prefetch_factor=4, persistent_workers=True, num_workers=4)
print("finishing init dataloader")
return train_dl_local_set, test_dl_local,train_ds_local_set,test_ds_local
def mkdirs(dirpath):
try:
os.makedirs(dirpath)
except Exception as _:
pass
def compute_accuracy(model, dataloader):
was_training = False
if model.training:
model.eval()
was_training = True
true_labels_list, pred_labels_list = np.array([]), np.array([])
correct, total = 0, 0
criterion = nn.CrossEntropyLoss().cuda()
loss_collector = []
with torch.no_grad():
for batch_idx, (x, target) in enumerate(dataloader):
x, target = x.cuda(), target.to(dtype=torch.int64).cuda()
_,_,out = model(x)
loss = criterion(out, target)
_, pred_label = torch.max(out.data, 1)
loss_collector.append(loss.item())
total += x.data.size()[0]
correct += (pred_label == target.data).sum().item()
pred_labels_list = np.append(pred_labels_list, pred_label.cpu().numpy())
true_labels_list = np.append(true_labels_list, target.data.cpu().numpy())
avg_loss = sum(loss_collector) / len(loss_collector)
if was_training:
model.train()
return correct / float(total), avg_loss
def compute_accuracy_proto(model, dataloader):
was_training = False
if model.training:
model.eval()
was_training = True
true_labels_list, pred_labels_list = np.array([]), np.array([])
correct, total = 0, 0
loss_collector = []
with torch.no_grad():
for batch_idx, (x, target) in enumerate(dataloader):
x, target = x.cuda(), target.to(dtype=torch.int64).cuda()
_,_,out = model(x)
out = torch.log(out)
loss = F.nll_loss(out, target)
_, pred_label = torch.max(out.data, 1)
loss_collector.append(loss.item())
total += x.data.size()[0]
correct += (pred_label == target.data).sum().item()
pred_labels_list = np.append(pred_labels_list, pred_label.cpu().numpy())
true_labels_list = np.append(true_labels_list, target.data.cpu().numpy())
avg_loss = sum(loss_collector) / len(loss_collector)
if was_training:
model.train()
return correct / float(total), avg_loss