forked from jahanxb/flcode
-
Notifications
You must be signed in to change notification settings - Fork 0
/
main_fed_old-svhn.py
328 lines (273 loc) · 13.1 KB
/
main_fed_old-svhn.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
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
# #!/usr/bin/env python
# # -*- coding: utf-8 -*-
# # Python version: 3.6
# import copy
# import numpy as np
# import time, math
# import torch
# from utils.data_utils import data_setup, DatasetSplit
# from utils.model_utils import *
# from utils.aggregation import *
# from options import call_parser
# from models.Update import LocalUpdate
# from models.test import test_img
# from torch.utils.data import DataLoader
# # from utils.rdp_accountant import compute_rdp, get_privacy_spent
# import warnings
# warnings.filterwarnings("ignore")
# torch.cuda.is_available()
# if __name__ == '__main__':
# ################################### hyperparameter setup ########################################
# args = call_parser()
# torch.manual_seed(args.seed+args.repeat)
# torch.cuda.manual_seed(args.seed+args.repeat)
# np.random.seed(args.seed+args.repeat)
# args, dataset_train, dataset_test, dict_users = data_setup(args)
# print("{:<50}".format("=" * 15 + " data setup " + "=" * 50)[0:60])
# print(
# 'length of dataset:{}'.format(len(dataset_train) + len(dataset_test)))
# print('num. of training data:{}'.format(len(dataset_train)))
# print('num. of testing data:{}'.format(len(dataset_test)))
# print('num. of classes:{}'.format(args.num_classes))
# print('num. of users:{}'.format(len(dict_users)))
# sample_per_users = int(sum([ len(dict_users[i]) for i in range(len(dict_users))])/len(dict_users))
# print('num. of samples per user:{}'.format(sample_per_users))
# if args.dataset == 'fmnist' or args.dataset == 'cifar':
# dataset_test, val_set = torch.utils.data.random_split(
# dataset_test, [9000, 1000])
# print(len(dataset_test), len(val_set))
# elif args.dataset == 'svhn':
# dataset_test, val_set = torch.utils.data.random_split(
# dataset_test, [len(dataset_test)-2000, 2000])
# print(len(dataset_test), len(val_set))
# print("{:<50}".format("=" * 15 + " log path " + "=" * 50)[0:60])
# log_path = set_log_path(args)
# print(log_path)
# args, net_glob = model_setup(args)
# print("{:<50}".format("=" * 15 + " model setup " + "=" * 50)[0:60])
# ###################################### model initialization ###########################
# print("{:<50}".format("=" * 15 + " training... " + "=" * 50)[0:60])
# t1 = time.time()
# net_glob.train()
# # copy weights
# global_model = copy.deepcopy(net_glob.state_dict())
# local_m = []
# train_local_loss = []
# test_acc = []
# norm_med = []
# ####################################### run experiment ##########################
# # initialize data loader
# data_loader_list = []
# print(len(dict_users))
# for i in range(args.num_users):
# dataset = DatasetSplit(dataset_train, dict_users[i])
# ldr_train = DataLoader(dataset, batch_size=args.batch_size, shuffle=True)
# data_loader_list.append(ldr_train)
# ldr_train_public = DataLoader(val_set, batch_size=args.batch_size, shuffle=True)
# m = max(int(args.frac * args.num_users), 1)
# #m = 10
# for t in range(args.round):
# args.local_lr = args.local_lr * args.decay_weight
# selected_idxs = list(np.random.choice(range(args.num_users), m, replace=False))
# print(selected_idxs)
# num_selected_users = len(selected_idxs)
# ###################### local training : SGD for selected users ######################
# loss_locals = []
# local_updates = []
# delta_norms = []
# for i in selected_idxs:
# l_solver = LocalUpdate(args=args)
# net_glob.load_state_dict(global_model)
# # choose local solver
# if args.local_solver == 'local_sgd':
# new_model, loss = l_solver.local_sgd(
# net=copy.deepcopy(net_glob).to(args.device),
# ldr_train=data_loader_list[i])
# # compute local delta
# model_update = {k: new_model[k] - global_model[k] for k in global_model.keys()}
# # compute local model norm
# delta_norm = torch.norm(
# torch.cat([
# torch.flatten(model_update[k])
# for k in model_update.keys()
# ]))
# delta_norms.append(delta_norm)
# # clipping local model or not ? : no clip for cifar10
# # threshold = delta_norm / args.clip
# # if threshold > 1.0:
# # for k in model_update.keys():
# # model_update[k] = model_update[k] / threshold
# local_updates.append(model_update)
# loss_locals.append(loss)
# print("local updates len",len(local_updates), "index",len(local_updates[0]))
# norm_med.append(torch.median(torch.stack(delta_norms)).cpu())
# ##################### communication: avg for all groups #######################
# model_update = {
# k: local_updates[0][k] * 0.0
# for k in local_updates[0].keys()
# }
# for i in range(num_selected_users):
# global_model = {
# k: global_model[k] + local_updates[i][k] / num_selected_users
# for k in global_model.keys()
# }
# ##################### testing on global model #######################
# net_glob.load_state_dict(global_model)
# net_glob.eval()
# test_acc_, _ = test_img(net_glob, dataset_test, args)
# test_acc.append(test_acc_)
# train_local_loss.append(sum(loss_locals) / len(loss_locals))
# # print('t {:3d}: '.format(t, ))
# print('t {:3d}: train_loss = {:.3f}, norm = {:.3f}, test_acc = {:.3f}'.
# format(t, train_local_loss[-1], norm_med[-1], test_acc[-1]))
# if math.isnan(train_local_loss[-1]) or train_local_loss[-1] > 1e8 or t == args.round - 1:
# np.savetxt(log_path + "_test_acc_repeat_" + str(args.repeat) + ".csv",
# test_acc,
# delimiter=",")
# np.savetxt(log_path + "_train_loss_repeat_" + str(args.repeat) + ".csv",
# train_local_loss,
# delimiter=",")
# np.savetxt(log_path + "_norm__repeat_" + str(args.repeat) + ".csv", norm_med, delimiter=",")
# break;
# t2 = time.time()
# hours, rem = divmod(t2-t1, 3600)
# minutes, seconds = divmod(rem, 60)
# print("training time: {:0>2}:{:0>2}:{:05.2f}".format(int(hours),int(minutes),seconds))
#!/usr/bin/env python
# -*- coding: utf-8 -*-
# Python version: 3.6
import copy
import numpy as np
import time, math
import torch
from utils.data_utils import data_setup, DatasetSplit
from utils.model_utils import *
from utils.aggregation import *
from options import call_parser
from models.Update import LocalUpdate
from models.test import test_img
from torch.utils.data import DataLoader
# from utils.rdp_accountant import compute_rdp, get_privacy_spent
import warnings
warnings.filterwarnings("ignore")
torch.cuda.is_available()
if __name__ == '__main__':
################################### hyperparameter setup ########################################
args = call_parser()
torch.manual_seed(args.seed+args.repeat)
torch.cuda.manual_seed(args.seed+args.repeat)
np.random.seed(args.seed+args.repeat)
args, dataset_train, dataset_test, dict_users = data_setup(args)
print("{:<50}".format("=" * 15 + " data setup " + "=" * 50)[0:60])
print(
'length of dataset:{}'.format(len(dataset_train) + len(dataset_test)))
print('num. of training data:{}'.format(len(dataset_train)))
print('num. of testing data:{}'.format(len(dataset_test)))
print('num. of classes:{}'.format(args.num_classes))
print('num. of users:{}'.format(len(dict_users)))
sample_per_users = int(sum([ len(dict_users[i]) for i in range(len(dict_users))])/len(dict_users))
#sample_per_users = 250
print('For SVHN we are going to reduce the training size:')
#dataset_train = 5000
#dataset_test = 2500
print('num. of samples per user:{}'.format(sample_per_users))
if args.dataset == 'fmnist' or args.dataset == 'cifar':
dataset_test, val_set = torch.utils.data.random_split(
dataset_test, [9000, 1000])
print(len(dataset_test), len(val_set))
elif args.dataset == 'svhn':
dataset_test, val_set = torch.utils.data.random_split(
dataset_test, [len(dataset_test)-2000, 200])
print(len(dataset_test), len(val_set))
# dataset_test, val_set = torch.utils.data.random_split(
# dataset_test, [2500, 200])
# print(len(dataset_test), len(val_set))
print("{:<50}".format("=" * 15 + " log path " + "=" * 50)[0:60])
log_path = set_log_path(args)
print(log_path)
args, net_glob = model_setup(args)
print("{:<50}".format("=" * 15 + " model setup " + "=" * 50)[0:60])
###################################### model initialization ###########################
print("{:<50}".format("=" * 15 + " training... " + "=" * 50)[0:60])
t1 = time.time()
net_glob.train()
# copy weights
global_model = copy.deepcopy(net_glob.state_dict())
local_m = []
train_local_loss = []
test_acc = []
norm_med = []
####################################### run experiment ##########################
# initialize data loader
data_loader_list = []
for i in range(args.num_users):
dataset = DatasetSplit(dataset_train, dict_users[i])
ldr_train = DataLoader(dataset, batch_size=args.batch_size, shuffle=True)
data_loader_list.append(ldr_train)
ldr_train_public = DataLoader(val_set, batch_size=args.batch_size, shuffle=True)
m = max(int(args.frac * args.num_users), 1)
for t in range(args.round):
args.local_lr = args.local_lr * args.decay_weight
selected_idxs = list(np.random.choice(range(args.num_users), m, replace=False))
num_selected_users = len(selected_idxs)
###################### local training : SGD for selected users ######################
loss_locals = []
local_updates = []
delta_norms = []
for i in selected_idxs:
l_solver = LocalUpdate(args=args)
net_glob.load_state_dict(global_model)
# choose local solver
if args.local_solver == 'local_sgd':
new_model, loss = l_solver.local_sgd(
net=copy.deepcopy(net_glob).to(args.device),
ldr_train=data_loader_list[i])
# compute local delta
model_update = {k: new_model[k] - global_model[k] for k in global_model.keys()}
# compute local model norm
delta_norm = torch.norm(
torch.cat([
torch.flatten(model_update[k])
for k in model_update.keys()
]))
delta_norms.append(delta_norm)
# clipping local model or not ? : no clip for cifar10
# threshold = delta_norm / args.clip
# if threshold > 1.0:
# for k in model_update.keys():
# model_update[k] = model_update[k] / threshold
local_updates.append(model_update)
loss_locals.append(loss)
norm_med.append(torch.median(torch.stack(delta_norms)).cpu())
##################### communication: avg for all groups #######################
model_update = {
k: local_updates[0][k] * 0.0
for k in local_updates[0].keys()
}
for i in range(num_selected_users):
global_model = {
k: global_model[k] + local_updates[i][k] / num_selected_users
for k in global_model.keys()
}
##################### testing on global model #######################
net_glob.load_state_dict(global_model)
net_glob.eval()
test_acc_, _ = test_img(net_glob, dataset_test, args)
test_acc.append(test_acc_)
train_local_loss.append(sum(loss_locals) / len(loss_locals))
# print('t {:3d}: '.format(t, ))
print('t {:3d}: train_loss = {:.3f}, norm = {:.3f}, test_acc = {:.3f}'.
format(t, train_local_loss[-1], norm_med[-1], test_acc[-1]))
if math.isnan(train_local_loss[-1]) or train_local_loss[-1] > 1e8 or t == args.round - 1:
np.savetxt(log_path + "_test_acc_repeat_" + str(args.repeat) + ".csv",
test_acc,
delimiter=",")
np.savetxt(log_path + "_train_loss_repeat_" + str(args.repeat) + ".csv",
train_local_loss,
delimiter=",")
np.savetxt(log_path + "_norm__repeat_" + str(args.repeat) + ".csv", norm_med, delimiter=",")
break;
t2 = time.time()
hours, rem = divmod(t2-t1, 3600)
minutes, seconds = divmod(rem, 60)
print("training time: {:0>2}:{:0>2}:{:05.2f}".format(int(hours),int(minutes),seconds))