-
Notifications
You must be signed in to change notification settings - Fork 11
/
Copy pathtrain_FedNoRo.py
203 lines (177 loc) · 8.33 KB
/
train_FedNoRo.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
import os
import copy
import logging
import numpy as np
from sklearn.metrics import balanced_accuracy_score, accuracy_score, confusion_matrix
from sklearn.mixture import GaussianMixture
from collections import Counter
import torch
import torch.backends.cudnn as cudnn
import torch.nn as nn
from tensorboardX import SummaryWriter
from torch.utils.data import DataLoader
from utils.options import args_parser
from utils.local_training import LocalUpdate, globaltest
from utils.FedAvg import FedAvg, DaAgg
from utils.utils import add_noise, set_seed, set_output_files, get_output, get_current_consistency_weight
from dataset.dataset import get_dataset
from model.build_model import build_model
np.set_printoptions(threshold=np.inf)
"""
Major framework of noise FL
"""
if __name__ == '__main__':
args = args_parser()
args.num_users = args.n_clients
os.environ['CUDA_VISIBLE_DEVICES'] = args.gpu
args.device = "cuda" if torch.cuda.is_available() else "cpu"
# ------------------------------ deterministic or not ------------------------------
if args.deterministic:
cudnn.benchmark = False
cudnn.deterministic = True
set_seed(args.seed)
# ------------------------------ output files ------------------------------
writer, models_dir = set_output_files(args)
# ------------------------------ dataset ------------------------------
dataset_train, dataset_test, dict_users = get_dataset(args)
logging.info(
f"train: {Counter(dataset_train.targets)}, total: {len(dataset_train.targets)}")
logging.info(
f"test: {Counter(dataset_test.targets)}, total: {len(dataset_test.targets)}")
# --------------------- Add Noise ---------------------------
y_train = np.array(dataset_train.targets)
y_train_noisy, gamma_s, real_noise_level = add_noise(
args, y_train, dict_users)
dataset_train.targets = y_train_noisy
# --------------------- Build Models ---------------------------
netglob = build_model(args)
user_id = list(range(args.n_clients))
trainer_locals = []
for id in user_id:
trainer_locals.append(LocalUpdate(
args, id, copy.deepcopy(dataset_train), dict_users[id]))
# ------------------------------ begin training ------------------------------
set_seed(args.seed)
logging.info("\n ---------------------begin training---------------------")
best_performance = 0.
# ------------------------ Stage 1: warm up ------------------------
if args.warm:
for rnd in range(args.s1):
w_locals, loss_locals = [], []
for idx in user_id: # training over the subset
local = trainer_locals[idx]
w_local, loss_local = local.train_LA(
net=copy.deepcopy(netglob).to(args.device), writer=writer)
# store every updated model
w_locals.append(copy.deepcopy(w_local))
loss_locals.append(copy.deepcopy(loss_local))
w_locals_last = copy.deepcopy(w_locals)
dict_len = [len(dict_users[idx]) for idx in user_id]
w_glob_fl = FedAvg(w_locals, dict_len)
netglob.load_state_dict(copy.deepcopy(w_glob_fl))
pred = globaltest(copy.deepcopy(netglob).to(
args.device), dataset_test, args)
acc = accuracy_score(dataset_test.targets, pred)
bacc = balanced_accuracy_score(dataset_test.targets, pred)
cm = confusion_matrix(dataset_test.targets, pred)
logging.info(
"******** round: %d, acc: %.4f, bacc: %.4f ********" % (rnd, acc, bacc))
logging.info(cm)
writer.add_scalar(f'test/acc', acc, rnd)
writer.add_scalar(f'test/bacc', bacc, rnd)
# save model
if bacc > best_performance:
best_performance = bacc
logging.info(f'best bacc: {best_performance}, now bacc: {bacc}')
logging.info('\n')
torch.save(netglob.state_dict(), models_dir +
f'/stage1_model_{rnd}.pth')
# ------------------------ client selection ------------------------
model_path = f"outputs_{args.dataset}_{args.level_n_system}_{args.level_n_lowerb}_{args.level_n_upperb}/FedNoRo_{args.level_n_system}_{args.level_n_lowerb}_{args.level_n_upperb}_{args.local_ep}/models/stage1_model_{args.s1-1}.pth"
logging.info(
f"********************** load model from: {model_path} **********************")
netglob.load_state_dict(torch.load(model_path))
loader = DataLoader(dataset=dataset_train, batch_size=32,
shuffle=False, num_workers=4)
criterion = nn.CrossEntropyLoss(reduction='none')
local_output, loss = get_output(
loader, netglob.to(args.device), args, False, criterion)
metrics = np.zeros((args.n_clients, args.n_classes)).astype("float")
num = np.zeros((args.n_clients, args.n_classes)).astype("float")
for id in range(args.n_clients):
idxs = dict_users[id]
for idx in idxs:
c = dataset_train.targets[idx]
num[id, c] += 1
metrics[id, c] += loss[idx]
metrics = metrics / num
for i in range(metrics.shape[0]):
for j in range(metrics.shape[1]):
if np.isnan(metrics[i, j]):
metrics[i, j] = np.nanmin(metrics[:, j])
for j in range(metrics.shape[1]):
metrics[:, j] = (metrics[:, j]-metrics[:, j].min()) / \
(metrics[:, j].max()-metrics[:, j].min())
logging.info("metrics:")
logging.info(metrics)
vote = []
for i in range(9):
gmm = GaussianMixture(n_components=2, random_state=i).fit(metrics)
gmm_pred = gmm.predict(metrics)
noisy_clients = np.where(gmm_pred == np.argmax(gmm.means_.sum(1)))[0]
noisy_clients = set(list(noisy_clients))
vote.append(noisy_clients)
cnt = []
for i in vote:
cnt.append(vote.count(i))
noisy_clients = list(vote[cnt.index(max(cnt))])
logging.info(
f"selected noisy clients: {noisy_clients}, real noisy clients: {np.where(gamma_s>0.)[0]}")
clean_clients = list(set(user_id) - set(noisy_clients))
logging.info(f"selected clean clients: {clean_clients}")
# ------------------------ Stage 2: ------------------------
BACC = []
for rnd in range(args.s1, args.rounds):
w_locals, loss_locals = [], []
weight_kd = get_current_consistency_weight(
rnd, args.begin, args.end) * args.a
writer.add_scalar(f'train/w_kd', weight_kd, rnd)
for idx in user_id: # training over the subset
local = trainer_locals[idx]
if idx in clean_clients:
w_local, loss_local = local.train_LA(
net=copy.deepcopy(netglob).to(args.device), writer=writer)
elif idx in noisy_clients:
w_local, loss_local = local.train_FedNoRo(
student_net=copy.deepcopy(netglob).to(args.device), teacher_net=copy.deepcopy(netglob).to(args.device), writer=writer, weight_kd=weight_kd)
# store every updated model
w_locals.append(copy.deepcopy(w_local))
loss_locals.append(copy.deepcopy(loss_local))
assert len(w_locals) == len(loss_locals) == idx+1
dict_len = [len(dict_users[idx]) for idx in user_id]
w_glob_fl = DaAgg(
w_locals, dict_len, clean_clients, noisy_clients)
netglob.load_state_dict(copy.deepcopy(w_glob_fl))
pred = globaltest(copy.deepcopy(netglob).to(
args.device), dataset_test, args)
acc = accuracy_score(dataset_test.targets, pred)
bacc = balanced_accuracy_score(dataset_test.targets, pred)
cm = confusion_matrix(dataset_test.targets, pred)
logging.info(
"******** round: %d, acc: %.4f, bacc: %.4f ********" % (rnd, acc, bacc))
logging.info(cm)
writer.add_scalar(f'test/acc', acc, rnd)
writer.add_scalar(f'test/bacc', bacc, rnd)
BACC.append(bacc)
# save model
if bacc > best_performance:
best_performance = bacc
logging.info(f'best bacc: {best_performance}, now bacc: {bacc}')
logging.info('\n')
torch.save(netglob.state_dict(), models_dir+'f/stage2_model_{rnd}.pth')
BACC = np.array(BACC)
logging.info("last:")
logging.info(BACC[-10:].mean())
logging.info("best:")
logging.info(BACC.max())
torch.cuda.empty_cache()