-
Notifications
You must be signed in to change notification settings - Fork 0
/
learner_FedAvg_SFD_densenet121.py
235 lines (204 loc) · 10.7 KB
/
learner_FedAvg_SFD_densenet121.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
import torch
from torchvision import models
import numpy as np
from torch.autograd import Variable
import torch.nn as nn
import pandas as pd
import os
from client_FedAvg_SFD_densenet121 import Client
import random
class Metanet(nn.Module):
def __init__(self, device,local_metatrain_epoch=1, local_test_epoch=3,outer_lr=0.001,inner_lr=0.001):
super(Metanet, self).__init__()
self.device = device
self.local_metatrain_epoch = local_metatrain_epoch
self.local_test_epoch = local_test_epoch
self.net = models.densenet121(pretrained=False).to(self.device)
self.loss_function = torch.nn.CrossEntropyLoss()
self.clients = []
self.test_clients = []
self.mode_1 = "fed_train"
self.mode_2 = "fed_test"
self.batch_size = 40
self.path_now = os.path.dirname(__file__)
self.last_path = '/0529_1_FedAvg_SFD_densenet121'#-----------------------------------------------------
train_path = self.path_now + r"/state-farm-distracted-driver-detection/data_client/train_client"
test_path = self.path_now + r"/state-farm-distracted-driver-detection/data_client/test_client"
train_file_set = os.listdir(train_path)
train_path_set = [os.path.join(train_path,i) for i in train_file_set]
test_file_set = os.listdir(test_path)
test_path_set = [os.path.join(test_path,i) for i in test_file_set]
self.time_accum = [0]
for index,path in enumerate(train_path_set):
model = models.densenet121(pretrained=False).to(self.device)
self.clients.append(Client(model,index,path,'train',local_metatrain_epoch,local_test_epoch,
inner_lr,outer_lr,self.device,self.mode_1,self.batch_size))
for index,path in enumerate(test_path_set):
model = models.densenet121(pretrained=False).to(self.device)
self.test_clients.append(Client(model,index,path,'test',local_metatrain_epoch,local_test_epoch,
inner_lr,outer_lr,self.device,self.mode_2,self.batch_size))
def forward(self):
pass
def save_time(self,save_path):
dataframe = pd.DataFrame(list(self.time_accum), columns=['time_accum'])
dataframe.to_excel(save_path, index=False)
def meta_training(self,round):
id_train_0 = list(range(len(self.clients)))
id_train = random.sample(id_train_0, int(len(id_train_0)*0.2)) #clients of this round
time_list_tr = []#time list this round
for id,j in enumerate(id_train):
time_list_tr.append(random.uniform(3, 20))
self.clients[j].refresh(self.net)
self.clients[j].local_fed_train()
self.clients[j].epoch = round
time_tr = max(time_list_tr)
#self.num_c_per_r.append(len(id_train))
self.time_accum.append(self.time_accum[-1]+time_tr)
weight = []
for id,j in enumerate(id_train):
#weight.append(self.clients[j].size / size_all)
weight.append(self.clients[j].size)
weight = np.array(weight)
weight = weight / weight.sum()
for id,j in enumerate(id_train):
for w,w_t in zip(self.net.parameters(),self.clients[j].net.parameters()):
if (w is None or id == 0):
w_tem = Variable(torch.zeros_like(w)).to(self.device)
w.data.copy_(w_tem.data)
if w_t is None:
w_t = Variable(torch.zeros_like(w)).to(self.device)
w.data.add_(w_t.data*weight[id])
def Testing(self,round):
id_test = list(range(len(self.test_clients)))
final_init_loss, final_init_acc, final_init_pre\
, final_init_recall, final_init_f1 = [], [], [], [], []
for a,id in enumerate(id_test):
self.test_clients[id].refresh(self.net)
fine_tuning_loss_list,test_loss_list,init_loss\
,fine_tuning_acc_list,test_acc_list,init_acc\
,fine_tuning_pre_list,test_pre_list,init_pre\
,fine_tuning_recall_list,test_recall_list,init_recall\
,fine_tuning_f1_list,test_f1_list,init_f1 = self.test_clients[id].test()
if a == 0:
final_train_loss = fine_tuning_loss_list.copy()
final_test_loss = test_loss_list.copy()
final_init_loss.append(init_loss)
final_train_acc = fine_tuning_acc_list.copy()
final_test_acc = test_acc_list.copy()
final_init_acc.append(init_acc)
final_train_pre = fine_tuning_pre_list.copy()
final_test_pre = test_pre_list.copy()
final_init_pre.append(init_pre)
final_train_recall = fine_tuning_recall_list.copy()
final_test_recall = test_recall_list.copy()
final_init_recall.append(init_recall)
final_train_f1 = fine_tuning_f1_list.copy()
final_test_f1 = test_f1_list.copy()
final_init_f1.append(init_f1)
else:
final_train_loss = np.concatenate((final_train_loss,fine_tuning_loss_list),axis = 0)
final_test_loss = np.concatenate((final_test_loss,test_loss_list),axis = 0)
final_init_loss.append(init_loss)
final_train_acc = np.concatenate((final_train_acc,fine_tuning_acc_list),axis = 0)
final_test_acc = np.concatenate((final_test_acc,test_acc_list),axis = 0)
final_init_acc.append(init_acc)
#---
final_train_pre = np.concatenate((final_train_pre,fine_tuning_pre_list),axis = 0)
final_test_pre = np.concatenate((final_test_pre,test_pre_list),axis = 0)
final_init_pre.append(init_pre)
#---
final_train_recall = np.concatenate((final_train_recall,fine_tuning_recall_list),axis = 0)
final_test_recall = np.concatenate((final_test_recall,test_recall_list),axis = 0)
final_init_recall.append(init_recall)
#---
final_train_f1 = np.concatenate((final_train_f1,fine_tuning_f1_list),axis = 0)
final_test_f1 = np.concatenate((final_test_f1,test_f1_list),axis = 0)
final_init_f1.append(init_f1)
last_path = self.last_path
loss_train_file_path = self.path_now + "/result_save/farm/loss/async_fomaml_train_loss" + last_path
loss_test_file_path = self.path_now + "/result_save/farm/loss/async_fomaml_test_loss" + last_path
loss_init_file_path = self.path_now + "/result_save/farm/loss/async_fomaml_init_loss" + last_path
acc_train_file_path = self.path_now + "/result_save/farm/acc/async_fomaml_train_acc" + last_path
acc_test_file_path = self.path_now + "/result_save/farm/acc/async_fomaml_test_acc" + last_path
acc_init_file_path = self.path_now + "/result_save/farm/acc/async_fomaml_init_acc" + last_path
#---
pre_train_file_path = self.path_now + "/result_save/farm/pre/async_fomaml_train_pre" + self.last_path
pre_test_file_path = self.path_now + "/result_save/farm/pre/async_fomaml_test_pre" + self.last_path
pre_init_file_path = self.path_now + "/result_save/farm/pre/async_fomaml_init_pre" + self.last_path
#---
recall_train_file_path = self.path_now + "/result_save/farm/recall/async_fomaml_train_recall" + self.last_path
recall_test_file_path = self.path_now + "/result_save/farm/recall/async_fomaml_test_recall" + self.last_path
recall_init_file_path = self.path_now + "/result_save/farm/recall/async_fomaml_init_recall" + self.last_path
#---
f1_train_file_path = self.path_now + "/result_save/farm/f1/async_fomaml_train_f1" + self.last_path
f1_test_file_path = self.path_now + "/result_save/farm/f1/async_fomaml_test_f1" + self.last_path
f1_init_file_path = self.path_now + "/result_save/farm/f1/async_fomaml_init_f1" + self.last_path
if not os.path.exists(loss_train_file_path):
os.mkdir(loss_train_file_path)
if not os.path.exists(loss_test_file_path):
os.mkdir(loss_test_file_path)
if not os.path.exists(loss_init_file_path):
os.mkdir(loss_init_file_path)
if not os.path.exists(acc_train_file_path):
os.mkdir(acc_train_file_path)
if not os.path.exists(acc_test_file_path):
os.mkdir(acc_test_file_path)
if not os.path.exists(acc_init_file_path):
os.mkdir(acc_init_file_path)
#---
if not os.path.exists(pre_train_file_path):
os.mkdir(pre_train_file_path)
if not os.path.exists(pre_test_file_path):
os.mkdir(pre_test_file_path)
if not os.path.exists(pre_init_file_path):
os.mkdir(pre_init_file_path)
#---
if not os.path.exists(recall_train_file_path):
os.mkdir(recall_train_file_path)
if not os.path.exists(recall_test_file_path):
os.mkdir(recall_test_file_path)
if not os.path.exists(recall_init_file_path):
os.mkdir(recall_init_file_path)
#---
if not os.path.exists(f1_train_file_path):
os.mkdir(f1_train_file_path)
if not os.path.exists(f1_test_file_path):
os.mkdir(f1_test_file_path)
if not os.path.exists(f1_init_file_path):
os.mkdir(f1_init_file_path)
loss_train_path = os.path.join(loss_train_file_path,"{}.npy".format(round))
loss_test_path = os.path.join(loss_test_file_path,"{}.npy".format(round))
loss_init_path = os.path.join(loss_init_file_path,"{}.npy".format(round))
acc_train_path = os.path.join(acc_train_file_path,"{}.npy".format(round))
acc_test_path = os.path.join(acc_test_file_path,"{}.npy".format(round))
acc_init_path = os.path.join(acc_init_file_path,"{}.npy".format(round))
#---
pre_train_path = os.path.join(pre_train_file_path,"{}.npy".format(round))
pre_test_path = os.path.join(pre_test_file_path,"{}.npy".format(round))
pre_init_path = os.path.join(pre_init_file_path,"{}.npy".format(round))
#---
recall_train_path = os.path.join(recall_train_file_path,"{}.npy".format(round))
recall_test_path = os.path.join(recall_test_file_path,"{}.npy".format(round))
recall_init_path = os.path.join(recall_init_file_path,"{}.npy".format(round))
#---
f1_train_path = os.path.join(f1_train_file_path,"{}.npy".format(round))
f1_test_path = os.path.join(f1_test_file_path,"{}.npy".format(round))
f1_init_path = os.path.join(f1_init_file_path,"{}.npy".format(round))
np.save(loss_train_path,final_train_loss)
np.save(loss_test_path,final_test_loss)
np.save(loss_init_path,final_init_loss)
np.save(acc_train_path,final_train_acc)
np.save(acc_test_path,final_test_acc)
np.save(acc_init_path,final_init_acc)
#---
np.save(pre_train_path,final_train_pre)
np.save(pre_test_path,final_test_pre)
np.save(pre_init_path,final_init_pre)
#---
np.save(recall_train_path,final_train_recall)
np.save(recall_test_path,final_test_recall)
np.save(recall_init_path,final_init_recall)
#---
np.save(f1_train_path,final_train_f1)
np.save(f1_test_path,final_test_f1)
np.save(f1_init_path,final_init_f1)