-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathmain.py
322 lines (248 loc) · 10.5 KB
/
main.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
import os
from PIL import Image
# from resnet18_bin import BinConv2d, resnet18_preact_bin
from xnor_net_revised import resnet18_XNOR
import numpy as np
import torch
from torch.utils.data import Dataset,DataLoader
import torch.nn as nn
import torchvision
import torchvision.transforms as transforms
import torch.optim as optim
import matplotlib
import matplotlib.pyplot as plt
from utils import imshow
import random
hyper_param_epoch = 80
hyper_param_batch = 64
hyper_param_learning_rate = 0.001
#Dataset and Dataloader
transforms_train=transforms.Compose([
transforms.RandomHorizontalFlip(),
transforms.RandomRotation(10),
transforms.ToTensor(),
transforms.Normalize((0.4914, 0.4822, 0.4465), (0.2023, 0.1994, 0.2010))
])
transforms_test=transforms.Compose([
transforms.ToTensor(),
transforms.Normalize((0.4914, 0.4822, 0.4465), (0.2023, 0.1994, 0.2010))
])
train_data_set=torchvision.datasets.CIFAR10(root='./data', train=True,
download=True, transform=transforms_train)
train_loader=DataLoader(train_data_set,batch_size=hyper_param_batch,shuffle=True) #배치 단위로 loader
test_data_set= torchvision.datasets.CIFAR10(root='./data', train=False,
download=True, transform=transforms_test)
test_loader=DataLoader(test_data_set,batch_size=hyper_param_batch,shuffle=False)
if not (train_data_set.classes ==test_data_set.classes):
print("error: Numbers of class in training set and test set are not equal")
exit()
device=torch.device('cuda:0' if torch.cuda.is_available() else 'cpu')
# Model, Loss, Optimizer
# #binarize=True이면 XNOR_Net++ False면 일반 ResNet18
net = resnet18_XNOR(num_classes=10, output_height=224, output_width=224, binarize=True).to(device) #output_height=32 output_width=32
# net=resnet18_preact_bin(num_classes=10,output_height=32,output_width=32,binarize=True).to(device)
criterion = nn.CrossEntropyLoss()
optimizer = optim.SGD(net.parameters(), lr=hyper_param_learning_rate, momentum=0.9, weight_decay=1e-5)
#학습 및 검증 결과를 저장할 리스트 초기화
train_losses=[]
train_accuracies=[]
test_losses=[]
test_accuracies=[]
#test와 train
for epoch in range(hyper_param_epoch):
net.train()
correct=0
running_loss=0
total=0
for num , data in enumerate(train_loader):
imgs,label =data # labels는 각 배치의 레이블을 담고 있는 1차원 텐서입니다.
imgs=imgs.to(device)
label=label.to(device)
optimizer.zero_grad()
out=net(imgs)
loss=criterion(out,label)
loss.backward()
optimizer.step()
running_loss+=loss.item()
_,predicted=out.max(1) #out의 크기는 (배치 크기, 클래스 수, 높이, 너비) => predicted: 클래스 인덱스
total+=label.size(0)
correct+=predicted.eq(label).sum().item()
train_loss=running_loss/len(train_loader)
train_accuracy=100*correct/total
train_losses.append(train_loss)
train_accuracies.append(train_accuracy)
net.eval()
test_loss=0.0
correct=0
total=0
with torch.no_grad():
for num,data in enumerate(test_loader):
imgs, label = data
imgs = imgs.to(device)
label = label.to(device)
outs=net(imgs)
loss=criterion(outs,label)
test_loss+=loss.item()
_,predicted=outs.max(1)
total+=label.size(0)
correct+=predicted.eq(label).sum().item()
test_loss=test_loss/len(test_loader)
test_accuracy=100*correct/total
test_losses.append(test_loss)
test_accuracies.append(test_accuracy)
print(f'Epoch {epoch + 1}, Train Loss: {train_loss:.4f}, Train Accuracy: {train_accuracy:.2f}%,Test Loss: {test_loss:.4f}, Test Accuracy: {test_accuracy:.2f}%')
print('Finished Training')
# 모델 저장하기
PATH = "xnornet_model_weights.pth"
torch.save(net.state_dict(), PATH)
#===================================================================================================================
# #결과 시각화
# epochs=range(1,51)
# plt.figure(figsize=(12,4))
# plt.subplot(1, 2, 1)
# plt.plot(epochs, train_losses, label='Train Loss')
# plt.plot(epochs, test_losses, label='Test Loss')
# plt.xlabel('Epoch')
# plt.ylabel('Loss')
# plt.title('Loss')
# plt.legend()# 모델을 평가 모드로 전환
# net.eval()
# #weight,activation(feature),scaling factor의 분포를 그래프로 분석
# #몇가지 레이어의 weight를 추출
# # 첫 번째 블록의 두 번째 conv 레이어 (layer1[0].conv1)의 weights와 scaling factors 추출
# block1_conv2_weights = net.layer3[0].conv2.weight.data.cpu().numpy().flatten()
# block1_conv2_alpha = net.layer3[0].conv2.alpha.data.cpu().numpy().flatten()
# block1_conv2_beta = net.layer3[0].conv2.beta.data.cpu().numpy().flatten()
# block1_conv2_gamma = net.layer3[0].conv2.gamma.data.cpu().numpy().flatten()
# # 히스토그램을 그려서 분포 시각화
# plt.figure(figsize=(18, 6))
# # Weights 분포 시각화
# plt.subplot(1, 4, 1)
# plt.hist(block1_conv2_weights, bins=100, alpha=0.75, color='blue')
# plt.title('weights')
# plt.xlabel('Weight value')
# plt.ylabel('Frequency')
# # Scaling factor (alpha) 분포 시각화
# plt.subplot(1, 4, 2)
# plt.hist(block1_conv2_alpha, bins=100, alpha=0.75, color='red')
# plt.title('alpha')
# plt.xlabel('Alpha value')
# plt.ylabel('Frequency')
# # Scaling factor (beta) 분포 시각화
# plt.subplot(1, 4, 3)
# plt.hist(block1_conv2_beta, bins=100, alpha=0.75, color='green')
# plt.title('beta')
# plt.xlabel('Beta value')
# plt.ylabel('Frequency')
# # Scaling factor (gamma) 분포 시각화
# plt.subplot(1, 4, 4)
# plt.hist(block1_conv2_gamma, bins=100, alpha=0.75, color='purple')
# plt.title('gamma')
# plt.xlabel('Gamma value')
# plt.ylabel('Frequency')
# plt.savefig('weight_distribution.jpg', format='jpg')
# plt.tight_layout()
# plt.show()
# # `layer3`을 통과한 후의 활성화 값을 저장할 리스트
# activations_layer3 = []
# # 활성화 값을 추출하기 위해 hook을 정의
# def hook_fn(module, input, output):
# activations_layer3.append(output.cpu().detach().numpy().flatten())
# # `layer3`에 hook을 등록
# hook = net.layer3.register_forward_hook(hook_fn)
# # 데이터 로더에서 첫 배치를 통과시켜 활성화 값을 추출
# with torch.no_grad():
# for images, _ in train_loader:
# images = images.to(device)
# _ = net(images) # forward pass를 통해 hook에서 활성화 값을 추출
# break # 한 배치만 사용
# # hook 제거
# hook.remove()
# # 추출한 활성화 값을 numpy 배열로 변환
# activations_layer3 = np.concatenate(activations_layer3)
# # 히스토그램을 그려서 분포 시각화 및 저장
# plt.figure(figsize=(6, 6))
# plt.hist(activations_layer3, bins=100, alpha=0.75, color='blue')
# plt.title('Distribution of activations in layer3')
# plt.xlabel('Activation value')
# plt.ylabel('Frequency')
# # 그래프를 JPG 파일로 저장
# plt.savefig('activations_layer3_histogram.jpg', format='jpg')
# # 그래프를 화면에 표시
# plt.show()
# plt.subplot(1, 2, 2)
# plt.plot(epochs, train_accuracies, label='Train Accuracy')
# plt.plot(epochs, test_accuracies, label='Test Accuracy')
# plt.xlabel('Epoch')
# plt.ylabel('Accuracy')
# plt.title('Accuracy')
# plt.legend()
# plt.tight_layout()
# plt.show()
# wandb.finish()
#=================================================================================================================
# # 모델을 평가 모드로 전환
# net.eval()
# #weight,activation(feature),scaling factor의 분포를 그래프로 분석
# #몇가지 레이어의 weight를 추출
# # 첫 번째 블록의 두 번째 conv 레이어 (layer1[0].conv1)의 weights와 scaling factors 추출
# block1_conv2_weights = net.layer3[0].conv2.weight.data.cpu().numpy().flatten()
# block1_conv2_alpha = net.layer3[0].conv2.alpha.data.cpu().numpy().flatten()
# block1_conv2_beta = net.layer3[0].conv2.beta.data.cpu().numpy().flatten()
# block1_conv2_gamma = net.layer3[0].conv2.gamma.data.cpu().numpy().flatten()
# # 히스토그램을 그려서 분포 시각화
# plt.figure(figsize=(18, 6))
# # Weights 분포 시각화
# plt.subplot(1, 4, 1)
# plt.hist(block1_conv2_weights, bins=100, alpha=0.75, color='blue')
# plt.title('weights')
# plt.xlabel('Weight value')
# plt.ylabel('Frequency')
# # Scaling factor (alpha) 분포 시각화
# plt.subplot(1, 4, 2)
# plt.hist(block1_conv2_alpha, bins=100, alpha=0.75, color='red')
# plt.title('alpha')
# plt.xlabel('Alpha value')
# plt.ylabel('Frequency')
# # Scaling factor (beta) 분포 시각화
# plt.subplot(1, 4, 3)
# plt.hist(block1_conv2_beta, bins=100, alpha=0.75, color='green')
# plt.title('beta')
# plt.xlabel('Beta value')
# plt.ylabel('Frequency')
# # Scaling factor (gamma) 분포 시각화
# plt.subplot(1, 4, 4)
# plt.hist(block1_conv2_gamma, bins=100, alpha=0.75, color='purple')
# plt.title('gamma')
# plt.xlabel('Gamma value')
# plt.ylabel('Frequency')
# plt.savefig('weight_distribution.jpg', format='jpg')
# plt.tight_layout()
# plt.show()
# # `layer3`을 통과한 후의 활성화 값을 저장할 리스트
# activations_layer3 = []
# # 활성화 값을 추출하기 위해 hook을 정의
# def hook_fn(module, input, output):
# activations_layer3.append(output.cpu().detach().numpy().flatten())
# # `layer3`에 hook을 등록
# hook = net.layer3.register_forward_hook(hook_fn)
# # 데이터 로더에서 첫 배치를 통과시켜 활성화 값을 추출
# with torch.no_grad():
# for images, _ in train_loader:
# images = images.to(device)
# _ = net(images) # forward pass를 통해 hook에서 활성화 값을 추출
# break # 한 배치만 사용
# # hook 제거
# hook.remove()
# # 추출한 활성화 값을 numpy 배열로 변환
# activations_layer3 = np.concatenate(activations_layer3)
# # 히스토그램을 그려서 분포 시각화 및 저장
# plt.figure(figsize=(6, 6))
# plt.hist(activations_layer3, bins=100, alpha=0.75, color='blue')
# plt.title('Distribution of activations in layer3')
# plt.xlabel('Activation value')
# plt.ylabel('Frequency')
# # 그래프를 JPG 파일로 저장
# plt.savefig('activations_layer3_histogram.jpg', format='jpg')
# # 그래프를 화면에 표시
# plt.show()