-
Notifications
You must be signed in to change notification settings - Fork 8
/
vanilla_vgg.py
246 lines (217 loc) · 9.14 KB
/
vanilla_vgg.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
import matplotlib
import matplotlib.pyplot as plt
from vgg_features import vgg11_features, vgg11_bn_features, vgg13_features, vgg13_bn_features, vgg16_features, vgg16_bn_features,\
vgg19_features, vgg19_bn_features
import argparse
import torch.nn as nn
from dataHelper import DatasetFolder
from torchvision import transforms
import torch
from torch import optim
from torch.utils.tensorboard import SummaryWriter
import numpy as np
from sklearn.metrics import roc_auc_score
import os
import random
# build model
class Vanilla_VGG(nn.Module):
def __init__(self, myfeatures, num_classes):
super(Vanilla_VGG, self).__init__()
self.features = myfeatures
self.avgpool = nn.AdaptiveAvgPool2d((7,7))
self.num_classes = num_classes
self.classifier = nn.Sequential(
nn.Linear(512*7*7, 4096),
nn.ReLU(True),
nn.Dropout(),
nn.Linear(4096, 4096),
nn.ReLU(True),
nn.Dropout(),
nn.Linear(4096, num_classes),
nn.LogSoftmax(dim=0)
)
def forward(self, x):
x = self.features(x)
x = self.avgpool(x)
x = torch.flatten(x, 1)
x = self.classifier(x)
return x
def main():
matplotlib.use("Agg")
parser = argparse.ArgumentParser()
parser.add_argument("-model", type=str)
parser.add_argument("-train_dir", type=str, default="/usr/project/xtmp/mammo/")
parser.add_argument("-test_dir", type=str, default="/usr/project/xtmp/mammo/")
parser.add_argument("-name", type=str)
parser.add_argument("-lr", type=lambda x: float(x))
parser.add_argument("-wd", type=lambda x: float(x))
parser.add_argument("-num_classes", type=lambda x: int(x))
args = parser.parse_args()
model_name = args.model
train_dir = args.train_dir
test_dir = args.test_dir
task_name = args.name
num_classes = args.num_classes
save_loc = '/usr/xtmp/mammo/saved_models/vanilla/'
os.makedirs(save_loc + task_name, exist_ok = True)
lr = args.lr
wd = args.wd
print(wd, lr)
print("Saving to: ", save_loc + task_name)
random_seed_number = 12
print("Random seed: ", random_seed_number)
torch.manual_seed(random_seed_number)
torch.cuda.manual_seed(random_seed_number)
np.random.seed(random_seed_number)
random.seed(random_seed_number)
torch.backends.cudnn.enabled=False
torch.backends.cudnn.deterministic=True
writer = SummaryWriter()
base_architecture_to_features = {'vgg11': vgg11_features,
'vgg11_bn': vgg11_bn_features,
'vgg13': vgg13_features,
'vgg13_bn': vgg13_bn_features,
'vgg16': vgg16_features,
'vgg16_bn': vgg16_bn_features,
'vgg19': vgg19_features,
'vgg19_bn': vgg19_bn_features}
features = base_architecture_to_features[model_name](pretrained=True)
model = Vanilla_VGG(features, num_classes)
# load data
# train set
train_dataset = DatasetFolder(
train_dir,
augmentation=False,
loader=np.load,
extensions=("npy",),
target_size=(224, 224),
transform = transforms.Compose([
torch.from_numpy,
]))
trainloader = torch.utils.data.DataLoader(
train_dataset, batch_size=100, shuffle=True,
num_workers=4, pin_memory=False)
# test set
test_dataset =DatasetFolder(
test_dir,
loader=np.load,
target_size=(224, 224),
extensions=("npy",),
transform = transforms.Compose([
torch.from_numpy,
]))
testloader = torch.utils.data.DataLoader(
test_dataset, batch_size=100, shuffle=False,
num_workers=4, pin_memory=False)
# start training
epochs = 250
criterion = nn.CrossEntropyLoss()
optimizer = optim.Adam(model.parameters(), lr=lr, weight_decay=wd)
device = torch.device("cuda")
model.to(device)
train_losses = []
test_losses = []
train_auc = []
test_auc = []
curr_best = 0
for epoch in range(epochs):
# train
confusion_matrix = np.zeros((num_classes, num_classes))
total_output = []
total_one_hot_label = []
running_loss = 0
model.train()
for inputs, labels, id in trainloader:
inputs, labels = inputs.to(device), labels.to(device)
optimizer.zero_grad()
logps = model.forward(inputs)
loss = criterion(logps, labels)
loss.backward()
optimizer.step()
running_loss += loss.item()
one_hot_label = np.zeros(shape=(len(labels), num_classes))
for k in range(len(labels)):
one_hot_label[k][labels[k].item()] = 1
# roc_auc_score()
total_output.extend(logps.cpu().detach().numpy())
total_one_hot_label.extend(one_hot_label)
# confusion matrix
_, predicted = torch.max(logps.data, 1)
for t_idx, t in enumerate(labels):
confusion_matrix[predicted[t_idx]][t] += 1 #row is predicted, col is true
# if predicted[t_idx] == t: # correct label
# confusion_matrix[t][t] += 1
# elif t == 0 and predicted[t_idx] == 1:
# confusion_matrix[1] += 1 # false positives
# elif t == 1 and predicted[t_idx] == 0:
# confusion_matrix[2] += 1 # false negative
# else:
# confusion_matrix[3] += 1
auc_score = roc_auc_score(np.array(total_one_hot_label), np.array(total_output))
train_losses.append(running_loss / len(trainloader))
train_auc.append(auc_score)
print("=======================================================")
print("\t at epoch {}".format(epoch))
print("\t train loss is {}".format(train_losses[-1]))
print("\t train auc is {}".format(auc_score))
print('\tthe confusion matrix is: \n{0}'.format(confusion_matrix))
# test
confusion_matrix = np.zeros((num_classes, num_classes))
test_loss = 0
total_output = []
total_one_hot_label = []
model.eval()
with torch.no_grad():
for inputs, labels, id in testloader:
inputs, labels = inputs.to(device), labels.to(device)
logps = model.forward(inputs)
batch_loss = criterion(logps, labels)
test_loss += batch_loss.item()
one_hot_label = np.zeros(shape=(len(labels), num_classes))
for k in range(len(labels)):
one_hot_label[k][labels[k].item()] = 1
# roc_auc_score()
total_output.extend(logps.cpu().numpy())
total_one_hot_label.extend(one_hot_label)
# confusion matrix
_, predicted = torch.max(logps.data, 1)
for t_idx, t in enumerate(labels):
confusion_matrix[predicted[t_idx]][t] += 1 #row is predicted, col is true
# if predicted[t_idx] == t and predicted[t_idx] == 1: # true positive
# confusion_matrix[0] += 1
# elif t == 0 and predicted[t_idx] == 1:
# confusion_matrix[1] += 1 # false positives
# elif t == 1 and predicted[t_idx] == 0:
# confusion_matrix[2] += 1 # false negative
# else:
# confusion_matrix[3] += 1
auc_score = roc_auc_score(np.array(total_one_hot_label), np.array(total_output))
test_losses.append(test_loss / len(testloader))
test_auc.append(auc_score)
print("===========================")
if auc_score > curr_best:
curr_best = auc_score
print("\t test loss is {}".format(test_losses[-1]))
print("\t test auc is {}".format(auc_score))
print("\t current best is {}".format(curr_best))
print('\tthe confusion matrix is: \n{0}'.format(confusion_matrix))
print("=======================================================")
# save model
if auc_score > 0:
torch.save(model, save_loc + task_name + "/" + str(auc_score) + "_at_epoch_" + str(epoch))
# plot graphs
plt.plot(train_losses, "b", label="train")
plt.plot(test_losses, "r", label="test")
#plt.ylim(0, 4)
plt.legend()
plt.savefig(save_loc + task_name + '/train_test_loss_vanilla' + ".png")
plt.close()
plt.plot(train_auc, "b", label="train")
plt.plot(test_auc, "r", label="test")
#plt.ylim(0.4, 1)
plt.legend()
plt.savefig(save_loc + task_name + '/train_test_auc_vanilla' + ".png")
plt.close()
writer.close()
if __name__=="__main__":
main()