-
Notifications
You must be signed in to change notification settings - Fork 0
/
train.py
223 lines (186 loc) · 8.71 KB
/
train.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
import os
import argparse
import torch
import pandas as pd
import numpy as np
import time
from dataset import COVID19DataSet
from model import mobilenet_v2, densenet121
import utils
from test import validate
NUM_COVID = 251
NUM_NONCOVID = 292
NTRAIN_RATIO = 0.8
def train(epoch, net, trainloader, criterion, optimizer, scheduler, model, device):
net.train() # train mode
train_loss = 0.
for batch_idx, (imgs, lungsegs, labels) in enumerate(trainloader):
imgs = imgs.to(device)
lungsegs = lungsegs.to(device)
labels = labels.to(device)
if model in ['densenet']:
if batch_idx%2 == 0:
optimizer.zero_grad()
else:
optimizer.zero_grad()
logits = net(imgs, lungsegs)
loss = criterion(logits, labels)
loss.backward()
if model in ['densenet']:
if batch_idx%2 == 1:
optimizer.step()
else:
optimizer.step()
train_loss += loss.item()
print(' Training... Epoch: %4d | Iter: %4d/%4d | Train Loss: %.4f'%(epoch, batch_idx+1, len(trainloader), train_loss/(batch_idx+1)), end = '\r')
print('')
scheduler.step()
return net
def split_dataset(dataset, logger):
imgpath_dict = {os.path.basename(img_fn): j for j, img_fn in enumerate(dataset.imgs)}
df_covid = pd.read_excel("./CT-MetaInfo.xlsx", sheet_name='COVID-CT-info')
df_covid = df_covid[['File name', 'Patient ID']]
# process COVID
fn_covid = df_covid['File name'].tolist()[:NUM_COVID]
pid_covid = df_covid['Patient ID'].tolist()[:NUM_COVID]
pid_fn_dict = {}
for pid, fn in zip(pid_covid, fn_covid):
if pid in pid_fn_dict:
fn_exist = pid_fn_dict[pid]
fn_exist.append(fn)
pid_fn_dict[pid] = fn_exist
else:
pid_fn_dict[pid] = [fn]
pid_covid_unique = sorted(list(set(pid_covid)))
indices = torch.randperm(len(pid_covid_unique))
pid_train = indices[:int(NTRAIN_RATIO*len(pid_covid_unique))]
pid_test = indices[int(NTRAIN_RATIO*len(pid_covid_unique)):]
indices_covid_train = []
for i in pid_train:
pid = pid_covid_unique[i]
fns = pid_fn_dict[pid]
for fn in fns:
if fn in imgpath_dict:
indices_covid_train.append(imgpath_dict[fn])
else:
raise ValueError
indices_covid_test = []
for i in pid_test:
pid = pid_covid_unique[i]
fns = pid_fn_dict[pid]
for fn in fns:
if fn in imgpath_dict:
indices_covid_test.append(imgpath_dict[fn])
else:
raise ValueError
assert len(indices_covid_train) + len(indices_covid_test) == NUM_COVID
# process NonCOVID
df_noncovid = pd.read_excel("CT-MetaInfo.xlsx", sheet_name='NonCOVID-CT-info')
df_noncovid = df_noncovid[['image name', 'patient id']]
fn_noncovid = df_noncovid['image name'].tolist()[:NUM_NONCOVID]
pid_noncovid = df_noncovid['patient id'].tolist()[:NUM_NONCOVID]
pid_fn_dict = {}
for pid, fn in zip(pid_noncovid, fn_noncovid):
if pid in pid_fn_dict:
fn_exist = pid_fn_dict[pid]
fn_exist.append(fn)
pid_fn_dict[pid] = fn_exist
else:
pid_fn_dict[pid] = [fn]
pid_noncovid_unique = sorted(list(set(pid_noncovid)))
indices = torch.randperm(len(pid_noncovid_unique))
pid_train = indices[:int(NTRAIN_RATIO*len(pid_noncovid_unique))]
pid_test = indices[int(NTRAIN_RATIO*len(pid_noncovid_unique)):]
indices_noncovid_train = []
for i in pid_train:
pid = pid_noncovid_unique[i]
fns = pid_fn_dict[pid]
for fn in fns:
if fn in imgpath_dict:
indices_noncovid_train.append(imgpath_dict[fn])
else:
raise ValueError
indices_noncovid_test = []
for i in pid_test:
pid = pid_noncovid_unique[i]
fns = pid_fn_dict[pid]
for fn in fns:
if fn in imgpath_dict:
indices_noncovid_test.append(imgpath_dict[fn])
else:
raise ValueError
assert len(indices_noncovid_train) + len(indices_noncovid_test) == NUM_NONCOVID
indices_train = indices_covid_train+indices_noncovid_train
indices_test = indices_covid_test+indices_noncovid_test
assert len(indices_train) == len(set(indices_train))
assert len(indices_test) == len(set(indices_test))
assert len(list(set(indices_test) & set(indices_train))) == 0
trainset = torch.utils.data.Subset(dataset, indices_train)
testset = torch.utils.data.Subset(dataset, indices_test)
dataset.set_indices_train(indices_train)
print("The number of training CT images:", len(indices_train))
print("The number of test CT images:", len(indices_test))
if logger is not None:
logger.write("The number of training CT images:%d\n"%(len(indices_train)))
logger.write("The number of test CT images:%d\n"%(len(indices_test)))
return trainset, testset
def main():
logger, result_dir, _ = utils.config_backup_get_log(args,__file__)
device = utils.get_device()
utils.set_seed(args.seed, device) # set random seed
dataset = COVID19DataSet(root = args.datapath, ctonly = args.ctonly) # load dataset
trainset, testset = split_dataset(dataset = dataset, logger = logger)
if args.model.lower() in ['mobilenet']:
net = mobilenet_v2(task = 'classification', moco = False, ctonly = args.ctonly).to(device)
elif args.model.lower() in ['densenet']:
net = densenet121(task = 'classification', moco = False, ctonly = args.ctonly).to(device)
else:
raise Exception
criterion = torch.nn.BCEWithLogitsLoss()
optimizer = torch.optim.Adam(net.parameters(), lr=args.lr, weight_decay=1e-3)
scheduler = torch.optim.lr_scheduler.StepLR(optimizer, step_size=25, gamma=0.1)
trainloader = torch.utils.data.DataLoader(trainset, batch_size=args.bstrain, shuffle=True, num_workers = args.nworkers)
testloader = torch.utils.data.DataLoader(testset, batch_size=args.bstest, shuffle=False, num_workers = args.nworkers)
best_auroc = 0.
print('==> Start training ..')
start = time.time()
for epoch in range(args.maxepoch):
net = train(epoch, net, trainloader, criterion, optimizer, scheduler, args.model, device)
scheduler.step()
if epoch%5 == 0:
auroc, aupr, f1_score, accuracy = validate(net, testloader, device)
logger.write('Epoch:%3d | AUROC: %5.4f | AUPR: %5.4f | F1_Score: %5.4f | Accuracy: %5.4f\n'%(epoch, auroc, aupr, f1_score, accuracy))
if auroc>best_auroc:
best_auroc = auroc
best_aupr = aupr
best_epoch = epoch
print("save checkpoint...")
torch.save(net.state_dict(), './%s/%s.pth'%(result_dir, args.model))
auroc, aupr, f1_score, accuracy = validate(net, testloader, device)
logger.write('Epoch:%3d | AUROC: %5.4f | AUPR: %5.4f | F1_Score: %5.4f | Accuracy: %5.4f\n'%(epoch, auroc, aupr, f1_score, accuracy))
if args.batchout:
with open('temp_result.txt', 'w') as f:
f.write("%10.8f\n"%(best_auroc))
f.write("%10.8f\n"%(best_aupr))
f.write("%d"%(best_epoch))
end = time.time()
hours, rem = divmod(end-start, 3600)
minutes, seconds = divmod(rem, 60)
print("Elapsed Time: {:0>2}:{:0>2}:{:05.2f}".format(int(hours),int(minutes),seconds))
logger.write("Elapsed Time: {:0>2}:{:0>2}:{:05.2f}\n".format(int(hours),int(minutes),seconds))
return True
if __name__ == "__main__":
parser = argparse.ArgumentParser()
parser.add_argument('--datapath', type=str, default="/home/sean/data/COVID-CT QSR Data Challenge/COVID-CT QSR Data Challenge", help='data path')
parser.add_argument('--seed', type=int, default=1, help='random seed')
parser.add_argument('--model', type=str, default='mobilenet', help='backbone architecture mobilenet|densenet')
parser.add_argument('--bstrain', type=int, default=32, help='batch size for training')
parser.add_argument('--bstest', type=int, default=64, help='batch size for testing')
parser.add_argument('--nworkers', type=int, default=2, help='the number of workers used in DataLoader')
parser.add_argument('--maxepoch', type=int, default=100, help='the number of epoches')
parser.add_argument('--lr', type=float, default=0.001, help='learning rate')
parser.add_argument('--ctonly', action='store_true', help='using ctimages only for the input of the network')
parser.add_argument('--suffix', type=str, default='test', help='suffix of result directory')
parser.add_argument('--batchout', action='store_true', help='batch out')
args = parser.parse_args()
main()