forked from ananyajana/fibrosis_code
/
train.py
executable file
·328 lines (269 loc) · 12 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
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
323
324
325
326
327
328
import shutil
import time
import os
import math
import json
import torch
import torch.nn as nn
import torch.nn.parallel
import torch.backends.cudnn as cudnn
import torch.optim
import torchvision.transforms as transforms
from models import BaselineNet, BaselineNet2
import numpy as np
from sklearn import metrics
from options import Options
from dataset import LiverDataset
import utils
def main():
global best_score, logger, logger_results, slide_weights
opt = Options(isTrain=True)
opt.parse()
opt.save_options()
# tb_writer = SummaryWriter('{:s}/tb_logs'.format(opt.train['save_dir']))
os.environ['CUDA_VISIBLE_DEVICES'] = ','.join(str(x) for x in opt.train['gpus'])
# set up logger
logger, logger_results = utils.setup_logger(opt)
opt.print_options(logger)
if opt.train['random_seed'] >= 0:
# logger.info("=> Using random seed {:d}".format(opt.random_seed))
torch.manual_seed(opt.train['random_seed'])
torch.cuda.manual_seed(opt.train['random_seed'])
torch.backends.cudnn.deterministic = True
torch.backends.cudnn.benchmark = False
np.random.seed(opt.train['random_seed'])
else:
torch.backends.cudnn.benchmark = True
# ---------- Create model ---------- #
#print('in train file', opt.model['use_resnet'])
if opt.model['use_resnet'] == 1:
#print('opt model use_resnet ', type(opt.model['use_resnet']))
#print('choosing baseline2')
model = BaselineNet2(opt.model['out_c'], opt.model['resnet_layers'], opt.model['train_res4'])
logger.info('choosing BaselineNet2(resnet based) model')
else:
model = BaselineNet(opt.model['in_c'], opt.model['out_c'])
logger.info('choosing BaselineNet model')
if opt.model['pre_train_sup'] == 1:
print("=> loading pre-trained model from path", opt.model['sup_model_path'])
model.set_fea_extractor(opt.model['sup_model_path'])
model = model.cuda()
# logger.info(model)
# ---------- End create model ---------- #
# define loss function (criterion) and optimizer
criterion = nn.CrossEntropyLoss().cuda()
optimizer = torch.optim.Adam(model.parameters(), opt.train['lr'], weight_decay=opt.train['weight_decay'])
# ---------- Data loading ---------- #
data_transform = transforms.Compose([
transforms.Resize((224, 224)),
transforms.ToTensor(),
transforms.Normalize(mean=[0.485],
std=[0.229])
])
fold_num = opt.exp_num.split('_')[-1]
logger.info('Fold number: {:s}'.format(fold_num))
if opt.model['use_resnet'] == 1:
train_set = LiverDataset('{:s}/train{:s}.h5'.format(opt.train['data_dir'], fold_num), data_transform, opt)
test_set = LiverDataset('{:s}/test{:s}.h5'.format(opt.train['data_dir'], fold_num), data_transform, opt)
else:
train_set = LiverDataset('{:s}/train{:s}.h5'.format(opt.train['data_dir'], fold_num), data_transform)
test_set = LiverDataset('{:s}/test{:s}.h5'.format(opt.train['data_dir'], fold_num), data_transform)
# loading from the pre-trained model
#print("=> loading pre-trained model from path", opt.model['sup_model_path'])
#model.set_fea_extractor(opt.model['sup_model_path'])
# ---------- End Data loading ---------- #
# ----- optionally load from a checkpoint ----- #
# if opt.train['checkpoint']:
# model_state_dict, optimizer_state_dict = load_checkpoint(opt.train['checkpoint'])
# model.load_state_dict(model_state_dict)
# optimizer.load_state_dict(optimizer_state_dict)
# ----- End checkpoint loading ----- #
# ----- Start training ---- #
best_score = 0
for epoch in range(opt.train['epochs']):
# train and validate for one epoch
train_loss, train_acc = train(opt, train_set, model, criterion, optimizer, epoch)
test_loss, test_acc, test_auc = test(opt, test_set, model, criterion, epoch)
# remember best accuracy and save checkpoint
is_best = test_auc > best_score
best_score = max(test_auc, best_score)
cp_flag = False
save_checkpoint({
'epoch': epoch + 1,
'state_dict': model.state_dict(),
'best_score': best_score,
'optimizer': optimizer.state_dict(),
}, is_best, opt.train['save_dir'], cp_flag, epoch+1)
# save training results
logger_results.info('{:<6d}| {:<12.4f}{:<12.4f}|| {:<12.4f}{:<12.4f}{:<12.4f}'
.format(epoch, train_loss, train_acc,
test_loss, test_acc, test_auc))
for i in list(logger.handlers):
logger.removeHandler(i)
i.flush()
i.close()
for i in list(logger_results.handlers):
logger_results.removeHandler(i)
i.flush()
i.close()
def train(opt, train_set, model, criterion, optimizer, epoch):
batch_time = utils.AverageMeter()
data_time = utils.AverageMeter()
losses = utils.AverageMeter()
acc = utils.AverageMeter()
# switch to train mode
model.train()
end = time.time()
idx_list = torch.randperm(len(train_set)).tolist()
# idx_list = torch.multinomial(torch.Tensor(slide_weights), len(train_set), replacement=True)
for i in range(0, len(idx_list), opt.train['batch_size']):
N = min(opt.train['batch_size'], len(idx_list) - i)
loss = 0
for k in range(N):
idx = idx_list[i + k]
ct_data, fib_label, nas_stea_label, nas_lob_label, nas_balloon_label = train_set[idx]
if opt.exp == 'fib':
label = fib_label
elif opt.exp == 'nas_stea':
label = nas_stea_label
elif opt.exp == 'nas_lob':
label = nas_lob_label
elif opt.exp == 'nas_balloon':
label = nas_balloon_label
else:
raise ValueError('Wrong label name')
input_data = ct_data
output = model(input_data.cuda())
#raise ValueError('Exit')
loss += criterion(output, label.cuda())
# measure accuracy
probs = nn.functional.softmax(output, dim=1)
pred = torch.argmax(probs, dim=1).cpu()
accuracy = (pred == label).sum().numpy()
acc.update(accuracy)
del output
loss /= N
losses.update(loss.item(), N)
# compute gradient and do SGD step
optimizer.zero_grad()
loss.backward()
optimizer.step()
# measure elapsed time
batch_time.update(time.time() - end)
end = time.time()
if i % opt.train['log_interval'] == 0:
logger.info('Epoch: [{0}][{1}/{2}]\t'
'Data Time: {data_time.avg:.3f}\t'
'Batch Time: {batch_time.avg:.3f}\t'
'Loss: {loss.avg:.3f}\t'
'Acc: {acc.avg:.4f}'
.format(epoch, i, len(train_set), data_time=data_time, batch_time=batch_time,
loss=losses, acc=acc))
logger.info('=> Train Avg: Loss: {loss.avg:.3f}\t\tAcc: {acc.avg:.4f}'
.format(loss=losses, acc=acc))
return losses.avg, acc.avg
def test(opt, test_set, model, criterion, epoch):
batch_time = utils.AverageMeter()
losses = utils.AverageMeter()
acc = utils.AverageMeter()
# switch to evaluate mode
model.eval()
end = time.time()
slide_probs_all = []
slide_targets_all = []
for i in range(len(test_set)):
ct_data, fib_label, nas_stea_label, nas_lob_label, nas_balloon_label = test_set[i]
if opt.exp == 'fib':
label = fib_label
elif opt.exp == 'nas_stea':
label = nas_stea_label
elif opt.exp == 'nas_lob':
label = nas_lob_label
elif opt.exp == 'nas_balloon':
label = nas_balloon_label
else:
raise ValueError('Wrong label name')
with torch.no_grad():
input_data = ct_data
output = model(input_data.cuda())
loss = criterion(output, label.cuda())
# measure accuracy and record loss
probs = nn.functional.softmax(output, dim=1)
pred = torch.argmax(probs, dim=1).detach().cpu()
accuracy = (pred == label).sum().detach().cpu().numpy()
# measure accuracy and record loss
losses.update(loss.item())
acc.update(accuracy)
slide_probs_all.append(probs.detach().cpu())
slide_targets_all.append(label.detach().cpu())
# measure elapsed time
batch_time.update(time.time() - end)
end = time.time()
del output, probs
if i % opt.train['log_interval'] == 0:
logger.info('Test: [{0}][{1}/{2}]\t'
'Time: {batch_time.avg:.3f}\t'
'Loss: {loss.avg:.3f}'
.format(epoch, i, len(test_set), batch_time=batch_time, loss=losses))
slide_probs_all = torch.cat(slide_probs_all, dim=0).numpy()
slide_targets_all = torch.cat(slide_targets_all, dim=0).numpy()
pred = np.argmax(slide_probs_all, axis=1)
acc = metrics.accuracy_score(slide_targets_all, pred)
if opt.exp in ['fib', 'nas_lob', 'nas_balloon']:
# Compute ROC curve and ROC area for each class
fpr = dict()
tpr = dict()
roc_auc = dict()
for i in range(3):
fpr[i], tpr[i], _ = metrics.roc_curve(slide_targets_all == i, slide_probs_all[:, i])
auc_i = metrics.auc(fpr[i], tpr[i])
roc_auc[i] = 0 if math.isnan(auc_i) else auc_i
auc = np.mean(np.array(list(roc_auc.values())))
logger.info('Test Avg: {}\tLoss: {:.3f}\tAcc: {:.4f}\tAUC: {:.4f}\n'
'AUC0: {:.4f}\tAUC1: {:.4f}\tAUC2: {:.4f}\n'
.format(epoch, losses.avg, acc, auc, roc_auc[0], roc_auc[1], roc_auc[2]))
else:
tp = np.sum((pred == 1) * (slide_targets_all == 1))
tn = np.sum((pred == 0) * (slide_targets_all == 0))
fp = np.sum((pred == 1) * (slide_targets_all == 0))
fn = np.sum((pred == 0) * (slide_targets_all == 1))
acc = metrics.accuracy_score(slide_targets_all, pred)
auc = metrics.roc_auc_score(slide_targets_all, slide_probs_all[:, 1])
logger.info('Test Avg: {}\tLoss: {:.3f}\tAcc: {:.4f}\tAUC: {:.4f}\n'
'TP: {:d}\tTN: {:d}\tFP: {:d}\tFN: {:d}\n'
.format(epoch, losses.avg, acc, auc, tp, tn, fp, fn))
return losses.avg, acc, auc
def save_checkpoint(state, is_best, save_dir, cp_flag, epoch):
if not os.path.exists(save_dir):
#os.mkdir(save_dir)
os.makedirs(save_dir, exist_ok=True)
if is_best:
print('saving mode at {}'.format(save_dir))
torch.save(state, '{:s}/checkpoint_best.pth.tar'.format(save_dir))
# filename = '{:s}/checkpoint.pth.tar'.format(save_dir)
# torch.save(state, filename)
# if cp_flag:
# shutil.copyfile(filename, '{:s}/checkpoint_{:d}.pth.tar'.format(save_dir, epoch))
# if is_best:
# shutil.copyfile(filename, '{:s}/checkpoint_best.pth.tar'.format(save_dir))
def load_checkpoint(checkpoint_path):
model_state_dict = None
optimizer_state_dict = None
if os.path.isfile(checkpoint_path):
logger.info("=> loading checkpoint '{}'".format(checkpoint_path))
checkpoint = torch.load(checkpoint_path)
model_state_dict = checkpoint['state_dict']
optimizer_state_dict = checkpoint['optimizer']
logger.info("=> loaded checkpoint '{}' (epoch {})".format(checkpoint_path, checkpoint['epoch']))
else:
logger.info("=> no checkpoint found at '{}'".format(checkpoint_path))
return model_state_dict, optimizer_state_dict
def assign_slide_weight(bags, weights):
slide_weights = np.zeros(len(bags))
for i in range(len(bags)):
bag = bags[i]
target = bag['label']
slide_weights[i] = weights[0] if target == 0 else weights[1]
return slide_weights
if __name__ == '__main__':
main()