-
Notifications
You must be signed in to change notification settings - Fork 1
/
validate_sweep.py
259 lines (219 loc) · 14.9 KB
/
validate_sweep.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
import os
import csv
import yaml
import numpy as np
import pandas as pd
import random
import torch
import torch.nn as nn
import torchio as tio
import monai
from tqdm import tqdm
from torch.utils.data import DataLoader
from sklearn.model_selection import train_test_split
from sklearn.metrics import roc_auc_score, accuracy_score, precision_score, recall_score
def get_yaml(f):
with open(f) as f:
yaml_dict = yaml.safe_load(f)
return yaml_dict
def set_seed(seed):
random.seed(seed)
os.environ["PYTHONHASHSEED"] = str(seed)
np.random.seed(seed)
torch.manual_seed(seed)
if torch.cuda.is_available():
torch.cuda.manual_seed_all(seed)
torch.backends.cudnn.deterministic = True
def main(config_path):
config = get_yaml(config_path)
transforms = [
tio.RescaleIntensity(out_min_max=(0.0,1.0)),
tio.CropOrPad(config['image_size']),
]
augment_transforms = [
tio.RandomFlip(axes='LR'),
tio.RandomAffine(scales=(0.9,1.2), degrees=10, isotropic=True, image_interpolation='nearest'),
]
train_transform = tio.Compose(transforms+augment_transforms)
test_transform = tio.Compose(transforms)
df = pd.read_csv(os.path.join(config['data_dir'], 'train_labels.csv'), index_col=0, dtype='string')
# start grid search validation
for seed in config['sweeps']['seed']:
for train_datasource in config['sweeps']['train_datasource']:
merged_subject_list = [s for s in os.listdir(os.path.join(config['data_dir'], 'nifti_reg', 'train')) if not s in config['exclusion']]
snuh_subject_list = [s for s in merged_subject_list if len(s)==8]
rsnamiccai_subject_list = [s for s in merged_subject_list if len(s)==5]
merged_label_list = [float(df.loc[df.index==subject]['MGMT_value']) for subject in merged_subject_list]
snuh_label_list = [float(df.loc[df.index==subject]['MGMT_value']) for subject in snuh_subject_list]
rsnamiccai_label_list = [float(df.loc[df.index==subject]['MGMT_value']) for subject in rsnamiccai_subject_list]
if train_datasource == 'rsnamiccai':
train_subject_list, val_subject_list, train_label_list, val_label_list = train_test_split(rsnamiccai_subject_list, rsnamiccai_label_list, test_size=0.1, random_state=seed, shuffle=True, stratify=rsnamiccai_label_list)
test_subject_list, test_label_list = snuh_subject_list, snuh_label_list
elif train_datasource == 'snuh':
train_subject_list, val_subject_list, train_label_list, val_label_list = train_test_split(snuh_subject_list, snuh_label_list, test_size=0.1, random_state=seed, shuffle=True, stratify=snuh_label_list)
test_subject_list, test_label_list = rsnamiccai_subject_list, rsnamiccai_label_list
else:
train_subject_list, test_subject_list, train_label_list, test_label_list = train_test_split(merged_subject_list, merged_label_list, test_size=0.1, random_state=seed, shuffle=True, stratify=merged_label_list)
train_subject_list, val_subject_list, train_label_list, val_label_list = train_test_split(train_subject_list, train_label_list, test_size=0.1, random_state=seed, shuffle=True, stratify=train_label_list)
for sequence in config['sweeps']['sequence']:
for model_name in config['sweeps']['model']:
model_class = getattr(monai.networks.nets, model_name)
set_seed(seed)
expname = f'seed-{seed}_traindatasource-{train_datasource}_model-{model_name}_sequence-{"".join(sequence)}'
print('='*70)
print(expname)
print('='*70)
result_dir = os.path.join(config['result_dir'], expname)
os.makedirs(result_dir, exist_ok=True)
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
criterion = nn.BCEWithLogitsLoss()
train_dataset = tio.SubjectsDataset([
tio.Subject(
FLAIR=tio.ScalarImage(os.path.join(config['data_dir'], 'nifti_reg', 'train', subject, 'flair.nii.gz')),
T1w=tio.ScalarImage(os.path.join(config['data_dir'], 'nifti_reg', 'train', subject, 't1w.nii.gz')),
T1wCE=tio.ScalarImage(os.path.join(config['data_dir'], 'nifti_reg', 'train', subject, 't1wce.nii.gz')),
T2w=tio.ScalarImage(os.path.join(config['data_dir'], 'nifti_reg', 'train', subject, 't2w.nii.gz')),
subject=subject,
label=label) for subject, label in zip(train_subject_list, train_label_list)], transform=train_transform)
val_dataset = tio.SubjectsDataset([
tio.Subject(
FLAIR=tio.ScalarImage(os.path.join(config['data_dir'], 'nifti_reg', 'train', subject, 'flair.nii.gz')),
T1w=tio.ScalarImage(os.path.join(config['data_dir'], 'nifti_reg', 'train', subject, 't1w.nii.gz')),
T1wCE=tio.ScalarImage(os.path.join(config['data_dir'], 'nifti_reg', 'train', subject, 't1wce.nii.gz')),
T2w=tio.ScalarImage(os.path.join(config['data_dir'], 'nifti_reg', 'train', subject, 't2w.nii.gz')),
subject=subject,
label=label) for subject, label in zip(val_subject_list, val_label_list)], transform=test_transform)
test_dataset = tio.SubjectsDataset([
tio.Subject(
FLAIR=tio.ScalarImage(os.path.join(config['data_dir'], 'nifti_reg', 'train', subject, 'flair.nii.gz')),
T1w=tio.ScalarImage(os.path.join(config['data_dir'], 'nifti_reg', 'train', subject, 't1w.nii.gz')),
T1wCE=tio.ScalarImage(os.path.join(config['data_dir'], 'nifti_reg', 'train', subject, 't1wce.nii.gz')),
T2w=tio.ScalarImage(os.path.join(config['data_dir'], 'nifti_reg', 'train', subject, 't2w.nii.gz')),
subject=subject,
label=label) for subject, label in zip(test_subject_list, test_label_list)], transform=test_transform)
train_dataloader = DataLoader(train_dataset, batch_size=config['batch_size'], shuffle=True, pin_memory=True, num_workers=config['num_workers'])
val_dataloader = DataLoader(val_dataset, batch_size=1, shuffle=False, pin_memory=True, num_workers=config['num_workers'])
test_dataloader = DataLoader(test_dataset, batch_size=1, shuffle=False, pin_memory=True, num_workers=config['num_workers'])
if model_class is monai.networks.nets.EfficientNetBN:
model = model_class('efficientnet-b0', in_channels=len(sequence), num_classes=1, spatial_dims=3, pretrained=False)
elif model_class is monai.networks.nets.DenseNet121:
model = model_class(in_channels=len(sequence), out_channels=1, spatial_dims=3, pretrained=False)
else:
model = model_class(in_channels=len(sequence), num_classes=1, spatial_dims=3, pretrained=False)
model.to(device)
optimizer = torch.optim.SGD(model.parameters(), lr=1e-3)
scheduler = torch.optim.lr_scheduler.ReduceLROnPlateau(optimizer, mode='min', factor=0.5, patience=3, verbose=True)
best_auroc = 0.0
best_loss = 1e9
best_accuracy = 0.0
stop_patience = 0
for epoch in range(100):
train_pred = []
train_prob = []
train_label = []
train_loss = []
model.train()
for x in tqdm(train_dataloader, ncols=60, desc=str(epoch)):
volume = torch.cat([x[s][tio.DATA] for s in sequence], axis=1).to(device)
label = x['label'].to(device)
# forward
logit = model(volume)
loss = criterion(logit.squeeze(1), (label-0.01).abs())
optimizer.zero_grad()
loss.backward()
optimizer.step()
train_loss.append(loss.detach().cpu().numpy())
train_pred.append(logit.sigmoid().round().detach().cpu().numpy())
train_prob.append(logit.sigmoid().detach().cpu().numpy())
train_label.append(label.detach().cpu().numpy())
torch.cuda.empty_cache()
train_loss = np.mean(train_loss)
train_auroc = roc_auc_score(np.concatenate(train_label), np.concatenate(train_prob))
train_accuracy = accuracy_score(np.concatenate(train_label), np.concatenate(train_pred))
val_prob = []
val_pred = []
val_label = []
val_loss = []
model.eval()
with torch.no_grad():
for x in val_dataloader:
volume = torch.cat([x[s][tio.DATA] for s in sequence], axis=1).to(device)
label = x['label'].to(device)
# forward
logit = model(volume)
loss = criterion(logit.squeeze(1), label)
val_loss.append(loss.detach().cpu().numpy())
val_pred.append(logit.sigmoid().round().detach().cpu().numpy())
val_prob.append(logit.sigmoid().detach().cpu().numpy())
val_label.append(label.detach().cpu().numpy())
torch.cuda.empty_cache()
val_loss = np.mean(val_loss)
val_auroc = roc_auc_score(np.concatenate(val_label), np.concatenate(val_prob))
val_accuracy = accuracy_score(np.concatenate(val_label), np.concatenate(val_pred))
val_precision = precision_score(np.concatenate(val_label), np.concatenate(val_pred))
val_recall = recall_score(np.concatenate(val_label), np.concatenate(val_pred))
scheduler.step(val_loss)
with open(os.path.join(config['result_dir'], 'train_log.csv'), 'a') as f:
f.write(','.join([str(epoch), str(train_loss), str(train_accuracy), str(train_auroc), str(val_loss), str(val_accuracy), str(val_auroc)]))
f.write('\n')
if val_accuracy > best_accuracy:
stop_patience = 0
best_loss = val_loss
best_auroc = val_auroc
best_accuracy = val_accuracy
best_precision = val_precision
best_recall = val_recall
torch.save(model.state_dict(), os.path.join(result_dir, f'model_epoch{epoch}_bestacc.pth'))
print(f'>>>>>> best model saved with loss {best_loss:.4f} / auroc {best_auroc:.4f} / acc {best_accuracy:.4f}')
test_prob = []
test_pred = []
test_label = []
test_loss = []
model.eval()
with torch.no_grad():
for x in test_dataloader:
volume = torch.cat([x[s][tio.DATA] for s in sequence], axis=1).to(device)
label = x['label'].to(device)
# forward
logit = model(volume)
loss = criterion(logit.squeeze(1), label)
test_loss.append(loss.detach().cpu().numpy())
test_pred.append(logit.sigmoid().round().detach().cpu().numpy())
test_prob.append(logit.sigmoid().detach().cpu().numpy())
test_label.append(label.detach().cpu().numpy())
torch.cuda.empty_cache()
test_loss = np.mean(test_loss)
test_auroc = roc_auc_score(np.concatenate(test_label), np.concatenate(test_prob))
test_accuracy = accuracy_score(np.concatenate(test_label), np.concatenate(test_pred))
test_precision = precision_score(np.concatenate(test_label), np.concatenate(test_pred))
test_recall = recall_score(np.concatenate(test_label), np.concatenate(test_pred))
print(f'>>>>>> test performance: loss {test_loss:.4f} / auroc {test_auroc:.4f} / acc {test_accuracy:.4f}')
result_dict = {
'seed': seed,
'model': repr(model_class),
'sequence': '-'.join(sequence),
'train_datasource': train_datasource,
'epoch': epoch,
'test_auroc': test_auroc,
'test_accuracy': test_accuracy,
'test_precision': test_precision,
'test_recall': test_recall,
'val_auroc': best_auroc,
'val_accuracy': best_accuracy,
'val_precision': best_precision,
'val_recall': best_recall,
}
with open(os.path.join(result_dir, 'result.csv'), 'w') as f:
w = csv.writer(f)
w.writerow(result_dict.keys())
w.writerow(result_dict.values())
else:
stop_patience += 1
if stop_patience >= 15:
print('='*70)
print(f'>>>>>> STOPPING AT EPOCH {epoch}')
print('='*70)
break
if __name__ == '__main__':
main('config.yaml')
exit(0)