-
Notifications
You must be signed in to change notification settings - Fork 2
/
CGNN_ours.py
375 lines (307 loc) · 15.6 KB
/
CGNN_ours.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
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
import numpy as np
from torch.utils.data import Dataset
import torch
from tqdm import tqdm
import os
import pandas as pd
import torch.optim as optim
from torch.optim import lr_scheduler
import argparse
import random
import shutil
from models.gcn import *
from patch_predict import AverageMeter
from sklearn.metrics import roc_auc_score, confusion_matrix, roc_curve
from utils import predict_binary
from tensorboardX import SummaryWriter
from time import gmtime, strftime, localtime
from itertools import cycle
from graph_construct import *
def seed_torch(seed=0):
random.seed(seed)
np.random.seed(seed)
os.environ['PYTHONHASHSEED'] = str(seed)
torch.manual_seed(seed)
torch.cuda.manual_seed(seed)
torch.backends.cudnn.deterministic = True
torch.backends.cudnn.benchmark = False
parser = argparse.ArgumentParser()
parser.add_argument('--model', default='GCN', type=str,
help='baseline of the model')
parser.add_argument('--start-epoch', default=0, type=int,
help='manual epoch number (useful on restarts)')
parser.add_argument('--n_epoch', default=10, type=int,
help='number of epoch to change')
parser.add_argument('--epoch', default=100, type=int,
help='number of total epochs to run')
parser.add_argument('--fold_index', default=0, type=int,
help='index of current fold')
parser.add_argument('--optimizer', default='Adam', type=str,
help='optimizer (Adam)')
parser.add_argument('--weight_decay', '--wd', default=1e-4, type=float,
help='weight decay (default: 1e-4)')
parser.add_argument('--num_classes', default=1, type=int,
help='numbers of classes (default: 2)')
parser.add_argument('--tensorboard', default=True,
help='Log progress to TensorBoard', action='store_true')
parser.add_argument('--use_cuda', default=True,
help='whether to use_cuda(default: True)')
parser.add_argument('--batch_size', default=20, type=int,
help='mini-batch size (default: 20)')
parser.add_argument('--lr', '--learning-rate', default=0.001, type=float,
help='initial learning rate')
parser.add_argument('--weight_bg', default=0.5, type=float,
help='weight of background negative data in loss function')
parser.add_argument("--result_path", default=None, type=str,
help="result path for cnn experiment.")
parser.add_argument("--result_path_gcn", default=None, type=str,
help="result path for gcn experiment.")
parser.add_argument("--gpu", type=str, default="0", metavar="N",
help="input visible devices for training (default: 0)")
parser.add_argument('--seed', default=2, type=int, help='random seed(default: 1)')
def train_gcn(train_loader, train_loader_bg, weight_bg, model, criterion, optimizer, epoch, use_cuda=True):
"""Train for one epoch on the training set"""
train_losses = AverageMeter()
train_acc = AverageMeter()
train_acc_bg = AverageMeter()
target_roc = torch.zeros((0, args.num_classes))
target_roc_bg = torch.zeros((0, args.num_classes))
# switch to train mode
model.train()
with tqdm(zip(train_loader, cycle(train_loader_bg)), ncols=130) as t:
for i, ([adjacency, feature, target, _], [adjacency_bg, feature_bg, target_bg, _]) in enumerate(t):
t.set_description("train epoch %s" % epoch)
if use_cuda:
target = target.type(torch.FloatTensor).cuda()
adjacency = adjacency.type(torch.FloatTensor).cuda()
feature = feature.type(torch.FloatTensor).cuda()
target_bg = target_bg.type(torch.FloatTensor).cuda()
adjacency_bg = adjacency_bg.type(torch.FloatTensor).cuda()
feature_bg = feature_bg.type(torch.FloatTensor).cuda()
optimizer.zero_grad()
output = model(feature, adjacency)
output_bg = model(feature_bg, adjacency_bg)
target_roc = torch.cat((target_roc, target.to(torch.float32).unsqueeze(1).data.cpu()), dim=0)
target_roc_bg = torch.cat((target_roc_bg, target_bg.to(torch.float32).unsqueeze(1).data.cpu()), dim=0)
train_loss = (criterion(output.view(output.shape[0]), target) + weight_bg * criterion(output_bg.view(output_bg.shape[0]), target_bg)) / (
weight_bg + 1)
train_losses.update(train_loss.item(), adjacency.size(0))
acc = accuracy(output.data, target)
train_acc.update(acc, adjacency.size(0))
acc_bg = accuracy(output_bg.data, target_bg)
train_acc_bg.update(acc_bg, adjacency_bg.size(0))
# compute gradient and do SGD step
train_loss.backward()
optimizer.step()
t.set_postfix({
'iter': '{i}'.format(i=i),
'loss': '{loss.val:.4f}({loss.avg:.4f})'.format(loss=train_losses),
'Acc': '{acc.val:.4f}({acc.avg:.4f})'.format(acc=train_acc)}
)
return train_losses, train_acc, train_acc_bg
def valid_gcn(val_loader, model, criterion, epoch, use_cuda=True):
"""valid for one epoch on the validation set"""
val_losses = AverageMeter()
val_acc = AverageMeter()
# switch to valid mode
model.eval()
target_roc = torch.zeros((0, args.num_classes))
output_roc = torch.zeros((0, args.num_classes))
with tqdm(val_loader, ncols=130) as t:
for i, (adjacency, feature, target, case_id) in enumerate(t):
t.set_description("train epoch %s" % epoch)
if use_cuda:
target = target.type(torch.FloatTensor).cuda()
adjacency = adjacency.type(torch.FloatTensor).cuda()
feature = feature.type(torch.FloatTensor).cuda()
# print('feature:', feature)
output = model(feature, adjacency)
val_loss = criterion(output.view(output.shape[0]), target)
val_losses.update(val_loss.item(), adjacency.size(0))
target_roc = torch.cat((target_roc, target.to(torch.float32).unsqueeze(1).data.cpu()), dim=0)
output_roc = torch.cat((output_roc, output.data.cpu()), dim=0)
acc = accuracy(output.data, target)
val_acc.update(acc, adjacency.size(0))
t.set_postfix({
'loss': '{loss.val:.4f}({loss.avg:.4f})'.format(loss=val_losses),
'Acc': '{acc.val:.4f}({acc.avg:.4f})'.format(acc=val_acc)}
)
AUROC = aucrocs(output_roc, target_roc)
print('The AUROC is %.4f' % AUROC)
return val_losses, val_acc, val_acc.avg, AUROC
def test(test_loader, model, mode, use_cuda=True):
probs = []
model.eval()
target_roc = torch.zeros((0, args.num_classes))
output_roc = torch.zeros((0, args.num_classes))
names = []
y_patient = []
with torch.no_grad():
with tqdm(test_loader, ncols=130) as t:
for i, (adjacency, feature, target, case_id) in enumerate(t):
t.set_description("Calculate {} probs:".format(mode))
if use_cuda:
adjacency = adjacency.type(torch.FloatTensor).cuda()
feature = feature.type(torch.FloatTensor).cuda()
names.extend(case_id)
# compute output
output = model(feature, adjacency)
output = output.detach().cpu()
target_roc = torch.cat((target_roc, target.to(torch.float32).unsqueeze(1).data.cpu()), dim=0)
output_roc = torch.cat((output_roc, output), dim=0)
for bs in range(output.shape[0]):
probs.append(output.numpy()[bs])
y_patient.append(target.numpy()[bs])
if mode != 'renmin' and mode != 'sixth':
AUROC = aucrocs(output_roc, target_roc)
print('The AUROC is %.4f' % AUROC)
y_patient = list(map(int, y_patient))
probs = np.array(probs)
if mode == 'valid':
return names, y_patient, probs, AUROC, target_roc, output_roc,
else:
return names, y_patient, probs
def aucrocs(output, target):
"""
Returns:
List of AUROCs of all classes.
"""
output_np = output.cpu().numpy()
target_np = target.cpu().numpy()
AUROCs=roc_auc_score(target_np[:, 0], output_np[:, 0])
return AUROCs
def save_checkpoint(state, is_best, epoch, fold):
"""Saves checkpoint to disk"""
# filename = 'checkpoint' + str(fold) + '_' + str(epoch) + '.pth.tar'
filename = 'checkpoint' + str(fold) + '.pth.tar'
directory = result_path_gcn + "/checkpoint_gcn/"
if not os.path.exists(directory):
os.makedirs(directory)
filename = directory + filename
torch.save(state, filename)
if is_best:
torch.save(state, filename)
shutil.copyfile(filename, directory + 'model_best' + str(fold) + '.pth.tar')
def save_probs_pred(result_path_gcn, names, y, probs, pred, mode):
patch_prob_filename = os.path.join(result_path_gcn, '{}_test_results_{}.csv'.format(args.fold_index, mode))
info = []
for i in range(len(names)):
info.append([names[i], y[i], int(pred[i]), probs[i][0]])
df = pd.DataFrame(info, columns=['id', 'label', 'pred', 'prob'])
df.to_csv(patch_prob_filename, index=False)
def save_result_matrix(true,pred,auc,mode,time,save_result):
tn,fp,fn,tp = confusion_matrix(true, pred, labels=[0, 1]).ravel()
acc = (tn+tp)/(tn+fp+fn+tp)
sen = tp/(tp+fn)
spe = tn/(tn+fp)
NPV = tn/(tn+fn)
PPV = tp/(tp+fp)
print('{}: {} tn:{},fp:{},fn:{},tp:{}'.format(mode,time,tn,fp,fn,tp))
print('{}: {} acc is {}'.format(mode,time,acc))
print('{}: {} sen is {}, spe is {}'.format(mode,time,sen,spe))
save_result.write(' '+','+mode+','+str(acc)+','+str(sen)+','+str(spe)
+','+str(tn)+','+str(tp)+','+str(fn)+','+str(fp)+','+str(NPV)+','+str(PPV)+','+str(auc)+'\n')
def accuracy(output, target):
output_np = output.cpu().numpy()
pred = predict_binary(output_np, 0.5)
target_np = target.cpu().numpy()
right = (pred.squeeze() == target_np)
acc = np.sum(right) / output.shape[0]
return acc
if __name__ == '__main__':
global use_cuda
args = parser.parse_args()
use_cuda = args.use_cuda and torch.cuda.is_available()
gpu = args.gpu
os.environ["CUDA_VISIBLE_DEVICES"] = gpu
learning_rate = args.lr
weight_decay = args.weight_decay
total_epoch = args.epoch
start_epoch = args.start_epoch
batch_size = args.batch_size
lr_factor = 0.1
patience = 10
seed_torch(args.seed)
weight_bg = args.weight_bg
result_basepath = args.result_path
run_name = strftime("%Y%m%d_%H%M%S", localtime())
result_path_gcn = os.path.join(result_basepath, args.result_path_gcn)
fold = args.fold_index
# get patches and create graphs
print("fold: ", fold)
model = GCN_noGaussian()
model = model.cuda()
criterion = nn.BCELoss(reduction='mean')
optimizer = optim.Adam(model.parameters(), lr=learning_rate, weight_decay=weight_decay)
model_path = os.path.join(result_path_gcn, 'checkpoint_gcn', 'checkpoint' + str(fold) + '.pth.tar')
if os.path.isfile(model_path):
print("=> loading checkpoint '{}'".format(model_path))
checkpoint = torch.load(model_path)
pretrained_dict = checkpoint['state_dict']
model.load_state_dict(pretrained_dict)
start_epoch = checkpoint['epoch']
print("============ start epoch:{} ==============".format(start_epoch))
else:
print("=> no checkpoint found at '{}'".format(model_path))
writer = SummaryWriter(result_path_gcn + '/tensorboard_gcn/')
print("Creat Graphs...")
patient_id_train, adjacency_train, node_feat_train, patient_y_train, id_bg, adjacency_bg, node_feat_bg, patient_y_bg = get_graph_data(fold, 'train', result_basepath)
patient_id_val, adjacency_val, node_feat_val, patient_y_val = get_graph_data(fold, 'valid', result_basepath)
train_datasets = graph_dataset(patient_id_train, adjacency_train, node_feat_train, patient_y_train)
train_datasets_bg = graph_dataset(id_bg, adjacency_bg, node_feat_bg, patient_y_bg)
valid_datasets = graph_dataset(patient_id_val, adjacency_val, node_feat_val, patient_y_val)
kwargs = {'num_workers': 0, 'pin_memory': True}
train_loader = torch.utils.data.DataLoader(dataset=train_datasets, batch_size=batch_size, shuffle=True, **kwargs)
train_loader_bg = torch.utils.data.DataLoader(dataset=train_datasets_bg, batch_size=batch_size, shuffle=True,
**kwargs)
val_loader = torch.utils.data.DataLoader(dataset=valid_datasets, batch_size=batch_size, shuffle=True, **kwargs)
print("Start Training...")
best_prec = 0
scheduler = lr_scheduler.ReduceLROnPlateau(optimizer, 'min', factor=lr_factor, patience=patience, min_lr=0.00001)
for epoch in range(total_epoch):
# train for one epoch
train_losses, train_acc, train_acc_bg = train_gcn(train_loader, train_loader_bg, weight_bg, model, criterion, optimizer, epoch)
val_losses, val_acc, prec1, val_AUC = valid_gcn(val_loader, model, criterion, epoch)
scheduler.step(val_losses.avg)
writer.add_scalars('data' + str(fold) + '/loss',
{'train_loss': train_losses.avg, 'val_loss': val_losses.avg}, epoch)
writer.add_scalars('data' + str(fold) + '/Accuracy',
{'train_acc': train_acc.avg, 'train_acc_bg': train_acc_bg.avg, 'val_acc': val_acc.avg}, epoch)
for param_group in optimizer.param_groups:
lr = param_group['lr']
print('learining rate:', lr)
writer.add_scalar('learning_rate', lr, epoch)
is_best = prec1 > best_prec
if is_best == 1:
best_prec = max(prec1, best_prec)
save_checkpoint({
'epoch': epoch + 1,
'state_dict': model.state_dict(),
'best_prec1': best_prec,
}, is_best, epoch, fold)
print("Start Testing...")
save_filename = result_path_gcn + '/result_gcn_{}.csv'.format(fold)
save_result = open(save_filename, 'a')
save_result.write('gcn classification result, ,acc,sen,spe,tn,tp,fn,fp,NPV,PPV,AUC' + '\n')
train_names, train_y, train_probs, train_AUC = test(train_loader, model, mode='train')
train_pred = predict_binary(train_probs, 0.5)
train_y = list(map(int, train_y))
train_pred = list(map(int, train_pred))
save_result_matrix(train_y, train_pred, train_AUC, 'train', 'gcn classify', save_result)
save_probs_pred(result_path_gcn, train_names, train_y, train_probs, train_pred, mode='train')
train_names_bg, train_y_bg, train_probs_bg, train_bg_AUC = test(train_loader_bg, model, mode='train_bg')
train_pred_bg = predict_binary(train_probs_bg, 0.5)
train_y_bg = list(map(int, train_y_bg))
train_pred_bg = list(map(int, train_pred_bg))
save_result_matrix(train_y_bg, train_pred_bg, train_bg_AUC, 'train_bg', 'gcn classify', save_result)
save_probs_pred(result_path_gcn, train_names_bg, train_y_bg, train_probs_bg, train_pred_bg,
mode='train_bg')
val_names, val_y, val_probs, val_AUC, val_target, val_output = test(val_loader, model, mode='valid')
val_pred = predict_binary(val_probs, 0.5)
val_y = list(map(int, val_y))
val_pred = list(map(int, val_pred))
save_result_matrix(val_y, val_pred, val_AUC, 'validate', 'gcn classify', save_result)
save_probs_pred(result_path_gcn, val_names, val_y, val_probs, val_pred, mode='valid')
np.save(os.path.join(result_path_gcn, 'target_val_fold{}.npy'.format(args.fold_index)), val_target)
np.save(os.path.join(result_path_gcn, 'pred_val_fold{}.npy'.format(args.fold_index)), val_output)
save_result.close()