-
Notifications
You must be signed in to change notification settings - Fork 2
/
Copy pathxadlime_adpen.py
400 lines (354 loc) · 14.8 KB
/
xadlime_adpen.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
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
import torch.optim as optim
import os
import datetime
import argparse
from tqdm import tqdm
import warnings
warnings.filterwarnings("ignore")
from models import ADPENVAE
from models import ADPENSOM
from losses import *
from helpers import *
# Define Arguments
parser = argparse.ArgumentParser()
parser.add_argument("--gpu_id", type=str, default="0")
parser.add_argument("--fold", type=int, default=1)
parser.add_argument("--epoch", type=int, default=500)
parser.add_argument("--lr", type=float, default=0.001)
parser.add_argument("--lambda1", type=float, default=0.00001)
parser.add_argument("--lambda2", type=float, default=0.0001)
parser.add_argument("--som_start_epoch", type=float, default=0)
parser.add_argument("--som_end_epoch", type=float, default=500)
# parser.add_argument("--som_map_size", nargs='+', type=int, default=[1, 1, 100]) # 1D Topology
parser.add_argument("--som_map_size", nargs='+', type=int, default=[1, 5, 20]) # 2D Topology
# parser.add_argument("--som_map_size", nargs='+', type=int, default=[2, 2, 25]) # 3D Topology
# parser.add_argument("--som_Tmin", type=float, default=2.0)
# parser.add_argument("--som_Tmax", type=float, default=20.0)
parser.add_argument("--som_dim", type=int, default=3)
parser.add_argument("--som_decay", type=str, default='exponential')
parser.add_argument("--z_dim", type=int, default=3)
parser.add_argument("--weight_vae", type=float, default=1.0)
parser.add_argument("--weight_order", type=float, default=1.0)
parser.add_argument("--weight_som", type=float, default=0.01)
parser.add_argument("--margin", type=float, default=1.0)
parser.add_argument("--dataset", type=str, default='adni_smri')
parser.add_argument("--load_images", type=bool, default=False)
parser.add_argument("--finetune", type=int, default=0) # 0: False (Pretrain), 1: True (Finetune)
parser.add_argument("--note", type=str, default='ADPEN')
args = parser.parse_args()
args.vae_hiddens = [10, 16, 8, 3]
args.task = ['CN', 'SMCI', 'PMCI', 'AD']
args.spectrum_dim = 4
args.batch_size = 4
if args.finetune:
args.som_Tmin = 0.8
args.som_Tmax = 2.0
args.weight_som = 1.0
if (args.som_map_size[0]==1 and args.som_map_size[1]==5 and args.som_map_size[2]==20):
directories_vaesom = ['-',
'-',
'-',
'-',
'-',]
log_dir_vaesom = directories_vaesom[args.fold-1]
else:
args.som_Tmin = 2.0
args.som_Tmax = 20.0
# GPU Configuration
gpu_id = args.gpu_id
os.environ["CUDA_VISIBLE_DEVICES"] = gpu_id
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
args.device = device
# Logging purpose
date_str = str(datetime.datetime.now().strftime('%Y%m%d.%H.%M.%S'))
if args.finetune:
directory = 'log/ADPEN/ADPEN_FINETUNE_%s_GPU%s_%d/' % (date_str, args.gpu_id, args.fold)
else:
directory = 'log/ADPEN/ADPEN_%s_GPU%s_%d/' % (date_str, args.gpu_id, args.fold)
if not os.path.exists(directory):
os.makedirs(directory)
os.makedirs(directory + 'img/')
os.makedirs(directory + 'img/train/')
os.makedirs(directory + 'img/valid/')
os.makedirs(directory + 'img/test/')
os.makedirs(directory + 'tflog/')
os.makedirs(directory + 'model/')
# Text Logging
f = open(directory + 'setting.log', 'a')
writelog(f, '======================')
writelog(f, 'Lambda1: %.5f' % args.lambda1)
writelog(f, 'Lambda2: %.5f' % args.lambda2)
writelog(f, '----------------------')
writelog(f, 'SOM')
writelog(f, 'SOM MAP Size: %d x %d x %d' % (args.som_map_size[0], args.som_map_size[1], args.som_map_size[2]))
writelog(f, 'SOM Dim: %d' % args.som_dim)
writelog(f, 'SOM Distance Metric: Euclidean')
writelog(f, 'SOM Tmin: %.5f' % args.som_Tmin)
writelog(f, 'SOM Tmax: %.5f' % args.som_Tmax)
writelog(f, 'SOM Start Epoch: %d' % args.som_start_epoch)
writelog(f, 'SOM End Epoch: %d' % args.som_end_epoch)
writelog(f, 'SOM Decay: %s' % args.som_decay)
for h in args.vae_hiddens:
writelog(f, 'VAE Hidden Units: %d' % h)
writelog(f, 'Z Dim: %.5f' % args.z_dim)
writelog(f, '----------------------')
writelog(f, 'WEIGHT')
writelog(f, 'Weight VAE: %.5f' % args.weight_vae)
writelog(f, 'Weight SOM: %.5f' % args.weight_som)
writelog(f, 'Weight Order: %.5f' % args.weight_order)
writelog(f, 'Margin: %.5f' % args.margin)
writelog(f, '----------------------')
writelog(f, 'Fold: %d' % args.fold)
writelog(f, 'Learning Rate: %.5f' % args.lr)
writelog(f, 'Batch Size: %d' % args.batch_size)
writelog(f, 'Epoch: %d' % args.epoch)
writelog(f, '----------------------')
writelog(f, 'Finetune: %d' % args.finetune)
writelog(f, 'Note: %s' % args.note)
writelog(f, '======================')
f.close()
f = open(directory + 'log.log', 'a')
# Tensorboard Logging
with tf.device('/cpu:0'):
tfw_train = tf.summary.create_file_writer(directory + 'tflog/kfold_' + str(args.fold) + '/train_')
tfw_valid = tf.summary.create_file_writer(directory + 'tflog/kfold_' + str(args.fold) + '/valid_')
tfw_test = tf.summary.create_file_writer(directory + 'tflog/kfold_' + str(args.fold) + '/test_')
# Tensor Seed
seed = 0
random.seed(seed)
np.random.seed(seed)
torch.manual_seed(seed)
torch.cuda.manual_seed(seed)
torch.cuda.manual_seed_all(seed)
torch.backends.cudnn.deterministic = True
torch.backends.cudnn.benchmark = False
# Define Loaders
writelog(f, 'Load data')
if args.finetune:
dataloaders, means, stds, mins, maxs = smri_batchloader(args, is_quartet=False)
else:
dataloaders, means, stds, mins, maxs = smri_batchloader(args, is_quartet=True)
f2 = open(directory + 'setting.log', 'a')
writelog(f2, 'MMSE Mean %.5f' % means[0])
writelog(f2, 'MMSE STD %.5f' % stds[0])
writelog(f2, 'MMSE Min %.5f' % mins[0])
writelog(f2, 'MMSE Max %.5f' % maxs[0])
writelog(f2, 'Age Mean %.5f' % means[1])
writelog(f2, 'Age STD %.5f' % stds[1])
writelog(f2, 'Age Min %.5f' % mins[1])
writelog(f2, 'Age Max %.5f' % maxs[1])
f2.close()
def train(dataloader, directory='.'):
# Set mode as training
if args.finetune:
vae.eval()
lssl.eval()
else:
vae.train()
lssl.train()
som.train()
# Define training variables
loss_total_ = 0
loss_vae_ = 0
loss_recon_ = 0
loss_kld_ = 0
loss_order_ = 0
loss_som_ = 0
n_samples = 0
# Loop over the minibatch
for i, xbatch in enumerate(tqdm(dataloader)):
# Data
x = xbatch['demographic'].float().to(device)
batch_size = x.shape[0]
if batch_size != args.batch_size:
continue
x_stage = x[:, :4] # Clinical Stage, Dimensionality: 4
x_mmse = x[:, 4].unsqueeze(1) # MMSE, Dimensionality: 1
x_age = x[:, 5].unsqueeze(1) # Age, Dimensionality: 1
# TRAIN VAESOM
optimizer.zero_grad()
z, z_mu, z_logvar, x_hat_stage, x_hat_mmse, x_hat_age = vae(x_stage, x_mmse, x_age)
loss_vae, loss_recon, loss_kld = criterion_vae(vae, x_stage, x_mmse, x_age,
z_mu, z_logvar,
x_hat_stage, x_hat_mmse, x_hat_age)
loss_order = lssl(z[0, :].unsqueeze(0), z[1, :].unsqueeze(0), args.margin) + \
lssl(z[1, :].unsqueeze(0), z[2, :].unsqueeze(0), args.margin) + \
lssl(z[2, :].unsqueeze(0), z[3, :].unsqueeze(0), args.margin)
w, distances = som(z, epoch - args.som_start_epoch, args.som_end_epoch - args.som_start_epoch)
loss_som = criterion_som(w, distances)
if args.finetune:
loss_total = (args.weight_som * loss_som)
loss_total.backward()
optimizer.step()
loss_total_ += loss_total.item()
loss_vae_ += 0
loss_recon_ += 0
loss_kld_ += 0
loss_order_ += 0
loss_som_ += args.weight_som * loss_som.item()
else:
loss_total = (args.weight_vae * loss_vae) + \
(args.weight_order * loss_order) + \
(args.weight_som * loss_som)
loss_total.backward()
optimizer.step()
loss_total_ += loss_total.item()
loss_vae_ += args.weight_vae * loss_vae.item()
loss_recon_ += loss_recon.item()
loss_kld_ += loss_kld.item()
loss_order_ += args.weight_order * loss_order.item()
loss_som_ += args.weight_som * loss_som.item()
n_samples += batch_size
# Take average
loss_total_ = loss_total_ / n_samples
loss_vae_ = loss_vae_ / n_samples
loss_recon_ = loss_recon_ / n_samples
loss_kld_ = loss_kld_ / n_samples
loss_order_ = loss_order_ / n_samples
loss_som_ = loss_som_ / n_samples
writelog(f, 'Loss Total: %.8f' % loss_total_)
writelog(f, 'Loss VAE: %.8f' % loss_vae_)
writelog(f, 'Loss Recon: %.8f' % loss_recon_)
writelog(f, 'Loss KLD: %.8f' % loss_kld_)
writelog(f, 'Loss Order: %.8f' % loss_order_)
writelog(f, 'Loss SOM: %.8f' % loss_som_)
# Tensorboard Logging
info = {'loss_total': loss_total_,
'loss_vae': loss_vae_,
'loss_recon': loss_recon_,
'loss_kld': loss_kld_,
'loss_order': loss_order_,
'loss_som': loss_som_,
'SOM.T': som.T}
for tag, value in info.items():
with tfw_train.as_default():
with tf.device('/cpu:0'):
tf.summary.scalar(tag,value, step=epoch)
# tfw_train.flush()
def evaluate(phase, dataloader, directory='.'):
# Set mode as training
vae.eval()
lssl.eval()
som.eval()
# Define training variables
loss_total_ = 0
loss_vae_ = 0
loss_recon_ = 0
loss_kld_ = 0
loss_order_ = 0
loss_som_ = 0
n_samples = 0
# No Grad
with torch.no_grad():
# Loop over the minibatch
for i, xbatch in enumerate(tqdm(dataloader)):
# Data
x = xbatch['demographic'].float().to(device)
batch_size = x.shape[0]
if batch_size != args.batch_size:
continue
x_stage = x[:, :4] # Clinical Stage, Dimensionality: 4
x_mmse = x[:, 4].unsqueeze(1) # MMSE, Dimensionality: 1
x_age = x[:, 5].unsqueeze(1) # Age, Dimensionality: 1
# TRAIN VAESOM
z, z_mu, z_logvar, x_hat_stage, x_hat_mmse, x_hat_age = vae(x_stage, x_mmse, x_age)
loss_vae, loss_recon, loss_kld = criterion_vae(vae, x_stage, x_mmse, x_age,
z_mu, z_logvar,
x_hat_stage, x_hat_mmse, x_hat_age)
loss_order = lssl(z_mu[0, :].unsqueeze(0), z_mu[1, :].unsqueeze(0), args.margin) + \
lssl(z_mu[1, :].unsqueeze(0), z_mu[2, :].unsqueeze(0), args.margin) + \
lssl(z_mu[2, :].unsqueeze(0), z_mu[3, :].unsqueeze(0), args.margin)
w, distances = som(z_mu, epoch - args.som_start_epoch, args.som_end_epoch - args.som_start_epoch)
loss_som = criterion_som(w, distances)
if args.finetune:
loss_total = (args.weight_som * loss_som)
loss_total_ += loss_total.item()
loss_vae_ += 0
loss_recon_ += 0
loss_kld_ += 0
loss_order_ += 0
loss_som_ += args.weight_som * loss_som.item()
else:
loss_total = (args.weight_vae * loss_vae) + \
(args.weight_order * loss_order) + \
(args.weight_som * loss_som)
loss_total_ += loss_total.item()
loss_vae_ += args.weight_vae * loss_vae.item()
loss_recon_ += loss_recon.item()
loss_kld_ += loss_kld.item()
loss_order_ += args.weight_order * loss_order.item()
loss_som_ += args.weight_som * loss_som.item()
n_samples += batch_size
# Take average
loss_total_ = loss_total_ / n_samples
loss_vae_ = loss_vae_ / n_samples
loss_recon_ = loss_recon_ / n_samples
loss_kld_ = loss_kld_ / n_samples
loss_order_ = loss_order_ / n_samples
loss_som_ = loss_som_ / n_samples
writelog(f, 'Loss Total: %.8f' % loss_total_)
writelog(f, 'Loss VAE: %.8f' % loss_vae_)
writelog(f, 'Loss Recon: %.8f' % loss_recon_)
writelog(f, 'Loss KLD: %.8f' % loss_kld_)
writelog(f, 'Loss Order: %.8f' % loss_order_)
writelog(f, 'Loss SOM: %.8f' % loss_som_)
# Tensorboard Logging
info = {'loss_total': loss_total_,
'loss_vae': loss_vae_,
'loss_recon': loss_recon_,
'loss_kld': loss_kld_,
'loss_order': loss_order_,
'loss_som': loss_som_}
for tag, value in info.items():
if phase == 'valid':
with tfw_valid.as_default():
with tf.device('/cpu:0'):
tf.summary.scalar(tag, value, step=epoch)
# tfw_valid.flush()
else:
with tfw_test.as_default():
with tf.device('/cpu:0'):
tf.summary.scalar(tag, value, step=epoch)
# tfw_test.flush()
return loss_total_
# Define Model
vae = ADPENVAE(args).to(device)
som = ADPENSOM(args).to(device)
if args.finetune:
vae.load_state_dict(torch.load('log/ADPEN/%s/model/ADPENVAE.pt' % (log_dir_vaesom)))
som.load_state_dict(torch.load('log/ADPEN/%s/model/ADPENSOM.pt' % (log_dir_vaesom)))
lssl = LSSLLoss(args.z_dim).to(device)
criterion_vae = VAELoss(args).to(device)
criterion_som = SOMLoss(args).to(device)
if args.finetune:
params = list(som.parameters())
else:
params = list(vae.parameters()) + list(lssl.parameters()) + list(som.parameters())
optimizer = optim.Adam(params, lr=args.lr, weight_decay=args.lambda2)
# Best epoch checking
valid = {'epoch': 0, 'loss': 99999}
# Train Epoch
for epoch in range(args.epoch):
writelog(f, '--- Epoch %d' % epoch)
writelog(f, 'Training')
train(dataloaders['train'], directory=directory)
writelog(f, 'Validation')
loss = evaluate('valid', dataloaders['valid'], directory=directory)
# Save Model
if loss < valid['loss']:
torch.save(vae.state_dict(), directory + '/model/ADPENVAE.pt')
torch.save(lssl.state_dict(), directory + '/model/LSSLLoss.pt')
torch.save(som.state_dict(), directory + '/model/ADPENSOM.pt')
writelog(f, 'Lowest validation loss is found! Validation Loss : %f' % loss)
writelog(f, 'Models at Epoch %d are saved!' % epoch)
valid['loss'] = loss
valid['epoch'] = epoch
f2 = open(directory + 'setting.log', 'a')
writelog(f2, 'Best Epoch %d' % epoch)
f2.close()
writelog(f, 'SOM T %.3f' % som.T)
writelog(f, 'Best model for testing: epoch %d-th' % valid['epoch'])
writelog(f, 'Testing')
loss = evaluate('test', dataloaders['test'], directory=directory)
writelog(f, 'END OF TRAINING')
f.close()