-
Notifications
You must be signed in to change notification settings - Fork 0
/
temp_model_tfc.py
288 lines (206 loc) · 11.1 KB
/
temp_model_tfc.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
import numpy as np
import os
import torch
import tqdm
from sklearn.metrics import roc_auc_score, average_precision_score, precision_score, recall_score, f1_score
from torch import nn
import torch.nn.functional as F
from models.tfc.config import Config
from models.tfc.dataloader import generate_freq, data_generator
from models.tfc.loss import NTXentLoss_poly
from models.tfc.model_mlp import TFC, target_classifier
from utils.model_size import get_model_size
from utils.path_utils import project_root
from utils.pretrain_utils.get_args import get_args
def model_pretrain(model, model_optimizer, train_loader, config, device):
total_loss = []
model.train()
model.to(device)
global loss, loss_t, loss_f, l_TF, loss_c, data_test, data_f_test
for batch_idx, (data, labels, aug1, data_f, aug1_f) in tqdm.tqdm(enumerate(train_loader), desc='Pre-training model',
total=len(train_loader)):
model_optimizer.zero_grad()
data, labels = data.float().to(device), labels.long().to(device)
aug1 = aug1.float().to(device)
data_f, aug1_f = data_f.float().to(device), aug1_f.float().to(device)
"""Produce embeddings"""
h_t, z_t, h_f, z_f = model(data, data_f, stage='train')
h_t_aug, z_t_aug, h_f_aug, z_f_aug = model(aug1, aug1_f, stage='train')
"""Compute Pre-train loss"""
"""NTXentLoss: normalized temperature-scaled cross entropy loss. From SimCLR"""
nt_xent_criterion = NTXentLoss_poly(device, config.batch_size, config.Context_Cont.temperature,
config.Context_Cont.use_cosine_similarity)
loss_t = nt_xent_criterion(h_t, h_t_aug)
loss_f = nt_xent_criterion(h_f, h_f_aug)
l_TF = nt_xent_criterion(z_t, z_f)
l_1, l_2, l_3 = nt_xent_criterion(z_t, z_f_aug), nt_xent_criterion(z_t_aug, z_f), nt_xent_criterion(z_t_aug,
z_f_aug)
loss_c = (1 + l_TF - l_1) + (1 + l_TF - l_2) + (1 + l_TF - l_3)
lam = 0.2
loss = lam * (loss_t + loss_f) + l_TF
total_loss.append(loss.item())
loss.backward()
model_optimizer.step()
print('Pretraining: overall loss:{}, l_t: {}, l_f:{}, l_c:{}'.format(loss, loss_t, loss_f, l_TF))
ave_loss = torch.tensor(total_loss).mean()
return ave_loss
def train(model, args, config, train_loader, device='cuda'):
model_optimizer = torch.optim.Adam(model.parameters(), lr=configs.lr, betas=(configs.beta1, configs.beta2),
weight_decay=config.weight_decay)
experiment_log_dir = os.path.join(project_root(), 'results', 'tfc')
os.makedirs(os.path.join(experiment_log_dir, f"gtn_mlp"), exist_ok=True)
log_file_path = 'pretrain_tfc.txt'
with open(log_file_path, 'a') as log_file:
for epoch in range(1, config.pretrain_epoch + 1):
train_loss = model_pretrain(model, model_optimizer, train_loader, config, device)
log_text = f'Pre-training Epoch: {epoch}\t Train Loss: {train_loss:.4f}\t'
print(log_text)
log_file.write(log_text)
if epoch % 2 == 0:
chkpoint = {'epoch': epoch, 'train_loss': train_loss, 'model_state_dict': model.state_dict()}
torch.save(chkpoint, os.path.join(experiment_log_dir, "gtn_mlp", f'ckp_ep{epoch}.pt'))
def build_model(args, lr, configs, device='cuda', chkpoint=None):
model = TFC(configs).to(device)
pretrained_dict = chkpoint
model_dict = model.state_dict()
model_dict.update(pretrained_dict)
model.load_state_dict(model_dict)
classifier = target_classifier(configs).to(device)
model_optimizer = torch.optim.Adam(model.parameters(), lr=configs.lr, betas=(configs.beta1, configs.beta2),
weight_decay=configs.weight_decay)
classifier_optimizer = torch.optim.Adam(model.parameters(), lr=configs.lr,
betas=(configs.beta1, configs.beta2), weight_decay=configs.weight_decay)
model_scheduler = torch.optim.lr_scheduler.ReduceLROnPlateau(model_optimizer, 'min')
return model, classifier, model_optimizer, classifier_optimizer, model_scheduler
def model_finetune(model, model_optimizer, val_dl, config, classifier=None, classifier_optimizer=None, device='cuda'):
global labels, pred_numpy, fea_concat_flat
model.train()
classifier.train()
total_loss = []
total_acc = []
total_auc = []
total_prc = []
criterion = nn.CrossEntropyLoss()
outs = np.array([])
trgs = np.array([])
feas = np.array([])
for data, labels, aug1, data_f, aug1_f in tqdm.tqdm(val_dl, desc="Fine-tuning model", total=len(val_dl)):
# print('Fine-tuning: {} of target samples'.format(labels.shape[0]))
data, labels = data.float().to(device), labels.long().to(device)
data_f = data_f.float().to(device)
aug1 = aug1.float().to(device)
aug1_f = aug1_f.float().to(device)
"""if random initialization:"""
model_optimizer.zero_grad() # The gradients are zero, but the parameters are still randomly initialized.
classifier_optimizer.zero_grad() # the classifier is newly added and randomly initialized
"""Produce embeddings"""
h_t, z_t, h_f, z_f = model(data, data_f, stage='train')
h_t_aug, z_t_aug, h_f_aug, z_f_aug = model(aug1, aug1_f, stage='train')
nt_xent_criterion = NTXentLoss_poly(device, config.batch_size, config.Context_Cont.temperature,
config.Context_Cont.use_cosine_similarity)
loss_t = nt_xent_criterion(h_t, h_t_aug)
loss_f = nt_xent_criterion(h_f, h_f_aug)
l_TF = nt_xent_criterion(z_t, z_f)
l_1, l_2, l_3 = nt_xent_criterion(z_t, z_f_aug), nt_xent_criterion(z_t_aug, z_f), \
nt_xent_criterion(z_t_aug, z_f_aug)
loss_c = (1 + l_TF - l_1) + (1 + l_TF - l_2) + (1 + l_TF - l_3) #
"""Add supervised classifier: 1) it's unique to finetuning. 2) this classifier will also be used in test."""
fea_concat = torch.cat((z_t, z_f), dim=1)
predictions = classifier(fea_concat)
fea_concat_flat = fea_concat.reshape(fea_concat.shape[0], -1)
loss_p = criterion(predictions, labels)
lam = 0.1
loss = loss_p + l_TF + lam * (loss_t + loss_f)
acc_bs = labels.eq(predictions.detach().argmax(dim=1)).float().mean()
onehot_label = F.one_hot(labels)
pred_numpy = predictions.detach().cpu().numpy()
try:
auc_bs = roc_auc_score(onehot_label.detach().cpu().numpy(), pred_numpy, average="macro", multi_class="ovr")
except:
auc_bs = np.float32(0)
try:
prc_bs = average_precision_score(onehot_label.detach().cpu().numpy(), pred_numpy)
except:
prc_bs = 0.0
total_acc.append(acc_bs)
total_auc.append(auc_bs)
total_prc.append(prc_bs)
total_loss.append(loss.item())
loss.backward()
model_optimizer.step()
classifier_optimizer.step()
# training_mode != 'pre_train'
pred = predictions.max(1, keepdim=True)[1] # get the index of the max log-probability
outs = np.append(outs, pred.cpu().numpy())
trgs = np.append(trgs, labels.data.cpu().numpy())
feas = np.append(feas, fea_concat_flat.data.cpu().numpy())
feas = feas.reshape([len(trgs), -1]) # produce the learned embeddings
labels_numpy = labels.detach().cpu().numpy()
pred_numpy = np.argmax(pred_numpy, axis=1)
precision = precision_score(labels_numpy, pred_numpy, average='macro', )
recall = recall_score(labels_numpy, pred_numpy, average='macro', )
F1 = f1_score(labels_numpy, pred_numpy, average='macro', )
ave_loss = torch.tensor(total_loss).mean()
ave_acc = torch.tensor(total_acc).mean()
ave_auc = torch.tensor(total_auc).mean()
ave_prc = torch.tensor(total_prc).mean()
print(' Finetune: loss = %.4f| Acc=%.4f | Precision = %.4f | Recall = %.4f | F1 = %.4f| AUROC=%.4f | AUPRC = %.4f'
% (ave_loss, ave_acc * 100, precision * 100, recall * 100, F1 * 100, ave_auc * 100, ave_prc * 100))
return model, classifier, ave_loss, ave_acc, ave_auc, ave_prc, feas, trgs, F1
def finetune(finetune_loader, args, config, chkpoint):
ft_model, ft_classifier, ft_model_optimizer, ft_classifier_optimizer, ft_scheduler = build_model(
args, args.lr, config, device='cuda', chkpoint=chkpoint)
experiment_log_dir = os.path.join(project_root(), 'results', 'tfc')
os.makedirs(os.path.join(experiment_log_dir, f"gtn_mlp"), exist_ok=True)
log_file_path = 'finetune_tfc.txt'
with open(log_file_path, 'a') as log_file:
for epoch in range(1, config.finetune_epoch + 1):
model, classifier, ave_loss, ave_acc, ave_auc, ave_prc, feas, trgs, F1 = model_finetune(
ft_model, ft_model_optimizer, finetune_loader, config, classifier=ft_classifier,
classifier_optimizer=ft_classifier_optimizer)
ft_scheduler.step(ave_loss)
log_text = (f"Fine-tuning ended ....\n"
f"{'=' * 100}\n"
f"epoch: {epoch}\n"
f"valid_auc: {ave_auc} valid_prc: {ave_prc} F1: {F1}\n"
f"valid_loss: {ave_loss} valid_acc: {ave_acc}\n"
f"{'=' * 100}\n"
)
print(log_text)
log_file.write(log_text)
if epoch % 2 == 0:
# Saving feature encoder and classifier after finetuning for testing.
chkpoint = {'seed': args.seed, 'epoch': epoch, 'train_loss': ave_loss,
'model_state_dict': model.state_dict(),
'classifier': ft_classifier.state_dict()}
torch.save(chkpoint, os.path.join(experiment_log_dir, f"gtn_mlp/", f'finetune_ep{epoch}.pt'))
def set_seed(seed):
SEED = seed
torch.manual_seed(SEED)
torch.backends.cudnn.deterministic = False
torch.backends.cudnn.benchmark = False
np.random.seed(SEED)
return seed
if __name__ == '__main__':
seed = set_seed(2024)
pretrain_exp = True
# Gathering args and configs
args, unknown = get_args()
configs = Config()
# Model
model = TFC(configs=configs)
# Model size
get_model_size(model)
# Gather datasets
tl_datasets = os.path.join(project_root(), 'data', 'tl_datasets')
if pretrain_exp:
pretrain = torch.load(os.path.join(tl_datasets, 'pretrain', 'pretrain.pt'))
train_loader = data_generator(pretrain, configs)
train(model, args, configs, train_loader)
else:
"""Fine-tuning and Test"""
finetune_dataset = torch.load(os.path.join(tl_datasets, 'finetune', 'finetune.pt'))
finetune_loader = data_generator(finetune_dataset, configs)
chkpoint = torch.load(os.path.join(project_root(), 'results', 'tfc', 'gtn_mlp', 'ckp_ep20.pt'))[
'model_state_dict']
finetune(finetune_loader, args, configs, chkpoint)