-
Notifications
You must be signed in to change notification settings - Fork 0
/
train.py
513 lines (409 loc) · 23.2 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
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
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
import os
import logging
import argparse
import random
import datetime
from tqdm import tqdm, trange
import numpy as np
import torch
from torch.utils.data import TensorDataset, DataLoader, RandomSampler, SequentialSampler
from itertools import cycle
from transformers import RobertaTokenizer, RobertaModel
from sklearn.metrics import precision_recall_fscore_support
from torch import optim
from torch.nn import CrossEntropyLoss
from DataProcessor import *
from model import Coarse2Fine
from boxes_utils import*
from optimization import BertAdam
def macro_f1(y_true, y_pred):
preds = np.argmax(y_pred, axis=-1)
true = y_true
p_macro, r_macro, f_macro, support_macro \
= precision_recall_fscore_support(true, preds, average='macro')
# f_macro = 2*p_macro*r_macro/(p_macro+r_macro)
return p_macro, r_macro, f_macro
def accuracy(out, labels):
outputs = np.argmax(out, axis=1)
return np.sum(outputs == labels)
def warmup_linear(x, warmup=0.002):
if x < warmup:
return x/warmup
return 1.0 - x
def post_dataloader(batch):
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
tokens,input_ids,input_mask, sentiment_label,\
img_id,img_shape,relation_label, GT_boxes,roi_boxes,img_feat,spatial_feat,box_labels=batch
input_ids=list(map(list, zip(*input_ids)))
input_mask=list(map(list, zip(*input_mask)))
img_shape=list(map(list, zip(*img_shape)))
input_ids=torch.tensor(input_ids,dtype=torch.long).to(device)
input_mask=torch.tensor(input_mask,dtype=torch.long).to(device)
img_shape=torch.tensor(img_shape,dtype=torch.float).to(device)
sentiment_label=sentiment_label.to(device).long()
relation_label=relation_label.to(device).long()
GT_boxes=GT_boxes.to(device).float()
roi_boxes=roi_boxes.to(device).float()
img_feat=img_feat.to(device).float()
spatial_feat=spatial_feat.to(device).float()
box_labels=box_labels.to(device).float()
return tokens,input_ids,input_mask, sentiment_label,\
img_id,img_shape,relation_label, GT_boxes,roi_boxes,img_feat,spatial_feat,box_labels
def main():
start_time = datetime.datetime.now().strftime('%m-%d-%Y-%H-%M-%S_')
parser = argparse.ArgumentParser()
## Required parameters
parser.add_argument("--dataset",
default='twitter2015',
type=str,
required=True,
help="The name of the task to train.")
parser.add_argument("--data_dir",
default= './data/',
type=str,
required=True,
help="The input data dir. Should contain the .tsv files for the task.")
parser.add_argument("--VG_data_dir",
default= './data/Image_Target_Matching',
type=str,
required=True,
help="The input data dir. Should contain the .tsv files for the task.")
parser.add_argument("--imagefeat_dir",
default = '/mnt/nfs-storage-titan/data/twitter_images/', # default ='./data/twitter_images/',
type=str,
required=True,
)
parser.add_argument("--VG_imagefeat_dir",
default = '/mnt/nfs-storage-titan/data/twitter_images/', # default ='./data/twitter_images/',
type=str,
required=True,
)
parser.add_argument("--output_dir",
default="./log/",
type=str,
required=True,
help="The output directory where the model predictions and checkpoints will be written.")
## Other parameters
parser.add_argument("--max_seq_length",
default=128,
type=int,
help="The maximum total input sequence length after WordPiece tokenization. \n"
"Sequences longer than this will be truncated, and sequences shorter \n"
"than this will be padded.")
parser.add_argument("--save",
default=True,
action='store_true',
help="Whether to save model.")
parser.add_argument("--train_batch_size",
default=32,
type=int,
help="Total batch size for training.")
parser.add_argument("--eval_batch_size",
default=32,
type=int,
help="Total batch size for eval.")
parser.add_argument("--SA_learning_rate",
default=1e-5,
type=float,
help="The initial learning rate for Adam.")
parser.add_argument("--VG_learning_rate",
default=1e-6,
type=float,
help="The initial learning rate for Adam.")
parser.add_argument("--ranking_loss_ratio",
default=0.5,
type=float)
parser.add_argument("--pred_loss_ratio",
default=1.,
type=float)
parser.add_argument("--num_train_epochs",
default=9.0,
type=float,
help="Total number of training epochs to perform.")
parser.add_argument("--warmup_proportion",
default=0.1,
type=float,
help="Proportion of training to perform linear learning rate warmup for. "
"E.g., 0.1 = 10% of training.")
parser.add_argument('--seed',
type=int,
default=2020, # 24
help="random seed for initialization")
parser.add_argument('--roi_num',
default=100,
type=int)
args = parser.parse_args()
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
args.data_dir=args.data_dir+str(args.dataset).lower()+ '/%s.pkl'
args.imagefeat_dir=args.imagefeat_dir+str(args.dataset).lower()
args.VG_data_dir= args.VG_data_dir + '/%s.pkl'
args.VG_imagefeat_dir = args.VG_imagefeat_dir+'twitter2017' # image-target-matching data from twitter2017
args.output_dir=args.output_dir + str(args.dataset) + "/"
if os.path.exists(args.output_dir) and os.listdir(args.output_dir) and args.do_train:
raise ValueError("Output directory ({}) already exists and is not empty.".format(args.output_dir))
if not os.path.exists(args.output_dir):
os.makedirs(args.output_dir)
output_model_file = os.path.join(args.output_dir, "pytorch_model.bin")
output_logger_file=os.path.join(args.output_dir,'log.txt')
logging.basicConfig(format = '%(asctime)s - %(levelname)s - %(name)s - %(message)s',
datefmt = '%m/%d/%Y %H:%M:%S',
level = logging.INFO,
filename=output_logger_file)
logger = logging.getLogger(__name__)
logger.info("dataset:{} num_train_epochs:{}".format(args.dataset, args.num_train_epochs))
logger.info("SA_learning_rate:{} warmup_proportion:{}".format(args.SA_learning_rate,args.warmup_proportion))
logger.info("VG_learning_rate:{} warmup_proportion:{}".format(args.VG_learning_rate,args.warmup_proportion))
logger.info(" ranking_loss_ratio:{} pred_loss_ratio:{} " .format(args.ranking_loss_ratio,args.pred_loss_ratio,))
logger.info(args)
random.seed(args.seed)
np.random.seed(args.seed)
torch.manual_seed(args.seed)
tokenizer = RobertaTokenizer.from_pretrained('roberta-base')
train_dataset_SA=MyDataset(args.data_dir%str('train'), args.imagefeat_dir, tokenizer, max_seq_len= args.max_seq_length, num_roi_boxes=100)
train_dataset_VG=MyDataset(args.VG_data_dir%str('VG_train'), args.VG_imagefeat_dir, tokenizer, max_seq_len= args.max_seq_length, num_roi_boxes=100)
train_dataloader_SA = Data.DataLoader(dataset=train_dataset_SA,shuffle=True, batch_size=args.train_batch_size,num_workers=0)
train_dataloader_VG = Data.DataLoader(dataset=train_dataset_VG,shuffle=True, batch_size=args.train_batch_size,num_workers=0)
eval_dataset_SA = MyDataset(args.data_dir%str('dev'),args.imagefeat_dir,tokenizer,max_seq_len=args.max_seq_length,num_roi_boxes=100)
eval_dataset_VG = MyDataset(args.VG_data_dir%str('VG_dev'),args.VG_imagefeat_dir,tokenizer,max_seq_len=args.max_seq_length,num_roi_boxes=100)
eval_dataloader_SA = Data.DataLoader(dataset=eval_dataset_SA,shuffle=False, batch_size=args.eval_batch_size,num_workers=0)
eval_dataloader_VG = Data.DataLoader(dataset=eval_dataset_VG,shuffle=False, batch_size=args.eval_batch_size,num_workers=0)
test_dataset = MyDataset(args.data_dir%str('test'),args.imagefeat_dir,tokenizer,max_seq_len=args.max_seq_length,num_roi_boxes=100)
test_dataloader = Data.DataLoader(dataset=test_dataset,shuffle=False, batch_size=args.eval_batch_size,num_workers=0)
train_number=max(train_dataset_SA.number,train_dataset_VG.number)
num_train_steps = int( train_number / args.train_batch_size * args.num_train_epochs)
model = Coarse2Fine(roberta_name='roberta-base', roi_num = args.roi_num )
model.to(device)
# new_state_dict=model.state_dict()
# logger.info(new_state_dict)
# Prepare optimizer
# optimizer BertAdam
param_optimizer = list(model.named_parameters())
no_decay = ['bias', 'LayerNorm.bias', 'LayerNorm.weight']
optimizer_grouped_parameters = [
{'params': [p for n, p in param_optimizer if not any(nd in n for nd in no_decay)], 'weight_decay': 0.01},
{'params': [p for n, p in param_optimizer if any(nd in n for nd in no_decay)], 'weight_decay': 0.0}
]
optimizer_VG = BertAdam(optimizer_grouped_parameters,
lr=args.VG_learning_rate,
warmup=args.warmup_proportion,
t_total=num_train_steps)
optimizer_SA = BertAdam(optimizer_grouped_parameters,
lr=args.SA_learning_rate,
warmup=args.warmup_proportion,
t_total=num_train_steps)
VG_global_step = 0
SA_global_step = 0
nb_tr_steps = 0
max_senti_acc = 0.0
best_epoch=-1
logger.info("*************** Running training ***************")
for train_idx in trange(int(args.num_train_epochs), desc="Epoch"):
logger.info("************************************************** Epoch: "+ str(train_idx) + " *************************************************************")
logger.info(" Num examples = %d", train_number)
logger.info(" Batch size = %d", args.train_batch_size)
logger.info(" Num steps = %d", num_train_steps)
### train
model.train()
vg_l,sa_l=0,0
pred_l=0
ranking_l,regression_l,senti_l,rel_l=0,0,0,0
for step,data in enumerate(tqdm(zip(cycle(train_dataloader_VG),train_dataloader_SA),desc="Iteration")):
batch_VG,batch_SA=data
#### VG
tokens,input_ids,input_mask, sentiment_label,\
img_id,img_shape,relation_label, GT_boxes,roi_boxes,img_feat,spatial_feat,box_labels=post_dataloader(batch_VG)
senti_pred,ranking_loss,pred_loss,pred_score,attn_map= model(img_id=img_id,
input_ids = input_ids,
input_mask = input_mask,
img_feat = img_feat,
relation_label = relation_label,
box_labels=box_labels,
ranking_loss_ratio=args.ranking_loss_ratio,
pred_loss_ratio=args.pred_loss_ratio,
)
loss_VG=ranking_loss + pred_loss
loss_VG.backward()
lr_this_step = args.VG_learning_rate * warmup_linear(VG_global_step/num_train_steps, args.warmup_proportion)
for param_group in optimizer_VG.param_groups:
param_group['lr'] = lr_this_step
optimizer_VG.step()
optimizer_VG.zero_grad()
VG_global_step += 1
ranking_l+=ranking_loss.item()
pred_l+=pred_loss.item()
#### SA
SA_tokens,SA_input_ids,SA_input_mask, SA_sentiment_label,\
SA_img_id,SA_img_shape,SA_relation_label, SA_GT_boxes,SA_roi_boxes,SA_img_feat,SA_spatial_feat,SA_box_labels=post_dataloader(batch_SA)
senti_pred,ranking_loss,pred_loss,pred_score,attn_map= model(img_id=SA_img_id,
input_ids = SA_input_ids,
input_mask = SA_input_mask,
img_feat = SA_img_feat,
relation_label = None,
box_labels = None
)
senti_loss_fct=CrossEntropyLoss()
sentiment_loss=senti_loss_fct(senti_pred.view(-1,3),SA_sentiment_label.view(-1))
loss_SA=sentiment_loss
loss_SA.backward()
senti_l+=sentiment_loss.item()
lr_this_step = args.SA_learning_rate * warmup_linear(SA_global_step/num_train_steps, args.warmup_proportion)
for param_group in optimizer_SA.param_groups:
param_group['lr'] = lr_this_step
optimizer_SA.step()
optimizer_SA.zero_grad()
SA_global_step += 1
nb_tr_steps += 1
ranking_l = ranking_l / nb_tr_steps
pred_l = pred_l / nb_tr_steps
senti_l= senti_l / nb_tr_steps
logger.info("ranking_loss:%s",ranking_l)
logger.info("pred_loss:%s",pred_l)
logger.info("sentiment_loss:%s",senti_l)
### dev
model.eval()
logger.info("***** Running evaluation on Dev Set*****")
logger.info(" SA Num examples = %d", eval_dataset_SA.number) #len(eval_examples)
logger.info(" Batch size = %d", args.eval_batch_size)
nb_eval_examples = 0
SA_nb_eval_examples = 0
senti_acc,rel_acc=0,0
ranking_vg_acc=0
senti_precision,senti_recall,senti_F_score=0,0,0
senti_true_label_list = []
senti_pred_label_list = []
num_right_vg=0
num_valid=0
#### VG
for s,batch_VG in enumerate(tqdm(eval_dataloader_VG, desc="Evaluating_VG")):
tokens,input_ids,input_mask, sentiment_label,\
img_id,img_shape,relation_label, GT_boxes,roi_boxes,img_feat,spatial_feat,box_labels=post_dataloader(batch_VG)
with torch.no_grad():
senti_pred,ranking_loss,pred_loss,pred_score,attn_map= model(img_id=img_id,
input_ids = input_ids,
input_mask = input_mask,
img_feat = img_feat,
relation_label = relation_label,
box_labels = box_labels
)
current_batch_size=input_ids.size()[0]
#-----evaluate
##### coarse-grained
pred_score = pred_score.detach().cpu().numpy() #[N*n, 100] #.reshape(current_batch_size,args.max_GT_boxes,-1) # [N*n, 100]->[N, n, 100]
relation_pred = np.argmax(pred_score, axis=1)
tmp_rel_accuracy=np.sum(relation_pred == relation_label.cpu().numpy())
rel_acc += tmp_rel_accuracy
roi_boxes=roi_boxes.detach().cpu().numpy() #[N, 100, 4]
GT_boxes=GT_boxes.detach().cpu() #[N, n,4]
attn_map=attn_map.detach().cpu().numpy()
##### fine-grained
for i in range(current_batch_size): #N
if relation_label[i]!=0:
num_valid+=1
ious=(torchvision.ops.box_iou(GT_boxes[i,0:1,:],torch.tensor(roi_boxes[i]))).numpy() #[1,4],[100,4]->[1,100] #如果GT是0，iou为0
sorted_index=np.argsort(-attn_map[i])[0]
pred_ids=sorted_index[:1] #top K=1
topk_max_iou=ious[0][pred_ids]
pred_iou=topk_max_iou.max()
if pred_iou>=0.5:
num_right_vg+=1
nb_eval_examples += current_batch_size
rel_acc = rel_acc/ nb_eval_examples
ranking_vg_acc = num_right_vg/num_valid
#### SA
for batch_SA in tqdm(eval_dataloader_SA, desc="Evaluating_SA"):
SA_tokens, SA_input_ids, SA_input_mask, SA_sentiment_label, \
SA_img_id, SA_img_shape, SA_relation_label, SA_GT_boxes, SA_roi_boxes, SA_img_feat, SA_spatial_feat, SA_box_labels = post_dataloader(batch_SA)
with torch.no_grad():
SA_senti_pred, SA_ranking_loss, SA_pred_loss,SA_pred_score,attn_map = model(
img_id=SA_img_id,
input_ids = SA_input_ids,
input_mask = SA_input_mask,
img_feat = SA_img_feat,
relation_label = None,
box_labels= None
)
SA_sentiment_label = SA_sentiment_label.cpu().numpy()
SA_senti_pred = SA_senti_pred.cpu().numpy()
senti_true_label_list.append(SA_sentiment_label)
senti_pred_label_list.append(SA_senti_pred)
tmp_senti_accuracy = accuracy(SA_senti_pred, SA_sentiment_label)
senti_acc += tmp_senti_accuracy
current_batch_size = SA_input_ids.size()[0]
SA_nb_eval_examples += current_batch_size
senti_acc = senti_acc / SA_nb_eval_examples
senti_true_label = np.concatenate(senti_true_label_list)
senti_pred_outputs = np.concatenate(senti_pred_label_list)
senti_precision, senti_recall, senti_F_score = macro_f1(senti_true_label, senti_pred_outputs)
result = { 'nb_eval_examples':nb_eval_examples,
'num_valid':num_valid,
'Dev_rel_acc':rel_acc,
'Dev_senti_acc':senti_acc,
'Dev_senti_precision':senti_precision,
'Dev_senti_recall':senti_recall,
'Dev_senti_F_score':senti_F_score,
'Dev_ranking_vg_acc':ranking_vg_acc, #num_right_vg/num_valid
}
logger.info("***** Dev Eval results *****")
for key in sorted(result.keys()):
logger.info(" %s = %s", key, str(result[key]))
### test
model.eval()
logger.info("***** Running evaluation on Test Set *****")
logger.info(" Num examples = %d", test_dataset.number)
logger.info(" Batch size = %d", args.eval_batch_size)
nb_test_examples = 0
test_senti_acc=0
test_senti_true_label_list = []
test_senti_pred_label_list = []
for batch in tqdm(test_dataloader, desc="Testing"):
tokens,input_ids,input_mask, sentiment_label,\
img_id,img_shape,relation_label, GT_boxes,roi_boxes,img_feat,spatial_feat,box_labels=post_dataloader(batch)
with torch.no_grad():
senti_pred,ranking_loss,pred_loss,pred_score,attn_map= model(
img_id=img_id,
input_ids = input_ids,
input_mask = input_mask,
img_feat = img_feat,
relation_label = None,
box_labels= None
)
sentiment_label=sentiment_label.cpu().numpy()
senti_pred=senti_pred.cpu().numpy()
test_senti_true_label_list.append(sentiment_label)
test_senti_pred_label_list.append(senti_pred)
tmp_senti_accuracy = accuracy(senti_pred, sentiment_label)
test_senti_acc += tmp_senti_accuracy
current_batch_size=input_ids.size()[0]
nb_test_examples += current_batch_size
test_senti_acc = test_senti_acc / nb_test_examples
senti_true_label = np.concatenate(test_senti_true_label_list)
senti_pred_outputs = np.concatenate(test_senti_pred_label_list)
test_senti_precision, test_senti_recall, test_senti_F_score = macro_f1(senti_true_label, senti_pred_outputs)
result = {
'Test_senti_acc':test_senti_acc,
'Test_senti_precision':test_senti_precision,
'Test_senti_recall':test_senti_recall,
'Test_senti_F_score':test_senti_F_score}
logger.info("***** Test Eval results *****")
for key in sorted(result.keys()):
logger.info(" %s = %s", key, str(result[key]))
# save model
if senti_acc >= max_senti_acc:
# Save a trained model
if args.save:
model_to_save = model.module if hasattr(model, 'module') else model
torch.save(model_to_save.state_dict(), output_model_file)
max_senti_acc=senti_acc
corresponding_test_acc=test_senti_acc
corresponding_test_p=test_senti_precision
corresponding_test_r=test_senti_recall
corresponding_test_f=test_senti_F_score
best_epoch=train_idx
logger.info("max_dev_senti_acc: %s ",max_senti_acc)
logger.info("corresponding_test_sentiment_acc: %s ",corresponding_test_acc)
logger.info("corresponding_test_sentiment_precision: %s ",test_senti_precision)
logger.info("corresponding_test_sentiment_recall: %s ",test_senti_recall)
logger.info("corresponding_test_sentiment_F_score: %s ",test_senti_F_score)
logger.info("best_epoch: %d",best_epoch)
if __name__ == "__main__":
main()