forked from loxs123/KBDPT
-
Notifications
You must be signed in to change notification settings - Fork 0
/
main.py
647 lines (557 loc) · 26.8 KB
/
main.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
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
# -*- coding: utf-8 -*-
from importlib import import_module
import torch
import torch.nn as nn
import os
import argparse
from transformers import AutoTokenizer
import transformers
import warnings
import numpy as np
from sklearn import metrics
import random
import time
from tqdm import tqdm
import json
import datetime
import copy
from config import DefaultConfig
import utils
from utils import all_metrics, print_metrics, write_result,topk_accuracy,model_optimizer
import re
import wandb
def softmax(x,axis):
exp_x = np.exp(x)
return exp_x/np.sum(exp_x,axis=axis,keepdims=True)
# os.environ['CUDA_VISIBLE_DEVICES'] = '0,1,2,3'
warnings.filterwarnings('ignore')
best_threshold = None
MODEL_TYPE1 = [
'ExperientialKnowledgeModel',
'ExperientialKnowledgeGATModel',
'ExperientialKnowledgeLastLayerAttnModel',
'ExperientialKnowledgeGATLastLayerAttnModel'
]
MODEL_TYPE2 = [
'CAML',
'MultiResCNN',
'BiGRU',
'DPCNN',
'RCNN',
'RNNAttn',
'TextCNN'
]
MODEL_TYPE3 = [
'TextCNNFuseModel',
'TextCNNFuseGATModel',
'TextCNNFuseLastLayerAttnModel',
'TextCNNFuseGATLastLayerAttnModel'
]
MODEL_TYPE4 = [
'LongFormer',
'AutoModel'
]
MODEL_TYPE5 = [
'LongFormerFuseModel',
'LongFormerFusePreTrainAsyModel',
'LongFormerFusePreTrainSynModel',
'LongFormerFuseGATModel'
]
MODEL_TYPE6 = [
'LongFormerPrompt',
'ErniePrompt',
'BertPrompt',
'ErniePrompt2',
'BertPrompt2'
]
MODEL_TYPE7 = [
'LongFormerPromptBase',
'AutoPromptBase',
'AutoPTuning',
'LongFormerPTuning'
]
MODEL_TYPE8 = [
'BertPromptField',
'ErniePromptField'
]
MODEL_TYPE9 = [
'AutoModelField'
]
def train(opt, train_data_loader, dev_data_loader,test_data_loader):
global adv_model, K
random.seed(opt.seed)
os.environ['PYTHONHASHSEED'] = str(opt.seed)
np.random.seed(opt.seed)
torch.manual_seed(opt.seed)
torch.cuda.manual_seed(opt.seed)
torch.backends.cudnn.deterministic = True
model = import_module('Model.' + opt.model_name).Model(opt)
if torch.cuda.is_available():
model = model.cuda(opt.gpu)
train_num = len(train_data_loader) * opt.batch_size
# 损失函数
loss_func = nn.BCEWithLogitsLoss()
loss_func2 = nn.BCEWithLogitsLoss()
optimizer = model_optimizer(model,opt.model_name,opt)
updates_total = len(train_data_loader) // (opt.accumulation_steps) * opt.epochs
scheduler = transformers.get_cosine_schedule_with_warmup(optimizer,
num_warmup_steps=opt.warmup_rate * updates_total,
num_training_steps=updates_total)
max_micro_f1 = -1.0 # the best micro F1
max_scores = [-1 for _ in range(opt.class_num)]
no_imp_valid = 0 # patience no improvement
for epoch in range(opt.epochs):
print("\n=== Epoch %d train ===" % epoch)
for i, data in enumerate(tqdm(train_data_loader)):
model.train()
if 'Prompt' not in opt.model_name and 'PTuning' not in opt.model_name and 'yangyang' not in opt.data_loader:
(sentence,), (mask,), nodes, adj_matrixs, labels = data
if torch.cuda.is_available():
sentence = sentence.cuda(opt.gpu)
mask = mask.cuda(opt.gpu)
labels = labels.cuda(opt.gpu)
for i in range(len(nodes)):
nodes[i] = nodes[i].cuda(opt.gpu)
adj_matrixs[i] = adj_matrixs[i].cuda(opt.gpu)
elif 'yangyang' in opt.data_loader:
(sentence1,sentence2,sentence3), (mask1,mask2,mask3),(mask_idx1,mask_idx2,mask_idx3),labels,label_prompts = data
if torch.cuda.is_available():
sentence1 = sentence1.cuda(opt.gpu)
sentence2 = sentence2.cuda(opt.gpu)
sentence3 = sentence3.cuda(opt.gpu)
mask1 = mask1.cuda(opt.gpu)
mask2 = mask2.cuda(opt.gpu)
mask3 = mask3.cuda(opt.gpu)
labels = labels.cuda(opt.gpu)
label_prompts = label_prompts.cuda(opt.gpu)
else:
(sentence,), (mask,),nodes,adj_matrixs,mask_idx, labels,label_prompts = data
if torch.cuda.is_available():
sentence = sentence.cuda(opt.gpu)
mask = mask.cuda(opt.gpu)
labels = labels.cuda(opt.gpu)
label_prompts = label_prompts.cuda(opt.gpu)
for i in range(len(nodes)):
nodes[i] = nodes[i].cuda(opt.gpu)
adj_matrixs[i] = adj_matrixs[i].cuda(opt.gpu)
if opt.model_name in MODEL_TYPE1 :
output = model(nodes,adj_matrixs)
loss = (loss_func(output, labels)) / opt.accumulation_steps
elif opt.model_name in MODEL_TYPE2:
output = model(sentence)
loss = (loss_func(output, labels)) / opt.accumulation_steps
elif opt.model_name in MODEL_TYPE3:
output = model(sentence,nodes,adj_matrixs)
loss = (loss_func(output, labels)) / opt.accumulation_steps
elif opt.model_name in MODEL_TYPE4:
output = model(sentence,mask)
loss = (loss_func(output, labels)) / opt.accumulation_steps
elif opt.model_name in MODEL_TYPE5:
output = model(sentence,mask,nodes,adj_matrixs)
loss = (loss_func(output, labels)) / opt.accumulation_steps
elif opt.model_name in MODEL_TYPE6:
output,output_prompt = model(sentence,mask,mask_idx)
loss = (loss_func(output, labels) + loss_func2(output_prompt, label_prompts)) / opt.accumulation_steps
elif opt.model_name in MODEL_TYPE7:
_,output_prompt = model(sentence,mask,mask_idx)
loss = (loss_func2(output_prompt, labels)) / opt.accumulation_steps
elif opt.model_name in MODEL_TYPE8:
output,output_prompt = model(sentence1,mask1,mask_idx1,sentence2,mask2,mask_idx2,sentence3,mask3,mask_idx3)
loss = (loss_func(output, labels) + loss_func2(output_prompt, label_prompts)) / opt.accumulation_steps
elif opt.model_name in MODEL_TYPE9:
output = model(sentence1,mask1,sentence2,mask2,sentence3,mask3)
loss = (loss_func(output, labels)) / opt.accumulation_steps
else:
raise Exception
loss.backward()
if opt.use_wandb:
# log metrics to wandb
wandb.log({"loss": loss})
if (i+1) % opt.accumulation_steps == 0:
optimizer.step()
scheduler.step()
optimizer.zero_grad()
# if (i+1) % 600 == 0:
# valid_macro_f1, report,scores = inference(model, dev_data_loader,opt,k_fold,batch_no = i//opt.accumulation_steps + updates_total//opt.epochs*epoch)
# if valid_macro_f1 > max_macro_f1:
# max_macro_f1 = valid_macro_f1
# torch.save(model.state_dict(),opt.save_model_path)
# print("目前最优验证集结果:{:.5f}".format(max_macro_f1))
print(f'epochs {epoch} end')
if (epoch + 1) % opt.test_freq == 0:
valid_micro_f1, valid_report = inference(model, dev_data_loader,opt)
test_micro_f1, test_report,_ = inference(model, test_data_loader,opt,test_set=True)
print("\n验证集micro f1: {:.5f}".format(valid_micro_f1))
print("\n测试集micro f1: {:.5f}".format(test_micro_f1))
if opt.use_wandb:
wandb.log({"dev-micro-f1":valid_micro_f1})
wandb.log({"test-micro-f1":test_micro_f1})
if valid_micro_f1 > max_micro_f1:
max_micro_f1 = valid_micro_f1
torch.save({'dict':model.state_dict()},opt.save_model_path)
print("目前最优验证集结果:{:.5f}".format(max_micro_f1))
print("\n=== Epoch %d end ===" % epoch)
def inference(model, data_loader, opt,test_set = False):
global best_threshold
"""validation"""
model.eval()
y, y_hat1 = [], []
y_hat2 = []
y_hat3 = []
y_hat4 = []
y_hat5 = []
y_hat6 = []
predict_ans = []
tokenizer = AutoTokenizer.from_pretrained(opt.bert_path)
id2labels = []
with open(opt.label_idx_path, "r", encoding="utf-8") as f:
for line in f:
lin = line.strip().split()
id2labels.append(lin[0])
with torch.no_grad():
for ii, data in enumerate(data_loader):
if 'Prompt' not in opt.model_name and 'PTuning' not in opt.model_name and 'yangyang' not in opt.data_loader:
(sentence,), (mask,), nodes, adj_matrixs, labels = data
if torch.cuda.is_available():
sentence = sentence.cuda(opt.gpu)
mask = mask.cuda(opt.gpu)
labels = labels.cuda(opt.gpu)
for i in range(len(nodes)):
nodes[i] = nodes[i].cuda(opt.gpu)
adj_matrixs[i] = adj_matrixs[i].cuda(opt.gpu)
elif 'yangyang' in opt.data_loader:
(sentence1,sentence2,sentence3), (mask1,mask2,mask3),(mask_idx1,mask_idx2,mask_idx3),labels,label_prompts = data
if torch.cuda.is_available():
sentence1 = sentence1.cuda(opt.gpu)
sentence2 = sentence2.cuda(opt.gpu)
sentence3 = sentence3.cuda(opt.gpu)
mask1 = mask1.cuda(opt.gpu)
mask2 = mask2.cuda(opt.gpu)
mask3 = mask3.cuda(opt.gpu)
labels = labels.cuda(opt.gpu)
label_prompts = label_prompts.cuda(opt.gpu)
else:
(sentence,), (mask,),nodes,adj_matrixs,mask_idx, labels,label_prompts = data
if torch.cuda.is_available():
sentence = sentence.cuda(opt.gpu)
mask = mask.cuda(opt.gpu)
labels = labels.cuda(opt.gpu)
label_prompts = label_prompts.cuda(opt.gpu)
for i in range(len(nodes)):
nodes[i] = nodes[i].cuda(opt.gpu)
adj_matrixs[i] = adj_matrixs[i].cuda(opt.gpu)
if opt.model_name in MODEL_TYPE1:
raw_output = model(nodes,adj_matrixs)
elif opt.model_name in MODEL_TYPE2:
raw_output = model(sentence)
elif opt.model_name in MODEL_TYPE3:
raw_output = model(sentence,nodes,adj_matrixs)
elif opt.model_name in MODEL_TYPE4:
raw_output = model(sentence,mask)
elif opt.model_name in MODEL_TYPE5:
raw_output = model(sentence,mask,nodes,adj_matrixs)
elif opt.model_name in MODEL_TYPE6:
raw_output,output_prompt = model(sentence,mask,mask_idx)
elif opt.model_name in MODEL_TYPE7:
_,raw_output = model(sentence,mask,mask_idx)
elif opt.model_name in MODEL_TYPE8:
raw_output,output_prompt = model(sentence1,mask1,mask_idx1,sentence2,mask2,mask_idx2,sentence3,mask3,mask_idx3)
elif opt.model_name in MODEL_TYPE9:
raw_output = model(sentence1,mask1,sentence2,mask2,sentence3,mask3)
else:
raise Exception
output = torch.sigmoid(raw_output).data.cpu().numpy() # [正负例]
raw_output = raw_output.data.cpu().numpy()
labels = labels.data.cpu().numpy()
if opt.model_name in MODEL_TYPE6 or opt.model_name in MODEL_TYPE8:
output_prompt1 = torch.sigmoid(output_prompt).data.cpu().numpy()
output_prompt2 = torch.softmax(output_prompt,dim = 0).data.cpu().numpy()
output_vote = np.argmax(output,axis = 1) # [路径标签数]
y.append(labels[0]) # 同一个batch都是一样的标签
# ----------------------- 方案一 ----------------------
y_hati = np.zeros((labels.shape[1]),dtype=np.float32)
for vote_label in output_vote:
y_hati[vote_label] += 1
y_hati = y_hati / (np.sum(y_hati) + 1e-4)
y_hati[np.argmax(y_hati)] = 1 # 票数最多的被预测出来
y_hat1.append(y_hati)
# ----------------------- 方案二 ----------------------
y_hati = np.zeros((labels.shape[1]),dtype=np.float32)
for i,vote_label in enumerate(output_vote):
if output_prompt1[i][1] > 0.5:
y_hati[vote_label] += 1
y_hati = y_hati / (np.sum(y_hati) + 1e-4)
y_hati[np.argmax(y_hati)] = 1 # 票数最多的被预测出来
y_hat2.append(y_hati)
# ----------------------- 方案三 ----------------------
y_hati = np.zeros((labels.shape[1]),dtype=np.float32)
for i,vote_label in enumerate(output_vote):
y_hati[vote_label] += output_prompt1[i][1] # 加入预测确信度
y_hati = y_hati / (np.sum(y_hati) + 1e-4)
y_hati[np.argmax(y_hati)] = 1 # 票数最多的被预测出来
y_hat3.append(y_hati)
if test_set and opt.test_only:
# case study
batch_text = tokenizer.batch_decode(sentence.tolist())
for batch_text_i in range(len(batch_text)):
batch_text[batch_text_i] = batch_text[batch_text_i].replace(tokenizer.pad_token,'')
item_dic = {}
item = []
for text,prompt_prob,cls_prob,vote_label in zip(batch_text,output_prompt1[:,1],output,output_vote):
item_item = {}
item_item['text'] = text
item_item['prompt_prob'] = float(prompt_prob)
item_item['cls_prob'] = {}
for label_id in range(len(id2labels)):
item_item['cls_prob'][id2labels[label_id]] = float(cls_prob[label_id])
item_item['vote_label'] = id2labels[vote_label]
item.append(item_item)
item_dic['info'] = item
item_dic['info'].sort(key = lambda x:x['prompt_prob'],reverse=True)
item_dic['label'] = id2labels[np.argmax(labels[0])]
item_dic['pred'] = {}
for label_id in range(len(id2labels)):
item_dic['pred'][id2labels[label_id]] = float(y_hati[label_id])
if np.argmax(labels[0]) == np.argmax(y_hati):
item_dic['ans'] = '正确'
else:
item_dic['ans'] = '错误'
predict_ans.append(item_dic)
# ----------------------- 方案四 ----------------------
y_hati = np.zeros((labels.shape[1]),dtype=np.float32)
for i,vote_label in enumerate(output):
y_hati += vote_label * output_prompt2[i][1] # 加入预测确信度
y_hat4.append(y_hati)
# ----------------------- 方案五 ----------------------
y_hati = np.zeros((labels.shape[1]),dtype=np.float32)
for i,vote_label in enumerate(raw_output):
y_hati += vote_label * output_prompt1[i][1] # 加入预测确信度
y_hat5.append(y_hati)
# ----------------------- 方案六 ----------------------
y_hati = np.zeros((labels.shape[1]),dtype=np.float32)
for i,vote_label in enumerate(output):
y_hati += (vote_label>0.5) * output_prompt2[i][1] # 加入预测确信度
y_hat6.append(y_hati)
else:
y.append(labels)
y_hat1.append(output)
if opt.model_name in MODEL_TYPE6 or opt.model_name in MODEL_TYPE8:
y = np.stack(y, axis=0)
y_hat1 = np.stack(y_hat1, axis=0)
y_hat2 = np.stack(y_hat2, axis=0)
y_hat3 = np.stack(y_hat3, axis=0)
y_hat4 = np.stack(y_hat4, axis=0)
y_hat5 = np.stack(y_hat5, axis=0)
y_hat6 = np.stack(y_hat5, axis=0)
topk1 = topk_accuracy(y_hat1,y)
topk2 = topk_accuracy(y_hat2,y)
topk3 = topk_accuracy(y_hat3,y)
topk4 = topk_accuracy(y_hat4,y)
topk5 = topk_accuracy(y_hat5,y)
topk6 = topk_accuracy(y_hat6,y)
#topk1 = 0
#topk2 = 0
#topk3 = 0
#topk4 = 0
#topk5 = 0
#topk6 = 0
if best_threshold is None:
best_threshold = utils.best_threshold(y,y_hat4)
print(best_threshold)
y_hat4_raw = y_hat4.copy()
y_hat8 = softmax(y_hat4_raw,axis = 1)
y_hat1[y_hat1>0.5] = 1
y_hat1[y_hat1<=0.5] = 0
y_hat2[y_hat2>0.5] = 1
y_hat2[y_hat2<=0.5] = 0
y_hat3[y_hat3>0.5] = 1
y_hat3[y_hat3<=0.5] = 0
y_hat5[y_hat5>0.0] = 1
y_hat5[y_hat5<=0.0] = 0
y_hat6[y_hat6>0.5] = 1
y_hat6[y_hat6<=0.5] = 0
y_hat7 = np.zeros((y.shape[0],y.shape[1]))
y_hat7[y_hat4>best_threshold] = 1
y_hat4[y_hat4>0.5] = 1
y_hat4[y_hat4<=0.5] = 0
print(np.mean(y_hat8))
print(np.max(y_hat8))
print('-'*10,'方案1','-'*10)
metrics_test = all_metrics(y_hat1, y)
print_metrics(metrics_test)
report = metrics.classification_report(y, y_hat1, digits=4,target_names = id2labels)
print(report)
print(topk1)
print('-'*10,'方案2','-'*10)
metrics_test = all_metrics(y_hat2, y)
print_metrics(metrics_test)
report = metrics.classification_report(y, y_hat2, digits=4,target_names = id2labels)
print(report)
print(topk2)
print('-'*10,'方案3','-'*10)
metrics_test = all_metrics(y_hat3, y)
print_metrics(metrics_test)
report = metrics.classification_report(y, y_hat3, digits=4,target_names = id2labels)
print(report)
print(topk3)
print('-'*10,'方案4','-'*10)
metrics_test = all_metrics(y_hat4, y)
print_metrics(metrics_test)
report = metrics.classification_report(y, y_hat4, digits=4,target_names = id2labels)
print(report)
print(topk4)
print('-'*10,'方案5','-'*10)
metrics_test = all_metrics(y_hat5, y)
print_metrics(metrics_test)
report = metrics.classification_report(y, y_hat5, digits=4,target_names = id2labels)
print(report)
print(topk5)
print('-'*10,'方案6','-'*10)
metrics_test = all_metrics(y_hat6, y)
print_metrics(metrics_test)
report = metrics.classification_report(y, y_hat6, digits=4,target_names = id2labels)
print(report)
print(topk6)
print('-'*10,'方案7','-'*10)
metrics_test = all_metrics(y_hat7, y)
print_metrics(metrics_test)
report = metrics.classification_report(y, y_hat7, digits=4,target_names = id2labels)
print(report)
print(topk6)
print('-'*10,'方案8','-'*10)
metrics_test = all_metrics((y_hat8>0.01).astype(int), y)
print_metrics(metrics_test)
report = metrics.classification_report(y, (y_hat8>0.01).astype(int), digits=4,target_names = id2labels)
print('-'*10,'方案9','-'*10)
metrics_test = all_metrics((y_hat8>0.02).astype(int), y)
print_metrics(metrics_test)
report = metrics.classification_report(y, (y_hat8>0.02).astype(int), digits=4,target_names = id2labels)
print('-'*10,'方案10','-'*10)
metrics_test = all_metrics((y_hat8>0.05).astype(int), y)
print_metrics(metrics_test)
report = metrics.classification_report(y, (y_hat8>0.05).astype(int), digits=4,target_names = id2labels)
print('-'*10,'方案11','-'*10)
metrics_test = all_metrics((y_hat8>0.015).astype(int), y)
print_metrics(metrics_test)
report = metrics.classification_report(y, (y_hat8>0.015).astype(int), digits=4,target_names = id2labels)
print('-'*10,'方案12','-'*10)
metrics_test = all_metrics((y_hat8>0.025).astype(int), y)
print_metrics(metrics_test)
report = metrics.classification_report(y, (y_hat8>0.025).astype(int), digits=4,target_names = id2labels)
print('-'*10,'方案13','-'*10)
metrics_test = all_metrics((y_hat8>0.035).astype(int), y)
print_metrics(metrics_test)
report = metrics.classification_report(y, (y_hat8>0.035).astype(int), digits=4,target_names = id2labels)
print('-'*10,'方案14','-'*10)
metrics_test = all_metrics((y_hat8>0.04).astype(int), y)
print_metrics(metrics_test)
report = metrics.classification_report(y, (y_hat8>0.04).astype(int), digits=4,target_names = id2labels)
print('-'*10,'方案15','-'*10)
metrics_test = all_metrics((y_hat8>0.45).astype(int), y)
print_metrics(metrics_test)
report = metrics.classification_report(y, (y_hat8>0.45).astype(int), digits=4,target_names = id2labels)
print('-'*10,'方案16','-'*10)
metrics_test = all_metrics((y_hat8>0.5).astype(int), y)
print_metrics(metrics_test)
report = metrics.classification_report(y, (y_hat8>0.5).astype(int), digits=4,target_names = id2labels)
print('-'*10,'方案17','-'*10)
metrics_test = all_metrics((y_hat4_raw>0.45).astype(int), y)
print_metrics(metrics_test)
report = metrics.classification_report(y, (y_hat4_raw>0.45).astype(int), digits=4,target_names = id2labels)
print('-'*10,'方案18','-'*10)
metrics_test = all_metrics((y_hat4_raw>0.55).astype(int), y)
print_metrics(metrics_test)
report = metrics.classification_report(y, (y_hat4_raw>0.55).astype(int), digits=4,target_names = id2labels)
else:
y = np.concatenate(y,axis=0)
y_hat = np.concatenate(y_hat1,axis=0)
topk = topk_accuracy(y_hat,y)
y_hat[y_hat>0.5] = 1
y_hat[y_hat<=0.5] = 0
print('-'*20)
metrics_test = all_metrics(y_hat, y)
print_metrics(metrics_test)
report = metrics.classification_report(y, y_hat, digits=4,target_names = id2labels)
print(report)
print(topk)
if not test_set:
return metrics_test["f1_micro"], report
else:
return metrics_test["f1_micro"], report, json.dumps(predict_ans,ensure_ascii=False,indent=4)
if __name__ == '__main__':
parser = argparse.ArgumentParser(description='Baseline')
parser.add_argument('--model_name', type=str, default='BertCNN_v1')
parser.add_argument('--bert_path', type=str, default='bert_chinese', help='pretrained path')
parser.add_argument('--data_path', type=str, default="data/electronic-medical-record")
parser.add_argument('--embedding_dim', type=int, default=768)
parser.add_argument('--dropout_rate', type=float, default=0.3)
parser.add_argument('--accumulation_steps',type=int ,default = 1)
parser.add_argument('--epochs', type=int, default=30)
parser.add_argument('--batch_size', type=int, default=8)
parser.add_argument('--bert_lr', type=float, default=1e-5)
parser.add_argument('--other_lr', type=float, default=5e-4)
parser.add_argument('--warmup_rate', type=float, default=0.3)
parser.add_argument('--patience', type=int, default=5)
parser.add_argument('--seed', type=int, default=1)
parser.add_argument('--result_path', type=str, default="result")
parser.add_argument('--gpu', type=int, default=1)
parser.add_argument('--graph', type=int, default=5)
parser.add_argument('--data_version', type=str, default='')
parser.add_argument('--use_wandb',action="store_true", default=False)
parser.add_argument('--data_loader',type=str,default='DataLoader_short')
parser.add_argument('--max_length',type=int,default=510)
parser.add_argument('--test_freq',type=int,default=1)
parser.add_argument('--test_only',action="store_true", default=False)
parser.add_argument('--test_model_path',type=str,default='')
# 隐藏数
parser.add_argument('--hidden_size',type=int,default=768)
# 标签平滑
parser.add_argument('--label_smooth_lambda',type=float,default=0)
parser.add_argument('--sample_radio',type=int,default=2)
args = parser.parse_args()
save_model_names = [args.bert_path.split('/')[-1], args.model_name, "seed", str(args.seed),
datetime.datetime.now().strftime("%Y-%m-%d-%H-%M-%S")]
save_model_path = os.path.join("checkpoints", '_'.join(save_model_names) + ".pth") # best model path
result_path = os.path.join("result", '_'.join(save_model_names) + ".txt") # the report of test dataset path
error_path = os.path.join("result", '_'.join(save_model_names) + "err.txt") # the report of test dataset path
score_path = os.path.join("result",'_'.join(save_model_names) + "score.txt") # 保存分数获取结果
correct_path = os.path.join(args.data_path,'train_correct.json') # the correct of test dataset path
opt = DefaultConfig(args, save_model_path)
print(opt.use_wandb)
if opt.use_wandb:
save_opt = copy.copy(opt)
save_opt.entity2id = None
save_opt.label2id = None
save_opt.adj_matrixs = None
save_opt.ent_tokens_matrix = None
# start a new wandb run to track this script
wandb.init(
# set the wandb project where this run will be logged
project="diagnose",
# track hyperparameters and run metadata
config=save_opt
)
import dataloader
data_loader = eval('dataloader.' + opt.data_loader)
print(opt)
train_data_loader = data_loader(opt.train_path, opt, shuffle=True)
dev_data_loader = data_loader(opt.dev_path, opt, shuffle=False)
test_data_loader = data_loader(opt.test_path, opt, shuffle=False)
if opt.test_only:
opt.save_model_path = opt.test_model_path
else:
train(opt, train_data_loader, dev_data_loader,test_data_loader)
model = import_module('Model.' + opt.model_name).Model(opt)
if torch.cuda.is_available():
model = model.cuda(opt.gpu)
save_dict = torch.load(opt.save_model_path,map_location=torch.device('cuda:%d'%opt.gpu))
model.load_state_dict(save_dict['dict'])
#micro_f1, report = inference(model, dev_data_loader, opt)
micro_f1, report,error_report = inference(model, test_data_loader, opt,test_set=True)
print('测试集micro-f1',micro_f1)
write_result(report, result_path)
write_result(error_report, error_path)
if opt.use_wandb:
# [optional] finish the wandb run, necessary in notebooks
wandb.finish()
print("==============Finish==============")