-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathtrain_classfication_the_EKnet.py
1257 lines (990 loc) · 53 KB
/
train_classfication_the_EKnet.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
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
import enum
import os
import torch
import pandas as pd
import numpy as np
# from torch.utils.data import Dataset, DataLoader
import torch.nn as nn
import torch.nn.functional as F
import torch.optim as optim
from torch.cuda import amp
import torchvision
from torchvision import transforms as T
# from torchvision.io import read_image
from tqdm import tqdm
from datetime import datetime
import time
from sklearn.metrics import classification_report,roc_auc_score,average_precision_score
# from dataset import dataloader_classfication_png as dataloader
from dataset import dataloader_classfication as dataloader
# from dataset import dataloader_PTB_classfication as dataloader
from tools import dir_utils,losses
from configs.load_yaml import load_yaml
from tools.lr_warmup_scheduler import GradualWarmupScheduler
import gc
import shutil
# import wandb
# from models.unet_e import Model
import multiprocessing as mp
import random
def onehot2pt_dict(seg_label):
'''
input: arr = np.array([0,0,0,1,1,1,1,2,2,2,2,2,3,3,3,3,1,1,1,1,2,2,2,3,3,3,0,2,2,1])
'''
def consecutive(data, stepsize=0):
'''
[array([0, 0, 0]),array([1, 1, 1, 1]),array([2, 2, 2, 2, 2]),
array([3, 3, 3, 3]),array([1, 1, 1, 1])]
'''
return np.split(data, np.where(np.diff(data) != stepsize)[0]+1)
if isinstance(seg_label, np.ndarray):
seg_label = np.array(seg_label)
seg_list = consecutive(seg_label, stepsize=0)
seg_dict = {}
passed_count = 0
for seg_list_i in seg_list:
key = str(seg_list_i[0])
if key not in seg_dict:
seg_dict[key] = []
# print('seg_list_i',seg_list_i)
start_pt = passed_count
end_pt = passed_count+len(seg_list_i)
seg_dict[key].append([start_pt,end_pt])
passed_count = end_pt
# print('passed_count',passed_count,'len(seg_list)',len(seg_label))
assert passed_count == len(seg_label)
return seg_dict
from copy import deepcopy
def calc_f1_threshs(val_pred_all,val_label_all,target_names,out_all_f1_csv,out_max_f1_csv,min_gap_thresh=0.01):
# df_label=val_label_all
# df_pred=val_pred_all
# target_names=target_names
# target_names_stats = target_names['micro avg','macro avg','weighted avg','samples avg']
# df=pd.DataFrame()
t_dict={}
threshs_cand=[]
for i in np.arange(0,1,min_gap_thresh):
threshs_cand.append(i)
t_dict['T']=threshs_cand
# 1. get 100 f1 results from every thresh
for target_name in target_names:
t_dict[target_name]=[]
for t in np.arange(0,1,min_gap_thresh):
df_pred_c=deepcopy(val_pred_all)
df_pred_c[df_pred_c<t]=0
df_pred_c[df_pred_c>=t]=1
val_dict = classification_report(val_label_all, df_pred_c,target_names=target_names,output_dict=True)
for target_name in target_names:
# t_dict[target_name].append(0)
f1_i = val_dict[target_name]['f1-score']
t_dict[target_name].append(f1_i)
# for k,v in val_dict.items():
# t_dict[k][-1]=val_dict[k]['f1-score']
df_all_thresh=pd.DataFrame.from_dict(t_dict)
df_all_thresh.to_csv(out_all_f1_csv,encoding='utf-8_sig',index=False)
# 2.get default 0.5 f1 results
# t_defult=[]
df_pred_c=deepcopy(val_pred_all)
df_pred_c[df_pred_c<0.5]=0
df_pred_c[df_pred_c>=0.5]=1
val_dict_default = classification_report(val_label_all, df_pred_c,target_names=target_names,output_dict=True)
# for k,v in val_dict_default.items():
# t_defult.append(val_dict_default[k]['f1-score'])
t_defult = []
for k in target_names:
t_defult.append(val_dict_default[k]['f1-score'])
# 3. get best thresh of each cls from results
best_f1s=[]
best_threshs=[]
for idx,target_name in enumerate(target_names):
f1_cls_thresh=df_all_thresh[target_name].tolist()
best_f1 = max(f1_cls_thresh)
default_f1 = t_defult[idx]
if abs(default_f1-best_f1) <= 0.03: # 差距太小不用
best_threshs.append(0.5)
else:
best_threshs.append(threshs_cand[f1_cls_thresh.index(best_f1)])
best_f1s.append(best_f1)
# 4. get best pred each cls in 0 1 from best thresh
# val_pred_all_best = []
val_pred_all_best=deepcopy(val_pred_all)
distributions = []
for idx in range(len(target_names)):
best_thresh = best_threshs[idx]
val_pred_all_best[:,idx][val_pred_all_best[:,idx]>=best_thresh]=1
val_pred_all_best[:,idx][val_pred_all_best[:,idx]<best_thresh]=0
distributions.append(sum(val_label_all[:,idx]))
# 5. calc the results again, why not
assert val_pred_all_best.shape == val_pred_all.shape, 'val_pred_all_best {} != val_pred_all {}'.format(val_pred_all_best.shape,val_pred_all.shape)
val_dict_best = classification_report(val_label_all, val_pred_all_best,target_names=target_names,output_dict=True)
best_f1s_re = []
for k in target_names:
best_f1s_re.append(val_dict_best[k]['f1-score'])
df_max=pd.DataFrame()
df_max['label']=['micro avg','macro avg']+target_names
df_max['num']=[np.nan,np.nan]+distributions
df_max['best_thresh']=[np.nan,np.nan]+best_threshs
# df_max['best_f1']=[val_dict_best['micro avg']['f1-score'],val_dict_best['macro avg']['f1-score']]+best_f1s
df_max['best_f1']=[val_dict_best['micro avg']['f1-score'],val_dict_best['macro avg']['f1-score']]+best_f1s_re
df_max['thresh=0.5']=[val_dict_default['micro avg']['f1-score'],val_dict_default['macro avg']['f1-score']]+t_defult
df_max.to_csv(out_max_f1_csv,encoding='utf-8_sig',index=False)
print(df_max)
print(out_max_f1_csv)
return df_pred_c,val_pred_all_best
import matplotlib.pyplot as plt
def draw_plt(inputs,label=None,outputs=None,idx=None,title='',save_path=None,sample_rates=500):
'''
inputs ([32, 1, 5000]) label/outputs ([32, 5000])
'''
# print('draw_plt inputs',inputs.shape,'idx',idx)
# print('inputs[idx,:,:]',inputs[idx,:,:].shape)
# print('-'*12)
inputs_i = np.squeeze(inputs[idx,:,:],0)
# label_i = label[idx,:]
# outputs_i = outputs[idx,:]
# save_path = './a.png'
#x横轴 间隔 大窗口
gap = int(0.04*sample_rates)
xgaps = [i for i in range(0,len(inputs_i),gap)]
gap_big = int(5*0.04*sample_rates)
xgaps_big = [i for i in range(0,len(inputs_i)+gap_big,gap_big)] #0.2s 一大格 900->1000
#y轴 间隔 大窗口
ygaps_big = [-1,-0.5,0,0.5,1,1.5,2,2.5] #0.5mv 一大格
x_labels = []
for i in xgaps_big:
#隔整数秒显示
if (i*0.002) % 1 == 0:
x_labels.append('{:.1f}s'.format(i*0.002))
else:
x_labels.append('')
y_labels = ['{:.1f}'.format(i) for i in ygaps_big]
fig = plt.figure(figsize=(90,8))
colors = ['k','r','b']
x = np.array([i for i in range(len(inputs_i))])
if label is not None:
values_loop = [outputs[idx,:],label[idx,:]]
else:
values_loop = [outputs[idx,:]]
for i,sem_value in enumerate(values_loop):
ax = fig.add_subplot(len(values_loop), 1, i+1)
ax.plot(inputs_i,'k',alpha=0.99,linewidth=2,label='inputs')
seg_dict = onehot2pt_dict(sem_value)
for k,v in seg_dict.items():
if len(v) > 0:
for v_i in v:
start_pt,end_pt = v_i[0],v_i[1]
ax.plot(x[start_pt:end_pt], inputs_i[start_pt:end_pt], color=colors[int(k)],linewidth=4)
ax.set_title(title,fontsize=35)
#画X轴 时间段 和 y轴 mv
ax.set_xticks(xgaps_big, minor=False)
ax.set_xticklabels(x_labels,fontsize=25)
ax.set_yticks(ygaps_big, minor=False)
ax.set_yticklabels(y_labels,fontsize=25)
ax.xaxis.grid(True, which='major',color='k', linewidth=2)
ax.yaxis.grid(True, which='major',color='k', linewidth=2)
ax.set_xlabel('time [s]')
ax.set_ylabel('signal mV')
ax.set_xlim(-gap_big, len(inputs_i)+gap_big)
ax.set_ylim(ygaps_big[0], ygaps_big[-1])
# plt.legend(['inputs','pred','label'])
plt.legend()
# plt.grid(True)
plt.savefig(save_path,dpi=70)
plt.close()
# return plt
return torchvision.io.read_image(save_path)
def draw_lr(train_lr,save_path):
'''
created by ws
'''
length_train=[]
length_all=[]
for i in range(len(train_lr)):
length_all.append(i)
if i==0 :
length_train.append(str(i))
elif i%10==0:
length_train.append(str(i))
else:
length_train.append('')
fig, ax = plt.subplots(figsize=(len(train_lr)/10,10))
ax.plot(length_all,train_lr,'r',alpha=0.99,linewidth=2,label='learning_rate')
ax.set_xticks(length_all, minor=False) #画线的index 不能有空值
ax.set_xticklabels(length_train,fontsize=25) #需要标出来的空值
ax.set_xlabel('epoch',fontsize=25)
ax.set_ylabel('learning_rate',fontsize=25)
plt.grid(True)
plt.legend()
plt.savefig(save_path,dpi=70)
plt.close()
def draw_loss(train_loss,val_loss,learning_rates,save_path,type='loss'):
'''
created by ws
'''
length_train=[]
length_all=[]
for i in range(len(train_loss)):
length_all.append(i)
if i==0 :
length_train.append(str(i))
elif i%10==0:
length_train.append(str(i))
else:
length_train.append('')
if type == 'loss':
min_train=min(train_loss)
min_val=min(val_loss)
else:
min_train=max(train_loss)
min_val=max(val_loss)
set_ylabel = type
min_t_index=train_loss.index(min_train)
min_v_index=val_loss.index(min_val)
# fig=plt.figure(figsize=(50,4))
fig, ax = plt.subplots(figsize=(len(train_loss)/10,10))
ax.plot(length_all[3:],train_loss[3:],'r',alpha=0.99,linewidth=2,label='train_loss')
ax.plot(length_all[3:],val_loss[3:],'b',alpha=0.99,linewidth=2,label='val_loss')
ax.plot(learning_rates,'g',alpha=0.7,linewidth=2,label='lr_curve_only')
ax.plot(min_t_index,min_train,'r*',markersize=16)
ax.plot(min_v_index,min_val,'b*',markersize=16)
plt.text(min_t_index,min_train,"{:.5f}".format(min_train)+' e'+str(min_t_index),ha='left',va='bottom',weight='bold',rotation=45,fontsize=20)
plt.text(min_v_index,min_val,"{:.5f}".format(min_val)+' e'+str(min_v_index),ha='left',va='bottom',weight='bold',rotation=45,fontsize=20)
ax.set_xticks(length_all, minor=False) #画线的index 不能有空值
ax.set_xticklabels(length_train,fontsize=25) #需要标出来的空值
ax.set_xlabel('epoch',fontsize=25)
ax.set_ylabel(set_ylabel,fontsize=25)
plt.grid(True)
plt.legend()
plt.savefig(save_path,dpi=70)
plt.close()
import ast
from PIL import Image, ImageFont, ImageDraw
# def draw_text(img_path,label_text,subject_valid_target_l,subject_valid_target_p,out_save):
def draw_text(args):
img_path,label_text,subject_valid_target_l,subject_valid_target_p,out_save = args
# creating a image object
im = Image.open(img_path)
draw = ImageDraw.Draw(im)
font = ImageFont.truetype('/home/raid_24T/qiaoran_data24T/SimHei.ttf',size=35)
draw.text((10, 10), '原标注: '+label_text, fill ="black", font = font, align ="left")
draw.text((10, 55), '用标注: '+subject_valid_target_l, fill ="blue", font = font, align ="left")
draw.text((10, 100), '预测为: '+subject_valid_target_p, fill ="red", font = font, align ="left")
im.save(out_save)
def load_pertrain(model,model_path='./'):
pertrain_model = torch.load(model_path)
try:
pertrain_dict = pertrain_model.state_dict()
except:
pertrain_dict = pertrain_model
model_dict = model.state_dict()
pertrained_dict = {}
for k,v in pertrain_dict.items():
if k in model_dict:
if pertrain_dict[k].size() == model_dict[k].size():
pertrained_dict[k]=v
else:
print('{} filter'.format(k))
else:
print('{} filter'.format(k))
model_dict.update(pertrained_dict)
model.load_state_dict(model_dict)
return model
def main(yaml_file,test_mode=False):
######### prepare environment ###########
if torch.cuda.is_available():
device = torch.device('cuda')
device_ids = [i for i in range(torch.cuda.device_count())]
print('===> using GPU {} '.format(device_ids))
else:
device = torch.device('cpu')
print('===> using CPU !!!!!')
opt = load_yaml(yaml_file,saveYaml2output=True)
try_time = opt.TRY_TIME+'_'+opt.RUN_DATE
epoch = opt.OPTIM.NUM_EPOCHS
model_dir = opt.SAVE_DIR+'models/'
visu_dir = opt.SAVE_DIR+'visu/'
dir_utils.mkdir_with_del(model_dir)
dir_utils.mkdir_with_del(visu_dir)
######### dataset ###########
# train_dataset = dataloader.Noise_Dataset(opt.DATASET.TRAIN_CSV, leads=opt.DATASET_CUSTOME.LEADS,
# date_len=opt.DATASET_CUSTOME.INPUT_LENGTH,
# n_max_cls=opt.DATASET_CUSTOME.OUT_C,
# random_crop=True,
# transform = dataloader.get_transform(train=True)
# )
# train_dataset = dataloader.Custome_Dataset(opt.DATASET.TRAIN_CSV, n_max_cls=opt.DATASET_CUSTOME.N_CLS,
# transform = dataloader.get_transform(train=True))
# val_dataset = dataloader.Custome_Dataset(opt.DATASET.VAL_CSV, n_max_cls=opt.DATASET_CUSTOME.N_CLS,
# transform = dataloader.get_transform(train=True))
df = pd.read_csv(opt.DATASET.STATS_CSV)
labels_names = df['names'].tolist()
if hasattr(opt.DATASET_CUSTOME,'operation_style'):
operation_style = opt.DATASET_CUSTOME.operation_style
else:
operation_style = 'ECG'
train_dataset = dataloader.Custome_Dataset(opt.DATASET.TRAIN_CSV,opt.DATASET.DATA_DIR,
labels_names=labels_names,
leads=opt.DATASET_CUSTOME.LEADS,
patch_size=opt.DATASET_CUSTOME.PATCH_SIZE,
operation_style=operation_style,
transform = dataloader.get_transform(train=True))
val_dataset = dataloader.Custome_Dataset(opt.DATASET.VAL_CSV,opt.DATASET.DATA_DIR,
labels_names=labels_names,
leads=opt.DATASET_CUSTOME.LEADS,
patch_size=opt.DATASET_CUSTOME.PATCH_SIZE,
operation_style=operation_style,
transform = dataloader.get_transform(train=False))
# train_dataset = dataloader.Noise_Dataset(opt.DATASET.TRAIN_CSV,
# leads=opt.DATASET_CUSTOME.LEADS,
# date_len=opt.DATASET_CUSTOME.INPUT_LENGTH,
# random_crop=True,
# normlize_singal=opt.DATASET_CUSTOME.NORMLIZE_SINGAL,
# transform = dataloader.get_transform(train=True)
# )
# val_dataset = dataloader.Noise_Dataset(opt.DATASET.VAL_CSV,
# leads=opt.DATASET_CUSTOME.LEADS,
# date_len=opt.DATASET_CUSTOME.INPUT_LENGTH,
# random_crop=False,
# normlize_singal=opt.DATASET_CUSTOME.NORMLIZE_SINGAL,
# transform = dataloader.get_transform(train=False)
# )
train_dataloader = torch.utils.data.DataLoader(train_dataset, batch_size=opt.OPTIM.BATCH_SIZE,
shuffle=True, num_workers=6,
prefetch_factor=3,
persistent_workers=False, #maintain woker alive even consumed
)
val_dataloader = torch.utils.data.DataLoader(val_dataset, batch_size=int(opt.OPTIM.BATCH_SIZE//4),
shuffle=False, num_workers=6,
prefetch_factor=3,
persistent_workers=False, #maintain woker alive even consumed
# drop_last=True,
)
dataset_sizes = {'train':len(train_dataset),
'val':len(val_dataset)}
df_stats = pd.read_csv(opt.DATASET.STATS_CSV)
target_names = df_stats['names'].tolist()
print('===> Loading datasets done')
######### model ###########
# if opt.MODEL.MODE == 'EKPnet':
# from models.EKPnet import Model
# elif opt.MODEL.MODE == 'EKPnet_v1':
# from models.EKPnet_v1 import Model
# elif opt.MODEL.MODE == 'EKPnet_v2s':
# from models.EKPnet_v2s import Model
# else:
# print('{} unrecoginze model'.format(opt.MODEL.MODE))
# assert 1>2
# model = Model(
# input_c=opt.DATASET_CUSTOME.INPUT_LEADS, \
# input_length=opt.DATASET_CUSTOME.INPUT_LENGTH, \
# patch_size=opt.DATASET_CUSTOME.PATCH_SIZE, \
# leads_input=opt.DATASET_CUSTOME.LEADS, \
# embedding_1d_2d=opt.MODEL.embedding_1d_2d, \
# embed_dim=opt.MODEL.embed_dim, \
# keep_ratio=opt.MODEL.keep_ratio, \
# local_depth=opt.MODEL.local_depth, \
# num_heads=opt.MODEL.num_heads, \
# pufication_style=opt.MODEL.pufication_style, \
# self_depth=opt.MODEL.self_depth, \
# normalize_before=opt.MODEL.normalize_before, \
# mode_stage=opt.MODEL.mode_stage, \
# cls_cross_depth=opt.MODEL.cls_cross_depth, \
# num_classes=opt.MODEL.num_classes, \
# ).to(device)
'''
orginal 不需要 knowledge_emb,pos_embed, 随便给一个占位
'''
if opt.MODEL.pos_emb_style == 'orginal':
time_emb,lead_emb = None,None
else:
finetune_path = opt.DATASET_CUSTOME.Pertrain_Path
pertrain_model = torch.load(finetune_path)
time_emb = pertrain_model['time_emb.weight'] #12*256
lead_emb = pertrain_model['lead_emb.weight'] #12*256
if opt.MODEL.MODE == 'EPKnet':
from models.EPKnet import Model
elif opt.MODEL.MODE == 'The_EKnet':
from models.The_EKnet import Model
else:
print('{} unrecoginze model'.format(opt.MODEL.MODE))
assert 1>2
if hasattr(opt.MODEL,'time_emb_order'):
time_emb_order = opt.MODEL.time_emb_order
else:
time_emb_order = 'orginal'
model = Model(
input_c=opt.DATASET_CUSTOME.INPUT_LEADS, \
input_length=opt.DATASET_CUSTOME.INPUT_LENGTH, \
patch_size=opt.DATASET_CUSTOME.PATCH_SIZE, \
leads_input=opt.DATASET_CUSTOME.LEADS, \
embedding_1d_2d=opt.MODEL.embedding_1d_2d, \
embed_dim=opt.MODEL.embed_dim, \
local_depth=opt.MODEL.local_depth, \
num_head=opt.MODEL.num_head, \
cls_cross_depth=opt.MODEL.cls_cross_depth, \
mode_stage=opt.MODEL.mode_stage, \
num_classes=opt.MODEL.num_classes, \
pos_emb_style=opt.MODEL.pos_emb_style,
lead_pos_embedding=(time_emb,lead_emb),
pos_merging_style=opt.MODEL.pos_merging_style,
classfierer_type=opt.MODEL.classfierer_type,
).to(device)
if opt.DATASET_CUSTOME.Using_Pertrain is True:
print('-'*10+'using pertrain!!')
model_path = opt.DATASET_CUSTOME.Pertrain_Path
model = load_pertrain(model,model_path=model_path)
# try:
# model_path = opt.DATASET_CUSTOME.Pertrain_Path
# model = load_pertrain(model,model_path=model_path)
# except Exception as e:
# print(e)
# from models.swin_transformer_1d import Model
# model = Model(in_c=1,out_c=opt.DATASET_CUSTOME.OUT_C,resdiual_output=False).to(device)
if torch.cuda.device_count() > 1:
print("Let's use", torch.cuda.device_count(), "GPUs!")
model = nn.DataParallel(model, device_ids = device_ids)
######### optim ########### d
new_lr = opt.OPTIM.LR_INITIAL
optimizer = optim.Adam(model.parameters(), lr=new_lr, betas=(0.9, 0.999),eps=1e-8)
# from optimizer.adan import Adan
# optimizer = Adan(model.parameters(),lr=new_lr, betas=(0.98, 0.92, 0.99), eps=1e-8,
# weight_decay=0.02, max_grad_norm=0.0, no_prox=False,)
warmup_epochs = 1
scheduler_cosine = optim.lr_scheduler.CosineAnnealingLR(optimizer, opt.OPTIM.NUM_EPOCHS-warmup_epochs, eta_min=opt.OPTIM.LR_MIN)
scheduler = GradualWarmupScheduler(optimizer, multiplier=1, total_epoch=warmup_epochs, after_scheduler=scheduler_cosine)
scheduler.step()
# scheduler = torch.optim.lr_scheduler.OneCycleLR(optimizer, max_lr=0.01, steps_per_epoch=len(train_dataloader), epochs=opt.OPTIM.NUM_EPOCHS)
# CE_criterion = torch.nn.CrossEntropyLoss(weight=torch.FloatTensor([1,1,1]).to(device))
# DiceLoss = losses.DiceLoss().to(device)
# criterion = losses.focal_loss(alpha=None,gamma=2.,reduction='mean',ignore_index=opt.OPTIM.Focal_ignore_idx,device=device)
# criterion = nn.BCELoss()
# criterion = torch.nn.BCEWithLogitsLoss()
print('===> model done')
grad_scaler = amp.GradScaler()
start_epoch = 1
since = time.time()
best_val_f1,best_macro_val_f1 = 0,0
best_save_path = ''
results_dict = {'epoch':[],
'train_loss':[],
'val_loss':[],
'lr':[],
# 'val_loss':[],
# 'val_accuracy':[]
"micro_f1_train_subs":[],
"micro_f1_val_subs":[],
"macro_f1_train_cls":[],
"macro_f1_val_cls":[],
"val_micro_roc":[],
"val_macro_roc":[],
"val_micro_ap":[],
"val_macro_ap":[],
}
results_dict_f1 = {'epoch':[],
"micro_f1_train_subs":[],
"micro_f1_val_subs":[],
"macro_f1_train_cls":[],
"macro_f1_val_cls":[],
}
results_dict_aoc = {'epoch':[],
"val_micro_roc":[],
"val_macro_roc":[],
}
results_dict_ap = {'epoch':[],
"val_micro_ap":[],
"val_macro_ap":[],
}
cls_name_pr = []
# opt.DRAW_EPOCHS =
epoch_split_n = int(opt.OPTIM.NUM_EPOCHS / opt.DRAW_EPOCHS)
torch.autograd.set_detect_anomaly(True)
# for epoch in range(start_epoch, 2 + 1):
for epoch in range(start_epoch, opt.OPTIM.NUM_EPOCHS + 1):
epoch_start_time = time.time()
epoch_train_loss,epoch_train_CE_loss,epoch_train_dice_loss,epoch_train_focal_loss = 0,0,0,0
# train_pred_all,train_label_all = [],[]
# val_pred_all,val_label_all = [],[]
train_pred_all,train_label_all = [],[]
val_pred_all,val_label_all = [],[]
val_pred_all_score = []
#### train ####
model.train()
for i, data in enumerate(train_dataloader):
# for i, data in enumerate(val_dataloader):
# for i, data in enumerate(tqdm(train_dataloader), 0):
# if i >=10:
# break
inputs = data['input'].to(device)
# labels = data['label'].to(device)
labels = data['label_onehot'].to(device)
data_file = data['data_file']
input_lead_loc = data['input_lead_loc'].to(device)
optimizer.zero_grad()
# with torch.set_grad_enabled(True):
torch.set_grad_enabled(True)
# with amp.autocast():
# print('inputs',inputs.shape)
# print('input_lead_loc',input_lead_loc.shape)
outputs = model(inputs,
mask_ratio=opt.DATASET_CUSTOME.MASK_RATIO,
# pure_ways=opt.MODEL.pure_ways,
# fast_pure_ways=opt.MODEL.fast_pure_ways,
)
# if opt.MODEL.MODE == 'vit_1d_cls_embedding':
if opt.MODEL.classfierer_type in ['cls_embeeing','cls_embeeing_seperate']:
criterion = nn.CrossEntropyLoss() # 函数自己有 logsfotmax 不需要额外加softmax
preds = outputs.reshape((outputs.shape[0]*outputs.shape[1], -1)) # [bs, n_cls, 2] -> [bs*n_cls,2]
labels_flatten = labels.reshape(labels.shape[0]*labels.shape[1]).to(torch.int64) # [bs, n_cls] -> [bs*n_cls]
loss = criterion(preds, labels_flatten)
outputs_idx = torch.max(preds, 1)[1]
outputs_idx = outputs_idx.reshape((outputs.shape[0],outputs.shape[1])) #[bs*n_cls,2] max ->[bs*n_cls] -> [bs,n_cls]
else:
criterion = nn.BCELoss()
outputs = torch.sigmoid(outputs)
loss = criterion(outputs, labels)
outputs_idx = outputs.detach().clone()
outputs_idx[outputs_idx>=0.5]=1
outputs_idx[outputs_idx<0.5]=0
loss.backward()
optimizer.step()
train_pred_all.append(outputs_idx.cpu().detach().numpy())
train_label_all.append(labels.cpu().detach().numpy())
epoch_train_loss += loss.item() * inputs.size(0)
train_loss_mean = epoch_train_loss / dataset_sizes['train']
train_pred_all = np.concatenate(train_pred_all, axis=0)
train_label_all = np.concatenate(train_label_all, axis=0)
#### Evaluation ####
model.eval()
epoch_val_loss,epoch_val_CE_loss,epoch_val_dice_loss,epoch_val_focalLoss = 0,0,0,0
# for data in val_dataloader:
for i, data in enumerate(val_dataloader):
# if i >=10:
# break
inputs = data['input'].to(device)
# labels = data['label'].to(device)
labels = data['label_onehot'].to(device)
data_file = data['data_file']
input_lead_loc = data['input_lead_loc'].to(device)
torch.set_grad_enabled(False)
if opt.MODEL.fast_pufication == 'random_mask':
val_mask_ratio = 0.
elif opt.MODEL.fast_pufication == 'score_embedding':
val_mask_ratio = opt.DATASET_CUSTOME.MASK_RATIO
else:
raise ValueError(f'fast_pufication {opt.MODEL.fast_pufication} not support')
outputs = model(inputs,
mask_ratio=0,
# pure_ways='atten',
# fast_pure_ways='atten',
)
# outputs = model(inputs)
# outputs = torch.sigmoid(outputs)
# loss = criterion(outputs, labels)
# if opt.MODEL.MODE == 'vit_1d_cls_embedding':
if opt.MODEL.classfierer_type in ['cls_embeeing','cls_embeeing_seperate']:
criterion = nn.CrossEntropyLoss() # one target index multi cls
preds = outputs.reshape((outputs.shape[0]*outputs.shape[1], -1)) # [bs, n_cls, 2] -> [bs*n_cls,2]
labels_flatten = labels.reshape(labels.shape[0]*labels.shape[1]).to(torch.int64) # [bs, n_cls] -> [bs*n_cls]
loss = criterion(preds, labels_flatten)
preds_score = torch.softmax(preds,1)
outputs_idx = torch.max(preds_score, 1)[1]
outputs_idx = outputs_idx.reshape((outputs.shape[0],outputs.shape[1])) #[bs*n_cls,2] max ->[bs*n_cls] -> [bs,n_cls]
# outputs_score = torch.max(preds_score, 1)[0]
outputs_score = preds_score[:,1]
outputs_score = outputs_score.reshape((outputs.shape[0],outputs.shape[1])) #[bs*n_cls,2] max ->[bs*n_cls] -> [bs,n_cls]
else:
criterion = nn.BCELoss()
outputs = torch.sigmoid(outputs)
outputs_score = outputs.clone()
loss = criterion(outputs, labels)
outputs[outputs>=0.5]=1
outputs[outputs<0.5]=0
outputs_idx = outputs
# outputs[outputs>=0.5]=1
# outputs[outputs<0.5]=0
val_pred_all.append(outputs_idx.cpu().detach().numpy())
val_pred_all_score.append(outputs_score.cpu().detach().numpy())
val_label_all.append(labels.cpu().detach().numpy())
epoch_val_loss += loss.item()* inputs.size(0)
val_pred_all = np.concatenate(val_pred_all, axis=0)
val_pred_all_score = np.concatenate(val_pred_all_score, axis=0)
val_label_all = np.concatenate(val_label_all, axis=0)
val_loss_mean = epoch_val_loss / dataset_sizes['val']
scheduler.step()
# if epoch >30:
# if best_val_loss == 0:
# best_val_loss = val_loss_mean
# if best_val_loss < val_loss_mean:
save_path = model_dir+'model_epoch_{}_val_{:.6f}.pth'.format(epoch,val_loss_mean)
# torch.save({'epoch': epoch,
# 'state_dict': model.state_dict(),
# 'optimizer' : optimizer.state_dict()
# }, save_path)
# torch.save(model, save_path)
if torch.cuda.device_count() > 1: #DataParallel 带有 module, save时候要去掉
torch.save(model.module.state_dict(), save_path)
else:
torch.save(model, save_path)
best_val_loss = val_loss_mean
# print(save_path)
train_dict = classification_report(train_label_all, train_pred_all,target_names=target_names,output_dict=True)
val_dict = classification_report(val_label_all, val_pred_all, target_names=target_names,output_dict=True)
# print('val_label_all',val_label_all,val_label_all.shape)
# print('val_pred_all',val_pred_all_score,val_pred_all_score.shape)
# np.save('./val_label_all', val_label_all)
# np.save('./val_pred_all', val_pred_all_score)
val_dict_roc = roc_auc_score(val_label_all, val_pred_all_score, average=None)
val_micro_roc = roc_auc_score(val_label_all, val_pred_all_score, average='micro')
val_macro_roc = roc_auc_score(val_label_all, val_pred_all_score, average='macro')
val_dict_ap = average_precision_score(val_label_all, val_pred_all_score, average=None)
val_micro_ap = average_precision_score(val_label_all, val_pred_all_score, average='micro')
val_macro_ap = average_precision_score(val_label_all, val_pred_all_score, average='macro')
# print(classification_report(val_label_all, val_pred_all, target_names=target_names))
# print('train_dict',train_dict)
# print('val_dict',val_dict)
train_micro_avg_f1 = train_dict['micro avg']['f1-score']
train_macro_avg_f1 = train_dict['macro avg']['f1-score']
val_micro_avg_f1 = val_dict['micro avg']['f1-score']
val_macro_avg_f1 = val_dict['macro avg']['f1-score']
'''
'''
if epoch == 1:
best_val_f1 = val_micro_avg_f1
best_save_path = save_path
best_epoch = epoch
if val_micro_avg_f1 > best_val_f1 :
best_val_f1 = val_micro_avg_f1
best_save_path = save_path
best_epoch = epoch
elif val_macro_avg_f1 > best_macro_val_f1 :
best_macro_val_f1 = val_macro_avg_f1
best_save_path = save_path
best_epoch = epoch
results_dict['macro_f1_train_cls'].append(train_macro_avg_f1*100)
results_dict['macro_f1_val_cls'].append(val_macro_avg_f1*100)
results_dict['micro_f1_train_subs'].append(train_micro_avg_f1*100)
results_dict['micro_f1_val_subs'].append(val_micro_avg_f1*100)
results_dict['val_micro_roc'].append(val_micro_roc*100)
results_dict['val_macro_roc'].append(val_macro_roc*100)
results_dict['val_micro_ap'].append(val_micro_ap*100)
results_dict['val_macro_ap'].append(val_macro_ap*100)
results_dict_f1['epoch'].append(epoch)
results_dict_aoc['epoch'].append(epoch)
results_dict_ap['epoch'].append(epoch)
results_dict_f1['macro_f1_train_cls'].append(train_macro_avg_f1*100)
results_dict_f1['macro_f1_val_cls'].append(val_macro_avg_f1*100)
results_dict_f1['micro_f1_train_subs'].append(train_micro_avg_f1*100)
results_dict_f1['micro_f1_val_subs'].append(val_micro_avg_f1*100)
results_dict_aoc['val_micro_roc'].append(val_micro_roc*100)
results_dict_aoc['val_macro_roc'].append(val_macro_roc*100)
results_dict_ap['val_micro_ap'].append(val_micro_roc*100)
results_dict_ap['val_macro_ap'].append(val_macro_roc*100)
for cls_idx,cls_name in enumerate(target_names):
r = train_dict[cls_name]['recall']*100
p = train_dict[cls_name]['precision']*100
f1 = train_dict[cls_name]['f1-score']*100
r_v = val_dict[cls_name]['recall']*100
p_v = val_dict[cls_name]['precision']*100
f1_v = val_dict[cls_name]['f1-score']*100
roc_v = val_dict_roc[cls_idx]*100
ap_v = val_dict_ap[cls_idx]*100
if '{}_f1'.format(cls_name) not in results_dict_f1:
results_dict_f1['{}_f1'.format(cls_name)]=[]
results_dict_f1['{}_p'.format(cls_name)]=[]
results_dict_f1['{}_r'.format(cls_name)]=[]
results_dict_aoc['{}'.format(cls_name)]=[]
results_dict_ap['{}'.format(cls_name)]=[]
cls_name_pr.append('{}_p'.format(cls_name))
cls_name_pr.append('{}_r'.format(cls_name))
results_dict_f1['{}_f1'.format(cls_name)].append(f1_v)
results_dict_f1['{}_p'.format(cls_name)].append(p_v)
results_dict_f1['{}_r'.format(cls_name)].append(r_v)
results_dict_aoc['{}'.format(cls_name)].append(roc_v)
results_dict_ap['{}'.format(cls_name)].append(ap_v)
if 'accuracy' not in train_dict:
train_dict['accuracy'] = 0.0
if 'accuracy' not in val_dict:
val_dict['accuracy'] = 0.0
# results_dict['val_accuracy'].append(val_dict['accuracy']*100)
results_dict['epoch'].append(epoch)
results_dict['train_loss'].append(train_loss_mean)
results_dict['val_loss'].append(val_loss_mean)
results_dict['lr'].append(scheduler.get_lr()[0])
# print('results_dict',results_dict)
if epoch > 5:
large_ratio = results_dict['train_loss'][4]/new_lr
draw_loss(results_dict['train_loss'],results_dict['val_loss'],np.array(results_dict['lr'])*large_ratio,opt.SAVE_DIR+'loss.png',type='loss')
draw_loss(results_dict['micro_f1_train_subs'],results_dict['micro_f1_val_subs'],np.array(results_dict['lr'])*(50/new_lr),opt.SAVE_DIR+'micro_f1.png',type='micro f1')
draw_loss(results_dict['macro_f1_train_cls'],results_dict['macro_f1_val_cls'],np.array(results_dict['lr'])*(50/new_lr),opt.SAVE_DIR+'macro_f1.png',type='macro f1')
draw_lr(results_dict['lr'],opt.SAVE_DIR+'lr.png')
# assert 1>2
print("------------------------------------------------------------------")
print("Epoch: {}\tTime: {:.4f}s \t train Loss: {:.6f} train macro f1: {:.2f} \t val loss: {:.6f} val macro f1: {:.2f} \t LearningRate {:.8f}".format(
epoch, time.time()-epoch_start_time, train_loss_mean,train_macro_avg_f1*100, val_loss_mean,val_macro_avg_f1*100, scheduler.get_lr()[0]))
# print('train CE loss {:.1f}: {:.6f}, val CE loss: {:.6f}, train Dice loss {:.1f}: {:.6f}, val Dice loss: {:.6f}, train focal loss {:.1f}: {:.6f}, val focal loss: {:.6f}, '.format(
# opt.OPTIM.CE_ratio,train_CE_loss_mean,val_CE_loss_mean,opt.OPTIM.Dice_ratio,train_dice_loss_mean,val_dice_loss_mean,opt.OPTIM.Focal_ratio,train_focal_loss_mean,val_focal_loss_mean)
# )
print("------------------------------------------------------------------")
# results_dict.pop('macro_f1_train_cls')
# results_dict.pop('micro_f1_train_subs')
# print(results_dict)
df = pd.DataFrame.from_dict(results_dict).round(6)
# df = df.sort_values(by=['timeint'])
df = df.drop(['macro_f1_train_cls'], axis=1)
df = df.drop(['micro_f1_train_subs'], axis=1)
df = df.drop(['lr'], axis=1)
df.to_csv(opt.SAVE_DIR+'results.csv',index=False,encoding='utf-8_sig')
# for k,v in results_dict_f1.items():
# print(k,v,len(v))
try:
df = pd.DataFrame.from_dict(results_dict_f1).round(6)
except:
for k,v in results_dict_f1.items():
print(k,v,len(v))
assert 1>2
df.to_csv(opt.SAVE_DIR+'results_pr_f1.csv',index=False,encoding='utf-8_sig')
df = df.drop(cls_name_pr, axis=1)
df.to_csv(opt.SAVE_DIR+'results_f1.csv',index=False,encoding='utf-8_sig')
df = pd.DataFrame.from_dict(results_dict_aoc).round(6)
df.to_csv(opt.SAVE_DIR+'results_aoc.csv',index=False,encoding='utf-8_sig')
df = pd.DataFrame.from_dict(results_dict_ap).round(6)
df.to_csv(opt.SAVE_DIR+'results_ap.csv',index=False,encoding='utf-8_sig')
analya_last = False
if analya_last is True:
from val_draw_atten_multi_label import analysis_atten
# best_epoch = 300
# best_save_path = '/home/raid_24T/qiaoran_data24T/All_project_model_output/ecg_clsffication/ECG_children_1100_res34_n_cls15_T0_2021_12_01-14_12/models/model_epoch_300_val_0.000023.pth'
# draw_dir = opt.SAVE_DIR+'e{}/'.format(best_epoch)
# '/home/raid_24T/qiaoran_data24T/儿科心电图/clean_data/xml数据/draw/'
# draw_dir = '/home/raid_24T/qiaoran_data24T/Ruijing_data/ECG_data/40W/orginal_png/'
# data_file = '{}{}.json'.format(self.data_dir,names_i)
print(best_save_path)
model = torch.load(best_save_path)
model.eval()
torch.set_grad_enabled(False)
val_pred_all,val_label_all,data_files = [],[],[]
val_pred_all_threshs = []
# for data in val_dataloader:
for i, data in enumerate(val_dataloader):
# if i >=10:
# break
inputs = data['input'].to(device)
labels = data['label_onehot'].to(device)
data_file = data['data_file']
input_lead_loc = data['input_lead_loc'].to(device)
outputs = model(inputs,
input_lead_loc,
mask_ratio=0.0,
random_mask_in_lead=False)
# outputs = model(inputs)
# if opt.MODEL.MODE == 'vit_1d_cls_embedding':
# if opt.MODEL.classfierer_type in ['cls_embeeing']:
if opt.MODEL.classfierer_type in ['cls_embeeing','cls_embeeing_seperate']:
preds = outputs.reshape((outputs.shape[0]*outputs.shape[1], -1)) # [bs, n_cls, 2] -> [bs*n_cls,2]
# labels_flatten = labels.reshape(labels.shape[0]*labels.shape[1]).to(torch.int64) # [bs, n_cls] -> [bs*n_cls]
outputs_idx = torch.max(preds, 1)[1]
outputs_idx = outputs_idx.reshape((outputs.shape[0],outputs.shape[1])) #[bs*n_cls,2] max ->[bs*n_cls] -> [bs,n_cls]
outputs_score = torch.max(preds, 1)[0]
outputs_score = outputs_score.reshape((outputs.shape[0],outputs.shape[1])) #[bs*n_cls,2] max ->[bs*n_cls] -> [bs,n_cls]
val_pred_all_threshs.append(outputs_score.cpu().detach().numpy())
else:
outputs = torch.sigmoid(outputs)
val_pred_all_threshs.append(outputs.cpu().detach().numpy())
outputs[outputs>=0.5]=1
outputs[outputs<0.5]=0
outputs_idx = outputs