-
Notifications
You must be signed in to change notification settings - Fork 58
/
Train_cnn_multi.cpp
1130 lines (1072 loc) · 28.3 KB
/
Train_cnn_multi.cpp
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
#include<iostream>
#include<cstring>
#include<cstdio>
#include<map>
#include<vector>
#include<string>
#include<ctime>
#include<cmath>
#include<cstdlib>
#include<sstream>
#include<pthread.h>
#include<ctime>
#include<sys/time.h>
#include<assert.h>
using namespace std;
//Ruobing Xie
//Representation Learning of Knowledge Graphs with Entity Descriptions
#define pi 3.1415926535897932384626433832795
#define THREADS_NUM 16
bool L1_flag=1;
//convolutional layer
int window_1 = 2; //1st window size
int window_2 = 1; //2nd window size
double **conv_w1;
double **conv_w1_tmp[THREADS_NUM];
double **conv_w2;
double **conv_w2_tmp[THREADS_NUM];
int n_pooling_1 = 4; //1st layer 4-pooling
//normal distribution
double rand(double min, double max)
{
return min+(max-min)*rand()/(RAND_MAX+1.0);
}
double normal(double x, double miu,double sigma)
{
return 1.0/sqrt(2*pi)/sigma*exp(-1*(x-miu)*(x-miu)/(2*sigma*sigma));
}
double randn(double miu,double sigma, double min ,double max)
{
double x,y,dScope;
do{
x=rand(min,max);
y=normal(x,miu,sigma);
dScope=rand(0.0,normal(miu,miu,sigma));
}while(dScope>y);
return x;
}
double sqr(double x)
{
return x*x;
}
double vec_len(vector<double> &a)
{
double res=0;
for (int i=0; i<a.size(); i++)
res+=a[i]*a[i];
res = sqrt(res);
return res;
}
double norm(vector<double> &a)
{
double x = vec_len(a);
if (x>1)
for (int ii=0; ii<a.size(); ii++)
a[ii]/=x;
return 0;
}
int rand_max(int x)
{
int res = (rand()*rand())%x;
while (res<0)
res+=x;
return res;
}
string version; //bern/unif
char buf[100000],buf1[100000];
int relation_num,entity_num,word_num;
map<string,int> relation2id,entity2id,word2id;
map<int,string> id2entity,id2relation,id2word;
vector<vector<int> > entityWords_vec;
int nbatches, batchsize;
map<int,map<int,int> > left_entity,right_entity;
map<int,double> left_num,right_num;
int n,n_1,n_w,method; //n£ºdimension of entity/relation, n_1: dimension of feature_vec in 1st-layer CNN, n_w: dimension of word
double res_triple,res_cnn,res_c_e,res_e_c,res_normal; //loss function value
double res_thread_triple[THREADS_NUM], res_thread_cnn[THREADS_NUM], res_thread_normal[THREADS_NUM]; //loss for each thread
double res_thread_c_e[THREADS_NUM], res_thread_e_c[THREADS_NUM];
double count,count1;
double rate,rate_c,rate_m,rate_n; //learning rate
double margin,margin_c,margin_m; //margin
vector<int> fb_h,fb_l,fb_r;
vector<vector<double> > relation_vec,word_vec,entity_vec; //embeddings
vector<vector<double> > relation_tmp,word_tmp,entity_tmp;
vector<vector<double> > cnn_vec; //cnn_vec for entity
vector<vector<double> > cnn_grad;
vector<vector<int> > l2_MaxPooling;
vector<int> cnnVecState; //entity state
vector<vector<vector<double> > > mid_vec; //1st-layer output
vector<vector<vector<int> > > l1_MaxPooling;
vector<vector<vector<double> > > mid_grad;
vector<double> posErrorVec[THREADS_NUM];
vector<double> negErrorVec[THREADS_NUM];
pthread_mutex_t mut_mutex;
void sgd();
void train_triple_mul(int, int, int, int, int, int, int);
void train_cnn_mul(int, int, int, int, int, int, int);
void train_e_c_mul(int, int, int, int, int, int, int);
void train_c_e_mul(int, int, int, int, int, int, int);
map<pair<int,int>, map<int,int> > ok; //mark of positive
void add(int x,int y,int z)
{
fb_h.push_back(x);
fb_r.push_back(z);
fb_l.push_back(y);
ok[make_pair(x,z)][y]=1;
}
void run(int n_in,int n_1_in,int n_w_in,double rate_in,double margin_in,int method_in)
{
n = n_in; //init
n_1 = n_1_in;
n_w = n_w_in;
rate = rate_in;
rate_c = 1*rate;
rate_n = 0.1*rate;
rate_m = 0.4*rate;
margin = margin_in;
margin_c = 1*margin;
margin_m = 0.8*margin;
method = method_in;
relation_vec.resize(relation_num);
for (int i=0; i<relation_vec.size(); i++)
relation_vec[i].resize(n);
entity_vec.resize(entity_num);
for (int i=0; i<entity_vec.size(); i++)
entity_vec[i].resize(n);
relation_tmp.resize(relation_num);
for (int i=0; i<relation_tmp.size(); i++)
relation_tmp[i].resize(n);
entity_tmp.resize(entity_num);
for (int i=0; i<entity_tmp.size(); i++)
entity_tmp[i].resize(n);
for (int i=0; i<relation_num; i++) //init
{
for (int ii=0; ii<n; ii++)
relation_vec[i][ii] = randn(0,1.0/n,-6/sqrt(n),6/sqrt(n)); //pre-trained embeddings are optional
}
for (int i=0; i<entity_num; i++)
{
for (int ii=0; ii<n; ii++)
entity_vec[i][ii] = randn(0,1.0/n,-6/sqrt(n),6/sqrt(n)); //pre-trained embeddings are optional
norm(entity_vec[i]);
}
//initialization for word embeddings
word_vec.resize(word_num);
for (int i=0; i<word_vec.size(); i++)
word_vec[i].resize(n_w);
for (int i=0; i<word_num; i++)
{
for (int ii=0; ii<n_w; ii++)
word_vec[i][ii] = randn(0,1.0/n_w,-6/sqrt(n_w),6/sqrt(n_w));
norm(word_vec[i]);
}
/*
// optional: init by pre-trained word embeddings
stringstream ss;
ss << n_w;
string n_w_string = "../../../word2vec-41/word_vector_" + ss.str() + ".txt";
FILE* f1 = fopen(n_w_string.c_str(), "r");
while (fscanf(f1,"%s",buf)==1)
{
string st=buf;
if(word2id.count(st)==0)
{
vector<double> tempx_word_vec;
tempx_word_vec.resize(n_w);
for (int ii=0; ii<n_w; ii++)
fscanf(f1,"%lf",&tempx_word_vec[ii]);
}
else
{
int tempWordID = word2id[st];
for (int ii=0; ii<n_w; ii++)
fscanf(f1,"%lf",&word_vec[tempWordID][ii]);
norm(word_vec[tempWordID]);
}
}
*/
//convolution init
conv_w1 = new double *[n_1];
for (int i=0; i<n_1; i++)
{
conv_w1[i] = new double[n_w*window_1];
for (int ii=0; ii<n_w*window_1; ii++)
{
conv_w1[i][ii] = randn(0,1.0/(n_w*window_1),-6/sqrt(n_w*window_1),6/sqrt(n_w*window_1));
}
}
conv_w2 = new double *[n];
for (int i=0; i<n; i++)
{
conv_w2[i] = new double[n_1*window_2];
for (int ii=0; ii<n_1*window_2; ii++)
{
conv_w2[i][ii] = randn(0,1.0/(n_1*window_2),-6/sqrt(n_1*window_2),6/sqrt(n_1*window_2));
}
}
for(int k=0; k<THREADS_NUM; k++)
{
conv_w1_tmp[k] = new double *[n_1];
for (int i=0; i<n_1; i++)
{
conv_w1_tmp[k][i] = new double[n_w*window_1];
for (int ii=0; ii<n_w*window_1; ii++)
{
conv_w1_tmp[k][i][ii] = 0;
}
}
}
for(int k = 0;k<THREADS_NUM;k++)
{
conv_w2_tmp[k] = new double *[n];
for (int i=0; i<n; i++)
{
conv_w2_tmp[k][i] = new double[n_1*window_2];
for (int ii=0; ii<n_1*window_2; ii++)
{
conv_w2_tmp[k][i][ii] = 0;
}
}
}
//cnn vec init
cnn_vec.resize(entity_num);
for (int i=0; i<cnn_vec.size(); i++)
cnn_vec[i].resize(n);
cnn_grad.resize(entity_num);
for (int i=0; i<cnn_grad.size(); i++)
{
cnn_grad[i].resize(n);
for(int j=0; j<n; j++)
{
cnn_grad[i][j] = 0;
}
}
l2_MaxPooling.resize(entity_num);
for (int i=0; i<l2_MaxPooling.size(); i++)
l2_MaxPooling[i].resize(n);
mid_vec.resize(entity_num);
for (int i=0; i<mid_vec.size(); i++)
{
int l1_length = (entityWords_vec[i].size()-1)/n_pooling_1+1;
mid_vec[i].resize(l1_length);
for(int j = 0;j<l1_length;j++)
mid_vec[i][j].resize(n_1);
}
l1_MaxPooling.resize(entity_num);
for (int i=0; i<l1_MaxPooling.size(); i++)
{
int l1_length = (entityWords_vec[i].size()-1)/n_pooling_1+1;
l1_MaxPooling[i].resize(l1_length);
for(int j = 0;j<l1_length;j++)
l1_MaxPooling[i][j].resize(n_1);
}
mid_grad.resize(entity_num);
for (int i=0; i<mid_grad.size(); i++)
{
int l1_length = (entityWords_vec[i].size()-1)/n_pooling_1+1;
mid_grad[i].resize(l1_length);
for(int j = 0;j<l1_length;j++)
mid_grad[i][j].resize(n_1);
}
cnnVecState.resize(entity_num);
for(int k = 0;k<THREADS_NUM;k++)
{
posErrorVec[k].resize(n);
negErrorVec[k].resize(n);
}
mut_mutex = PTHREAD_MUTEX_INITIALIZER;
sgd();
}
void *rand_sel(void *tid_void) //multi-thread train
{
long tid = (long) tid_void;
for (int k=0; k<batchsize; k++)
{
int i=rand_max(fb_h.size());
int j=rand_max(entity_num);
double pr = 1000*right_num[fb_r[i]]/(right_num[fb_r[i]]+left_num[fb_r[i]]);
if (method ==0)
pr = 500;
//||entity_h+r-entity_t||
int flag_num = rand_max(1000);
if (flag_num<pr)
{
while (ok.count(make_pair(fb_h[i],fb_r[i]))>0&&ok[make_pair(fb_h[i],fb_r[i])].count(j)>0)
j=rand_max(entity_num);
train_triple_mul(fb_h[i],fb_l[i],fb_r[i],fb_h[i],j,fb_r[i],tid);
}
else
{
while (ok.count(make_pair(j,fb_r[i]))>0&&ok[make_pair(j,fb_r[i])].count(fb_l[i])>0)
j=rand_max(entity_num);
train_triple_mul(fb_h[i],fb_l[i],fb_r[i],j,fb_l[i],fb_r[i],tid);
}
int rel_neg = rand_max(relation_num); //negative relation
while (ok.count(make_pair(fb_h[i], rel_neg))>0&& ok[make_pair(fb_h[i], rel_neg)].count(fb_l[i]) > 0)
rel_neg = rand_max(relation_num);
train_triple_mul(fb_h[i],fb_l[i],fb_r[i],fb_h[i],fb_l[i],rel_neg,tid);
//||cnn_h+r-cnn_t||
if (flag_num<pr)
train_cnn_mul(fb_h[i],fb_l[i],fb_r[i],fb_h[i],j,fb_r[i],tid);
else
train_cnn_mul(fb_h[i],fb_l[i],fb_r[i],j,fb_l[i],fb_r[i],tid);
train_cnn_mul(fb_h[i], fb_l[i], fb_r[i], fb_h[i], fb_l[i],rel_neg,tid);
//||entity_h+r-cnn_t||
if (flag_num<pr)
train_e_c_mul(fb_h[i],fb_l[i],fb_r[i],fb_h[i],j,fb_r[i],tid);
else
train_e_c_mul(fb_h[i],fb_l[i],fb_r[i],j,fb_l[i],fb_r[i],tid);
train_e_c_mul(fb_h[i], fb_l[i], fb_r[i], fb_h[i], fb_l[i],rel_neg,tid);
//||cnn_h+r-entity_t||
if (flag_num<pr)
train_c_e_mul(fb_h[i],fb_l[i],fb_r[i],fb_h[i],j,fb_r[i],tid);
else
train_c_e_mul(fb_h[i],fb_l[i],fb_r[i],j,fb_l[i],fb_r[i],tid);
train_c_e_mul(fb_h[i], fb_l[i], fb_r[i], fb_h[i], fb_l[i],rel_neg,tid);
//normalization
norm(relation_tmp[fb_r[i]]);
norm(relation_tmp[rel_neg]);
norm(entity_tmp[fb_h[i]]);
norm(entity_tmp[fb_l[i]]);
norm(entity_tmp[j]);
}
}
void *update_grad(void *tid_void) //multi-thread update
{
long tid = (long) tid_void;
int block_size = entity_num / THREADS_NUM;
int begin_e = tid * block_size;
int end_e = (tid+1) * block_size;
if(tid == THREADS_NUM-1)
end_e = entity_num;
//updata conv & word_vec
double nonLinear_1 = 0, der_1 = 0, der_2 = 0;
for(int ent=begin_e; ent<end_e; ent++)
{
if(cnnVecState[ent] != 2) //no need to update
continue;
//init
for(int i=0; i<mid_grad[ent].size(); i++)
for(int j=0; j<n_1; j++)
mid_grad[ent][i][j] = 0;
for(int i=0; i<n; i++)
{
der_2 = cnn_grad[ent][i]/mid_vec[ent].size();
cnn_grad[ent][i] = 0;
for(int h = 0;h<mid_vec[ent].size();h++)
{
for(int ii = 0;ii<window_2;ii++)
{
int tempMidID = h+ii;
if(tempMidID >= mid_vec[ent].size()) //zero-padding
break;
for(int k = 0;k<n_1;k++)
{
conv_w2_tmp[tid][i][ii*n_1+k] += der_2*mid_vec[ent][tempMidID][k];
mid_grad[ent][tempMidID][k] += der_2*conv_w2[i][ii*n_1+k];
}
}
}
}
for(int tempMidID=0; tempMidID<mid_grad[ent].size(); tempMidID++)
{
for(int k=0; k<n_1; k++)
{
nonLinear_1 = 1-mid_vec[ent][tempMidID][k]*mid_vec[ent][tempMidID][k];
der_1 = mid_grad[ent][tempMidID][k]*nonLinear_1;
for(int iii = 0;iii<window_1;iii++)
{
if(l1_MaxPooling[ent][tempMidID][k]+iii >= entityWords_vec[ent].size())
break;
int tempWordID = entityWords_vec[ent][l1_MaxPooling[ent][tempMidID][k]+iii];
for(int j = 0;j<n_w;j++)
{
conv_w1_tmp[tid][k][iii*n_w+j] += der_1*word_vec[tempWordID][j];
word_tmp[tempWordID][j] += der_1*conv_w1[k][iii*n_w+j];
}
}
}
}
for(int ii=0;ii<entityWords_vec[ent].size();ii++)
{
norm(word_tmp[entityWords_vec[ent][ii]]);
}
}
}
void update_multithread() //update others
{
//update
relation_vec = relation_tmp;
entity_vec = entity_tmp;
word_vec = word_tmp;
for(int k = 0;k<THREADS_NUM;k++)
{
for (int i=0; i<n_1; i++)
{
for (int ii=0; ii<n_w*window_1; ii++)
conv_w1[i][ii] += conv_w1_tmp[k][i][ii];
}
for (int i=0; i<n; i++)
{
for (int ii=0; ii<n_1*window_2; ii++)
conv_w2[i][ii] += conv_w2_tmp[k][i][ii];
}
res_triple += res_thread_triple[k];
res_cnn += res_thread_cnn[k];
res_c_e += res_thread_c_e[k];
res_e_c += res_thread_e_c[k];
res_normal += res_thread_normal[k];
}
}
void sgd() //SGD training
{
res_triple=0;
res_cnn = 0;
res_c_e = 0;
res_e_c = 0;
res_normal = 0;
nbatches=100; //block number
int nepoch = 1000; //iter
batchsize = fb_h.size()/nbatches/THREADS_NUM; //mini_batch size for each thread
cout << "batchsize : " << batchsize << endl;
for (int epoch=0; epoch<nepoch; epoch++)
{
res_triple=0;
res_cnn = 0;
res_c_e = 0;
res_e_c = 0;
res_normal = 0;
for (int batch = 0; batch<nbatches; batch++)
{
for(int i = 0;i<entity_num;i++)
cnnVecState[i] = 0;
for(int k = 0;k<THREADS_NUM;k++) //init
{
for (int i=0; i<n_1; i++)
for (int ii=0; ii<n_w*window_1; ii++)
conv_w1_tmp[k][i][ii] = 0;
for (int i=0; i<n; i++)
for (int ii=0; ii<n_1*window_2; ii++)
conv_w2_tmp[k][i][ii] = 0;
res_thread_triple[k] = 0;
res_thread_cnn[k] = 0;
res_thread_c_e[k] = 0;
res_thread_e_c[k] = 0;
res_thread_normal[k] = 0;
}
relation_tmp = relation_vec;
entity_tmp = entity_vec;
word_tmp = word_vec;
//multi-thread for train
pthread_t threads[THREADS_NUM];
for(int k = 0; k < THREADS_NUM; k ++){
pthread_create(&threads[k], NULL, rand_sel, (void *)k); //train
}
for(int k = 0; k < THREADS_NUM; k ++){
pthread_join(threads[k], NULL);
}
//multi-thread for update
pthread_t threads2[THREADS_NUM];
for(int k = 0; k < THREADS_NUM; k ++){
pthread_create(&threads2[k], NULL, update_grad, (void *)k); //update
}
for(int k = 0; k < THREADS_NUM; k ++){
pthread_join(threads2[k], NULL);
}
//update other
update_multithread();
//cout << "update once : " << batch << endl;
}
cout<<"epoch:"<<epoch<<' '<<res_triple<< ' ' << res_cnn << ' ';
cout << res_c_e << ' ' << res_e_c << ' ' << res_normal << endl;
//output
FILE* f2 = fopen(("../res/relation2vec."+version).c_str(),"w");
FILE* f3 = fopen(("../res/entity2vec."+version).c_str(),"w");
FILE* f4 = fopen(("../res/word2vec."+version).c_str(),"w");
FILE* f5 = fopen(("../res/weight1."+version).c_str(),"w");
FILE* f6 = fopen(("../res/weight2."+version).c_str(),"w");
for (int i=0; i<relation_num; i++)
{
for (int ii=0; ii<n; ii++)
fprintf(f2,"%.6lf\t",relation_vec[i][ii]);
fprintf(f2,"\n");
}
for (int i=0; i<entity_num; i++)
{
for (int ii=0; ii<n; ii++)
fprintf(f3,"%.6lf\t",entity_vec[i][ii]);
fprintf(f3,"\n");
}
for (int i=0; i<word_num; i++)
{
for (int ii=0; ii<n_w; ii++)
fprintf(f4,"%.6lf\t",word_vec[i][ii]);
fprintf(f4,"\n");
}
for (int i=0; i<n_1; i++)
{
for (int ii=0; ii<n_w*window_1; ii++)
fprintf(f5,"%.6lf\t",conv_w1[i][ii]);
fprintf(f5,"\n");
}
for (int i=0; i<n; i++)
{
for (int ii=0; ii<n_1*window_2; ii++)
fprintf(f6,"%.6lf\t",conv_w2[i][ii]);
fprintf(f6,"\n");
}
fclose(f2);
fclose(f3);
fclose(f4);
fclose(f5);
fclose(f6);
}
}
//for cnn_vec length
void gradient_normalization(int e1,int tid)
{
for (int i=0; i<n; i++)
{
double nonLinear_2 = 1-cnn_vec[e1][i]*cnn_vec[e1][i];
cnn_grad[e1][i] -= rate_n*2*cnn_vec[e1][i]*nonLinear_2;
}
}
//calc cnn_vec
void calc_cnn_vec(int ent) //use CNN to get entity description-based embedding
{
//1st convolutional layer & max-pooling layer
int l1_length = (entityWords_vec[ent].size()-1)/n_pooling_1+1;
for(int k = 0;k<l1_length;k++)
{
for(int j = 0;j<n_1;j++)
{
double tempMax = -2147483647; //small number
l1_MaxPooling[ent][k][j] = -1;
for(int i = n_pooling_1*k;i<n_pooling_1*(k+1);i++)
{
if(i >= entityWords_vec[ent].size())
break;
double tempTokenValue = 0;
for(int ii = 0;ii<window_1;ii++)
{
if(i+ii >= entityWords_vec[ent].size())
break;
int tempWordID = entityWords_vec[ent][i+ii];
for(int iii = 0;iii<n_w;iii++)
{
tempTokenValue += word_vec[tempWordID][iii] * conv_w1[j][ii*n_w+iii];
}
}
if(tempMax < tempTokenValue)
{
tempMax = tempTokenValue;
l1_MaxPooling[ent][k][j] = i;
}
}
//hyperpolic tangent
double tempExpo = exp(-2*tempMax);
mid_vec[ent][k][j] = (1-tempExpo) / (1+tempExpo);
}
}
//2nd convolutional layer & max-pooling layer
for(int j = 0;j<n;j++)
{
double tempExpo = 0;
int tempWordNum = 0;
for(int i = 0;i<l1_length;i++)
{
double tempTokenValue = 0;
for(int ii = 0;ii<window_2;ii++)
{
if(i+ii >= l1_length)
break;
for(int iii = 0;iii<n_1;iii++)
{
tempTokenValue += mid_vec[ent][i+ii][iii] * conv_w2[j][ii*n_1+iii];
}
}
//mean pooling
tempExpo += tempTokenValue;
tempWordNum++;
}
//hyperpolic tangent
tempExpo /= tempWordNum;
tempExpo = exp(-2*tempExpo);
cnn_vec[ent][j] = (1-tempExpo) / (1+tempExpo);
}
}
double calc_sum_cnn(int e1,int e2,int rel, int flag, int tid) //similarity
{
double sum=0;
//CNN part
if(cnnVecState[e1] == 0)
{
calc_cnn_vec(e1);
double sum1 = vec_len(cnn_vec[e1]);
if(sum1 > 1)
{
res_thread_normal[tid] += sum1-1;
gradient_normalization(e1,tid);
}
cnnVecState[e1] = 1;
}
if(cnnVecState[e2] == 0)
{
calc_cnn_vec(e2);
double sum1 = vec_len(cnn_vec[e2]);
if(sum1 > 1)
{
res_thread_normal[tid] += sum1-1;
gradient_normalization(e2,tid);
}
cnnVecState[e2] = 1;
}
if(flag == 1) //positive_sign
{
if (L1_flag) //L1
{
for (int ii=0; ii<n; ii++)
{
double tempSum = cnn_vec[e2][ii]-cnn_vec[e1][ii]-relation_vec[rel][ii];
sum+=fabs(tempSum);
if(tempSum > 0)
posErrorVec[tid][ii] = 1;
else
posErrorVec[tid][ii] = -1;
}
}
else //L2
{
for (int ii=0; ii<n; ii++)
{
double tempSum = cnn_vec[e2][ii]-cnn_vec[e1][ii]-relation_vec[rel][ii];
sum+=sqr(tempSum);
posErrorVec[tid][ii] = 2*tempSum;
}
}
return sum;
}
else //negative_sign
{
if (L1_flag) //L1
{
for (int ii=0; ii<n; ii++)
{
double tempSum = cnn_vec[e2][ii]-cnn_vec[e1][ii]-relation_vec[rel][ii];
sum+=fabs(tempSum);
if(tempSum > 0)
negErrorVec[tid][ii] = 1;
else
negErrorVec[tid][ii] = -1;
}
}
else //L2
{
for (int ii=0; ii<n; ii++)
{
double tempSum = cnn_vec[e2][ii]-cnn_vec[e1][ii]-relation_vec[rel][ii];
sum+=sqr(tempSum);
negErrorVec[tid][ii] = 2*tempSum;
}
}
return sum;
}
}
double calc_sum_c_e(int e1,int e2,int rel, int flag, int tid)
{
double sum=0;
if(flag == 1) //positive_sign
{
if (L1_flag) //L1
{
for (int ii=0; ii<n; ii++)
{
double tempSum = entity_vec[e2][ii]-cnn_vec[e1][ii]-relation_vec[rel][ii];
sum+=fabs(tempSum);
if(tempSum > 0)
posErrorVec[tid][ii] = 1;
else
posErrorVec[tid][ii] = -1;
}
}
else //L2
{
for (int ii=0; ii<n; ii++)
{
double tempSum = entity_vec[e2][ii]-cnn_vec[e1][ii]-relation_vec[rel][ii];
sum+=sqr(tempSum);
posErrorVec[tid][ii] = 2*tempSum;
}
}
return sum;
}
else //negative_sign
{
if (L1_flag) //L1
{
for (int ii=0; ii<n; ii++)
{
double tempSum = entity_vec[e2][ii]-cnn_vec[e1][ii]-relation_vec[rel][ii];
sum+=fabs(tempSum);
if(tempSum > 0)
negErrorVec[tid][ii] = 1;
else
negErrorVec[tid][ii] = -1;
}
}
else //L2
{
for (int ii=0; ii<n; ii++)
{
double tempSum = entity_vec[e2][ii]-cnn_vec[e1][ii]-relation_vec[rel][ii];
sum+=sqr(tempSum);
negErrorVec[tid][ii] = 2*tempSum;
}
}
return sum;
}
}
double calc_sum_e_c(int e1,int e2,int rel, int flag, int tid)
{
double sum=0;
if(flag == 1) //positive_sign
{
if (L1_flag) //L1
{
for (int ii=0; ii<n; ii++)
{
double tempSum = cnn_vec[e2][ii]-entity_vec[e1][ii]-relation_vec[rel][ii];
sum+=fabs(tempSum);
if(tempSum > 0)
posErrorVec[tid][ii] = 1;
else
posErrorVec[tid][ii] = -1;
}
}
else //L2
{
for (int ii=0; ii<n; ii++)
{
double tempSum = cnn_vec[e2][ii]-entity_vec[e1][ii]-relation_vec[rel][ii];
sum+=sqr(tempSum);
posErrorVec[tid][ii] = 2*tempSum;
}
}
return sum;
}
else //negative_sign
{
if (L1_flag) //L1
{
for (int ii=0; ii<n; ii++)
{
double tempSum = cnn_vec[e2][ii]-entity_vec[e1][ii]-relation_vec[rel][ii];
sum+=fabs(tempSum);
if(tempSum > 0)
negErrorVec[tid][ii] = 1;
else
negErrorVec[tid][ii] = -1;
}
}
else //L2
{
for (int ii=0; ii<n; ii++)
{
double tempSum = cnn_vec[e2][ii]-entity_vec[e1][ii]-relation_vec[rel][ii];
sum+=sqr(tempSum);
negErrorVec[tid][ii] = 2*tempSum;
}
}
return sum;
}
}
void gradient_cnn(int e1_a,int e2_a,int rel_a,int e1_b,int e2_b,int rel_b,int tid) //SGD update
{
double nonLinear_2 = 0;
//update state
cnnVecState[e1_a] = 2;
cnnVecState[e2_a] = 2;
cnnVecState[e1_b] = 2;
cnnVecState[e2_b] = 2;
for (int i = 0;i<n;i++)
{
//relation
relation_tmp[rel_a][i] += rate_c*posErrorVec[tid][i];
//head
nonLinear_2 = 1-cnn_vec[e1_a][i]*cnn_vec[e1_a][i];
cnn_grad[e1_a][i] += rate_c*posErrorVec[tid][i]*nonLinear_2;
//tail
nonLinear_2 = 1-cnn_vec[e2_a][i]*cnn_vec[e2_a][i];
cnn_grad[e2_a][i] -= rate_c*posErrorVec[tid][i]*nonLinear_2;
//relation
relation_tmp[rel_b][i] -= rate_c*negErrorVec[tid][i];
//head
nonLinear_2 = 1-cnn_vec[e1_b][i]*cnn_vec[e1_b][i];
cnn_grad[e1_b][i] -= rate_c*negErrorVec[tid][i]*nonLinear_2;
//tail
nonLinear_2 = 1-cnn_vec[e2_b][i]*cnn_vec[e2_b][i];
cnn_grad[e2_b][i] += rate_c*negErrorVec[tid][i]*nonLinear_2;
}
}
void gradient_c_e(int e1_a,int e2_a,int rel_a,int e1_b,int e2_b,int rel_b,int tid) //SGD update
{
double nonLinear_2 = 0;
//update state
cnnVecState[e1_a] = 2;
cnnVecState[e2_a] = 2;
cnnVecState[e1_b] = 2;
cnnVecState[e2_b] = 2;
for (int i = 0;i<n;i++)
{
//relation
relation_tmp[rel_a][i] += rate_m*posErrorVec[tid][i];
//head
nonLinear_2 = 1-cnn_vec[e1_a][i]*cnn_vec[e1_a][i];
cnn_grad[e1_a][i] += rate_m*posErrorVec[tid][i]*nonLinear_2;
//tail
entity_tmp[e2_a][i] -= rate_m*posErrorVec[tid][i];
//relation
relation_tmp[rel_b][i] -= rate_m*negErrorVec[tid][i];
//head
nonLinear_2 = 1-cnn_vec[e1_b][i]*cnn_vec[e1_b][i];
cnn_grad[e1_b][i] -= rate_m*negErrorVec[tid][i]*nonLinear_2;
//tail
entity_tmp[e2_b][i] += rate_m*negErrorVec[tid][i];
}
}
void gradient_e_c(int e1_a,int e2_a,int rel_a,int e1_b,int e2_b,int rel_b,int tid) //SGD update
{
double nonLinear_2 = 0;
//update state
cnnVecState[e1_a] = 2;
cnnVecState[e2_a] = 2;
cnnVecState[e1_b] = 2;
cnnVecState[e2_b] = 2;
for (int i = 0;i<n;i++)
{
//relation
relation_tmp[rel_a][i] += rate_m*posErrorVec[tid][i];
//head
entity_tmp[e1_a][i] += rate_m*posErrorVec[tid][i];
//tail
nonLinear_2 = 1-cnn_vec[e2_a][i]*cnn_vec[e2_a][i];
cnn_grad[e2_a][i] -= rate_m*posErrorVec[tid][i]*nonLinear_2;
//relation
relation_tmp[rel_b][i] -= rate_m*negErrorVec[tid][i];
//head
entity_tmp[e1_b][i] -= rate_m*negErrorVec[tid][i];
//tail
nonLinear_2 = 1-cnn_vec[e2_b][i]*cnn_vec[e2_b][i];
cnn_grad[e2_b][i] += rate_m*negErrorVec[tid][i]*nonLinear_2;
}
}
void train_cnn_mul(int e1_a,int e2_a,int rel_a,int e1_b,int e2_b,int rel_b,int tid) //||entity_vec-cnn_vec||
{
double sum1 = calc_sum_cnn(e1_a,e2_a,rel_a,1,tid);
double sum2 = calc_sum_cnn(e1_b,e2_b,rel_b,0,tid);
if (sum1+margin_c>sum2)
{
res_thread_cnn[tid]+=margin_c+sum1-sum2;
gradient_cnn( e1_a, e2_a, rel_a, e1_b, e2_b, rel_b, tid);
}
}
void train_c_e_mul(int e1_a,int e2_a,int rel_a,int e1_b,int e2_b,int rel_b,int tid) //||cnn_h+r-entity_t||
{
double sum1 = calc_sum_c_e(e1_a,e2_a,rel_a,1,tid);
double sum2 = calc_sum_c_e(e1_b,e2_b,rel_b,0,tid);
if (sum1+margin_m>sum2)
{
res_thread_c_e[tid]+=margin_m+sum1-sum2;
gradient_c_e( e1_a, e2_a, rel_a, e1_b, e2_b, rel_b, tid);
}
}
void train_e_c_mul(int e1_a,int e2_a,int rel_a,int e1_b,int e2_b,int rel_b,int tid) //||entity_h+r-cnn_t||
{
double sum1 = calc_sum_e_c(e1_a,e2_a,rel_a,1,tid);
double sum2 = calc_sum_e_c(e1_b,e2_b,rel_b,0,tid);
if (sum1+margin_m>sum2)
{
res_thread_e_c[tid]+=margin_m+sum1-sum2;
gradient_e_c( e1_a, e2_a, rel_a, e1_b, e2_b, rel_b, tid);
}
}
double calc_triple_sum(int e1,int e2,int rel) //for ||h+r-t||
{
double sum=0;
if (L1_flag)
for (int ii=0; ii<n; ii++)
sum+=fabs(entity_vec[e2][ii]-entity_vec[e1][ii]-relation_vec[rel][ii]);
else
for (int ii=0; ii<n; ii++)
sum+=sqr(entity_vec[e2][ii]-entity_vec[e1][ii]-relation_vec[rel][ii]);
return sum;
}
void gradient_triple(int e1_a,int e2_a,int rel_a,int e1_b,int e2_b,int rel_b) //SGD update
{
for (int ii=0; ii<n; ii++)
{
double x = 2*(entity_vec[e2_a][ii]-entity_vec[e1_a][ii]-relation_vec[rel_a][ii]);
if (L1_flag)
if (x>0)
x=1;
else
x=-1;
relation_tmp[rel_a][ii]-=-1*rate*x;
entity_tmp[e1_a][ii]-=-1*rate*x;
entity_tmp[e2_a][ii]+=-1*rate*x;
x = 2*(entity_vec[e2_b][ii]-entity_vec[e1_b][ii]-relation_vec[rel_b][ii]);
if (L1_flag)
if (x>0)
x=1;
else
x=-1;
relation_tmp[rel_b][ii]-=rate*x;
entity_tmp[e1_b][ii]-=rate*x;
entity_tmp[e2_b][ii]+=rate*x;
}
}
void train_triple_mul(int e1_a,int e2_a,int rel_a,int e1_b,int e2_b,int rel_b,int tid)
{
double sum1 = calc_triple_sum(e1_a,e2_a,rel_a);
double sum2 = calc_triple_sum(e1_b,e2_b,rel_b);
if (sum1+margin>sum2)
{
res_thread_triple[tid]+=margin+sum1-sum2;
gradient_triple( e1_a, e2_a, rel_a, e1_b, e2_b, rel_b);
}
}
void prepare()
{
FILE* f1 = fopen("../data/entity2id.txt","r");
FILE* f2 = fopen("../data/relation2id.txt","r");
FILE* f3 = fopen("../data/word2id.txt","r");
FILE* f4 = fopen("../data/entityWords.txt","r");
int x;