forked from snap-stanford/snap
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathbowlearn.cpp
1203 lines (1186 loc) · 49.7 KB
/
bowlearn.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
void TBowLearn::TrainMajorityMd(const PBowDocBs& BowDocBs, TFltV& MdCatExWgtV){
// P(C) -> MdCatExWgtV
// prepare shortcuts
int AllCats=BowDocBs->GetCats();
// training
// collecting category example weights
double SumCatExWgt=0;
{for (int DIdN=0; DIdN<BowDocBs->GetTrainDocs(); DIdN++){
int DId=BowDocBs->GetTrainDId(DIdN);
int CIds=BowDocBs->GetDocCIds(DId);
for (int CIdN=0; CIdN<CIds; CIdN++){
int CId=BowDocBs->GetDocCId(DId, CIdN);
MdCatExWgtV[CId]++; SumCatExWgt++;
}
}}
if (SumCatExWgt>0){
for (int CId=0; CId<AllCats; CId++){
MdCatExWgtV[CId]/=SumCatExWgt;
TStr CatNm=BowDocBs->GetCatNm(CId);
// if (MdCatExWgtV[CId]>0){printf("[%s]-%d:%.5f ", CatNm(), CId, MdCatExWgtV[CId]);} //**
}
// printf("\n");
} else {printf("No training examples!\n");}
}
double TBowLearn::TestMajorityMd(const PBowDocBs& BowDocBs,
TFltV& MdCatExWgtV, FILE* OutFile, int& TestExs){
// P(C|Doc)=argmx_C P(C)
// prepare shortcuts
int AllCats=BowDocBs->GetCats();
// testing
double CorrectExs=0; TestExs=0;
int MajorCId=-1; TFlt MajorCatWgt=0;
{for (int CId=0; CId<AllCats; CId++){
if (MdCatExWgtV[CId]>MajorCatWgt){MajorCId=CId; MajorCatWgt=MdCatExWgtV[CId];}
}}
if (MajorCId>-1){
TStr CatNm=BowDocBs->GetCatNm(MajorCId);
printf("\nMajority category P(%d:%s)=%.5f\n", MajorCId, CatNm(), MajorCatWgt()); //**
fprintf(OutFile, "\nMajority category P(%d:%s)=%.5f\n", MajorCId, CatNm(), MajorCatWgt());
} else {
printf("No category with probability > 0\n"); //**
fprintf(OutFile, "No category with probability > 0\n");
return CorrectExs;
}
for (int DIdN=0; DIdN<BowDocBs->GetTestDocs(); DIdN++){
int DId=BowDocBs->GetTestDId(DIdN);
int WIds=BowDocBs->GetDocWIds(DId);
int CIds=BowDocBs->GetDocCIds(DId);
// write best categories
TestExs++;
// printf("Document vector: "); //**
fprintf(OutFile, "Document vector:");
{for (int WIdN=0; WIdN<WIds; WIdN++){
int WId; double WordFq; BowDocBs->GetDocWIdFq(DId, WIdN, WId, WordFq);
TStr WordStr=BowDocBs->GetWordStr(WId);
for (int WordFqN=0; WordFqN<WordFq; WordFqN++){
// printf("%s ", WordStr.CStr()); //**
fprintf(OutFile, "%s ", WordStr.CStr()); //**
}
}}
// printf("\n");
fprintf(OutFile, "\n");
{for (int CIdN=0; CIdN<CIds; CIdN++){
int CId=BowDocBs->GetDocCId(DId, CIdN);
TStr CatNm=BowDocBs->GetCatNm(CId);
// printf("[%d:%s] ", CId, CatNm()); //**
fprintf(OutFile, "[%d:%s] ", CId, CatNm());
if (CId==MajorCId){CorrectExs+=1/double(CIds);}
}}
// printf(" --> "); //**
fprintf(OutFile, " --> ");
int CId=MajorCId;
TStr CatNm=BowDocBs->GetCatNm(CId);
// printf("\n [%d:%s]:%.5f ", CId, CatNm(), MajorCatWgt); //**
fprintf(OutFile, "\n [%d:%s]:%.5f ", CId, CatNm(), MajorCatWgt());
// printf("\n Model evidence: majority class ");
fprintf(OutFile, "\n Model evidence: majority class value");
// printf("\n"); //**
fprintf(OutFile, "\n");
}
printf("Testing documents:%d, Non-empty documents:%d ",
BowDocBs->GetTestDocs(), TestExs);
fprintf(OutFile, "Testing documents:%d, Non-empty documents:%d ",
BowDocBs->GetTestDocs(), TestExs);
if (TestExs>0){
printf("CorrectExs:%.5f\n", CorrectExs);
fprintf(OutFile, "CorrectExs:%.5f\n", CorrectExs);
} else {
printf("CorrectExs:%.5f\n", CorrectExs);
fprintf(OutFile, "CorrectExs:%.5f\n", CorrectExs);
}
return CorrectExs;
}
void TBowLearn::TrainInvertedMd(const PBowDocBs& BowDocBs,
TFltVV& MdWordCatWgtVV, TFltV& MdWordWgtV, TFltV& MdCatWgtV){
// prepare shortcuts
int AllWords=BowDocBs->GetWords();
int AllCats=BowDocBs->GetCats();
// collecting word-category weights
for (int DIdN=0; DIdN<BowDocBs->GetTrainDocs(); DIdN++){
int DId=BowDocBs->GetTrainDId(DIdN);
int WIds=BowDocBs->GetDocWIds(DId);
int CIds=BowDocBs->GetDocCIds(DId);
for (int WIdN=0; WIdN<WIds; WIdN++){
int WId; double WordFq; BowDocBs->GetDocWIdFq(DId, WIdN, WId, WordFq);
for (int CIdN=0; CIdN<CIds; CIdN++){
int CId=BowDocBs->GetDocCId(DId, CIdN);
double CatWgt=WordFq*(1/double(CIds));
MdWordCatWgtVV.At(WId, CId)+=CatWgt;
}
}
}
// collecting word weights
for (int WId=0; WId<AllWords; WId++){
double SumWordCatWgt=0;
for (int CId=0; CId<AllCats; CId++){
SumWordCatWgt+=MdWordCatWgtVV.At(WId, CId);}
if (SumWordCatWgt>0){
for (int CId=0; CId<AllCats; CId++){
MdWordCatWgtVV.At(WId, CId)/=SumWordCatWgt;}
}
double WordWgt=0;
{for (int CId=0; CId<AllCats; CId++){
double WordCatWgt=MdWordCatWgtVV.At(WId, CId);
if (WordCatWgt>=0){
WordWgt-=WordCatWgt*log(WordCatWgt);}
}}
TStr WordStr=BowDocBs->GetWordStr(WId);
MdWordWgtV[WId]=1/(1+TMath::Sqr(WordWgt));
//MdWordWgtV[WId]=1;
// if (WordWgt>0){printf("MdWordWgtV[%d:%s]:%.3f ", WId, WordStr.CStr(), MdWordWgtV[WId]);} //**
}
printf("\n"); //**
// collecting category weights
double SumCatExWgt=0;
{for (int DIdN=0; DIdN<BowDocBs->GetTrainDocs(); DIdN++){
int DId=BowDocBs->GetTrainDId(DIdN);
int CIds=BowDocBs->GetDocCIds(DId);
for (int CIdN=0; CIdN<CIds; CIdN++){
int CId=BowDocBs->GetDocCId(DId, CIdN);
MdCatWgtV[CId]++; SumCatExWgt++;
}
}}
if (SumCatExWgt>0){
for (int CId=0; CId<AllCats; CId++){
TStr CatNm=BowDocBs->GetCatNm(CId); //**
// if (MdCatWgtV[CId]>0){printf("MdCatWgtV[%d:%s]:%.3f", CId, CatNm(), MdCatWgtV[CId]);} //**
MdCatWgtV[CId]=1/log(1+1+TMath::Sqr(MdCatWgtV[CId]));
//MdCatWgtV[CId]=1;
}
printf("\n"); //**
} else {printf("No training examples!\n");}
}
double TBowLearn::TestInvertedMd(const PBowDocBs& BowDocBs,
TFltVV& MdWordCatWgtVV, TFltV& MdWordWgtV, TFltV& MdCatWgtV,
FILE* OutFile, int& TestExs){
// testing
// prepare shortcuts
int AllCats=BowDocBs->GetCats();
// testing
double CorrectExs=0; TestExs=0;
int MajorCId=-1; TFlt MajorCatWgt=0;
{for (int CId=0; CId<AllCats; CId++){
if (MdCatWgtV[CId]>MajorCatWgt){MajorCId=CId; MajorCatWgt=MdCatWgtV[CId];}
}}
if (MajorCId>-1){
TStr CatNm=BowDocBs->GetCatNm(MajorCId);
printf("\nMajority category P(%d:%s])=%.5f\n", MajorCId, CatNm(), MajorCatWgt()); //**
fprintf(OutFile, "\nMajority category P(%d:%s])=%.5f\n", MajorCId, CatNm(), MajorCatWgt());
} else {
printf("No category with probability > 0\n"); //**
fprintf(OutFile, "No category with probability > 0\n");
}
for (int DIdN=0; DIdN<BowDocBs->GetTestDocs(); DIdN++){
int DId=BowDocBs->GetTestDId(DIdN);
int WIds=BowDocBs->GetDocWIds(DId);
int CIds=BowDocBs->GetDocCIds(DId);
if (WIds==0){printf("It is hard to classify an empty document!\n"); continue;}
// create category vector
TFltIntKdV CatWgtCIdKdV(AllCats);
{for (int CId=0; CId<AllCats; CId++){
CatWgtCIdKdV[CId].Dat=CId; CatWgtCIdKdV[CId].Key=0;
}}
{for (int WIdN=0; WIdN<WIds; WIdN++){
int WId; double WordFq; BowDocBs->GetDocWIdFq(DId, WIdN, WId, WordFq);
printf("MdWordWgtV[%d]:%.3f ", WId, MdWordWgtV[WId]());//**
for (int CId=0; CId<AllCats; CId++){
double CatWgt=
WordFq*MdCatWgtV[CId]*MdWordWgtV[WId]*MdWordCatWgtVV.At(WId, CId);
CatWgtCIdKdV[CId].Key+=CatWgt;
if ((CId==5)||(CId==130)){printf("CatWgt[%d]=%.1f*%.3f*%.3f*%.3f\n",
CId, WordFq, MdCatWgtV[CId](), MdWordWgtV[WId](), MdWordCatWgtVV.At(WId, CId)());} //**
if ((CId==5)||(CId==130)){printf("CatWgt.Key[%d]=%.5f\n", CId, CatWgtCIdKdV[CId].Key());} //**
}
}}
double SumCatWgt=0;
{for (int CId=0; CId<AllCats; CId++){
SumCatWgt+=CatWgtCIdKdV[CId].Key;
if ((CId==5)||(CId==130)){ //**
printf("\nMdCatWgtV[%d]=%.5f CatWgt[%d]=%.5f\n", CId, MdCatWgtV[CId](),
CId, CatWgtCIdKdV[CId].Key());
}
}}
printf("\n");
if (SumCatWgt>0){
for (int CId=0; CId<AllCats; CId++){CatWgtCIdKdV[CId].Key/=SumCatWgt;}
}
// choose best categories
CatWgtCIdKdV.Sort(false);
printf("BestCat:%d, prob:%.5f\n", CatWgtCIdKdV[0].Dat(), CatWgtCIdKdV[0].Key()); //**
if (CatWgtCIdKdV[0].Key<=0.0001){ // check the most probable category
printf("No category found for %d!\n", DId); //**
fprintf(OutFile, "No category found for %d!\n", DId);
continue;
}
double SumCatWgtSF=0; TIntV BestCIdV;
for (int CIdN=0; CIdN<AllCats; CIdN++){
double CatWgt=CatWgtCIdKdV[CIdN].Key;
int CId=CatWgtCIdKdV[CIdN].Dat;
SumCatWgtSF+=CatWgt; BestCIdV.Add(CId);
if (SumCatWgtSF>0.25){break;}
}
// write best categories
TestExs++;
printf("Document vector: "); //**
fprintf(OutFile, "Document vector:");
{for (int WIdN=0; WIdN<WIds; WIdN++){
int WId; double WordFq; BowDocBs->GetDocWIdFq(DId, WIdN, WId, WordFq);
TStr WordStr=BowDocBs->GetWordStr(WId);
for (int WordFqN=0; WordFqN<WordFq; WordFqN++){
printf("%s ", WordStr.CStr()); //**
fprintf(OutFile, "%s ", WordStr.CStr()); //**
}
}}
printf("\n"); //**
fprintf(OutFile, "\n");
{for (int CIdN=0; CIdN<CIds; CIdN++){
int CId=BowDocBs->GetDocCId(DId, CIdN);
TStr CatNm=BowDocBs->GetCatNm(CId);
printf("[%d:%s] ", CId, CatNm()); //**
fprintf(OutFile, "[%d:%s] ", CId, CatNm());
if (BestCIdV.IsIn(CId)){CorrectExs+=1/double(CIds);}
}}
printf(" --> "); //**
fprintf(OutFile, " --> ");
{for (int CIdN=0; CIdN<BestCIdV.Len(); CIdN++){
int CId=BestCIdV[CIdN];
TStr CatNm=BowDocBs->GetCatNm(CId);
printf("\n [%d:%s]:%.5f ", CId, CatNm(), CatWgtCIdKdV[CIdN].Key()); //**
fprintf(OutFile, "\n [%d:%s]:%.5f ", CId, CatNm(), CatWgtCIdKdV[CIdN].Key());
printf("\n Model evidence: "); //**
fprintf(OutFile, "\n Model evidence: ");
for (int WIdN=0; WIdN<WIds; WIdN++){
int WId; double WordFq; BowDocBs->GetDocWIdFq(DId, WIdN, WId, WordFq);
if (MdWordCatWgtVV.At(WId, CId)>0){
TStr WordStr=BowDocBs->GetWordStr(WId);
printf("TF(%s)=%.1f ", WordStr.CStr(), MdWordCatWgtVV.At(WId, CId)()); //**
fprintf(OutFile, "TF(%s)=%.1f ", WordStr.CStr(), MdWordCatWgtVV.At(WId, CId)());
}
}
printf("\n"); //**
fprintf(OutFile, "\n");
}}
printf("\n"); //**
fprintf(OutFile, "\n");
}
printf("Testing documents:%d, Non-empty documents:%d ",
BowDocBs->GetTestDocs(), TestExs);
fprintf(OutFile, "Testing documents:%d, Non-empty documents:%d ",
BowDocBs->GetTestDocs(), TestExs);
if (TestExs>0){
printf("CorrectExs:%.5f\n", CorrectExs);
fprintf(OutFile, "CorrectExs:%.5f\n", CorrectExs);
} else {
printf("CorrectExs:%.5f\n", CorrectExs);
fprintf(OutFile, "CorrectExs:%.5f\n", CorrectExs);
}
return CorrectExs;
}
void TBowLearn::TrainSimMd(const PBowDocBs& BowDocBs,
const TVec<TIntV>& DIdVV, TBowSpVV& ConceptSpVV){
// P(C|Doc) = argmx_C CosSim(TFIDF(Doc),TFIDF(C))
// TFIDF(DId) -> SpV=BowDocWgtBs->GetSpVFromDIdN(DId), Doc -> DId
// TF(CIdN) -> BowDocWgtBs->GetConceptSpV(BowSim, DIdVV[CIdN], ConceptSpVV[CIdN]), C -> CIdN
// prepare shortcuts
int AllCats=BowDocBs->GetCats();
// get train doc-ids
TIntV CtxTrainDIdV; BowDocBs->GetTrainDIdV(CtxTrainDIdV);
// make TFIDF vectors for training documents
PBowDocWgtBs BowDocWgtBs= //** Dunja: error, CtxTrainDIdV has 389 Ids, but their range is 0..3175
TBowDocWgtBs::New(BowDocBs, bwwtNrmTFIDF, 0, 0, CtxTrainDIdV);
// get representative concept vectors
PBowSim BowSim=TBowSim::New(bstCos);
for (int CIdN=0; CIdN<AllCats; CIdN++){
ConceptSpVV[CIdN]=TBowClust::GetConceptSpV(BowDocWgtBs, BowSim, DIdVV[CIdN]);
}
}
double TBowLearn::TestSimMd(const PBowDocBs& BowDocBs,
const TBowSpVV& ConceptSpVV,
FILE* OutFile, int& TestExs){
// P(C|Doc) = argmx_C CosSim(TFIDF(Doc),TFIDF(C))
// TFIDF(DIdN) -> SpV=BowDocWgtBs->GetSpVFromDIdN(DIdN), Doc -> DIdN
// TF(CIdN) -> BowDocWgtBs->GetConceptSpV(BowSim, DIdVV[CIdN], ConceptSpVV[CIdN]), C -> CIdN
// prepare shortcuts
int AllCats=BowDocBs->GetCats();
// make TFIDF vectors for testing documents
TIntV CtxTestDIdNV; BowDocBs->GetTestDIdV(CtxTestDIdNV);
PBowDocWgtBs BowDocWgtBs=
TBowDocWgtBs::New(BowDocBs, bwwtNrmTFIDF, 0, 0, CtxTestDIdNV);
// testing
double CorrectExs=0; TestExs=0;
// for each document find closest concept vector
for (int DIdN=0; DIdN<BowDocBs->GetTestDocs(); DIdN++){
int DId=BowDocWgtBs->GetDId(DIdN);
PBowSpV SpV=BowDocWgtBs->GetSpV(DId);
int DIdx=BowDocBs->GetTestDId(DIdN);
printf("DIdx:%d = DId:%d ", DIdx, DId); // are they the same?
int WIdsx=BowDocBs->GetDocWIds(DId);
int WIds=SpV->GetWIds();
printf("WIdsx:%d = WIds:%d", WIdsx, WIds); // are they the same?
int CIds=BowDocBs->GetDocCIds(DId);
if (WIds==0){printf("It is hard to classify an empty document!\n"); continue;}
for (int WIdN=0; WIdN<WIds; WIdN++){
int WId=SpV->GetWId(WIdN); double WordWgt=SpV->GetWgt(WIdN);
printf("WId:%d ", WId);
TStr WordStr=BowDocBs->GetWordStr(WId);
printf("%s:%.2f ", WordStr.CStr(), WordWgt); //**
}
printf("\n");
// create category vector
PBowSim BowSim=TBowSim::New(bstCos);
TFltIntKdV CatWgtCIdKdV(AllCats);
for (int CId=0; CId<AllCats; CId++){
CatWgtCIdKdV[CId].Dat=CId;
CatWgtCIdKdV[CId].Key=BowSim->GetSim(ConceptSpVV[CId], SpV);
printf("Sim[%d]=%.3f ", CId, CatWgtCIdKdV[CId].Key());
}
// choose best categories
CatWgtCIdKdV.Sort(false);
printf("BestCat:%d, prob:%.5f\n", CatWgtCIdKdV[0].Dat(), CatWgtCIdKdV[0].Key());
if (CatWgtCIdKdV[0].Key<=0.0001){ // check the most probable category
printf("No category found for %d!\n", DId); //**
fprintf(OutFile, "No category found for %d!\n", DId);
continue;
}
double SumCatWgtSF=0; TIntV BestCIdV;
for (int CIdN=0; CIdN<AllCats; CIdN++){
double CatWgt=CatWgtCIdKdV[CIdN].Key;
int CId=CatWgtCIdKdV[CIdN].Dat;
SumCatWgtSF+=CatWgt; BestCIdV.Add(CId);
if (SumCatWgtSF>0.25){break;}
}
// write best categories
TestExs++;
{for (int CIdN=0; CIdN<CIds; CIdN++){
int CId=BowDocBs->GetDocCId(DId, CIdN);
TStr CatNm=BowDocBs->GetCatNm(CId);
printf("[%s]-%d ", CatNm(), CId); //**
fprintf(OutFile, "[%s]-%d ", CatNm(), CId);
if (BestCIdV.IsIn(CId)){CorrectExs+=1/double(CIds);}
}}
printf(" Document: "); //**
fprintf(OutFile, " Document: ");
{for (int WIdN=0; WIdN<WIds; WIdN++){
int WId=SpV->GetWId(WIdN); double WordWgt=SpV->GetWgt(WIdN);
TStr WordStr=BowDocBs->GetWordStr(WId);
printf("%s:%.2f ", WordStr.CStr(), WordWgt);
fprintf(OutFile, "%s:%.2f ", WordStr.CStr(), WordWgt);
}}
printf("\n--> "); //**
fprintf(OutFile, "\n--> ");
{for (int CIdN=0; CIdN<BestCIdV.Len(); CIdN++){
int CId=BestCIdV[CIdN];
TStr CatNm=BowDocBs->GetCatNm(CId);
printf("[%s]:%.5f ", CatNm(), CatWgtCIdKdV[CIdN].Key());
fprintf(OutFile, "[%s]:%.5f ", CatNm(), CatWgtCIdKdV[CIdN].Key());
for (int WIdN=0; WIdN<WIds; WIdN++){
int WId=SpV->GetWId(WIdN); double WordWgt=ConceptSpVV[CId]->GetWgt(WIdN);
if (WordWgt>0){
TStr WordStr=BowDocBs->GetWordStr(WId);
printf("TF(%s,%d)=%.5f ", WordStr.CStr(), CId, WordWgt);
fprintf(OutFile, "TF(%s,%d)=%.5f ", WordStr.CStr(), CId, WordWgt);
}
}
}}
printf("\n"); //**
fprintf(OutFile, "\n");
}
return CorrectExs;
}
void TBowLearn::TrainNBayesMd(const PBowDocBs& BowDocBs,
TFltVV& MdWordCatWgtVV, TFltV& MdCatWgtV,
TFltV& MdCatExWgtV){
// log P(C|Doc)= argmx_C log P(C) + sum_i TF(W_i,Doc) * log P(W_i|C)
// P(W_i|C) = (TF(W_i,C)+1)/(sum_j P(W_j,C) + AllWords
// TF(W_i,C) = sum_k TF(W_i, D_k) * P(D_k,C) -> MdWordCatWgtVV
// sum_j TF(W_j|C) -> MdCatWgtV
// P(C) -> MdCatExWgtV
// prepare shortcuts
int AllWords=BowDocBs->GetWords();
int AllCats=BowDocBs->GetCats();
int TrainDocs=BowDocBs->GetTrainDocs();
// training
// collecting word-category weights
for (int DIdN=0; DIdN<TrainDocs; DIdN++){
int DId=BowDocBs->GetTrainDId(DIdN);
int WIds=BowDocBs->GetDocWIds(DId);
int CIds=BowDocBs->GetDocCIds(DId);
for (int WIdN=0; WIdN<WIds; WIdN++){
int WId; double WordFq; BowDocBs->GetDocWIdFq(DId, WIdN, WId, WordFq);
for (int CIdN=0; CIdN<CIds; CIdN++){
int CId=BowDocBs->GetDocCId(DId, CIdN);
double CatWgt=WordFq*(1/double(CIds));
MdWordCatWgtVV.At(WId, CId)+=CatWgt;
}
}
}
// collecting category weights
{for (int CId=0; CId<AllCats; CId++){
double SumWordCatWgt=0;
for (int WId=0; WId<AllWords; WId++){
SumWordCatWgt+=MdWordCatWgtVV.At(WId, CId);
/* if (MdWordCatWgtVV.At(WId, CId)>0){
printf("TF(W%d,C%d):%.5f ", WId, CId, MdWordCatWgtVV.At(WId, CId));} */
}
MdCatWgtV[CId]=SumWordCatWgt;
}}
// collecting category example weights
{for (int DIdN=0; DIdN<TrainDocs; DIdN++){
int DId=BowDocBs->GetTrainDId(DIdN);
int CIds=BowDocBs->GetDocCIds(DId);
for (int CIdN=0; CIdN<CIds; CIdN++){
int CId=BowDocBs->GetDocCId(DId, CIdN);
MdCatExWgtV[CId]++;
}
}}
if (TrainDocs>0){
for (int CId=0; CId<AllCats; CId++){
MdCatExWgtV[CId]/=TrainDocs;
TStr CatNm=BowDocBs->GetCatNm(CId);
// if (MdCatExWgtV[CId]>0){printf("[%s]-%d:%.5f ", CatNm(), CId, MdCatExWgtV[CId]);} //**
}
// printf("\n");
} else {printf("No training examples!\n");}
}
double TBowLearn::TestNBayesMd(const PBowDocBs& BowDocBs,
TFltVV& MdWordCatWgtVV, TFltV& MdCatWgtV,
TFltV& MdCatExWgtV, FILE* OutFile, FILE* ErrFile, int& TestExs){
// log P(C|Doc)= argmx_C sum_i [log P(W_i|C) * TF(W_i,Doc)] + log P(C)
// P(W_i|C) = (TF(W_i,C)+1)/(sum_j P(W_j,C) + AllWords)
// TF(W_i,C) -> MdWordCatWgtVV
// sum_j TF(W_j|C) -> MdCatWgtV
// P(C) -> MdCatExWgtV
//
// handles multilabeled examples
//
// skip empty documents and docs where all categories have low prob. (instead of giving majority category)
// set TF(W_i,Doc)=1
//
//double MnProb=0.00000001;
// prepare shortcuts
int AllWords=BowDocBs->GetWords();
int AllCats=BowDocBs->GetCats();
// find majority category
int MajorCId=-1; TFlt MajorCatWgt=0;
{for (int CId=0; CId<AllCats; CId++){
if (MdCatExWgtV[CId]>MajorCatWgt){MajorCId=CId; MajorCatWgt=MdCatExWgtV[CId];}
}}
if (MajorCId>-1){
TStr CatNm=BowDocBs->GetCatNm(MajorCId);
// printf(" Majority P(%d:%s])=%.5f\n", MajorCId, CatNm(), MajorCatWgt); //**
fprintf(OutFile, " Majority P(%d:%s])=%.5f\n", MajorCId, CatNm(), MajorCatWgt());
} else {
// printf(" No category with probability > 0\n"); //**
fprintf(OutFile, " No category with probability > 0\n");
}
// testing on testing examples
double CorrectExs=0; TestExs=0;
//int FalsePos=0; int FalseNeg=0; int TruePos=0; int TrueNeg=0;
for (int DIdN=0; DIdN<BowDocBs->GetTestDocs(); DIdN++){
int DId=BowDocBs->GetTestDId(DIdN);
int WIds=BowDocBs->GetDocWIds(DId);
int CIds=BowDocBs->GetDocCIds(DId);
if (WIds==0){printf("It is hard to classify an empty document!\n"); continue;}
// create category vector
TIntV ShareWordsCat(AllCats); // 1 if the category share any word with document
TFltIntKdV CatWgtCIdKdV(AllCats);
{for (int CId=0; CId<AllCats; CId++){
ShareWordsCat[CId]=0;
CatWgtCIdKdV[CId].Dat=CId; CatWgtCIdKdV[CId].Key=0;
}}
{for (int CId=0; CId<AllCats; CId++){
if(MdCatExWgtV[CId]>0){ // prior probability of category > 0
for (int WIdN=0; WIdN<WIds; WIdN++){
int WId; double WordFq; BowDocBs->GetDocWIdFq(DId, WIdN, WId, WordFq);
// different contexts compose one documents, WordFq can be as high as 80
WordFq=1;
if(MdWordCatWgtVV.At(WId, CId)>0){
ShareWordsCat[CId]=1;
// printf("P(%d,%d)=%0.4f ", WId, CId, MdWordCatWgtVV.At(WId, CId));
}
double CatWgt=(MdWordCatWgtVV.At(WId, CId)+1)/(MdCatWgtV[CId]+double(AllWords));
CatWgtCIdKdV[CId].Key+=(log(CatWgt)*WordFq);
/* if(MdWordCatWgtVV.At(WId, CId)>0){
if ((CId==8)||(CId==35)||(CId==20)){
printf("DIdN=%d ", DIdN);
printf("log CatWgt[%d]=(%.3f+1)/(%.3f+%d) = %.5f\n", CId,
MdWordCatWgtVV.At(WId, CId), MdCatWgtV[CId], AllWords, CatWgtCIdKdV[CId].Key);
}
} //*/
}
// add prior probabilty only to categories that have prior porbability > 0
/* if ((CId==8)||(CId==35)||(CId==20)){
printf("log(MdCatExWgtV[%d])=%.4f, CatWgtCIdKdV[%d])=%.4f ", CId, log(MdCatExWgtV[CId]), CId, CatWgtCIdKdV[CId].Key);
} //*/
CatWgtCIdKdV[CId].Key+=log(double(MdCatExWgtV[CId]));
/* if ((CId==8)||(CId==35)||(CId==20)){
printf("CatWgtCIdKdV.Key[%d]=%.20f\n", CId, CatWgtCIdKdV[CId].Key);
} //*/
} else {
CatWgtCIdKdV[CId].Key=TFlt::NInf;
}
}}
// sort categories according to log(probability)
CatWgtCIdKdV.Sort(false);
// printf("BestCat:%d, prob:%.5f\n", CatWgtCIdKdV[0].Dat, CatWgtCIdKdV[0].Key);
if(ShareWordsCat[CatWgtCIdKdV[0].Dat]==0){ // category does not share a word with document
continue; // skip the document instead of returning prior probability
}
/* if (CatWgtCIdKdV[0].Key<=log(MnProb)){ // check the most probable category
printf("No category found for %d!\n", DId);
fprintf(OutFile, "No category found for %d!\n", DId);
continue; // skip the document instead of returning prior probability
} */
// back from log and calc prob. for each category
double SumCatWgt=0;
{for (int CIdN=0; CIdN<AllCats; CIdN++){
// printf("\nMdCatExWgtV[%d]=%.5f log(CatWgt[%d])=%.5f ", CatWgtCIdKdV[CIdN].Dat,
// MdCatExWgtV[CIdN], CatWgtCIdKdV[CIdN].Dat, CatWgtCIdKdV[CIdN].Key);
// set prob. to 0 for categories that do not share a word with document
if (ShareWordsCat[CatWgtCIdKdV[CIdN].Dat]==1){// (CatWgtCIdKdV[CId].Key>TFlt::NInf)
CatWgtCIdKdV[CIdN].Key=exp(double(CatWgtCIdKdV[CIdN].Key));
// printf("Shared: CatWgtCIdKdV.Key[%d]=%g\n", CatWgtCIdKdV[CIdN].Dat, CatWgtCIdKdV[CIdN].Key);
} else {CatWgtCIdKdV[CIdN].Key=0;}
SumCatWgt+=CatWgtCIdKdV[CIdN].Key;
/* if ((CatWgtCIdKdV[CIdN].Dat==8)||(CatWgtCIdKdV[CIdN].Dat==35)||(CatWgtCIdKdV[CIdN].Dat==20)){
printf("CatWgtCIdKdV.Key[%d]=%g\n", CatWgtCIdKdV[CIdN].Dat, CatWgtCIdKdV[CIdN].Key);
} //*/
}}
// printf("\n");
if (SumCatWgt>0){
// printf("SumCatWgt=%g ", SumCatWgt);
{for (int CIdN=0; CIdN<AllCats; CIdN++){
/* if ((CatWgtCIdKdV[CIdN].Dat==8)||(CatWgtCIdKdV[CIdN].Dat==35)||(CatWgtCIdKdV[CIdN].Dat==20)){
printf("CatWgtCIdKdV[CIdN].Key[%d]=%g ", CatWgtCIdKdV[CIdN].Dat, CatWgtCIdKdV[CIdN].Key);
} //*/
CatWgtCIdKdV[CIdN].Key/=SumCatWgt;
/* if ((CatWgtCIdKdV[CIdN].Dat==8)||(CatWgtCIdKdV[CIdN].Dat==35)||(CatWgtCIdKdV[CIdN].Dat==20)){
printf("CatWgtCIdKdV[CIdN].Key[%d]=%g ", CatWgtCIdKdV[CIdN].Dat, CatWgtCIdKdV[CIdN].Key);
} //*/
}}
}
// collect the set of best categories having summary prob. weight 25% of the total weight
double SumCatWgtSF=0; TIntV BestCIdV;
{for (int CIdN=0; CIdN<AllCats; CIdN++){
double CatWgt=CatWgtCIdKdV[CIdN].Key;
int CId=CatWgtCIdKdV[CIdN].Dat;
SumCatWgtSF+=CatWgt; BestCIdV.Add(CId);
if (SumCatWgtSF>0.25){break;}
}}
// record wrongly classified examples
bool ErrorInClass=true;
{for (int CIdN=0; CIdN<CIds; CIdN++){
int CId=BowDocBs->GetDocCId(DId, CIdN);
if (BestCIdV.IsIn(CId)){ErrorInClass=false;}
}}
if (ErrorInClass){fprintf(ErrFile, "%s\n", BowDocBs->GetDocNm(DId).CStr());}
// write best categories
TestExs++;
// printf("Document vector: "); //**
fprintf(OutFile, "Document vector:");
{for (int WIdN=0; WIdN<WIds; WIdN++){
int WId; double WordFq; BowDocBs->GetDocWIdFq(DId, WIdN, WId, WordFq);
TStr WordStr=BowDocBs->GetWordStr(WId);
for (int WordFqN=0; WordFqN<WordFq; WordFqN++){
// printf("%s ", WordStr.CStr()); //**
fprintf(OutFile, "%s ", WordStr.CStr()); //**
}
}}
// printf("\n");
fprintf(OutFile, "\n");
{for (int CIdN=0; CIdN<CIds; CIdN++){
int CId=BowDocBs->GetDocCId(DId, CIdN);
TStr CatNm=BowDocBs->GetCatNm(CId);
// printf("[%d:%s] ", CId, CatNm()); //**
fprintf(OutFile, "[%d:%s] ", CId, CatNm());
if (BestCIdV.IsIn(CId)){CorrectExs+=1/double(CIds);}
}}
// printf(" --> "); //**
fprintf(OutFile, " --> ");
{for (int CIdN=0; CIdN<BestCIdV.Len(); CIdN++){
int CId=BestCIdV[CIdN];
TStr CatNm=BowDocBs->GetCatNm(CId);
// printf("\n [%d:%s]:%.5f ", CId, CatNm(), CatWgtCIdKdV[CIdN].Key); //**
fprintf(OutFile, "\n [%d:%s]:%.5f ", CId, CatNm(), CatWgtCIdKdV[CIdN].Key());
// printf("\n Model evidence: ");
fprintf(OutFile, "\n Model evidence: ");
for (int WIdN=0; WIdN<WIds; WIdN++){
int WId; double WordFq; BowDocBs->GetDocWIdFq(DId, WIdN, WId, WordFq);
if (MdWordCatWgtVV.At(WId, CId)>0){
TStr WordStr=BowDocBs->GetWordStr(WId);
// printf("TF(%s)=%.1f ", WordStr.CStr(), MdWordCatWgtVV.At(WId, CId)); //**
fprintf(OutFile, "TF(%s)=%.1f ", WordStr.CStr(), MdWordCatWgtVV.At(WId, CId)());
}
}
// printf("\n"); //**
fprintf(OutFile, "\n");
}}
// printf("\n"); //**
fprintf(OutFile, "\n");
}
// printf("Testing documents:%d, Non-empty documents:%d ", BowDocBs->GetTestDocs(), TestExs);
fprintf(OutFile, "Testing documents:%d, Non-empty documents:%d ", BowDocBs->GetTestDocs(), TestExs);
if (TestExs>0){
// printf("CorrectExs:%.5f\n", CorrectExs);
fprintf(OutFile, "CorrectExs:%.5f\n", CorrectExs);
} else {
// printf("CorrectExs:%.5f\n", CorrectExs);
fprintf(OutFile, "CorrectExs:%.5f\n", CorrectExs);
}
return CorrectExs;
}
void TBowLearn::TrainBinNBayesMd(const PBowDocBs& BowDocBs,
TFltVV& MdWordCatWgtVV, TFltV& MdCatWgtV,
TFltV& MdCatExWgtV, const TFileId& OutFile){
// log P(C|Doc)= argmx_C log P(C) + sum_i TF(W_i,Doc) * log P(W_i|C)
// P(W_i|C) = (TF(W_i,C)+1)/(sum_j P(W_j,C) + AllWords
// TF(W_i,C) = sum_k TF(W_i, D_k) * P(D_k,C) -> MdWordCatWgtVV
// sum_j TF(W_j|C) -> MdCatWgtV
// P(C) -> MdCatExWgtV
// prepare shortcuts
int AllWords=BowDocBs->GetWords();
int AllCats=BowDocBs->GetCats();
int TrainDocs=BowDocBs->GetTrainDocs();
// training
// collecting word-category weights
for (int DIdN=0; DIdN<TrainDocs; DIdN++){
int DId=BowDocBs->GetTrainDId(DIdN);
int WIds=BowDocBs->GetDocWIds(DId);
int CIds=BowDocBs->GetDocCIds(DId);
for (int WIdN=0; WIdN<WIds; WIdN++){
int WId; double WordFq; BowDocBs->GetDocWIdFq(DId, WIdN, WId, WordFq);
for (int CIdN=0; CIdN<CIds; CIdN++){
int CId=BowDocBs->GetDocCId(DId, CIdN);
double CatWgt=WordFq*(1/double(CIds));
MdWordCatWgtVV.At(WId, CId)+=CatWgt;
}
}
}
// collecting category weights
{for (int CId=0; CId<AllCats; CId++){
double SumWordCatWgt=0;
for (int WId=0; WId<AllWords; WId++){
SumWordCatWgt+=MdWordCatWgtVV.At(WId, CId);
/* if (MdWordCatWgtVV.At(WId, CId)>0){
printf("TF(W%d,C%d):%.5f ", WId, CId, MdWordCatWgtVV.At(WId, CId));} */
}
MdCatWgtV[CId]=SumWordCatWgt;
}}
// collecting category example weights
{for (int DIdN=0; DIdN<TrainDocs; DIdN++){
int DId=BowDocBs->GetTrainDId(DIdN);
int CIds=BowDocBs->GetDocCIds(DId);
for (int CIdN=0; CIdN<CIds; CIdN++){
int CId=BowDocBs->GetDocCId(DId, CIdN);
MdCatExWgtV[CId]++;
}
}}
if (TrainDocs>0){
for (int CId=0; CId<AllCats; CId++){
MdCatExWgtV[CId]/=TrainDocs;
TStr CatNm=BowDocBs->GetCatNm(CId);
// if (MdCatExWgtV[CId]>0){printf("[%s]-%d:%.5f ", CatNm(), CId, MdCatExWgtV[CId]);} //**
}
// printf("\n");
} else {printf("No training examples!\n");}
// write the model
fprintf(OutFile, "\n# NBayes Model\n");
fprintf(OutFile, "# Feature");
for (int CId=0; CId<AllCats; CId++){
TStr CatNm=BowDocBs->GetCatNm(CId);
fprintf(OutFile, ", %s_Weight", CatNm.CStr());
}
fprintf(OutFile, "\n");
{for (int WId=0; WId<AllWords; WId++){
fprintf(OutFile, "%s", BowDocBs->GetWordStr(WId).CStr());
for (int CId=0; CId<AllCats; CId++){
fprintf(OutFile, ", %0.5f", MdWordCatWgtVV.At(WId, CId)());
}
fprintf(OutFile, "\n");
}}
}
void TBowLearn::TestBinNBayesMd(const PBowDocBs& BowDocBs,
const TStr& TargetCatNm,
TFltVV& MdWordCatWgtVV, TFltV& MdCatWgtV, TFltV& MdCatExWgtV,
const TFileId& TrainOutFile, const TFileId& TestOutFile, const TFileId& ErrFile){
// log P(C|Doc)= argmx_C sum_i [log P(W_i|C) * TF(W_i,Doc)] + log P(C)
// P(W_i|C) = (TF(W_i,C)+1)/(sum_j P(W_j,C) + AllWords)
// TF(W_i,C) = sum_k TF(W_i, D_k) * P(D_k,C) -> MdWordCatWgtVV
// sum_j TF(W_j|C) -> MdCatWgtV
// P(C) -> MdCatExWgtV
double MnProb=0.00000001;
// prepare shortcuts
int AllWords=BowDocBs->GetWords();
int AllCats=BowDocBs->GetCats();
// prepare shortcuts
int TargetCId=BowDocBs->GetCId(TargetCatNm);
int TrainDocs=BowDocBs->GetTrainDocs();
int TestDocs=BowDocBs->GetTestDocs();
// find majority category
int MajorCId=-1; TFlt MajorCatWgt=0;
{for (int CId=0; CId<AllCats; CId++){
if (MdCatExWgtV[CId]>MajorCatWgt){MajorCId=CId; MajorCatWgt=MdCatExWgtV[CId];}
}}
if (MajorCId>-1){
TStr CatNm=BowDocBs->GetCatNm(MajorCId);
printf(" Majority P([%d]:%s)=%.5f\n", MajorCId, CatNm(), MajorCatWgt()); //**
fprintf(ErrFile, " Majority P([%d]:%s)=%.5f\n", MajorCId, CatNm(), MajorCatWgt());
} else {
printf(" No category with probability > 0\n"); //**
fprintf(ErrFile, " No category with probability > 0\n");
}
// test on training examples
{int FalsePos=0; int FalseNeg=0; int TruePos=0; int TrueNeg=0;
for (int DIdN=0; DIdN<TrainDocs; DIdN++){
int DId=BowDocBs->GetTrainDId(DIdN);
bool ClassVal=BowDocBs->IsCatInDoc(DId, TargetCId);
int DocWIds=BowDocBs->GetDocWIds(DId);
if (DocWIds==0){
printf("It is hard to classify an empty document %s!\n", BowDocBs->GetDocNm(DId).CStr());
}
// create category vector
TFltIntKdV CatWgtCIdKdV(AllCats);
{for (int CId=0; CId<AllCats; CId++){
CatWgtCIdKdV[CId].Dat=CId; CatWgtCIdKdV[CId].Key=0;
}}
{for (int WIdN=0; WIdN<DocWIds; WIdN++){
int WId; double WordFq; BowDocBs->GetDocWIdFq(DId, WIdN, WId, WordFq);
WordFq=1; // use boolean weight insted of TF
// use LaPlace estimate smoothing of probability
for (int CId=0; CId<AllCats; CId++){
double CatWgt=(MdWordCatWgtVV.At(WId, CId)+1)/(MdCatWgtV[CId]+double(AllWords));
CatWgtCIdKdV[CId].Key+=(log(CatWgt)*WordFq);
}
}}
// add prior probabilty to all categories
{for (int CId=0; CId<AllCats; CId++){
if(MdCatExWgtV[CId]>0){
CatWgtCIdKdV[CId].Key+=log(double(MdCatExWgtV[CId]));
} else {CatWgtCIdKdV[CId].Key=TFlt::NInf;}
}}
// find the predicted category
CatWgtCIdKdV.Sort(false); // stil in log space
// shift to have mx. value in 0
{for (int CId=1; CId<AllCats; CId++){
CatWgtCIdKdV[CId].Key-=CatWgtCIdKdV[0].Key;
}}
CatWgtCIdKdV[0].Key-=CatWgtCIdKdV[0].Key;
// back from log and calc prob. for each category
double SumCatWgt=0;
for (int CId=0; CId<AllCats; CId++){
CatWgtCIdKdV[CId].Key=exp(double(CatWgtCIdKdV[CId].Key));
SumCatWgt+=CatWgtCIdKdV[CId].Key;
}
if (SumCatWgt>0){
for (int CId=0; CId<AllCats; CId++){CatWgtCIdKdV[CId].Key/=SumCatWgt;}
}
// prepare the results
if (CatWgtCIdKdV[0].Key<=MnProb){ // all categories have very low probability
CatWgtCIdKdV[0].Dat=MajorCId; CatWgtCIdKdV[0].Key=MajorCatWgt;
}
bool PredClassVal;
PredClassVal=(CatWgtCIdKdV[0].Dat==TargetCId);
if (PredClassVal==ClassVal){
if (PredClassVal){TruePos++;} else {TrueNeg++;}
} else {// Error In Classification
if (PredClassVal){FalsePos++;} else {FalseNeg++;}
}
// write results in the same form as SVM
int ClassValN=(ClassVal)?1:-1;
int PredClassValN=(PredClassVal)?1:-1;
fprintf(TrainOutFile, "%0.6f # true=%d # docDesc=%s\n",
PredClassValN*CatWgtCIdKdV[0].Key, ClassValN, BowDocBs->GetDocNm(DId).CStr());
}}
// now the real test on testing examples
{int FalsePos=0; int FalseNeg=0; int TruePos=0; int TrueNeg=0;
for (int DIdN=0; DIdN<TestDocs; DIdN++){
int DId=BowDocBs->GetTestDId(DIdN);
bool ClassVal=BowDocBs->IsCatInDoc(DId, TargetCId);
int DocWIds=BowDocBs->GetDocWIds(DId);
if (DocWIds==0){
printf("It is hard to classify an empty document %s!\n", BowDocBs->GetDocNm(DId).CStr());
}
// create category vector
TFltIntKdV CatWgtCIdKdV(AllCats);
{for (int CId=0; CId<AllCats; CId++){
CatWgtCIdKdV[CId].Dat=CId; CatWgtCIdKdV[CId].Key=0;
}}
{for (int WIdN=0; WIdN<DocWIds; WIdN++){
int WId; double WordFq; BowDocBs->GetDocWIdFq(DId, WIdN, WId, WordFq);
WordFq=1; // use boolean weight insted of TF
// use LaPlace estimate smoothing of probability
for (int CId=0; CId<AllCats; CId++){
double CatWgt=(MdWordCatWgtVV.At(WId, CId)+1)/(MdCatWgtV[CId]+double(AllWords));
CatWgtCIdKdV[CId].Key+=(log(CatWgt)*WordFq);
}
}}
// add prior probabilty to all categories
{for (int CId=0; CId<AllCats; CId++){
CatWgtCIdKdV[CId].Key+=log(double(MdCatExWgtV[CId]));
}}
// find the predicted category
CatWgtCIdKdV.Sort(false); // stil in log space
// shift to have mx. value in 0
{for (int CId=1; CId<AllCats; CId++){
CatWgtCIdKdV[CId].Key-=CatWgtCIdKdV[0].Key;
}}
CatWgtCIdKdV[0].Key-=CatWgtCIdKdV[0].Key;
// back from log and calc prob. for each category
double SumCatWgt=0;
for (int CId=0; CId<AllCats; CId++){
CatWgtCIdKdV[CId].Key=exp(double(CatWgtCIdKdV[CId].Key));
SumCatWgt+=CatWgtCIdKdV[CId].Key;
}
if (SumCatWgt>0){
for (int CId=0; CId<AllCats; CId++){CatWgtCIdKdV[CId].Key/=SumCatWgt;}
}
// prepare the results
if (CatWgtCIdKdV[0].Key<=MnProb){ // all categories have very low probability
CatWgtCIdKdV[0].Dat=MajorCId; CatWgtCIdKdV[0].Key=MajorCatWgt;
}
bool PredClassVal;
PredClassVal=(CatWgtCIdKdV[0].Dat==TargetCId);
if (PredClassVal==ClassVal){
if (PredClassVal){TruePos++;} else {TrueNeg++;}
} else {// Error In Classification
fprintf(ErrFile, "%s\n", BowDocBs->GetDocNm(DId).CStr());
if (PredClassVal){FalsePos++;} else {FalseNeg++;}
}
// write results in the same form as SVM
int ClassValN=(ClassVal)?1:-1;
int PredClassValN=(PredClassVal)?1:-1;
fprintf(TestOutFile, "%0.6f # true=%d # docDesc=%s\n",
PredClassValN*CatWgtCIdKdV[0].Key, ClassValN, BowDocBs->GetDocNm(DId).CStr());
}
double Acc=100*(TruePos+TrueNeg)/double(TestDocs);
double Prec=100*(TruePos/double(TruePos+FalsePos));
double Rec=100*(TruePos/double(TruePos+FalseNeg));
printf("Precision %0.2f%%, Recall %0.2f%%, Accuracy %0.2f%%\n", Prec, Rec, Acc);
// fprintf(ErrFile, "%0.2f, %0.2f, %0.2f, %0.2f\n", Beta, Prec, Rec, Acc);
}
}
void TBowLearn::TrainWinnowMd(const PBowDocBs& BowDocBs,
const TStr& TargetCatNm, TFltV& ExpWgtV,
const TFileId& OutFile, const double& Beta){
// prepare Winnow parameters
double VoteTsh=0.5;
// prepare shortcuts
int TargetCId=BowDocBs->GetCId(TargetCatNm);
int TrainDocs=BowDocBs->GetTrainDocs();
// prepare mini-experts, two for each word
ExpWgtV.PutAll(1);
const double MnWgt=1e-15;
double PrevAcc=0; double PrevPrec=0; double PrevRec=0; double PrevF1=0;
const double MxDiff=-0.005; const int MxWorseIters=3; int WorseIters=0;
const int MxIters=50; int IterN=0;
while ((IterN<MxIters)&&(WorseIters<MxWorseIters)){
IterN++;
int FalsePos=0; int FalseNeg=0; int TruePos=0; int TrueNeg=0;
for (int DIdN=0; DIdN<TrainDocs; DIdN++){
int DId=BowDocBs->GetTrainDId(DIdN);
bool ClassVal=BowDocBs->IsCatInDoc(DId, TargetCId);
double PosWgt=0; double NegWgt=0;
double OldSum=0; double NewSum=0;
int DocWIds=BowDocBs->GetDocWIds(DId);
// change only experts of words that occur in the document
for (int DocWIdN=0; DocWIdN<DocWIds; DocWIdN++){
int DocWId=BowDocBs->GetDocWId(DId, DocWIdN);
int PosExpN=2*DocWId; int NegExpN=2*DocWId+1;
OldSum+=ExpWgtV[PosExpN]+ExpWgtV[NegExpN];
// penalize expert giving wrong class prediction
int PenN=(ClassVal)?NegExpN:PosExpN;
ExpWgtV[PenN]=ExpWgtV[PenN]*Beta;
NewSum+=ExpWgtV[PosExpN]+ExpWgtV[NegExpN];
PosWgt+=ExpWgtV[PosExpN]; NegWgt+=ExpWgtV[NegExpN];
}
// normalize all experts
if (NewSum>MnWgt){
for (int DocWIdN=0; DocWIdN<DocWIds; DocWIdN++){
int DocWId=BowDocBs->GetDocWId(DId, DocWIdN);
ExpWgtV[2*DocWId]=ExpWgtV[2*DocWId]*OldSum/NewSum;
ExpWgtV[2*DocWId+1]=ExpWgtV[2*DocWId+1]*OldSum/NewSum;
}
}
bool PredClassVal;
if (PosWgt+NegWgt==0){PredClassVal=TBool::GetRnd();}
else {PredClassVal=(PosWgt/(PosWgt+NegWgt))>VoteTsh;}
if (PredClassVal==ClassVal){
if (PredClassVal){TruePos++;} else {TrueNeg++;}
} else {
if (PredClassVal){FalsePos++;} else {FalseNeg++;}
}
//printf(PredClassVal==ClassVal ? "." : (PredClassVal ? "+" : "-"));
}
// output temporary results
double Acc=100*(TruePos+TrueNeg)/double(TrainDocs);
double Prec=(TruePos/double(TruePos+FalsePos));
double Rec=(TruePos/double(TruePos+FalseNeg));
double F1=(2*Prec*Rec/(Prec+Rec));
// check if the current iteration gave worse results then the previous
if (((Acc-PrevAcc)<MxDiff)||((F1-PrevF1)<MxDiff)||(((Prec-PrevPrec)<MxDiff)&&
((Rec-PrevRec)<MxDiff))){WorseIters++;}
else {WorseIters=0;}
PrevAcc=Acc; PrevPrec=Prec; PrevRec=Rec; PrevF1=F1;
// if we want to see contingency table info.
/* printf("%d] %d True Pos, %d True Neg, %d False Pos, %d False Neg, Accuracy %0.2f%%\n",
IterN, TruePos, TrueNeg, FalsePos, FalseNeg, Acc);
fprintf(OutFile,
"%d] %d True Pos, %d True Neg, %d False Pos, %d False Neg, Accuracy %0.2f%%\n",
IterN, TruePos, TrueNeg, FalsePos, FalseNeg, Acc);
*/ if(IterN==1){fprintf(OutFile, "# IterN, Precision, Recall, F1, Accuracy\n");}
printf("%d] Precision %0.2f, Recall %0.2f, F1 %0.2f, Accuracy %0.2f%%\n",
IterN, Prec, Rec, F1, Acc);
fprintf(OutFile,
"%d %0.2f, %0.2f, %0.2f, %0.2f%%\n", IterN, Prec, Rec, F1, Acc);
}
// write the model
/* fprintf(OutFile, "\n# Winnow Model\n");
fprintf(OutFile, "# Feature, PosClass_Weight, NegClass_Weight\n");
for (int WId=0; WId<BowDocBs->GetWords(); WId++){
if((ExpWgtV[2*WId]>MnWgt)||(ExpWgtV[2*WId+1]>MnWgt)){
fprintf(OutFile, "%s, %0.5f, %0.5f\n",
BowDocBs->GetWordStr(WId).CStr(), ExpWgtV[2*WId], ExpWgtV[2*WId+1]);
}
}
*/
// used for testing paramatere range
// fprintf(OutFile,
// "%0.2f, %0.2f, %0.2f, %0.2f, %0.2f\n", Beta, PrevPrec, PrevRec, PrevF1, PrevAcc);
}
void TBowLearn::TestWinnowMd(const PBowDocBs& BowDocBs,
const TStr& TargetCatNm, const TFltV& ExpWgtV,
const TFileId& TrainOutFile, const TFileId& TestOutFile, const TFileId& ErrFile,
const double& /*Beta*/){
// prepare Winnow threshold
double VoteTsh=0.5;
// prepare shortcuts
int TargetCId=BowDocBs->GetCId(TargetCatNm);
int TrainDocs=BowDocBs->GetTrainDocs();
int TestDocs=BowDocBs->GetTestDocs();
// run through training examples
{int FalsePos=0; int FalseNeg=0; int TruePos=0; int TrueNeg=0;
for (int DIdN=0; DIdN<TrainDocs; DIdN++){
int DId=BowDocBs->GetTrainDId(DIdN);
bool ClassVal=BowDocBs->IsCatInDoc(DId, TargetCId);
double PosWgt=0; double NegWgt=0;
int DocWIds=BowDocBs->GetDocWIds(DId);
if (DocWIds==0){
printf("It is hard to classify an empty document %s!\n", BowDocBs->GetDocNm(DId).CStr());
}
// check experts of words that occur in the document
for (int DocWIdN=0; DocWIdN<DocWIds; DocWIdN++){
int DocWId=BowDocBs->GetDocWId(DId, DocWIdN);
int PosExpN=2*DocWId; int NegExpN=2*DocWId+1;
PosWgt+=ExpWgtV[PosExpN]; NegWgt+=ExpWgtV[NegExpN];
}
bool PredClassVal;
if (PosWgt+NegWgt==0){PredClassVal=TBool::GetRnd();}
else {PredClassVal=(PosWgt/(PosWgt+NegWgt))>VoteTsh;}
if (PredClassVal==ClassVal){
if (PredClassVal){TruePos++;} else {TrueNeg++;}
} else {// Error In Classification
if (PredClassVal){FalsePos++;} else {FalseNeg++;}
}
int ClassValN=(ClassVal)?1:-1;
fprintf(TrainOutFile, "%0.6f # true=%d # docDesc=%s\n",
(PosWgt/(PosWgt+NegWgt))-VoteTsh, ClassValN, BowDocBs->GetDocNm(DId).CStr());