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
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
|
// $Id$
// vim:tabstop=2
/***********************************************************************
Moses - factored phrase-based language decoder
Copyright (C) 2006 University of Edinburgh
This library is free software; you can redistribute it and/or
modify it under the terms of the GNU Lesser General Public
License as published by the Free Software Foundation; either
version 2.1 of the License, or (at your option) any later version.
This library is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
Lesser General Public License for more details.
You should have received a copy of the GNU Lesser General Public
License along with this library; if not, write to the Free Software
Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
***********************************************************************/
#include <string>
#include <cassert>
#include "PhraseDictionaryMemory.h"
#include "DecodeStepTranslation.h"
#include "DecodeStepGeneration.h"
#include "GenerationDictionary.h"
#include "DummyScoreProducers.h"
#include "StaticData.h"
#include "Util.h"
#include "FactorCollection.h"
#include "Timer.h"
#include "LanguageModelFactory.h"
#include "LexicalReordering.h"
#include "GlobalLexicalModel.h"
#include "SentenceStats.h"
#include "PhraseDictionary.h"
#include "UserMessage.h"
#include "TranslationOption.h"
#include "DecodeGraph.h"
#include "InputFileStream.h"
using namespace std;
namespace Moses
{
static size_t CalcMax(size_t x, const vector<size_t>& y)
{
size_t max = x;
for (vector<size_t>::const_iterator i=y.begin(); i != y.end(); ++i)
if (*i > max) max = *i;
return max;
}
static size_t CalcMax(size_t x, const vector<size_t>& y, const vector<size_t>& z)
{
size_t max = x;
for (vector<size_t>::const_iterator i=y.begin(); i != y.end(); ++i)
if (*i > max) max = *i;
for (vector<size_t>::const_iterator i=z.begin(); i != z.end(); ++i)
if (*i > max) max = *i;
return max;
}
StaticData StaticData::s_instance;
StaticData::StaticData()
:m_numLinkParams(1)
,m_fLMsLoaded(false)
,m_sourceStartPosMattersForRecombination(false)
,m_inputType(SentenceInput)
,m_numInputScores(0)
,m_detailedTranslationReportingFilePath()
,m_onlyDistinctNBest(false)
,m_factorDelimiter("|") // default delimiter between factors
,m_isAlwaysCreateDirectTranslationOption(false)
{
m_maxFactorIdx[0] = 0; // source side
m_maxFactorIdx[1] = 0; // target side
// memory pools
Phrase::InitializeMemPool();
}
bool StaticData::LoadData(Parameter *parameter)
{
ResetUserTime();
m_parameter = parameter;
// verbose level
m_verboseLevel = 1;
if (m_parameter->GetParam("verbose").size() == 1) {
m_verboseLevel = Scan<size_t>( m_parameter->GetParam("verbose")[0]);
}
// to cube or not to cube
m_searchAlgorithm = (m_parameter->GetParam("search-algorithm").size() > 0) ?
(SearchAlgorithm) Scan<size_t>(m_parameter->GetParam("search-algorithm")[0]) : Normal;
if (m_searchAlgorithm == ChartDecoding)
LoadChartDecodingParameters();
else
LoadPhraseBasedParameters();
// input type has to be specified BEFORE loading the phrase tables!
if(m_parameter->GetParam("inputtype").size())
m_inputType= (InputTypeEnum) Scan<int>(m_parameter->GetParam("inputtype")[0]);
std::string s_it = "text input";
if (m_inputType == 1) {
s_it = "confusion net";
}
if (m_inputType == 2) {
s_it = "word lattice";
}
VERBOSE(2,"input type is: "<<s_it<<"\n");
if(m_parameter->GetParam("recover-input-path").size()) {
m_recoverPath = Scan<bool>(m_parameter->GetParam("recover-input-path")[0]);
if (m_recoverPath && m_inputType == SentenceInput) {
TRACE_ERR("--recover-input-path should only be used with confusion net or word lattice input!\n");
m_recoverPath = false;
}
}
// factor delimiter
if (m_parameter->GetParam("factor-delimiter").size() > 0) {
m_factorDelimiter = m_parameter->GetParam("factor-delimiter")[0];
}
SetBooleanParameter( &m_continuePartialTranslation, "continue-partial-translation", false );
//word-to-word alignment
SetBooleanParameter( &m_UseAlignmentInfo, "use-alignment-info", false );
SetBooleanParameter( &m_PrintAlignmentInfo, "print-alignment-info", false );
SetBooleanParameter( &m_PrintAlignmentInfoNbest, "print-alignment-info-in-n-best", false );
SetBooleanParameter( &m_outputHypoScore, "output-hypo-score", false );
if (!m_UseAlignmentInfo && m_PrintAlignmentInfo) {
TRACE_ERR("--print-alignment-info should only be used together with \"--use-alignment-info true\". Continue forcing to false.\n");
m_PrintAlignmentInfo=false;
}
if (!m_UseAlignmentInfo && m_PrintAlignmentInfoNbest) {
TRACE_ERR("--print-alignment-info-in-n-best should only be used together with \"--use-alignment-info true\". Continue forcing to false.\n");
m_PrintAlignmentInfoNbest=false;
}
if (m_parameter->GetParam("alignment-output-file").size() > 0) {
m_alignmentOutputFile = Scan<std::string>(m_parameter->GetParam("alignment-output-file")[0]);
}
// n-best
if (m_parameter->GetParam("n-best-list").size() >= 2) {
m_nBestFilePath = m_parameter->GetParam("n-best-list")[0];
m_nBestSize = Scan<size_t>( m_parameter->GetParam("n-best-list")[1] );
m_onlyDistinctNBest=(m_parameter->GetParam("n-best-list").size()>2 && m_parameter->GetParam("n-best-list")[2]=="distinct");
} else if (m_parameter->GetParam("n-best-list").size() == 1) {
UserMessage::Add(string("ERROR: wrong format for switch -n-best-list file size"));
return false;
} else {
m_nBestSize = 0;
}
if (m_parameter->GetParam("n-best-factor").size() > 0) {
m_nBestFactor = Scan<size_t>( m_parameter->GetParam("n-best-factor")[0]);
} else {
m_nBestFactor = 20;
}
// word graph
if (m_parameter->GetParam("output-word-graph").size() == 2)
m_outputWordGraph = true;
else
m_outputWordGraph = false;
// search graph
if (m_parameter->GetParam("output-search-graph").size() > 0) {
if (m_parameter->GetParam("output-search-graph").size() != 1) {
UserMessage::Add(string("ERROR: wrong format for switch -output-search-graph file"));
return false;
}
m_outputSearchGraph = true;
}
// ... in extended format
else if (m_parameter->GetParam("output-search-graph-extended").size() > 0) {
if (m_parameter->GetParam("output-search-graph-extended").size() != 1) {
UserMessage::Add(string("ERROR: wrong format for switch -output-search-graph-extended file"));
return false;
}
m_outputSearchGraph = true;
m_outputSearchGraphExtended = true;
} else
m_outputSearchGraph = false;
#ifdef HAVE_PROTOBUF
if (m_parameter->GetParam("output-search-graph-pb").size() > 0) {
if (m_parameter->GetParam("output-search-graph-pb").size() != 1) {
UserMessage::Add(string("ERROR: wrong format for switch -output-search-graph-pb path"));
return false;
}
m_outputSearchGraphPB = true;
} else
m_outputSearchGraphPB = false;
#endif
// include feature names in the n-best list
SetBooleanParameter( &m_labeledNBestList, "labeled-n-best-list", true );
// include word alignment in the n-best list
SetBooleanParameter( &m_nBestIncludesAlignment, "include-alignment-in-n-best", false );
// printing source phrase spans
SetBooleanParameter( &m_reportSegmentation, "report-segmentation", false );
// print all factors of output translations
SetBooleanParameter( &m_reportAllFactors, "report-all-factors", false );
// print all factors of output translations
SetBooleanParameter( &m_reportAllFactorsNBest, "report-all-factors-in-n-best", false );
//
if (m_inputType == SentenceInput) {
SetBooleanParameter( &m_useTransOptCache, "use-persistent-cache", true );
m_transOptCacheMaxSize = (m_parameter->GetParam("persistent-cache-size").size() > 0)
? Scan<size_t>(m_parameter->GetParam("persistent-cache-size")[0]) : DEFAULT_MAX_TRANS_OPT_CACHE_SIZE;
} else {
m_useTransOptCache = false;
}
//input factors
const vector<string> &inputFactorVector = m_parameter->GetParam("input-factors");
for(size_t i=0; i<inputFactorVector.size(); i++) {
m_inputFactorOrder.push_back(Scan<FactorType>(inputFactorVector[i]));
}
if(m_inputFactorOrder.empty()) {
UserMessage::Add(string("no input factor specified in config file"));
return false;
}
//output factors
const vector<string> &outputFactorVector = m_parameter->GetParam("output-factors");
for(size_t i=0; i<outputFactorVector.size(); i++) {
m_outputFactorOrder.push_back(Scan<FactorType>(outputFactorVector[i]));
}
if(m_outputFactorOrder.empty()) {
// default. output factor 0
m_outputFactorOrder.push_back(0);
}
//source word deletion
SetBooleanParameter( &m_wordDeletionEnabled, "phrase-drop-allowed", false );
//Disable discarding
SetBooleanParameter(&m_disableDiscarding, "disable-discarding", false);
//Print All Derivations
SetBooleanParameter( &m_printAllDerivations , "print-all-derivations", false );
// additional output
if (m_parameter->isParamSpecified("translation-details")) {
const vector<string> &args = m_parameter->GetParam("translation-details");
if (args.size() == 1) {
m_detailedTranslationReportingFilePath = args[0];
} else {
UserMessage::Add(string("the translation-details option requires exactly one filename argument"));
return false;
}
}
// word penalties
for (size_t i = 0; i < m_parameter->GetParam("weight-w").size(); ++i) {
float weightWordPenalty = Scan<float>( m_parameter->GetParam("weight-w")[i] );
m_wordPenaltyProducers.push_back(new WordPenaltyProducer(m_scoreIndexManager));
m_allWeights.push_back(weightWordPenalty);
}
float weightUnknownWord = (m_parameter->GetParam("weight-u").size() > 0) ? Scan<float>(m_parameter->GetParam("weight-u")[0]) : 1;
m_unknownWordPenaltyProducer = new UnknownWordPenaltyProducer(m_scoreIndexManager);
m_allWeights.push_back(weightUnknownWord);
// reordering constraints
m_maxDistortion = (m_parameter->GetParam("distortion-limit").size() > 0) ?
Scan<int>(m_parameter->GetParam("distortion-limit")[0])
: -1;
SetBooleanParameter( &m_reorderingConstraint, "monotone-at-punctuation", false );
// settings for pruning
m_maxHypoStackSize = (m_parameter->GetParam("stack").size() > 0)
? Scan<size_t>(m_parameter->GetParam("stack")[0]) : DEFAULT_MAX_HYPOSTACK_SIZE;
m_minHypoStackDiversity = 0;
if (m_parameter->GetParam("stack-diversity").size() > 0) {
if (m_maxDistortion > 15) {
UserMessage::Add("stack diversity > 0 is not allowed for distortion limits larger than 15");
return false;
}
if (m_inputType == WordLatticeInput) {
UserMessage::Add("stack diversity > 0 is not allowed for lattice input");
return false;
}
m_minHypoStackDiversity = Scan<size_t>(m_parameter->GetParam("stack-diversity")[0]);
}
m_beamWidth = (m_parameter->GetParam("beam-threshold").size() > 0) ?
TransformScore(Scan<float>(m_parameter->GetParam("beam-threshold")[0]))
: TransformScore(DEFAULT_BEAM_WIDTH);
m_earlyDiscardingThreshold = (m_parameter->GetParam("early-discarding-threshold").size() > 0) ?
TransformScore(Scan<float>(m_parameter->GetParam("early-discarding-threshold")[0]))
: TransformScore(DEFAULT_EARLY_DISCARDING_THRESHOLD);
m_translationOptionThreshold = (m_parameter->GetParam("translation-option-threshold").size() > 0) ?
TransformScore(Scan<float>(m_parameter->GetParam("translation-option-threshold")[0]))
: TransformScore(DEFAULT_TRANSLATION_OPTION_THRESHOLD);
m_maxNoTransOptPerCoverage = (m_parameter->GetParam("max-trans-opt-per-coverage").size() > 0)
? Scan<size_t>(m_parameter->GetParam("max-trans-opt-per-coverage")[0]) : DEFAULT_MAX_TRANS_OPT_SIZE;
m_maxNoPartTransOpt = (m_parameter->GetParam("max-partial-trans-opt").size() > 0)
? Scan<size_t>(m_parameter->GetParam("max-partial-trans-opt")[0]) : DEFAULT_MAX_PART_TRANS_OPT_SIZE;
m_maxPhraseLength = (m_parameter->GetParam("max-phrase-length").size() > 0)
? Scan<size_t>(m_parameter->GetParam("max-phrase-length")[0]) : DEFAULT_MAX_PHRASE_LENGTH;
m_cubePruningPopLimit = (m_parameter->GetParam("cube-pruning-pop-limit").size() > 0)
? Scan<size_t>(m_parameter->GetParam("cube-pruning-pop-limit")[0]) : DEFAULT_CUBE_PRUNING_POP_LIMIT;
m_cubePruningDiversity = (m_parameter->GetParam("cube-pruning-diversity").size() > 0)
? Scan<size_t>(m_parameter->GetParam("cube-pruning-diversity")[0]) : DEFAULT_CUBE_PRUNING_DIVERSITY;
// unknown word processing
SetBooleanParameter( &m_dropUnknown, "drop-unknown", false );
// minimum Bayes risk decoding
SetBooleanParameter( &m_mbr, "minimum-bayes-risk", false );
m_mbrSize = (m_parameter->GetParam("mbr-size").size() > 0) ?
Scan<size_t>(m_parameter->GetParam("mbr-size")[0]) : 200;
m_mbrScale = (m_parameter->GetParam("mbr-scale").size() > 0) ?
Scan<float>(m_parameter->GetParam("mbr-scale")[0]) : 1.0f;
//lattice mbr
SetBooleanParameter( &m_useLatticeMBR, "lminimum-bayes-risk", false );
if (m_useLatticeMBR && m_mbr) {
cerr << "Errror: Cannot use both n-best mbr and lattice mbr together" << endl;
exit(1);
}
if (m_useLatticeMBR) m_mbr = true;
m_lmbrPruning = (m_parameter->GetParam("lmbr-pruning-factor").size() > 0) ?
Scan<size_t>(m_parameter->GetParam("lmbr-pruning-factor")[0]) : 30;
m_lmbrThetas = Scan<float>(m_parameter->GetParam("lmbr-thetas"));
SetBooleanParameter( &m_useLatticeHypSetForLatticeMBR, "lattice-hypo-set", false );
m_lmbrPrecision = (m_parameter->GetParam("lmbr-p").size() > 0) ?
Scan<float>(m_parameter->GetParam("lmbr-p")[0]) : 0.8f;
m_lmbrPRatio = (m_parameter->GetParam("lmbr-r").size() > 0) ?
Scan<float>(m_parameter->GetParam("lmbr-r")[0]) : 0.6f;
m_lmbrMapWeight = (m_parameter->GetParam("lmbr-map-weight").size() >0) ?
Scan<float>(m_parameter->GetParam("lmbr-map-weight")[0]) : 0.0f;
//consensus decoding
SetBooleanParameter( &m_useConsensusDecoding, "consensus-decoding", false );
if (m_useConsensusDecoding && m_mbr) {
cerr<< "Error: Cannot use consensus decoding together with mbr" << endl;
exit(1);
}
if (m_useConsensusDecoding) m_mbr=true;
m_timeout_threshold = (m_parameter->GetParam("time-out").size() > 0) ?
Scan<size_t>(m_parameter->GetParam("time-out")[0]) : -1;
m_timeout = (GetTimeoutThreshold() == (size_t)-1) ? false : true;
m_lmcache_cleanup_threshold = (m_parameter->GetParam("clean-lm-cache").size() > 0) ?
Scan<size_t>(m_parameter->GetParam("clean-lm-cache")[0]) : 1;
// Read in constraint decoding file, if provided
if(m_parameter->GetParam("constraint").size()) {
if (m_parameter->GetParam("search-algorithm").size() > 0
&& Scan<size_t>(m_parameter->GetParam("search-algorithm")[0]) != 0) {
cerr << "Can use -constraint only with stack-based search (-search-algorithm 0)" << endl;
exit(1);
}
m_constraintFileName = m_parameter->GetParam("constraint")[0];
InputFileStream constraintFile(m_constraintFileName);
std::string line;
long sentenceID = -1;
while (getline(constraintFile, line)) {
vector<string> vecStr = Tokenize(line, "\t");
if (vecStr.size() == 1) {
sentenceID++;
Phrase phrase(Output, 0);
phrase.CreateFromString(GetOutputFactorOrder(), vecStr[0], GetFactorDelimiter());
m_constraints.insert(make_pair(sentenceID,phrase));
} else if (vecStr.size() == 2) {
sentenceID = Scan<long>(vecStr[0]);
Phrase phrase(Output, 0);
phrase.CreateFromString(GetOutputFactorOrder(), vecStr[1], GetFactorDelimiter());
m_constraints.insert(make_pair(sentenceID,phrase));
} else {
assert(false);
}
}
}
// use of xml in input
if (m_parameter->GetParam("xml-input").size() == 0) m_xmlInputType = XmlPassThrough;
else if (m_parameter->GetParam("xml-input")[0]=="exclusive") m_xmlInputType = XmlExclusive;
else if (m_parameter->GetParam("xml-input")[0]=="inclusive") m_xmlInputType = XmlInclusive;
else if (m_parameter->GetParam("xml-input")[0]=="ignore") m_xmlInputType = XmlIgnore;
else if (m_parameter->GetParam("xml-input")[0]=="pass-through") m_xmlInputType = XmlPassThrough;
else {
UserMessage::Add("invalid xml-input value, must be pass-through, exclusive, inclusive, or ignore");
return false;
}
if (!LoadLexicalReorderingModel()) return false;
if (!LoadLanguageModels()) return false;
if (!LoadGenerationTables()) return false;
if (!LoadPhraseTables()) return false;
if (!LoadGlobalLexicalModel()) return false;
if (!LoadDecodeGraphs()) return false;
//configure the translation systems with these tables
vector<string> tsConfig = m_parameter->GetParam("translation-systems");
if (!tsConfig.size()) {
//use all models in default system.
tsConfig.push_back(TranslationSystem::DEFAULT + " R * D * L * G *");
}
if (m_wordPenaltyProducers.size() != tsConfig.size()) {
UserMessage::Add(string("Mismatch between number of word penalties and number of translation systems"));
return false;
}
if (m_searchAlgorithm == ChartDecoding) {
//insert some null distortion score producers
m_distortionScoreProducers.assign(tsConfig.size(), NULL);
} else {
if (m_distortionScoreProducers.size() != tsConfig.size()) {
UserMessage::Add(string("Mismatch between number of distortion scores and number of translation systems"));
return false;
}
}
for (size_t i = 0; i < tsConfig.size(); ++i) {
vector<string> config = Tokenize(tsConfig[i]);
if (config.size() % 2 != 1) {
UserMessage::Add(string("Incorrect number of fields in Translation System config. Should be an odd number"));
}
m_translationSystems.insert(pair<string, TranslationSystem>(config[0],
TranslationSystem(config[0],m_wordPenaltyProducers[i],m_unknownWordPenaltyProducer,m_distortionScoreProducers[i])));
for (size_t j = 1; j < config.size(); j += 2) {
const string& id = config[j];
const string& tables = config[j+1];
set<size_t> tableIds;
if (tables != "*") {
//selected tables
vector<string> tableIdStrings = Tokenize(tables,",");
vector<size_t> tableIdList;
Scan<size_t>(tableIdList, tableIdStrings);
copy(tableIdList.begin(), tableIdList.end(), inserter(tableIds,tableIds.end()));
}
if (id == "D") {
for (size_t k = 0; k < m_decodeGraphs.size(); ++k) {
if (!tableIds.size() || tableIds.find(k) != tableIds.end()) {
VERBOSE(2,"Adding decoder graph " << k << " to translation system " << config[0] << endl);
m_translationSystems.find(config[0])->second.AddDecodeGraph(m_decodeGraphs[k],m_decodeGraphBackoff[k]);
}
}
} else if (id == "R") {
for (size_t k = 0; k < m_reorderModels.size(); ++k) {
if (!tableIds.size() || tableIds.find(k) != tableIds.end()) {
m_translationSystems.find(config[0])->second.AddReorderModel(m_reorderModels[k]);
VERBOSE(2,"Adding reorder table " << k << " to translation system " << config[0] << endl);
}
}
} else if (id == "G") {
for (size_t k = 0; k < m_globalLexicalModels.size(); ++k) {
if (!tableIds.size() || tableIds.find(k) != tableIds.end()) {
m_translationSystems.find(config[0])->second.AddGlobalLexicalModel(m_globalLexicalModels[k]);
VERBOSE(2,"Adding global lexical model " << k << " to translation system " << config[0] << endl);
}
}
} else if (id == "L") {
size_t lmid = 0;
for (LMList::const_iterator k = m_languageModel.begin(); k != m_languageModel.end(); ++k, ++lmid) {
if (!tableIds.size() || tableIds.find(lmid) != tableIds.end()) {
m_translationSystems.find(config[0])->second.AddLanguageModel(*k);
VERBOSE(2,"Adding language model " << lmid << " to translation system " << config[0] << endl);
}
}
} else {
UserMessage::Add(string("Incorrect translation system identifier: ") + id);
return false;
}
}
//Instigate dictionary loading
m_translationSystems.find(config[0])->second.ConfigDictionaries();
//Add any other features here.
}
m_scoreIndexManager.InitFeatureNames();
return true;
}
void StaticData::SetBooleanParameter( bool *parameter, string parameterName, bool defaultValue )
{
// default value if nothing is specified
*parameter = defaultValue;
if (! m_parameter->isParamSpecified( parameterName ) ) {
return;
}
// if parameter is just specified as, e.g. "-parameter" set it true
if (m_parameter->GetParam( parameterName ).size() == 0) {
*parameter = true;
}
// if paramter is specified "-parameter true" or "-parameter false"
else if (m_parameter->GetParam( parameterName ).size() == 1) {
*parameter = Scan<bool>( m_parameter->GetParam( parameterName )[0]);
}
}
StaticData::~StaticData()
{
RemoveAllInColl(m_phraseDictionary);
RemoveAllInColl(m_generationDictionary);
RemoveAllInColl(m_reorderModels);
RemoveAllInColl(m_globalLexicalModels);
RemoveAllInColl(m_decodeGraphs);
RemoveAllInColl(m_wordPenaltyProducers);
RemoveAllInColl(m_distortionScoreProducers);
m_languageModel.CleanUp();
// delete trans opt
map<std::pair<size_t, Phrase>, std::pair< TranslationOptionList*, clock_t > >::iterator iterCache;
for (iterCache = m_transOptCache.begin() ; iterCache != m_transOptCache.end() ; ++iterCache) {
TranslationOptionList *transOptList = iterCache->second.first;
delete transOptList;
}
// small score producers
delete m_unknownWordPenaltyProducer;
//delete m_parameter;
// memory pools
Phrase::FinalizeMemPool();
}
bool StaticData::LoadLexicalReorderingModel()
{
VERBOSE(1, "Loading lexical distortion models...");
const vector<string> fileStr = m_parameter->GetParam("distortion-file");
bool hasWeightlr = (m_parameter->GetParam("weight-lr").size() != 0);
vector<string> weightsStr;
if (hasWeightlr) {
weightsStr = m_parameter->GetParam("weight-lr");
} else {
weightsStr = m_parameter->GetParam("weight-d");
}
std::vector<float> weights;
size_t w = 1; //cur weight
if (hasWeightlr) {
w = 0; // if reading from weight-lr, don't have to count first as distortion penalty
}
size_t f = 0; //cur file
//get weights values
VERBOSE(1, "have " << fileStr.size() << " models" << std::endl);
for(size_t j = 0; j < weightsStr.size(); ++j) {
weights.push_back(Scan<float>(weightsStr[j]));
}
//load all models
for(size_t i = 0; i < fileStr.size(); ++i) {
vector<string> spec = Tokenize<string>(fileStr[f], " ");
++f; //mark file as consumed
if(spec.size() != 4) {
UserMessage::Add("Invalid Lexical Reordering Model Specification: " + fileStr[f]);
return false;
}
// spec[0] = factor map
// spec[1] = name
// spec[2] = num weights
// spec[3] = fileName
// decode factor map
vector<FactorType> input, output;
vector<string> inputfactors = Tokenize(spec[0],"-");
if(inputfactors.size() == 2) {
input = Tokenize<FactorType>(inputfactors[0],",");
output = Tokenize<FactorType>(inputfactors[1],",");
} else if(inputfactors.size() == 1) {
//if there is only one side assume it is on e side... why?
output = Tokenize<FactorType>(inputfactors[0],",");
} else {
//format error
return false;
}
string modelType = spec[1];
// decode num weights and fetch weights from array
std::vector<float> mweights;
size_t numWeights = atoi(spec[2].c_str());
for(size_t k = 0; k < numWeights; ++k, ++w) {
if(w >= weights.size()) {
UserMessage::Add("Lexicalized distortion model: Not enough weights, add to [weight-d]");
return false;
} else
mweights.push_back(weights[w]);
}
string filePath = spec[3];
m_reorderModels.push_back(new LexicalReordering(input, output, modelType, filePath, mweights));
}
return true;
}
bool StaticData::LoadGlobalLexicalModel()
{
const vector<float> &weight = Scan<float>(m_parameter->GetParam("weight-lex"));
const vector<string> &file = m_parameter->GetParam("global-lexical-file");
if (weight.size() != file.size()) {
std::cerr << "number of weights and models for the global lexical model does not match ("
<< weight.size() << " != " << file.size() << ")" << std::endl;
return false;
}
for (size_t i = 0; i < weight.size(); i++ ) {
vector<string> spec = Tokenize<string>(file[i], " ");
if ( spec.size() != 2 ) {
std::cerr << "wrong global lexical model specification: " << file[i] << endl;
return false;
}
vector< string > factors = Tokenize(spec[0],"-");
if ( factors.size() != 2 ) {
std::cerr << "wrong factor definition for global lexical model: " << spec[0] << endl;
return false;
}
vector<FactorType> inputFactors = Tokenize<FactorType>(factors[0],",");
vector<FactorType> outputFactors = Tokenize<FactorType>(factors[1],",");
m_globalLexicalModels.push_back( new GlobalLexicalModel( spec[1], weight[i], inputFactors, outputFactors ) );
}
return true;
}
bool StaticData::LoadLanguageModels()
{
if (m_parameter->GetParam("lmodel-file").size() > 0) {
// weights
vector<float> weightAll = Scan<float>(m_parameter->GetParam("weight-l"));
for (size_t i = 0 ; i < weightAll.size() ; i++) {
m_allWeights.push_back(weightAll[i]);
}
// dictionary upper-bounds fo all IRST LMs
vector<int> LMdub = Scan<int>(m_parameter->GetParam("lmodel-dub"));
if (m_parameter->GetParam("lmodel-dub").size() == 0) {
for(size_t i=0; i<m_parameter->GetParam("lmodel-file").size(); i++)
LMdub.push_back(0);
}
// initialize n-gram order for each factor. populated only by factored lm
const vector<string> &lmVector = m_parameter->GetParam("lmodel-file");
//prevent language models from being loaded twice
map<string,LanguageModel*> languageModelsLoaded;
for(size_t i=0; i<lmVector.size(); i++) {
LanguageModel* lm = NULL;
if (languageModelsLoaded.find(lmVector[i]) != languageModelsLoaded.end()) {
lm = new LanguageModel(m_scoreIndexManager, languageModelsLoaded[lmVector[i]]);
} else {
vector<string> token = Tokenize(lmVector[i]);
if (token.size() != 4 && token.size() != 5 ) {
UserMessage::Add("Expected format 'LM-TYPE FACTOR-TYPE NGRAM-ORDER filePath [mapFilePath (only for IRSTLM)]'");
return false;
}
// type = implementation, SRI, IRST etc
LMImplementation lmImplementation = static_cast<LMImplementation>(Scan<int>(token[0]));
// factorType = 0 = Surface, 1 = POS, 2 = Stem, 3 = Morphology, etc
vector<FactorType> factorTypes = Tokenize<FactorType>(token[1], ",");
// nGramOrder = 2 = bigram, 3 = trigram, etc
size_t nGramOrder = Scan<int>(token[2]);
string &languageModelFile = token[3];
if (token.size() == 5) {
if (lmImplementation==IRST)
languageModelFile += " " + token[4];
else {
UserMessage::Add("Expected format 'LM-TYPE FACTOR-TYPE NGRAM-ORDER filePath [mapFilePath (only for IRSTLM)]'");
return false;
}
}
IFVERBOSE(1)
PrintUserTime(string("Start loading LanguageModel ") + languageModelFile);
lm = LanguageModelFactory::CreateLanguageModel(
lmImplementation
, factorTypes
, nGramOrder
, languageModelFile
, m_scoreIndexManager
, LMdub[i]);
if (lm == NULL) {
UserMessage::Add("no LM created. We probably don't have it compiled");
return false;
}
languageModelsLoaded[lmVector[i]] = lm;
}
m_languageModel.Add(lm);
}
}
// flag indicating that language models were loaded,
// since phrase table loading requires their presence
m_fLMsLoaded = true;
IFVERBOSE(1)
PrintUserTime("Finished loading LanguageModels");
return true;
}
bool StaticData::LoadGenerationTables()
{
if (m_parameter->GetParam("generation-file").size() > 0) {
const vector<string> &generationVector = m_parameter->GetParam("generation-file");
const vector<float> &weight = Scan<float>(m_parameter->GetParam("weight-generation"));
IFVERBOSE(1) {
TRACE_ERR( "weight-generation: ");
for (size_t i = 0 ; i < weight.size() ; i++) {
TRACE_ERR( weight[i] << "\t");
}
TRACE_ERR(endl);
}
size_t currWeightNum = 0;
for(size_t currDict = 0 ; currDict < generationVector.size(); currDict++) {
vector<string> token = Tokenize(generationVector[currDict]);
vector<FactorType> input = Tokenize<FactorType>(token[0], ",")
,output = Tokenize<FactorType>(token[1], ",");
m_maxFactorIdx[1] = CalcMax(m_maxFactorIdx[1], input, output);
string filePath;
size_t numFeatures;
numFeatures = Scan<size_t>(token[2]);
filePath = token[3];
if (!FileExists(filePath) && FileExists(filePath + ".gz")) {
filePath += ".gz";
}
VERBOSE(1, filePath << endl);
m_generationDictionary.push_back(new GenerationDictionary(numFeatures, m_scoreIndexManager, input,output));
assert(m_generationDictionary.back() && "could not create GenerationDictionary");
if (!m_generationDictionary.back()->Load(filePath, Output)) {
delete m_generationDictionary.back();
return false;
}
for(size_t i = 0; i < numFeatures; i++) {
assert(currWeightNum < weight.size());
m_allWeights.push_back(weight[currWeightNum++]);
}
}
if (currWeightNum != weight.size()) {
TRACE_ERR( " [WARNING] config file has " << weight.size() << " generation weights listed, but the configuration for generation files indicates there should be " << currWeightNum << "!\n");
}
}
return true;
}
/* Doesn't load phrase tables any more. Just creates the features. */
bool StaticData::LoadPhraseTables()
{
VERBOSE(2,"Creating phrase table features" << endl);
// language models must be loaded prior to loading phrase tables
assert(m_fLMsLoaded);
// load phrase translation tables
if (m_parameter->GetParam("ttable-file").size() > 0) {
// weights
vector<float> weightAll = Scan<float>(m_parameter->GetParam("weight-t"));
const vector<string> &translationVector = m_parameter->GetParam("ttable-file");
vector<size_t> maxTargetPhrase = Scan<size_t>(m_parameter->GetParam("ttable-limit"));
if(maxTargetPhrase.size() == 1 && translationVector.size() > 1) {
VERBOSE(1, "Using uniform ttable-limit of " << maxTargetPhrase[0] << " for all translation tables." << endl);
for(size_t i = 1; i < translationVector.size(); i++)
maxTargetPhrase.push_back(maxTargetPhrase[0]);
} else if(maxTargetPhrase.size() != 1 && maxTargetPhrase.size() < translationVector.size()) {
stringstream strme;
strme << "You specified " << translationVector.size() << " translation tables, but only " << maxTargetPhrase.size() << " ttable-limits.";
UserMessage::Add(strme.str());
return false;
}
size_t index = 0;
size_t weightAllOffset = 0;
bool oldFileFormat = false;
for(size_t currDict = 0 ; currDict < translationVector.size(); currDict++) {
vector<string> token = Tokenize(translationVector[currDict]);
if(currDict == 0 && token.size() == 4) {
VERBOSE(1, "Warning: Phrase table specification in old 4-field format. Assuming binary phrase tables (type 1)!" << endl);
oldFileFormat = true;
}
if((!oldFileFormat && token.size() < 5) || (oldFileFormat && token.size() != 4)) {
UserMessage::Add("invalid phrase table specification");
return false;
}
PhraseTableImplementation implementation = (PhraseTableImplementation) Scan<int>(token[0]);
if(oldFileFormat) {
token.push_back(token[3]);
token[3] = token[2];
token[2] = token[1];
token[1] = token[0];
token[0] = "1";
implementation = Binary;
} else
implementation = (PhraseTableImplementation) Scan<int>(token[0]);
assert(token.size() >= 5);
//characteristics of the phrase table
vector<FactorType> input = Tokenize<FactorType>(token[1], ",")
,output = Tokenize<FactorType>(token[2], ",");
m_maxFactorIdx[0] = CalcMax(m_maxFactorIdx[0], input);
m_maxFactorIdx[1] = CalcMax(m_maxFactorIdx[1], output);
m_maxNumFactors = std::max(m_maxFactorIdx[0], m_maxFactorIdx[1]) + 1;
size_t numScoreComponent = Scan<size_t>(token[3]);
string filePath= token[4];
assert(weightAll.size() >= weightAllOffset + numScoreComponent);
// weights for this phrase dictionary
// first InputScores (if any), then translation scores
vector<float> weight;
if(currDict==0 && (m_inputType == ConfusionNetworkInput || m_inputType == WordLatticeInput)) {
// TODO. find what the assumptions made by confusion network about phrase table output which makes
// it only work with binrary file. This is a hack
m_numInputScores=m_parameter->GetParam("weight-i").size();
for(unsigned k=0; k<m_numInputScores; ++k)
weight.push_back(Scan<float>(m_parameter->GetParam("weight-i")[k]));
if(m_parameter->GetParam("link-param-count").size())
m_numLinkParams = Scan<size_t>(m_parameter->GetParam("link-param-count")[0]);
//print some info about this interaction:
if (m_numLinkParams == m_numInputScores) {
VERBOSE(1,"specified equal numbers of link parameters and insertion weights, not using non-epsilon 'real' word link count.\n");
} else if ((m_numLinkParams + 1) == m_numInputScores) {
VERBOSE(1,"WARN: "<< m_numInputScores << " insertion weights found and only "<< m_numLinkParams << " link parameters specified, applying non-epsilon 'real' word link count for last feature weight.\n");
} else {
stringstream strme;
strme << "You specified " << m_numInputScores
<< " input weights (weight-i), but you specified " << m_numLinkParams << " link parameters (link-param-count)!";
UserMessage::Add(strme.str());
return false;
}
}
if (!m_inputType) {
m_numInputScores=0;
}
//this number changes depending on what phrase table we're talking about: only 0 has the weights on it
size_t tableInputScores = (currDict == 0 ? m_numInputScores : 0);
for (size_t currScore = 0 ; currScore < numScoreComponent; currScore++)
weight.push_back(weightAll[weightAllOffset + currScore]);
if(weight.size() - tableInputScores != numScoreComponent) {
stringstream strme;
strme << "Your phrase table has " << numScoreComponent
<< " scores, but you specified " << (weight.size() - tableInputScores) << " weights!";
UserMessage::Add(strme.str());
return false;
}
weightAllOffset += numScoreComponent;
numScoreComponent += tableInputScores;
string targetPath, alignmentsFile;
if (implementation == SuffixArray) {
targetPath = token[5];
alignmentsFile= token[6];
}
assert(numScoreComponent==weight.size());
std::copy(weight.begin(),weight.end(),std::back_inserter(m_allWeights));
//This is needed for regression testing, but the phrase table
//might not really be loading here
IFVERBOSE(1)
PrintUserTime(string("Start loading PhraseTable ") + filePath);
VERBOSE(1,"filePath: " << filePath <<endl);
PhraseDictionaryFeature* pdf = new PhraseDictionaryFeature(
implementation
, numScoreComponent
, (currDict==0 ? m_numInputScores : 0)
, input
, output
, filePath
, weight
, maxTargetPhrase[index]
, targetPath, alignmentsFile);
m_phraseDictionary.push_back(pdf);
index++;
}
}
IFVERBOSE(1)
PrintUserTime("Finished loading phrase tables");
return true;
}
void StaticData::LoadNonTerminals()
{
string defaultNonTerminals;
if (m_parameter->GetParam("non-terminals").size() == 0) {
defaultNonTerminals = "X";
} else {
vector<std::string> tokens = Tokenize(m_parameter->GetParam("non-terminals")[0]);
defaultNonTerminals = tokens[0];
}
FactorCollection &factorCollection = FactorCollection::Instance();
m_inputDefaultNonTerminal.SetIsNonTerminal(true);
const Factor *sourceFactor = factorCollection.AddFactor(Input, 0, defaultNonTerminals);
m_inputDefaultNonTerminal.SetFactor(0, sourceFactor);
m_outputDefaultNonTerminal.SetIsNonTerminal(true);
const Factor *targetFactor = factorCollection.AddFactor(Output, 0, defaultNonTerminals);
m_outputDefaultNonTerminal.SetFactor(0, targetFactor);
// for unknwon words
if (m_parameter->GetParam("unknown-lhs").size() == 0) {
UnknownLHSEntry entry(defaultNonTerminals, 0.0f);
m_unknownLHS.push_back(entry);
} else {
const string &filePath = m_parameter->GetParam("unknown-lhs")[0];
InputFileStream inStream(filePath);
string line;
while(getline(inStream, line)) {
vector<string> tokens = Tokenize(line);
assert(tokens.size() == 2);
UnknownLHSEntry entry(tokens[0], Scan<float>(tokens[1]));
m_unknownLHS.push_back(entry);
}
}
}
void StaticData::LoadChartDecodingParameters()
{
LoadNonTerminals();
// source label overlap
if (m_parameter->GetParam("source-label-overlap").size() > 0) {
m_sourceLabelOverlap = (SourceLabelOverlap) Scan<int>(m_parameter->GetParam("source-label-overlap")[0]);
} else {
m_sourceLabelOverlap = SourceLabelOverlapAdd;
}
m_ruleLimit = (m_parameter->GetParam("rule-limit").size() > 0)
? Scan<size_t>(m_parameter->GetParam("rule-limit")[0]) : DEFAULT_MAX_TRANS_OPT_SIZE;
}
void StaticData::LoadPhraseBasedParameters()
{
const vector<string> distortionWeights = m_parameter->GetParam("weight-d");
size_t distortionWeightCount = distortionWeights.size();
//if there's a lex-reordering model, and no separate weight set, then
//take just one of these weights for linear distortion
if (!m_parameter->GetParam("weight-lr").size() && m_parameter->GetParam("distortion-file").size()) {
distortionWeightCount = 1;
}
for (size_t i = 0; i < distortionWeightCount; ++i) {
float weightDistortion = Scan<float>(distortionWeights[i]);
m_distortionScoreProducers.push_back(new DistortionScoreProducer(m_scoreIndexManager));
m_allWeights.push_back(weightDistortion);
}
}
bool StaticData::LoadDecodeGraphs()
{
const vector<string> &mappingVector = m_parameter->GetParam("mapping");
const vector<size_t> &maxChartSpans = Scan<size_t>(m_parameter->GetParam("max-chart-span"));
DecodeStep *prev = 0;
size_t prevDecodeGraphInd = 0;
for(size_t i=0; i<mappingVector.size(); i++) {
vector<string> token = Tokenize(mappingVector[i]);
size_t decodeGraphInd;
DecodeType decodeType;
size_t index;
if (token.size() == 2) {
decodeGraphInd = 0;
decodeType = token[0] == "T" ? Translate : Generate;
index = Scan<size_t>(token[1]);
} else if (token.size() == 3) {
// For specifying multiple translation model
decodeGraphInd = Scan<size_t>(token[0]);
//the vectorList index can only increment by one
assert(decodeGraphInd == prevDecodeGraphInd || decodeGraphInd == prevDecodeGraphInd + 1);
if (decodeGraphInd > prevDecodeGraphInd) {
prev = NULL;
}
decodeType = token[1] == "T" ? Translate : Generate;
index = Scan<size_t>(token[2]);
} else {
UserMessage::Add("Malformed mapping!");
assert(false);
}
DecodeStep* decodeStep = NULL;
switch (decodeType) {
case Translate:
if(index>=m_phraseDictionary.size()) {
stringstream strme;
strme << "No phrase dictionary with index "
<< index << " available!";
UserMessage::Add(strme.str());
assert(false);
}
decodeStep = new DecodeStepTranslation(m_phraseDictionary[index], prev);
break;
case Generate:
if(index>=m_generationDictionary.size()) {
stringstream strme;
strme << "No generation dictionary with index "
<< index << " available!";
UserMessage::Add(strme.str());
assert(false);
}
decodeStep = new DecodeStepGeneration(m_generationDictionary[index], prev);
break;
case InsertNullFertilityWord:
assert(!"Please implement NullFertilityInsertion.");
break;
}
assert(decodeStep);
if (m_decodeGraphs.size() < decodeGraphInd + 1) {
DecodeGraph *decodeGraph;
if (m_searchAlgorithm == ChartDecoding) {
size_t maxChartSpan = (decodeGraphInd < maxChartSpans.size()) ? maxChartSpans[decodeGraphInd] : DEFAULT_MAX_CHART_SPAN;
decodeGraph = new DecodeGraph(m_decodeGraphs.size(), maxChartSpan);
} else {
decodeGraph = new DecodeGraph(m_decodeGraphs.size());
}
m_decodeGraphs.push_back(decodeGraph); // TODO max chart span
}
m_decodeGraphs[decodeGraphInd]->Add(decodeStep);
prev = decodeStep;
prevDecodeGraphInd = decodeGraphInd;
}
// set maximum n-gram size for backoff approach to decoding paths
// default is always use subsequent paths (value = 0)
for(size_t i=0; i<m_decodeGraphs.size(); i++) {
m_decodeGraphBackoff.push_back( 0 );
}
// if specified, record maxmimum unseen n-gram size
const vector<string> &backoffVector = m_parameter->GetParam("decoding-graph-backoff");
for(size_t i=0; i<m_decodeGraphs.size() && i<backoffVector.size(); i++) {
m_decodeGraphBackoff[i] = Scan<size_t>(backoffVector[i]);
}
return true;
}
void StaticData::SetWeightsForScoreProducer(const ScoreProducer* sp, const std::vector<float>& weights)
{
const size_t id = sp->GetScoreBookkeepingID();
const size_t begin = m_scoreIndexManager.GetBeginIndex(id);
const size_t end = m_scoreIndexManager.GetEndIndex(id);
assert(end - begin == weights.size());
if (m_allWeights.size() < end)
m_allWeights.resize(end);
std::vector<float>::const_iterator weightIter = weights.begin();
for (size_t i = begin; i < end; i++)
m_allWeights[i] = *weightIter++;
}
const TranslationOptionList* StaticData::FindTransOptListInCache(const DecodeGraph &decodeGraph, const Phrase &sourcePhrase) const
{
std::pair<size_t, Phrase> key(decodeGraph.GetPosition(), sourcePhrase);
#ifdef WITH_THREADS
boost::mutex::scoped_lock lock(m_transOptCacheMutex);
#endif
std::map<std::pair<size_t, Phrase>, std::pair<TranslationOptionList*,clock_t> >::iterator iter
= m_transOptCache.find(key);
if (iter == m_transOptCache.end())
return NULL;
iter->second.second = clock(); // update last used time
return iter->second.first;
}
void StaticData::ReduceTransOptCache() const
{
if (m_transOptCache.size() <= m_transOptCacheMaxSize) return; // not full
clock_t t = clock();
// find cutoff for last used time
priority_queue< clock_t > lastUsedTimes;
std::map<std::pair<size_t, Phrase>, std::pair<TranslationOptionList*,clock_t> >::iterator iter;
iter = m_transOptCache.begin();
while( iter != m_transOptCache.end() ) {
lastUsedTimes.push( iter->second.second );
iter++;
}
for( size_t i=0; i < lastUsedTimes.size()-m_transOptCacheMaxSize/2; i++ )
lastUsedTimes.pop();
clock_t cutoffLastUsedTime = lastUsedTimes.top();
// remove all old entries
iter = m_transOptCache.begin();
while( iter != m_transOptCache.end() ) {
if (iter->second.second < cutoffLastUsedTime) {
std::map<std::pair<size_t, Phrase>, std::pair<TranslationOptionList*,clock_t> >::iterator iterRemove = iter++;
delete iterRemove->second.first;
m_transOptCache.erase(iterRemove);
} else iter++;
}
VERBOSE(2,"Reduced persistent translation option cache in " << ((clock()-t)/(float)CLOCKS_PER_SEC) << " seconds." << std::endl);
}
void StaticData::AddTransOptListToCache(const DecodeGraph &decodeGraph, const Phrase &sourcePhrase, const TranslationOptionList &transOptList) const
{
if (m_transOptCacheMaxSize == 0) return;
std::pair<size_t, Phrase> key(decodeGraph.GetPosition(), sourcePhrase);
TranslationOptionList* storedTransOptList = new TranslationOptionList(transOptList);
#ifdef WITH_THREADS
boost::mutex::scoped_lock lock(m_transOptCacheMutex);
#endif
m_transOptCache[key] = make_pair( storedTransOptList, clock() );
ReduceTransOptCache();
}
}
|