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
|
// $Id$
/***********************************************************************
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 <cassert>
#include <limits>
#include <iostream>
#include <fstream>
#include "dictionary.h"
#include "n_gram.h"
#include "lmtable.h"
#include "lmmacro.h"
#include "LanguageModelIRST.h"
#include "TypeDef.h"
#include "Util.h"
#include "FactorCollection.h"
#include "Phrase.h"
#include "InputFileStream.h"
#include "StaticData.h"
using namespace std;
LanguageModelIRST::LanguageModelIRST(bool registerScore, ScoreIndexManager &scoreIndexManager)
:LanguageModelSingleFactor(registerScore, scoreIndexManager)
,m_lmtb(0)
{
}
LanguageModelIRST::~LanguageModelIRST()
{
delete m_lmtb;
delete m_lmtb_ng;
}
bool LanguageModelIRST::Load(const std::string &filePath,
FactorType factorType,
float weight,
size_t nGramOrder)
{
char *SepString = " \t\n";
cerr << "In LanguageModelIRST::Load: nGramOrder = " << nGramOrder << "\n";
FactorCollection &factorCollection = FactorCollection::Instance();
m_factorType = factorType;
m_weight = weight;
m_nGramOrder = nGramOrder;
// get name of LM file and, if any, of the micro-macro map file
char *filenames = strdup(filePath.c_str());
m_filePath = strsep(&filenames, SepString);
// Open the input file (possibly gzipped)
InputFileStream inp(m_filePath);
if (filenames) {
// case LMfile + MAPfile: create an object of lmmacro class and load both LM file and map
cerr << "Loading LM file + MAP\n";
m_mapFilePath = strsep(&filenames, SepString);
if (!FileExists(m_mapFilePath)) {
cerr << "ERROR: Map file <" << m_mapFilePath << "> does not exist\n";
return false;
}
InputFileStream inpMap(m_mapFilePath);
m_lmtb = new lmmacro(m_filePath, inp, inpMap);
} else {
// case (standard) LMfile only: create an object of lmtable
cerr << "Loading LM file (no MAP)\n";
m_lmtb = (lmtable *)new lmtable;
// Load the (possibly binary) model
#ifdef WIN32
m_lmtb->load(inp); //don't use memory map
#else
if (m_filePath.compare(m_filePath.size()-3,3,".mm")==0)
m_lmtb->load(inp,m_filePath.c_str(),NULL,1);
else
m_lmtb->load(inp,m_filePath.c_str(),NULL,0);
#endif
}
m_lmtb_ng=new ngram(m_lmtb->getDict()); // ngram of words/micro tags
m_lmtb_size=m_lmtb->maxlevel();
// LM can be ok, just outputs warnings
// Mauro: in the original, the following two instructions are wrongly switched:
m_unknownId = m_lmtb->getDict()->oovcode(); // at the level of micro tags
CreateFactors(factorCollection);
VERBOSE(1, "IRST: m_unknownId=" << m_unknownId << std::endl);
//install caches
m_lmtb->init_probcache();
m_lmtb->init_statecache();
m_lmtb->init_lmtcaches(m_lmtb->maxlevel()>2?m_lmtb->maxlevel()-1:2);
return true;
}
void LanguageModelIRST::CreateFactors(FactorCollection &factorCollection)
{ // add factors which have srilm id
// code copied & paste from SRI LM class. should do template function
std::map<size_t, int> lmIdMap;
size_t maxFactorId = 0; // to create lookup vector later on
dict_entry *entry;
dictionary_iter iter(m_lmtb->getDict()); // at the level of micro tags
while ( (entry = iter.next()) != NULL)
{
size_t factorId = factorCollection.AddFactor(Output, m_factorType, entry->word)->GetId();
lmIdMap[factorId] = entry->code;
maxFactorId = (factorId > maxFactorId) ? factorId : maxFactorId;
}
size_t factorId;
m_sentenceStart = factorCollection.AddFactor(Output, m_factorType, BOS_);
factorId = m_sentenceStart->GetId();
m_lmtb_sentenceStart=lmIdMap[factorId] = GetLmID(BOS_);
maxFactorId = (factorId > maxFactorId) ? factorId : maxFactorId;
m_sentenceStartArray[m_factorType] = m_sentenceStart;
m_sentenceEnd = factorCollection.AddFactor(Output, m_factorType, EOS_);
factorId = m_sentenceEnd->GetId();
m_lmtb_sentenceEnd=lmIdMap[factorId] = GetLmID(EOS_);
maxFactorId = (factorId > maxFactorId) ? factorId : maxFactorId;
m_sentenceEndArray[m_factorType] = m_sentenceEnd;
// add to lookup vector in object
m_lmIdLookup.resize(maxFactorId+1);
fill(m_lmIdLookup.begin(), m_lmIdLookup.end(), m_unknownId);
map<size_t, int>::iterator iterMap;
for (iterMap = lmIdMap.begin() ; iterMap != lmIdMap.end() ; ++iterMap)
{
m_lmIdLookup[iterMap->first] = iterMap->second;
}
}
int LanguageModelIRST::GetLmID( const std::string &str ) const
{
return m_lmtb->getDict()->encode( str.c_str() ); // at the level of micro tags
}
float LanguageModelIRST::GetValue(const vector<const Word*> &contextFactor, State* finalState, unsigned int* len) const
{
unsigned int dummy;
if (!len) { len = &dummy; }
FactorType factorType = GetFactorType();
// set up context
size_t count = contextFactor.size();
m_lmtb_ng->size=0;
if (count< (size_t)(m_lmtb_size-1)) m_lmtb_ng->pushc(m_lmtb_sentenceEnd);
if (count< (size_t)m_lmtb_size) m_lmtb_ng->pushc(m_lmtb_sentenceStart);
for (size_t i = 0 ; i < count ; i++)
{
//int lmId = GetLmID((*contextFactor[i])[factorType]);
#ifdef DEBUG
cout << "i=" << i << " -> " << (*contextFactor[i])[factorType]->GetString() << "\n";
#endif
int lmId = GetLmID((*contextFactor[i])[factorType]->GetString());
m_lmtb_ng->pushc(lmId);
}
if (finalState){
*finalState=(State *)m_lmtb->cmaxsuffptr(*m_lmtb_ng);
// back off stats not currently available
*len = 0;
}
return TransformIRSTScore((float) m_lmtb->clprob(*m_lmtb_ng));
}
void LanguageModelIRST::CleanUpAfterSentenceProcessing(){
TRACE_ERR( "reset caches\n");
m_lmtb->reset_caches();
#ifndef WIN32
TRACE_ERR( "reset mmap\n");
m_lmtb->reset_mmap();
#endif
}
void LanguageModelIRST::InitializeBeforeSentenceProcessing(){
//nothing to do
#ifdef TRACE_CACHE
m_lmtb->sentence_id++;
#endif
}
|
