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
|
// vim:tabstop=2
#include "ProbingPT.h"
#include "moses/StaticData.h"
#include "moses/FactorCollection.h"
#include "moses/TargetPhraseCollection.h"
#include "moses/InputFileStream.h"
#include "moses/TranslationModel/CYKPlusParser/ChartRuleLookupManagerSkeleton.h"
#include "quering.hh"
using namespace std;
namespace Moses
{
ProbingPT::ProbingPT(const std::string &line)
: PhraseDictionary(line,true)
,m_engine(NULL)
{
ReadParameters();
assert(m_input.size() == 1);
assert(m_output.size() == 1);
}
ProbingPT::~ProbingPT()
{
delete m_engine;
}
void ProbingPT::Load(AllOptions::ptr const& opts)
{
m_options = opts;
SetFeaturesToApply();
m_engine = new QueryEngine(m_filePath.c_str());
m_unkId = 456456546456;
FactorCollection &vocab = FactorCollection::Instance();
// source vocab
const std::map<uint64_t, std::string> &sourceVocab =
m_engine->getSourceVocab();
std::map<uint64_t, std::string>::const_iterator iterSource;
for (iterSource = sourceVocab.begin(); iterSource != sourceVocab.end();
++iterSource) {
string wordStr = iterSource->second;
//cerr << "wordStr=" << wordStr << endl;
const Factor *factor = vocab.AddFactor(wordStr);
uint64_t probingId = iterSource->first;
size_t factorId = factor->GetId();
if (factorId >= m_sourceVocab.size()) {
m_sourceVocab.resize(factorId + 1, m_unkId);
}
m_sourceVocab[factorId] = probingId;
}
// target vocab
InputFileStream targetVocabStrme(m_filePath + "/TargetVocab.dat");
string line;
while (getline(targetVocabStrme, line)) {
vector<string> toks = Tokenize(line, "\t");
UTIL_THROW_IF2(toks.size() != 2, string("Incorrect format:") + line + "\n");
//cerr << "wordStr=" << toks[0] << endl;
const Factor *factor = vocab.AddFactor(toks[0]);
uint32_t probingId = Scan<uint32_t>(toks[1]);
if (probingId >= m_targetVocab.size()) {
m_targetVocab.resize(probingId + 1);
}
m_targetVocab[probingId] = factor;
}
// alignments
CreateAlignmentMap(m_filePath + "/Alignments.dat");
// memory mapped file to tps
string filePath = m_filePath + "/TargetColl.dat";
file.open(filePath.c_str());
if (!file.is_open()) {
throw "Couldn't open file ";
}
data = file.data();
//size_t size = file.size();
// cache
//CreateCache(system);
}
void ProbingPT::CreateAlignmentMap(const std::string path)
{
const std::vector< std::vector<unsigned char> > &probingAlignColl = m_engine->getAlignments();
m_aligns.resize(probingAlignColl.size(), NULL);
for (size_t i = 0; i < probingAlignColl.size(); ++i) {
AlignmentInfo::CollType aligns;
const std::vector<unsigned char> &probingAligns = probingAlignColl[i];
for (size_t j = 0; j < probingAligns.size(); j += 2) {
size_t startPos = probingAligns[j];
size_t endPos = probingAligns[j+1];
//cerr << "startPos=" << startPos << " " << endPos << endl;
aligns.insert(std::pair<size_t,size_t>(startPos, endPos));
}
const AlignmentInfo *align = AlignmentInfoCollection::Instance().Add(aligns);
m_aligns[i] = align;
//cerr << "align=" << align->Debug(system) << endl;
}
}
void ProbingPT::InitializeForInput(ttasksptr const& ttask)
{
}
void ProbingPT::GetTargetPhraseCollectionBatch(const InputPathList &inputPathQueue) const
{
InputPathList::const_iterator iter;
for (iter = inputPathQueue.begin(); iter != inputPathQueue.end(); ++iter) {
InputPath &inputPath = **iter;
const Phrase &sourcePhrase = inputPath.GetPhrase();
if (sourcePhrase.GetSize() > m_options->search.max_phrase_length) {
continue;
}
TargetPhraseCollection::shared_ptr tpColl = CreateTargetPhrase(sourcePhrase);
inputPath.SetTargetPhrases(*this, tpColl, NULL);
}
}
TargetPhraseCollection::shared_ptr ProbingPT::CreateTargetPhrase(const Phrase &sourcePhrase) const
{
// create a target phrase from the 1st word of the source, prefix with 'ProbingPT:'
assert(sourcePhrase.GetSize());
std::pair<bool, uint64_t> keyStruct = GetKey(sourcePhrase);
if (!keyStruct.first) {
return TargetPhraseCollection::shared_ptr();
}
// check in cache
CachePb::const_iterator iter = m_cachePb.find(keyStruct.second);
if (iter != m_cachePb.end()) {
//cerr << "FOUND IN CACHE " << keyStruct.second << " " << sourcePhrase.Debug(mgr.system) << endl;
TargetPhraseCollection *tps = iter->second;
return TargetPhraseCollection::shared_ptr(tps);
}
// query pt
TargetPhraseCollection *tps = CreateTargetPhrases(sourcePhrase,
keyStruct.second);
return TargetPhraseCollection::shared_ptr(tps);
}
std::pair<bool, uint64_t> ProbingPT::GetKey(const Phrase &sourcePhrase) const
{
std::pair<bool, uint64_t> ret;
// create a target phrase from the 1st word of the source, prefix with 'ProbingPT:'
size_t sourceSize = sourcePhrase.GetSize();
assert(sourceSize);
uint64_t probingSource[sourceSize];
GetSourceProbingIds(sourcePhrase, ret.first, probingSource);
if (!ret.first) {
// source phrase contains a word unknown in the pt.
// We know immediately there's no translation for it
}
else {
ret.second = m_engine->getKey(probingSource, sourceSize);
}
return ret;
}
void ProbingPT::GetSourceProbingIds(const Phrase &sourcePhrase,
bool &ok, uint64_t probingSource[]) const
{
size_t size = sourcePhrase.GetSize();
for (size_t i = 0; i < size; ++i) {
const Word &word = sourcePhrase.GetWord(i);
uint64_t probingId = GetSourceProbingId(word);
if (probingId == m_unkId) {
ok = false;
return;
}
else {
probingSource[i] = probingId;
}
}
ok = true;
}
uint64_t ProbingPT::GetSourceProbingId(const Word &word) const
{
uint64_t ret = 0;
for (size_t i = 0; i < m_input.size(); ++i) {
FactorType factorType = m_input[i];
const Factor *factor = word[factorType];
size_t factorId = factor->GetId();
if (factorId >= m_sourceVocab.size()) {
return m_unkId;
}
ret += m_sourceVocab[factorId];
}
return ret;
}
TargetPhraseCollection *ProbingPT::CreateTargetPhrases(
const Phrase &sourcePhrase, uint64_t key) const
{
TargetPhraseCollection *tps = NULL;
//Actual lookup
std::pair<bool, uint64_t> query_result; // 1st=found, 2nd=target file offset
query_result = m_engine->query(key);
//cerr << "key2=" << query_result.second << endl;
if (query_result.first) {
const char *offset = data + query_result.second;
uint64_t *numTP = (uint64_t*) offset;
tps = new TargetPhraseCollection();
offset += sizeof(uint64_t);
for (size_t i = 0; i < *numTP; ++i) {
TargetPhrase *tp = CreateTargetPhrase(offset);
assert(tp);
tp->EvaluateInIsolation(sourcePhrase, GetFeaturesToApply());
tps->Add(tp);
}
tps->Prune(true, m_tableLimit);
//cerr << *tps << endl;
}
return tps;
}
TargetPhrase *ProbingPT::CreateTargetPhrase(
const char *&offset) const
{
TargetPhraseInfo *tpInfo = (TargetPhraseInfo*) offset;
size_t numRealWords = tpInfo->numWords / m_output.size();
TargetPhrase *tp = new TargetPhrase(this);
offset += sizeof(TargetPhraseInfo);
// scores
float *scores = (float*) offset;
size_t totalNumScores = m_engine->num_scores + m_engine->num_lex_scores;
if (m_engine->logProb) {
// set pt score for rule
tp->GetScoreBreakdown().PlusEquals(this, scores);
// save scores for other FF, eg. lex RO. Just give the offset
/*
if (m_engine->num_lex_scores) {
tp->scoreProperties = scores + m_engine->num_scores;
}
*/
}
else {
// log score 1st
float logScores[totalNumScores];
for (size_t i = 0; i < totalNumScores; ++i) {
logScores[i] = FloorScore(TransformScore(scores[i]));
}
// set pt score for rule
tp->GetScoreBreakdown().PlusEquals(this, logScores);
// save scores for other FF, eg. lex RO.
/*
tp->scoreProperties = pool.Allocate<SCORE>(m_engine->num_lex_scores);
for (size_t i = 0; i < m_engine->num_lex_scores; ++i) {
tp->scoreProperties[i] = logScores[i + m_engine->num_scores];
}
*/
}
offset += sizeof(float) * totalNumScores;
// words
for (size_t targetPos = 0; targetPos < numRealWords; ++targetPos) {
for (size_t i = 0; i < m_output.size(); ++i) {
FactorType factorType = m_output[i];
uint32_t *probingId = (uint32_t*) offset;
const Factor *factor = GetTargetFactor(*probingId);
assert(factor);
Word &word = tp->GetWord(targetPos);
word[factorType] = factor;
offset += sizeof(uint32_t);
}
}
// align
uint32_t alignTerm = tpInfo->alignTerm;
//cerr << "alignTerm=" << alignTerm << endl;
UTIL_THROW_IF2(alignTerm >= m_aligns.size(), "Unknown alignInd");
tp->SetAlignTerm(m_aligns[alignTerm]);
// properties TODO
return tp;
}
//////////////////////////////////////////////////////////////////
ChartRuleLookupManager *ProbingPT::CreateRuleLookupManager(
const ChartParser &,
const ChartCellCollectionBase &,
std::size_t)
{
abort();
return NULL;
}
TO_STRING_BODY(ProbingPT);
// friend
ostream& operator<<(ostream& out, const ProbingPT& phraseDict)
{
return out;
}
}
|