bool RuleTableLoaderStandard::Load(FormatType format
, const std::vector<FactorType> &input
, const std::vector<FactorType> &output
, const std::string &inFile
, const std::vector<float> &weight
, size_t /* tableLimit */
, const LMList &languageModels
, const WordPenaltyProducer* wpProducer
, RuleTableTrie &ruleTable)
{
PrintUserTime(string("Start loading text SCFG phrase table. ") + (format==MosesFormat?"Moses ":"Hiero ") + " format");
const StaticData &staticData = StaticData::Instance();
const std::string& factorDelimiter = staticData.GetFactorDelimiter();
string lineOrig;
size_t count = 0;
std::ostream *progress = NULL;
IFVERBOSE(1) progress = &std::cerr;
util::FilePiece in(inFile.c_str(), progress);
// reused variables
vector<float> scoreVector;
StringPiece line;
std::string hiero_before, hiero_after;
while(true) {
try {
line = in.ReadLine();
} catch (const util::EndOfFileException &e) { break; }
if (format == HieroFormat) { // inefficiently reformat line
hiero_before.assign(line.data(), line.size());
ReformatHieroRule(hiero_before, hiero_after);
line = hiero_after;
}
util::TokenIter<util::MultiCharacter> pipes(line, "|||");
StringPiece sourcePhraseString(*pipes);
StringPiece targetPhraseString(*++pipes);
StringPiece scoreString(*++pipes);
StringPiece alignString(*++pipes);
// TODO(bhaddow) efficiently handle default instead of parsing this string every time.
StringPiece ruleCountString = ++pipes ? *pipes : StringPiece("1 1");
if (++pipes) {
stringstream strme;
strme << "Syntax error at " << ruleTable.GetFilePath() << ":" << count;
UserMessage::Add(strme.str());
abort();
}
bool isLHSEmpty = (sourcePhraseString.find_first_not_of(" \t", 0) == string::npos);
if (isLHSEmpty && !staticData.IsWordDeletionEnabled()) {
TRACE_ERR( ruleTable.GetFilePath() << ":" << count << ": pt entry contains empty target, skipping\n");
continue;
}
scoreVector.clear();
for (util::TokenIter<util::AnyCharacter, true> s(scoreString, " \t"); s; ++s) {
char *err_ind;
scoreVector.push_back(strtod(s->data(), &err_ind));
UTIL_THROW_IF(err_ind == s->data(), util::Exception, "Bad score " << *s << " on line " << count);
}
const size_t numScoreComponents = ruleTable.GetFeature()->GetNumScoreComponents();
if (scoreVector.size() != numScoreComponents) {
stringstream strme;
strme << "Size of scoreVector != number (" << scoreVector.size() << "!="
<< numScoreComponents << ") of score components on line " << count;
UserMessage::Add(strme.str());
abort();
}
// parse source & find pt node
// constituent labels
Word sourceLHS, targetLHS;
// source
Phrase sourcePhrase( 0);
sourcePhrase.CreateFromStringNewFormat(Input, input, sourcePhraseString, factorDelimiter, sourceLHS);
// create target phrase obj
TargetPhrase *targetPhrase = new TargetPhrase(Output);
targetPhrase->CreateFromStringNewFormat(Output, output, targetPhraseString, factorDelimiter, targetLHS);
targetPhrase->SetSourcePhrase(sourcePhrase);
// rest of target phrase
targetPhrase->SetAlignmentInfo(alignString, sourcePhrase);
targetPhrase->SetTargetLHS(targetLHS);
targetPhrase->SetRuleCount(ruleCountString, scoreVector[0]);
//targetPhrase->SetDebugOutput(string("New Format pt ") + line);
// component score, for n-best output
std::transform(scoreVector.begin(),scoreVector.end(),scoreVector.begin(),TransformScore);
std::transform(scoreVector.begin(),scoreVector.end(),scoreVector.begin(),FloorScore);
targetPhrase->SetScoreChart(ruleTable.GetFeature(), scoreVector, weight, languageModels,wpProducer);
TargetPhraseCollection &phraseColl = GetOrCreateTargetPhraseCollection(ruleTable, sourcePhrase, *targetPhrase, sourceLHS);
phraseColl.Add(targetPhrase);
count++;
}
// sort and prune each target phrase collection
SortAndPrune(ruleTable);
return true;
}
bool RuleTableLoaderStandard::Load(FormatType format
, const std::vector<FactorType> &input
, const std::vector<FactorType> &output
, const std::string &inFile
, size_t /* tableLimit */
, RuleTableTrie &ruleTable)
{
PrintUserTime(string("Start loading text SCFG phrase table. ") + (format==MosesFormat?"Moses ":"Hiero ") + " format");
const StaticData &staticData = StaticData::Instance();
const std::string& factorDelimiter = staticData.GetFactorDelimiter();
string lineOrig;
size_t count = 0;
std::ostream *progress = NULL;
IFVERBOSE(1) progress = &std::cerr;
util::FilePiece in(inFile.c_str(), progress);
// reused variables
vector<float> scoreVector;
StringPiece line;
std::string hiero_before, hiero_after;
double_conversion::StringToDoubleConverter converter(double_conversion::StringToDoubleConverter::NO_FLAGS, NAN, NAN, "inf", "nan");
while(true) {
try {
line = in.ReadLine();
} catch (const util::EndOfFileException &e) {
break;
}
if (format == HieroFormat) { // inefficiently reformat line
hiero_before.assign(line.data(), line.size());
ReformatHieroRule(hiero_before, hiero_after);
line = hiero_after;
}
util::TokenIter<util::MultiCharacter> pipes(line, "|||");
StringPiece sourcePhraseString(*pipes);
StringPiece targetPhraseString(*++pipes);
StringPiece scoreString(*++pipes);
StringPiece alignString;
if (++pipes) {
StringPiece temp(*pipes);
alignString = temp;
}
if (++pipes) {
StringPiece str(*pipes); //counts
}
bool isLHSEmpty = (sourcePhraseString.find_first_not_of(" \t", 0) == string::npos);
if (isLHSEmpty && !staticData.IsWordDeletionEnabled()) {
TRACE_ERR( ruleTable.GetFilePath() << ":" << count << ": pt entry contains empty target, skipping\n");
continue;
}
scoreVector.clear();
for (util::TokenIter<util::AnyCharacter, true> s(scoreString, " \t"); s; ++s) {
int processed;
float score = converter.StringToFloat(s->data(), s->length(), &processed);
UTIL_THROW_IF(isnan(score), util::Exception, "Bad score " << *s << " on line " << count);
scoreVector.push_back(FloorScore(TransformScore(score)));
}
const size_t numScoreComponents = ruleTable.GetNumScoreComponents();
if (scoreVector.size() != numScoreComponents) {
stringstream strme;
strme << "Size of scoreVector != number (" << scoreVector.size() << "!="
<< numScoreComponents << ") of score components on line " << count;
UserMessage::Add(strme.str());
abort();
}
// parse source & find pt node
// constituent labels
Word *sourceLHS;
Word *targetLHS;
// create target phrase obj
TargetPhrase *targetPhrase = new TargetPhrase();
targetPhrase->CreateFromString(Output, output, targetPhraseString, factorDelimiter, &targetLHS);
// source
Phrase sourcePhrase;
sourcePhrase.CreateFromString(Input, input, sourcePhraseString, factorDelimiter, &sourceLHS);
// rest of target phrase
targetPhrase->SetAlignmentInfo(alignString);
targetPhrase->SetTargetLHS(targetLHS);
//targetPhrase->SetDebugOutput(string("New Format pt ") + line);
if (++pipes) {
StringPiece sparseString(*pipes);
targetPhrase->SetSparseScore(&ruleTable, sparseString);
}
if (++pipes) {
StringPiece propertiesString(*pipes);
targetPhrase->SetProperties(propertiesString);
}
targetPhrase->GetScoreBreakdown().Assign(&ruleTable, scoreVector);
targetPhrase->Evaluate(sourcePhrase, ruleTable.GetFeaturesToApply());
TargetPhraseCollection &phraseColl = GetOrCreateTargetPhraseCollection(ruleTable, sourcePhrase, *targetPhrase, sourceLHS);
phraseColl.Add(targetPhrase);
count++;
}
// sort and prune each target phrase collection
SortAndPrune(ruleTable);
return true;
}
bool RuleTableLoaderCompact::LoadRuleSection(
LineReader &reader,
const std::vector<Word> &vocab,
const std::vector<Phrase> &sourcePhrases,
const std::vector<Phrase> &targetPhrases,
const std::vector<size_t> &targetLhsIds,
const std::vector<const AlignmentInfo *> &alignmentSets,
RuleTableTrie &ruleTable)
{
// Read rule count.
reader.ReadLine();
const size_t ruleCount = std::atoi(reader.m_line.c_str());
// Read rules and add to table.
const size_t numScoreComponents = ruleTable.GetNumScoreComponents();
std::vector<float> scoreVector(numScoreComponents);
std::vector<size_t> tokenPositions;
for (size_t i = 0; i < ruleCount; ++i) {
reader.ReadLine();
tokenPositions.clear();
FindTokens(tokenPositions, reader.m_line);
const char *charLine = reader.m_line.c_str();
// The first three tokens are IDs for the source phrase, target phrase,
// and alignment set.
const int sourcePhraseId = std::atoi(charLine+tokenPositions[0]);
const int targetPhraseId = std::atoi(charLine+tokenPositions[1]);
const int alignmentSetId = std::atoi(charLine+tokenPositions[2]);
const Phrase &sourcePhrase = sourcePhrases[sourcePhraseId];
const Phrase &targetPhrasePhrase = targetPhrases[targetPhraseId];
const Word *targetLhs = new Word(vocab[targetLhsIds[targetPhraseId]]);
Word sourceLHS("X"); // TODO not implemented for compact
const AlignmentInfo *alignNonTerm = alignmentSets[alignmentSetId];
// Then there should be one score for each score component.
for (size_t j = 0; j < numScoreComponents; ++j) {
float score = std::atof(charLine+tokenPositions[3+j]);
scoreVector[j] = FloorScore(TransformScore(score));
}
if (reader.m_line[tokenPositions[3+numScoreComponents]] != ':') {
std::stringstream msg;
msg << "Size of scoreVector != number ("
<< scoreVector.size() << "!=" << numScoreComponents
<< ") of score components on line " << reader.m_lineNum;
UserMessage::Add(msg.str());
return false;
}
// The remaining columns are currently ignored.
// Create and score target phrase.
TargetPhrase *targetPhrase = new TargetPhrase(targetPhrasePhrase);
targetPhrase->SetAlignNonTerm(alignNonTerm);
targetPhrase->SetTargetLHS(targetLhs);
targetPhrase->SetSourcePhrase(sourcePhrase);
targetPhrase->Evaluate(sourcePhrase, ruleTable.GetFeaturesToApply());
// Insert rule into table.
TargetPhraseCollection &coll = GetOrCreateTargetPhraseCollection(
ruleTable, sourcePhrase, *targetPhrase, &sourceLHS);
coll.Add(targetPhrase);
}
return true;
}
PhraseDictionary* PhraseDictionaryFeature::LoadPhraseTable(const TranslationSystem* system)
{
const StaticData& staticData = StaticData::Instance();
if (m_implementation == Memory) {
// memory phrase table
VERBOSE(2,"using standard phrase tables" << std::endl);
if (!FileExists(m_filePath) && FileExists(m_filePath + ".gz")) {
m_filePath += ".gz";
VERBOSE(2,"Using gzipped file" << std::endl);
}
if (staticData.GetInputType() != SentenceInput) {
UserMessage::Add("Must use binary phrase table for this input type");
CHECK(false);
}
PhraseDictionaryMemory* pdm = new PhraseDictionaryMemory(m_numScoreComponent,this);
bool ret = pdm->Load(GetInput(), GetOutput()
, m_filePath
, m_weight
, m_tableLimit
, system->GetLanguageModels()
, system->GetWeightWordPenalty());
CHECK(ret);
return pdm;
} else if (m_implementation == Binary) {
PhraseDictionaryTreeAdaptor* pdta = new PhraseDictionaryTreeAdaptor(m_numScoreComponent, m_numInputScores,this);
bool ret = pdta->Load( GetInput()
, GetOutput()
, m_filePath
, m_weight
, m_tableLimit
, system->GetLanguageModels()
, system->GetWeightWordPenalty());
CHECK(ret);
return pdta;
} else if (m_implementation == SCFG || m_implementation == Hiero) {
// memory phrase table
if (m_implementation == Hiero) {
VERBOSE(2,"using Hiero format phrase tables" << std::endl);
} else {
VERBOSE(2,"using Moses-formatted SCFG phrase tables" << std::endl);
}
if (!FileExists(m_filePath) && FileExists(m_filePath + ".gz")) {
m_filePath += ".gz";
VERBOSE(2,"Using gzipped file" << std::endl);
}
RuleTableTrie *dict;
if (staticData.GetParsingAlgorithm() == ParseScope3) {
dict = new RuleTableUTrie(m_numScoreComponent, this);
} else {
dict = new PhraseDictionarySCFG(m_numScoreComponent, this);
}
bool ret = dict->Load(GetInput()
, GetOutput()
, m_filePath
, m_weight
, m_tableLimit
, system->GetLanguageModels()
, system->GetWordPenaltyProducer());
assert(ret);
return dict;
} else if (m_implementation == ALSuffixArray) {
// memory phrase table
VERBOSE(2,"using Hiero format phrase tables" << std::endl);
if (!FileExists(m_filePath) && FileExists(m_filePath + ".gz")) {
m_filePath += ".gz";
VERBOSE(2,"Using gzipped file" << std::endl);
}
PhraseDictionaryALSuffixArray* pdm = new PhraseDictionaryALSuffixArray(m_numScoreComponent,this);
bool ret = pdm->Load(GetInput()
, GetOutput()
, m_filePath
, m_weight
, m_tableLimit
, system->GetLanguageModels()
, system->GetWordPenaltyProducer());
CHECK(ret);
return pdm;
} else if (m_implementation == OnDisk) {
PhraseDictionaryOnDisk* pdta = new PhraseDictionaryOnDisk(m_numScoreComponent, this);
bool ret = pdta->Load(GetInput()
, GetOutput()
, m_filePath
, m_weight
, m_tableLimit
, system->GetLanguageModels()
, system->GetWordPenaltyProducer());
CHECK(ret);
return pdta;
} else if (m_implementation == SuffixArray) {
#ifndef WIN32
PhraseDictionaryDynSuffixArray *pd = new PhraseDictionaryDynSuffixArray(m_numScoreComponent, this);
if(!(pd->Load(
GetInput()
,GetOutput()
,m_filePath
,m_targetFile
, m_alignmentsFile
, m_weight, m_tableLimit
, system->GetLanguageModels()
, system->GetWeightWordPenalty()))) {
std::cerr << "FAILED TO LOAD\n" << endl;
delete pd;
pd = NULL;
}
std::cerr << "Suffix array phrase table loaded" << std::endl;
return pd;
#else
CHECK(false);
#endif
} else if (m_implementation == FuzzyMatch) {
PhraseDictionaryFuzzyMatch *dict = new PhraseDictionaryFuzzyMatch(m_numScoreComponent, this);
bool ret = dict->Load(GetInput()
, GetOutput()
, m_filePath
, m_weight
, m_tableLimit
, system->GetLanguageModels()
, system->GetWordPenaltyProducer());
assert(ret);
return dict;
} else if (m_implementation == Compact) {
#ifndef WIN32
VERBOSE(2,"Using compact phrase table" << std::endl);
PhraseDictionaryCompact* pd = new PhraseDictionaryCompact(m_numScoreComponent, m_implementation, this);
bool ret = pd->Load(GetInput(), GetOutput()
, m_filePath
, m_weight
, m_tableLimit
, system->GetLanguageModels()
, system->GetWeightWordPenalty());
assert(ret);
return pd;
#else
CHECK(false);
#endif
}
else {
std::cerr << "Unknown phrase table type " << m_implementation << endl;
CHECK(false);
}
}
bool RuleTableLoaderStandard::Load(FormatType format
, const std::vector<FactorType> &input
, const std::vector<FactorType> &output
, std::istream &inStream
, const std::vector<float> &weight
, size_t /* tableLimit */
, const LMList &languageModels
, const WordPenaltyProducer* wpProducer
, RuleTableTrie &ruleTable)
{
PrintUserTime(string("Start loading text SCFG phrase table. ") + (format==MosesFormat?"Moses ":"Hiero ") + " format");
const StaticData &staticData = StaticData::Instance();
const std::string& factorDelimiter = staticData.GetFactorDelimiter();
string lineOrig;
size_t count = 0;
while(getline(inStream, lineOrig)) {
const string *line;
if (format == HieroFormat) { // reformat line
line = ReformatHieroRule(lineOrig);
}
else
{ // do nothing to format of line
line = &lineOrig;
}
vector<string> tokens;
vector<float> scoreVector;
TokenizeMultiCharSeparator(tokens, *line , "|||" );
if (tokens.size() != 4 && tokens.size() != 5) {
stringstream strme;
strme << "Syntax error at " << ruleTable.GetFilePath() << ":" << count;
UserMessage::Add(strme.str());
abort();
}
const string &sourcePhraseString = tokens[0]
, &targetPhraseString = tokens[1]
, &scoreString = tokens[2]
, &alignString = tokens[3];
bool isLHSEmpty = (sourcePhraseString.find_first_not_of(" \t", 0) == string::npos);
if (isLHSEmpty && !staticData.IsWordDeletionEnabled()) {
TRACE_ERR( ruleTable.GetFilePath() << ":" << count << ": pt entry contains empty target, skipping\n");
continue;
}
Tokenize<float>(scoreVector, scoreString);
const size_t numScoreComponents = ruleTable.GetFeature()->GetNumScoreComponents();
if (scoreVector.size() != numScoreComponents) {
stringstream strme;
strme << "Size of scoreVector != number (" << scoreVector.size() << "!="
<< numScoreComponents << ") of score components on line " << count;
UserMessage::Add(strme.str());
abort();
}
CHECK(scoreVector.size() == numScoreComponents);
// parse source & find pt node
// constituent labels
Word sourceLHS, targetLHS;
// source
Phrase sourcePhrase( 0);
sourcePhrase.CreateFromStringNewFormat(Input, input, sourcePhraseString, factorDelimiter, sourceLHS);
// create target phrase obj
TargetPhrase *targetPhrase = new TargetPhrase(Output);
targetPhrase->CreateFromStringNewFormat(Output, output, targetPhraseString, factorDelimiter, targetLHS);
// rest of target phrase
targetPhrase->SetAlignmentInfo(alignString);
targetPhrase->SetTargetLHS(targetLHS);
//targetPhrase->SetDebugOutput(string("New Format pt ") + line);
// component score, for n-best output
std::transform(scoreVector.begin(),scoreVector.end(),scoreVector.begin(),TransformScore);
std::transform(scoreVector.begin(),scoreVector.end(),scoreVector.begin(),FloorScore);
targetPhrase->SetScoreChart(ruleTable.GetFeature(), scoreVector, weight, languageModels,wpProducer);
TargetPhraseCollection &phraseColl = GetOrCreateTargetPhraseCollection(ruleTable, sourcePhrase, *targetPhrase, sourceLHS);
phraseColl.Add(targetPhrase);
count++;
if (format == HieroFormat) { // reformat line
delete line;
}
else
{ // do nothing
}
}
// sort and prune each target phrase collection
SortAndPrune(ruleTable);
return true;
}
