void OutputNBest(std::ostream& out, const Moses::TrellisPathList &nBestList, const std::vector<Moses::FactorType>& outputFactorOrder, const TranslationSystem* system, long translationId)
{
const StaticData &staticData = StaticData::Instance();
bool labeledOutput = staticData.IsLabeledNBestList();
bool reportAllFactors = staticData.GetReportAllFactorsNBest();
bool includeAlignment = staticData.NBestIncludesAlignment();
bool includeWordAlignment = staticData.PrintAlignmentInfoInNbest();
TrellisPathList::const_iterator iter;
for (iter = nBestList.begin() ; iter != nBestList.end() ; ++iter) {
const TrellisPath &path = **iter;
const std::vector<const Hypothesis *> &edges = path.GetEdges();
// print the surface factor of the translation
out << translationId << " ||| ";
for (int currEdge = (int)edges.size() - 1 ; currEdge >= 0 ; currEdge--) {
const Hypothesis &edge = *edges[currEdge];
OutputSurface(out, edge.GetCurrTargetPhrase(), outputFactorOrder, reportAllFactors);
}
out << " |||";
std::string lastName = "";
const vector<const StatefulFeatureFunction*>& sff = system->GetStatefulFeatureFunctions();
for( size_t i=0; i<sff.size(); i++ ) {
if( labeledOutput && lastName != sff[i]->GetScoreProducerWeightShortName() ) {
lastName = sff[i]->GetScoreProducerWeightShortName();
out << " " << lastName << ":";
}
vector<float> scores = path.GetScoreBreakdown().GetScoresForProducer( sff[i] );
for (size_t j = 0; j<scores.size(); ++j) {
out << " " << scores[j];
}
}
const vector<const StatelessFeatureFunction*>& slf = system->GetStatelessFeatureFunctions();
for( size_t i=0; i<slf.size(); i++ ) {
if( labeledOutput && lastName != slf[i]->GetScoreProducerWeightShortName() ) {
lastName = slf[i]->GetScoreProducerWeightShortName();
out << " " << lastName << ":";
}
vector<float> scores = path.GetScoreBreakdown().GetScoresForProducer( slf[i] );
for (size_t j = 0; j<scores.size(); ++j) {
out << " " << scores[j];
}
}
// translation components
const vector<PhraseDictionaryFeature*>& pds = system->GetPhraseDictionaries();
if (pds.size() > 0) {
for( size_t i=0; i<pds.size(); i++ ) {
size_t pd_numinputscore = pds[i]->GetNumInputScores();
vector<float> scores = path.GetScoreBreakdown().GetScoresForProducer( pds[i] );
for (size_t j = 0; j<scores.size(); ++j){
if (labeledOutput && (i == 0) ){
if ((j == 0) || (j == pd_numinputscore)){
lastName = pds[i]->GetScoreProducerWeightShortName(j);
out << " " << lastName << ":";
}
}
out << " " << scores[j];
}
}
}
// generation
const vector<GenerationDictionary*>& gds = system->GetGenerationDictionaries();
if (gds.size() > 0) {
for( size_t i=0; i<gds.size(); i++ ) {
size_t pd_numinputscore = gds[i]->GetNumInputScores();
vector<float> scores = path.GetScoreBreakdown().GetScoresForProducer( gds[i] );
for (size_t j = 0; j<scores.size(); ++j){
if (labeledOutput && (i == 0) ){
if ((j == 0) || (j == pd_numinputscore)){
lastName = gds[i]->GetScoreProducerWeightShortName(j);
out << " " << lastName << ":";
}
}
out << " " << scores[j];
}
}
}
// total
out << " ||| " << path.GetTotalScore();
//phrase-to-phrase alignment
if (includeAlignment) {
out << " |||";
for (int currEdge = (int)edges.size() - 2 ; currEdge >= 0 ; currEdge--) {
const Hypothesis &edge = *edges[currEdge];
const WordsRange &sourceRange = edge.GetCurrSourceWordsRange();
WordsRange targetRange = path.GetTargetWordsRange(edge);
out << " " << sourceRange.GetStartPos();
if (sourceRange.GetStartPos() < sourceRange.GetEndPos()) {
out << "-" << sourceRange.GetEndPos();
}
out<< "=" << targetRange.GetStartPos();
if (targetRange.GetStartPos() < targetRange.GetEndPos()) {
out<< "-" << targetRange.GetEndPos();
}
}
}
if (includeWordAlignment) {
out << " ||| ";
for (int currEdge = (int)edges.size() - 2 ; currEdge >= 0 ; currEdge--) {
const Hypothesis &edge = *edges[currEdge];
const WordsRange &sourceRange = edge.GetCurrSourceWordsRange();
WordsRange targetRange = path.GetTargetWordsRange(edge);
const int sourceOffset = sourceRange.GetStartPos();
const int targetOffset = targetRange.GetStartPos();
const AlignmentInfo &ai = edge.GetCurrTargetPhrase().GetAlignmentInfo();
OutputAlignment(out, ai, sourceOffset, targetOffset);
}
}
if (StaticData::Instance().IsPathRecoveryEnabled()) {
out << "|||";
OutputInput(out, edges[0]);
}
out << endl;
}
out <<std::flush;
}
void OutputNBest(std::ostream& out
, const Moses::TrellisPathList &nBestList
, const std::vector<Moses::FactorType>& outputFactorOrder
, long translationId
, char reportSegmentation)
{
const StaticData &staticData = StaticData::Instance();
bool reportAllFactors = staticData.GetReportAllFactorsNBest();
bool includeSegmentation = staticData.NBestIncludesSegmentation();
bool includeWordAlignment = staticData.PrintAlignmentInfoInNbest();
TrellisPathList::const_iterator iter;
for (iter = nBestList.begin() ; iter != nBestList.end() ; ++iter) {
const TrellisPath &path = **iter;
const std::vector<const Hypothesis *> &edges = path.GetEdges();
// print the surface factor of the translation
out << translationId << " ||| ";
for (int currEdge = (int)edges.size() - 1 ; currEdge >= 0 ; currEdge--) {
const Hypothesis &edge = *edges[currEdge];
OutputSurface(out, edge, outputFactorOrder, reportSegmentation, reportAllFactors);
}
out << " |||";
// print scores with feature names
OutputAllFeatureScores(path.GetScoreBreakdown(), out );
// total
out << " ||| " << path.GetTotalScore();
//phrase-to-phrase segmentation
if (includeSegmentation) {
out << " |||";
for (int currEdge = (int)edges.size() - 2 ; currEdge >= 0 ; currEdge--) {
const Hypothesis &edge = *edges[currEdge];
const WordsRange &sourceRange = edge.GetCurrSourceWordsRange();
WordsRange targetRange = path.GetTargetWordsRange(edge);
out << " " << sourceRange.GetStartPos();
if (sourceRange.GetStartPos() < sourceRange.GetEndPos()) {
out << "-" << sourceRange.GetEndPos();
}
out<< "=" << targetRange.GetStartPos();
if (targetRange.GetStartPos() < targetRange.GetEndPos()) {
out<< "-" << targetRange.GetEndPos();
}
}
}
if (includeWordAlignment) {
out << " ||| ";
for (int currEdge = (int)edges.size() - 2 ; currEdge >= 0 ; currEdge--) {
const Hypothesis &edge = *edges[currEdge];
const WordsRange &sourceRange = edge.GetCurrSourceWordsRange();
WordsRange targetRange = path.GetTargetWordsRange(edge);
const int sourceOffset = sourceRange.GetStartPos();
const int targetOffset = targetRange.GetStartPos();
const AlignmentInfo &ai = edge.GetCurrTargetPhrase().GetAlignTerm();
OutputAlignment(out, ai, sourceOffset, targetOffset);
}
}
if (StaticData::Instance().IsPathRecoveryEnabled()) {
out << " ||| ";
OutputInput(out, edges[0]);
}
out << endl;
}
out << std::flush;
}
void OutputNBest(std::ostream& out, const Moses::TrellisPathList &nBestList, const std::vector<Moses::FactorType>& outputFactorOrder,long translationId)
{
const StaticData &staticData = StaticData::Instance();
bool labeledOutput = staticData.IsLabeledNBestList();
bool reportAllFactors = staticData.GetReportAllFactorsNBest();
bool includeAlignment = staticData.NBestIncludesAlignment();
//bool includeWordAlignment = staticData.PrintAlignmentInfoInNbest();
TrellisPathList::const_iterator iter;
for (iter = nBestList.begin() ; iter != nBestList.end() ; ++iter)
{
const TrellisPath &path = **iter;
const std::vector<const Hypothesis *> &edges = path.GetEdges();
// print the surface factor of the translation
out << translationId << " ||| ";
for (int currEdge = (int)edges.size() - 1 ; currEdge >= 0 ; currEdge--)
{
const Hypothesis &edge = *edges[currEdge];
OutputSurface(out, edge.GetCurrTargetPhrase(), outputFactorOrder, reportAllFactors);
}
out << " |||";
std::string lastName = "";
const vector<const StatefulFeatureFunction*>& sff =
staticData.GetScoreIndexManager().GetStatefulFeatureFunctions();
for( size_t i=0; i<sff.size(); i++ )
{
if( labeledOutput && lastName != sff[i]->GetScoreProducerWeightShortName() )
{
lastName = sff[i]->GetScoreProducerWeightShortName();
out << " " << lastName << ":";
}
vector<float> scores = path.GetScoreBreakdown().GetScoresForProducer( sff[i] );
for (size_t j = 0; j<scores.size(); ++j)
{
out << " " << scores[j];
}
}
const vector<const StatelessFeatureFunction*>& slf =
staticData.GetScoreIndexManager().GetStatelessFeatureFunctions();
for( size_t i=0; i<slf.size(); i++ )
{
if( labeledOutput && lastName != slf[i]->GetScoreProducerWeightShortName() )
{
lastName = slf[i]->GetScoreProducerWeightShortName();
out << " " << lastName << ":";
}
vector<float> scores = path.GetScoreBreakdown().GetScoresForProducer( slf[i] );
for (size_t j = 0; j<scores.size(); ++j)
{
out << " " << scores[j];
}
}
// translation components
if (StaticData::Instance().GetInputType()==SentenceInput){
// translation components for text input
vector<PhraseDictionaryFeature*> pds = StaticData::Instance().GetPhraseDictionaries();
if (pds.size() > 0) {
if (labeledOutput)
out << " tm:";
vector<PhraseDictionaryFeature*>::iterator iter;
for (iter = pds.begin(); iter != pds.end(); ++iter) {
vector<float> scores = path.GetScoreBreakdown().GetScoresForProducer(*iter);
for (size_t j = 0; j<scores.size(); ++j)
out << " " << scores[j];
}
}
}
else{
// translation components for Confusion Network input
// first translation component has GetNumInputScores() scores from the input Confusion Network
// at the beginning of the vector
vector<PhraseDictionaryFeature*> pds = StaticData::Instance().GetPhraseDictionaries();
if (pds.size() > 0) {
vector<PhraseDictionaryFeature*>::iterator iter;
iter = pds.begin();
vector<float> scores = path.GetScoreBreakdown().GetScoresForProducer(*iter);
size_t pd_numinputscore = (*iter)->GetNumInputScores();
if (pd_numinputscore){
if (labeledOutput)
out << " I:";
for (size_t j = 0; j < pd_numinputscore; ++j)
out << " " << scores[j];
}
for (iter = pds.begin() ; iter != pds.end(); ++iter) {
vector<float> scores = path.GetScoreBreakdown().GetScoresForProducer(*iter);
size_t pd_numinputscore = (*iter)->GetNumInputScores();
if (iter == pds.begin() && labeledOutput)
out << " tm:";
for (size_t j = pd_numinputscore; j < scores.size() ; ++j)
out << " " << scores[j];
}
}
}
// generation
vector<GenerationDictionary*> gds = StaticData::Instance().GetGenerationDictionaries();
if (gds.size() > 0) {
if (labeledOutput)
out << " g: ";
vector<GenerationDictionary*>::iterator iter;
for (iter = gds.begin(); iter != gds.end(); ++iter) {
vector<float> scores = path.GetScoreBreakdown().GetScoresForProducer(*iter);
for (size_t j = 0; j<scores.size(); j++) {
out << scores[j] << " ";
}
}
}
// total
out << " ||| " << path.GetTotalScore();
//phrase-to-phrase alignment
if (includeAlignment) {
out << " |||";
for (int currEdge = (int)edges.size() - 2 ; currEdge >= 0 ; currEdge--)
{
const Hypothesis &edge = *edges[currEdge];
const WordsRange &sourceRange = edge.GetCurrSourceWordsRange();
WordsRange targetRange = path.GetTargetWordsRange(edge);
out << " " << sourceRange.GetStartPos();
if (sourceRange.GetStartPos() < sourceRange.GetEndPos()) {
out << "-" << sourceRange.GetEndPos();
}
out<< "=" << targetRange.GetStartPos();
if (targetRange.GetStartPos() < targetRange.GetEndPos()) {
out<< "-" << targetRange.GetEndPos();
}
}
}
if (StaticData::Instance().IsPathRecoveryEnabled()) {
out << "|||";
OutputInput(out, edges[0]);
}
out << endl;
}
out <<std::flush;
}