support for binary models with multimodelcounts

This commit is contained in:
Rico Sennrich 2013-05-02 12:14:17 +02:00
parent 0450fd6776
commit adb7de6e61
4 changed files with 33 additions and 17 deletions

View File

@ -292,6 +292,14 @@ protected:
} }
} }
if (scoreVector.size() != m_obj->GetNumScoreComponentMultiModel()) {
//PhraseDictionaryMultiModel may use input phrase dictionaries with a different number of features than it is assigned in the log-linear model;
//filling extra slots with zeroes to prevent error messages on the way
if (m_obj->GetNumScoreComponentMultiModel() > 0 && scoreVector.size() < m_obj->GetNumScoreComponentMultiModel()) {
const_cast<Scores &>(scoreVector).resize(m_obj->GetNumScoreComponentMultiModel());
}
}
targetPhrase.SetScore(m_obj->GetFeature(), scoreVector, sparseFeatures, weights, weightWP, *m_languageModels); targetPhrase.SetScore(m_obj->GetFeature(), scoreVector, sparseFeatures, weights, weightWP, *m_languageModels);
targetPhrase.SetSourcePhrase(*srcPtr); targetPhrase.SetSourcePhrase(*srcPtr);
} }

View File

@ -264,6 +264,7 @@ PhraseDictionary* PhraseDictionaryFeature::LoadPhraseTable(const TranslationSyst
, m_config , m_config
, weightT , weightT
, m_tableLimit , m_tableLimit
, m_numInputScores
, system->GetLanguageModels() , system->GetLanguageModels()
, system->GetWeightWordPenalty()); , system->GetWeightWordPenalty());
CHECK(ret); CHECK(ret);

View File

@ -16,6 +16,7 @@ You should have received a copy of the GNU Lesser General Public
License along with this library; if not, write to the Free Software License along with this library; if not, write to the Free Software
Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
***********************************************************************/ ***********************************************************************/
#include "util/exception.hh"
#include "moses/TranslationModel/PhraseDictionaryMultiModelCounts.h" #include "moses/TranslationModel/PhraseDictionaryMultiModelCounts.h"
@ -73,6 +74,7 @@ bool PhraseDictionaryMultiModelCounts::Load(const vector<FactorType> &input
, const vector<string> &config , const vector<string> &config
, const vector<float> &weight , const vector<float> &weight
, size_t tableLimit , size_t tableLimit
, size_t numInputScores
, const LMList &languageModels , const LMList &languageModels
, float weightWP) , float weightWP)
{ {
@ -104,10 +106,7 @@ bool PhraseDictionaryMultiModelCounts::Load(const vector<FactorType> &input
string delim = ":"; string delim = ":";
size_t delim_pos = files[i].find(delim); size_t delim_pos = files[i].find(delim);
if (delim_pos >= files[i].size()) { UTIL_THROW_IF(delim_pos >= files[i].size(), util::Exception, "Phrase table must be specified in this format: Implementation:Path");
UserMessage::Add("Phrase table must be specified in this format: Implementation:Path");
CHECK(false);
}
impl = files[i].substr(0,delim_pos); impl = files[i].substr(0,delim_pos);
file = files[i].substr(delim_pos+1,files[i].size()); file = files[i].substr(delim_pos+1,files[i].size());
@ -119,11 +118,11 @@ bool PhraseDictionaryMultiModelCounts::Load(const vector<FactorType> &input
PhraseTableImplementation implementation = (PhraseTableImplementation) Scan<int>(impl); PhraseTableImplementation implementation = (PhraseTableImplementation) Scan<int>(impl);
if (implementation == Memory) { //how many actual scores there are in the phrase tables
size_t numScoresCounts = 3;
size_t numScoresTargetCounts = 1;
//how many actual scores there are in the phrase tables if (implementation == Memory) {
size_t numScoresCounts = 3;
size_t numScoresTargetCounts = 1;
if (!FileExists(main_table) && FileExists(main_table + ".gz")) main_table += ".gz"; if (!FileExists(main_table) && FileExists(main_table + ".gz")) main_table += ".gz";
if (!FileExists(target_table) && FileExists(target_table + ".gz")) target_table += ".gz"; if (!FileExists(target_table) && FileExists(target_table + ".gz")) target_table += ".gz";
@ -137,8 +136,17 @@ bool PhraseDictionaryMultiModelCounts::Load(const vector<FactorType> &input
pdm_inverse->SetNumScoreComponentMultiModel(numScoresTargetCounts); pdm_inverse->SetNumScoreComponentMultiModel(numScoresTargetCounts);
pdm_inverse->Load( input, output, target_table, m_weight, componentTableLimit, languageModels, m_weightWP); pdm_inverse->Load( input, output, target_table, m_weight, componentTableLimit, languageModels, m_weightWP);
m_inverse_pd.push_back(pdm_inverse); m_inverse_pd.push_back(pdm_inverse);
} } else if (implementation == Binary) {
else if (implementation == Compact) { PhraseDictionaryTreeAdaptor* pdta = new PhraseDictionaryTreeAdaptor(m_numScoreComponent, numInputScores , m_feature_load);
pdta->SetNumScoreComponentMultiModel(m_numScoreComponent); //for binary models, we need to pass number of log-linear components to correctly resize the score vector
pdta->Load(input, output, main_table, m_weight, m_componentTableLimit, languageModels, m_weightWP);
m_pd.push_back(pdta);
PhraseDictionaryTreeAdaptor* pdta_inverse = new PhraseDictionaryTreeAdaptor(m_numScoreComponent, numInputScores , m_feature_load);
pdta_inverse->SetNumScoreComponentMultiModel(m_numScoreComponent);
pdta_inverse->Load(input, output, target_table, m_weight, m_componentTableLimit, languageModels, m_weightWP);
m_inverse_pd.push_back(pdta_inverse);
} else if (implementation == Compact) {
#ifndef WIN32 #ifndef WIN32
PhraseDictionaryCompact* pdc = new PhraseDictionaryCompact(m_numScoreComponent, implementation, m_feature_load); PhraseDictionaryCompact* pdc = new PhraseDictionaryCompact(m_numScoreComponent, implementation, m_feature_load);
pdc->SetNumScoreComponentMultiModel(m_numScoreComponent); //for compact models, we need to pass number of log-linear components to correctly resize the score vector pdc->SetNumScoreComponentMultiModel(m_numScoreComponent); //for compact models, we need to pass number of log-linear components to correctly resize the score vector
@ -150,12 +158,11 @@ bool PhraseDictionaryMultiModelCounts::Load(const vector<FactorType> &input
pdc_inverse->Load( input, output, target_table, m_weight, componentTableLimit, languageModels, m_weightWP); pdc_inverse->Load( input, output, target_table, m_weight, componentTableLimit, languageModels, m_weightWP);
m_inverse_pd.push_back(pdc_inverse); m_inverse_pd.push_back(pdc_inverse);
#else #else
CHECK(false); UTIL_THROW(util::Exception, "Compact phrase table not supported in windows");
#endif #endif
} }
else { else {
UserMessage::Add("phrase table type unknown to multi-model mode"); UTIL_THROW(util::Exception,"PhraseDictionaryMultiModel does not support phrase table type " << implementation);
CHECK(false);
} }
lexicalTable* e2f = new lexicalTable; lexicalTable* e2f = new lexicalTable;
@ -257,8 +264,7 @@ TargetPhraseCollection* PhraseDictionaryMultiModelCounts::CreateTargetPhraseColl
multiModelCountsStatistics * statistics = iter->second; multiModelCountsStatistics * statistics = iter->second;
if (statistics->targetPhrase->GetAlignTerm().GetSize() == 0) { if (statistics->targetPhrase->GetAlignTerm().GetSize() == 0) {
UserMessage::Add(" alignment information empty\ncount-tables need to include alignment information for computation of lexical weights.\nUse --phrase-word-alignment during training; for on-disk tables, also set -alignment-info when creating on-disk tables."); UTIL_THROW(util::Exception, " alignment information empty\ncount-tables need to include alignment information for computation of lexical weights.\nUse --phrase-word-alignment during training; for on-disk tables, also set -alignment-info when creating on-disk tables.");
CHECK(false);
} }
try { try {
@ -619,8 +625,7 @@ double CrossEntropyCounts::operator() ( const dlib::matrix<double,0,1>& arg) con
} }
else { else {
score = 0; score = 0;
UserMessage::Add("Trying to optimize feature that I don't know. Aborting"); UTIL_THROW(util::Exception, "Trying to optimize feature that I don't know. Aborting");
CHECK(false);
} }
total -= (FloorScore(TransformScore(score))/TransformScore(2))*f; total -= (FloorScore(TransformScore(score))/TransformScore(2))*f;
n += f; n += f;

View File

@ -22,6 +22,7 @@ Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
#include "moses/TranslationModel/PhraseDictionaryMultiModel.h" #include "moses/TranslationModel/PhraseDictionaryMultiModel.h"
#include "moses/TranslationModel/PhraseDictionaryMemory.h" #include "moses/TranslationModel/PhraseDictionaryMemory.h"
#include "moses/TranslationModel/PhraseDictionaryTreeAdaptor.h"
#ifndef WIN32 #ifndef WIN32
#include "moses/TranslationModel/CompactPT/PhraseDictionaryCompact.h" #include "moses/TranslationModel/CompactPT/PhraseDictionaryCompact.h"
#endif #endif
@ -90,6 +91,7 @@ public:
, const std::vector<std::string> &files , const std::vector<std::string> &files
, const std::vector<float> &weight , const std::vector<float> &weight
, size_t tableLimit , size_t tableLimit
, size_t numInputScores
, const LMList &languageModels , const LMList &languageModels
, float weightWP); , float weightWP);
TargetPhraseCollection* CreateTargetPhraseCollectionCounts(const Phrase &src, std::vector<float> &fs, std::map<std::string,multiModelCountsStatistics*>* allStats, std::vector<std::vector<float> > &multimodelweights) const; TargetPhraseCollection* CreateTargetPhraseCollectionCounts(const Phrase &src, std::vector<float> &fs, std::map<std::string,multiModelCountsStatistics*>* allStats, std::vector<std::vector<float> > &multimodelweights) const;