mirror of
https://github.com/moses-smt/mosesdecoder.git
synced 2024-12-29 06:52:34 +03:00
f223f5a276
M mert/BleuScorer.h M mert/ScorerFactory.h M mert/Scorer.h M mert/PerScorer.h M mert/TerScorer.h M mert/Makefile.am AM scripts/training/mert-moses-multi.pl git-svn-id: https://mosesdecoder.svn.sourceforge.net/svnroot/mosesdecoder/trunk@4299 1f5c12ca-751b-0410-a591-d2e778427230
112 lines
3.2 KiB
C++
112 lines
3.2 KiB
C++
#include "TerScorer.h"
|
|
#include "TERsrc/tercalc.h"
|
|
#include "TERsrc/terAlignment.h"
|
|
|
|
const int TerScorer::LENGTH = 2;
|
|
using namespace TERCpp;
|
|
using namespace std;
|
|
|
|
|
|
void TerScorer::setReferenceFiles ( const vector<string>& referenceFiles )
|
|
{
|
|
// for each line in the reference file, create a multiset of the
|
|
// word ids
|
|
for ( int incRefs = 0; incRefs < ( int ) referenceFiles.size(); incRefs++ ) {
|
|
stringstream convert;
|
|
m_references.clear();
|
|
|
|
_reftokens.clear();
|
|
_reflengths.clear();
|
|
ifstream in ( referenceFiles.at ( incRefs ).c_str() );
|
|
if ( !in ) {
|
|
throw runtime_error ( "Unable to open " + referenceFiles.at ( incRefs ) );
|
|
}
|
|
string line;
|
|
int sid = 0;
|
|
while ( getline ( in, line ) ) {
|
|
vector<int> tokens;
|
|
encode ( line, tokens );
|
|
m_references.push_back ( tokens );
|
|
TRACE_ERR ( "." );
|
|
++sid;
|
|
}
|
|
m_multi_references.push_back ( m_references );
|
|
}
|
|
|
|
TRACE_ERR ( endl );
|
|
m_references=m_multi_references.at(0);
|
|
}
|
|
|
|
void TerScorer::prepareStats ( size_t sid, const string& text, ScoreStats& entry )
|
|
{
|
|
|
|
terAlignment result;
|
|
result.numEdits = 0.0 ;
|
|
result.numWords = 0.0 ;
|
|
result.averageWords = 0.0;
|
|
|
|
for ( int incRefs = 0; incRefs < ( int ) m_multi_references.size(); incRefs++ ) {
|
|
if ( sid >= m_multi_references.at(incRefs).size() ) {
|
|
stringstream msg;
|
|
msg << "Sentence id (" << sid << ") not found in reference set";
|
|
throw runtime_error ( msg.str() );
|
|
}
|
|
|
|
vector<int> testtokens;
|
|
vector<int> reftokens;
|
|
reftokens = m_multi_references.at ( incRefs ).at ( sid );
|
|
double averageLength=0.0;
|
|
for ( int incRefsBis = 0; incRefsBis < ( int ) m_multi_references.size(); incRefsBis++ ) {
|
|
if ( sid >= m_multi_references.at(incRefsBis).size() ) {
|
|
stringstream msg;
|
|
msg << "Sentence id (" << sid << ") not found in reference set";
|
|
throw runtime_error ( msg.str() );
|
|
}
|
|
averageLength+=(double)m_multi_references.at ( incRefsBis ).at ( sid ).size();
|
|
}
|
|
averageLength=averageLength/( double ) m_multi_references.size();
|
|
encode ( text, testtokens );
|
|
terCalc * evaluation=new terCalc();
|
|
evaluation->setDebugMode ( false );
|
|
terAlignment tmp_result = evaluation->TER ( reftokens, testtokens );
|
|
tmp_result.averageWords=averageLength;
|
|
if ( ( result.numEdits == 0.0 ) && ( result.averageWords == 0.0 ) ) {
|
|
result = tmp_result;
|
|
} else if ( result.scoreAv() > tmp_result.scoreAv() ) {
|
|
result = tmp_result;
|
|
}
|
|
|
|
}
|
|
ostringstream stats;
|
|
// multiplication by 100 in order to keep the average precision in the TER calculation
|
|
stats << result.numEdits*100.0 << " " << result.averageWords*100.0 << " " << result.scoreAv()*100.0 << " " ;
|
|
string stats_str = stats.str();
|
|
entry.set ( stats_str );
|
|
}
|
|
|
|
float TerScorer::calculateScore ( const vector<int>& comps )
|
|
{
|
|
float denom = 1.0 * comps[1];
|
|
float num = -1.0 * comps[0];
|
|
if ( denom == 0 ) {
|
|
// shouldn't happen!
|
|
return 1.0;
|
|
} else {
|
|
return (1.0+(num / denom));
|
|
}
|
|
}
|
|
|
|
/*
|
|
float TerScorer::calculateScore ( const vector<float>& comps )
|
|
{
|
|
float denom = 1.0 * comps[1];
|
|
float num = -1.0 * comps[0];
|
|
if ( denom == 0 ) {
|
|
// shouldn't happen!
|
|
return 1.0;
|
|
} else {
|
|
return (1.0+(num / denom));
|
|
}
|
|
}
|
|
*/
|