2014-09-08 02:18:05 +04:00
|
|
|
// $Id$
|
|
|
|
// vim:tabstop=2
|
|
|
|
|
|
|
|
/***********************************************************************
|
|
|
|
Moses - factored phrase-based language decoder
|
|
|
|
Copyright (C) 2014- University of Edinburgh
|
|
|
|
|
|
|
|
This library is free software; you can redistribute it and/or
|
|
|
|
modify it under the terms of the GNU Lesser General Public
|
|
|
|
License as published by the Free Software Foundation; either
|
|
|
|
version 2.1 of the License, or (at your option) any later version.
|
|
|
|
|
|
|
|
This library is distributed in the hope that it will be useful,
|
|
|
|
but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
|
|
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
|
|
|
Lesser General Public License for more details.
|
|
|
|
|
|
|
|
You should have received a copy of the GNU Lesser General Public
|
|
|
|
License along with this library; if not, write to the Free Software
|
|
|
|
Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
|
|
|
|
***********************************************************************/
|
|
|
|
|
|
|
|
|
|
|
|
/**
|
|
|
|
Prune the phrase table using the same translation pruning that Moses uses during decoding.
|
|
|
|
**/
|
|
|
|
|
|
|
|
#include <cstring>
|
|
|
|
#include <iostream>
|
|
|
|
#include <fstream>
|
2014-10-21 18:11:09 +04:00
|
|
|
#include <map>
|
2014-09-08 02:18:05 +04:00
|
|
|
#include <string>
|
|
|
|
#include <vector>
|
|
|
|
|
|
|
|
#include <boost/program_options.hpp>
|
|
|
|
#include <boost/scoped_ptr.hpp>
|
|
|
|
|
2014-09-08 18:08:18 +04:00
|
|
|
#include "moses/InputPath.h"
|
2014-09-08 02:18:05 +04:00
|
|
|
#include "moses/Parameter.h"
|
2014-09-08 18:08:18 +04:00
|
|
|
#include "moses/TranslationModel/PhraseDictionary.h"
|
2014-09-08 02:18:05 +04:00
|
|
|
#include "moses/StaticData.h"
|
|
|
|
|
|
|
|
#include "util/file_piece.hh"
|
|
|
|
#include "util/string_piece.hh"
|
|
|
|
#include "util/tokenize_piece.hh"
|
|
|
|
#include "util/double-conversion/double-conversion.h"
|
|
|
|
#include "util/exception.hh"
|
|
|
|
|
|
|
|
|
|
|
|
using namespace Moses;
|
|
|
|
using namespace std;
|
|
|
|
|
|
|
|
namespace po = boost::program_options;
|
2014-10-21 18:11:09 +04:00
|
|
|
typedef multimap<float,string> Lines;
|
2014-09-08 02:18:05 +04:00
|
|
|
|
2015-01-14 14:07:42 +03:00
|
|
|
static void usage(const po::options_description& desc, char** argv)
|
|
|
|
{
|
|
|
|
cerr << "Usage: " + string(argv[0]) + " [options] input-file output-file" << endl;
|
|
|
|
cerr << desc << endl;
|
2014-09-08 02:18:05 +04:00
|
|
|
}
|
|
|
|
|
|
|
|
//Find top n translations of source, and send them to output
|
2015-01-14 14:07:42 +03:00
|
|
|
static void outputTopN(Lines lines, size_t maxPhrases, ostream& out)
|
|
|
|
{
|
2014-10-21 18:11:09 +04:00
|
|
|
size_t count = 0;
|
|
|
|
for (Lines::const_reverse_iterator i = lines.rbegin(); i != lines.rend(); ++i) {
|
|
|
|
out << i->second << endl;
|
|
|
|
++count;
|
|
|
|
if (count >= maxPhrases) break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
/*
|
|
|
|
static void outputTopN(const Phrase& sourcePhrase, const multimap<float,const TargetPhrase*>& targetPhrases,
|
|
|
|
size_t maxPhrases, const PhraseDictionary* phraseTable,
|
|
|
|
const vector<FactorType> & input, const vector<FactorType> & output, ostream& out) {
|
|
|
|
size_t count = 0;
|
|
|
|
for (multimap<float,const TargetPhrase*>::const_reverse_iterator i
|
|
|
|
= targetPhrases.rbegin(); i != targetPhrases.rend() && count < maxPhrases; ++i, ++count) {
|
|
|
|
const TargetPhrase* targetPhrase = i->second;
|
2014-09-08 18:08:18 +04:00
|
|
|
out << sourcePhrase.GetStringRep(input);
|
|
|
|
out << " ||| ";
|
|
|
|
out << targetPhrase->GetStringRep(output);
|
|
|
|
out << " ||| ";
|
|
|
|
const ScoreComponentCollection scores = targetPhrase->GetScoreBreakdown();
|
|
|
|
vector<float> phraseScores = scores.GetScoresForProducer(phraseTable);
|
|
|
|
for (size_t j = 0; j < phraseScores.size(); ++j) {
|
|
|
|
out << exp(phraseScores[j]) << " ";
|
|
|
|
}
|
|
|
|
out << "||| ";
|
|
|
|
const AlignmentInfo& align = targetPhrase->GetAlignTerm();
|
|
|
|
for (AlignmentInfo::const_iterator j = align.begin(); j != align.end(); ++j) {
|
|
|
|
out << j->first << "-" << j->second << " ";
|
|
|
|
}
|
|
|
|
out << endl;
|
|
|
|
}
|
2014-10-21 18:11:09 +04:00
|
|
|
}*/
|
2015-01-14 14:07:42 +03:00
|
|
|
int main(int argc, char** argv)
|
2014-09-08 02:18:05 +04:00
|
|
|
{
|
|
|
|
bool help;
|
|
|
|
string input_file;
|
|
|
|
string config_file;
|
2014-10-21 18:11:09 +04:00
|
|
|
size_t maxPhrases = 100;
|
2014-09-08 02:18:05 +04:00
|
|
|
|
|
|
|
|
|
|
|
po::options_description desc("Allowed options");
|
|
|
|
desc.add_options()
|
|
|
|
("help,h", po::value(&help)->zero_tokens()->default_value(false), "Print this help message and exit")
|
|
|
|
("input-file,i", po::value<string>(&input_file), "Input file")
|
|
|
|
("config-file,f", po::value<string>(&config_file), "Config file")
|
2014-10-21 18:11:09 +04:00
|
|
|
("max-phrases,n", po::value<size_t>(&maxPhrases), "Maximum target phrases per source phrase")
|
2014-09-08 02:18:05 +04:00
|
|
|
;
|
|
|
|
|
|
|
|
po::options_description cmdline_options;
|
|
|
|
cmdline_options.add(desc);
|
|
|
|
po::variables_map vm;
|
2014-09-08 18:08:18 +04:00
|
|
|
po::parsed_options parsed = po::command_line_parser(argc,argv).
|
2015-01-14 14:07:42 +03:00
|
|
|
options(cmdline_options).run();
|
2014-09-08 18:08:18 +04:00
|
|
|
po::store(parsed, vm);
|
2014-09-08 02:18:05 +04:00
|
|
|
po::notify(vm);
|
|
|
|
if (help) {
|
|
|
|
usage(desc, argv);
|
|
|
|
exit(0);
|
|
|
|
}
|
|
|
|
if (input_file.empty()) {
|
|
|
|
cerr << "ERROR: Please specify an input file" << endl << endl;
|
|
|
|
usage(desc, argv);
|
|
|
|
exit(1);
|
|
|
|
}
|
|
|
|
if (config_file.empty()) {
|
|
|
|
cerr << "ERROR: Please specify a config file" << endl << endl;
|
|
|
|
usage(desc, argv);
|
|
|
|
exit(1);
|
|
|
|
}
|
|
|
|
|
|
|
|
vector<string> mosesargs;
|
|
|
|
mosesargs.push_back(argv[0]);
|
|
|
|
mosesargs.push_back("-f");
|
|
|
|
mosesargs.push_back(config_file);
|
|
|
|
|
2015-01-14 14:07:42 +03:00
|
|
|
boost::scoped_ptr<Parameter> params(new Parameter());
|
2014-09-08 02:18:05 +04:00
|
|
|
char** mosesargv = new char*[mosesargs.size()];
|
|
|
|
for (size_t i = 0; i < mosesargs.size(); ++i) {
|
|
|
|
mosesargv[i] = new char[mosesargs[i].length() + 1];
|
|
|
|
strcpy(mosesargv[i], mosesargs[i].c_str());
|
|
|
|
}
|
|
|
|
|
|
|
|
if (!params->LoadParam(mosesargs.size(), mosesargv)) {
|
|
|
|
params->Explain();
|
|
|
|
exit(1);
|
|
|
|
}
|
|
|
|
|
|
|
|
if (!StaticData::LoadDataStatic(params.get(),argv[0])) {
|
|
|
|
exit(1);
|
|
|
|
}
|
|
|
|
|
2014-09-08 18:08:18 +04:00
|
|
|
const StaticData &staticData = StaticData::Instance();
|
|
|
|
|
2014-10-21 18:11:09 +04:00
|
|
|
//Find the phrase table to manage the target phrases
|
2014-09-08 18:08:18 +04:00
|
|
|
PhraseDictionary* phraseTable = NULL;
|
|
|
|
const vector<FeatureFunction*>& ffs = FeatureFunction::GetFeatureFunctions();
|
|
|
|
for (size_t i = 0; i < ffs.size(); ++i) {
|
|
|
|
PhraseDictionary* maybePhraseTable = dynamic_cast< PhraseDictionary*>(ffs[i]);
|
|
|
|
if (maybePhraseTable) {
|
|
|
|
UTIL_THROW_IF(phraseTable,util::Exception,"Can only score translations with one phrase table");
|
|
|
|
phraseTable = maybePhraseTable;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
UTIL_THROW_IF(!phraseTable,util::Exception,"Unable to find scoring phrase table");
|
|
|
|
|
2014-09-08 02:18:05 +04:00
|
|
|
|
|
|
|
//
|
|
|
|
//Load and prune the phrase table. This is taken (with mods) from moses/TranslationModel/RuleTable/LoaderStandard.cpp
|
|
|
|
//
|
|
|
|
|
|
|
|
std::ostream *progress = NULL;
|
|
|
|
IFVERBOSE(1) progress = &std::cerr;
|
|
|
|
util::FilePiece in(input_file.c_str(), progress);
|
|
|
|
|
|
|
|
// reused variables
|
|
|
|
vector<float> scoreVector;
|
|
|
|
StringPiece line;
|
|
|
|
|
|
|
|
double_conversion::StringToDoubleConverter converter(double_conversion::StringToDoubleConverter::NO_FLAGS, NAN, NAN, "inf", "nan");
|
|
|
|
|
2014-10-21 18:11:09 +04:00
|
|
|
string previous;
|
|
|
|
Lines lines;
|
|
|
|
|
2014-09-08 02:18:05 +04:00
|
|
|
|
|
|
|
while(true) {
|
|
|
|
try {
|
|
|
|
line = in.ReadLine();
|
|
|
|
} catch (const util::EndOfFileException &e) {
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
|
|
|
|
util::TokenIter<util::MultiCharacter> pipes(line, "|||");
|
|
|
|
StringPiece sourcePhraseString(*pipes);
|
2014-10-21 18:11:09 +04:00
|
|
|
StringPiece targetPhraseString(*++pipes);
|
|
|
|
StringPiece scoreString(*++pipes);
|
|
|
|
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_IF2(isnan(score), "Bad score " << *s);
|
|
|
|
scoreVector.push_back(FloorScore(TransformScore(score)));
|
|
|
|
}
|
|
|
|
|
2014-09-08 02:18:05 +04:00
|
|
|
if (sourcePhraseString != previous) {
|
2014-10-21 18:11:09 +04:00
|
|
|
outputTopN(lines, maxPhrases, cout);
|
|
|
|
previous = sourcePhraseString.as_string();
|
|
|
|
lines.clear();
|
2014-09-08 02:18:05 +04:00
|
|
|
}
|
2014-10-21 18:11:09 +04:00
|
|
|
|
|
|
|
ScoreComponentCollection scores;
|
|
|
|
scores.Assign(phraseTable,scoreVector);
|
|
|
|
float score = scores.InnerProduct(staticData.GetAllWeights());
|
|
|
|
lines.insert(pair<float,string>(score,line.as_string()));
|
|
|
|
|
|
|
|
}
|
|
|
|
if (!lines.empty()) {
|
|
|
|
outputTopN(lines, maxPhrases, cout);
|
2014-09-08 02:18:05 +04:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|