mosesdecoder/mert/Data.h

110 lines
2.5 KiB
C
Raw Normal View History

/*
* Data.h
2012-02-20 03:29:53 +04:00
* mert - Minimum Error Rate Training
*
* Created by Nicola Bertoldi on 13/05/08.
*
*/
#ifndef MERT_DATA_H_
#define MERT_DATA_H_
#include <vector>
2012-03-10 12:47:01 +04:00
#include <boost/shared_ptr.hpp>
#include "Util.h"
#include "FeatureData.h"
#include "ScoreData.h"
namespace MosesTuning
{
class Scorer;
2012-02-08 21:11:56 +04:00
typedef boost::shared_ptr<ScoreData> ScoreDataHandle;
typedef boost::shared_ptr<FeatureData> FeatureDataHandle;
2012-03-10 12:47:01 +04:00
// NOTE: there is no copy constructor implemented, so only the
// compiler synthesised shallow copy is available.
class Data
{
private:
Scorer* m_scorer;
std::string m_score_type;
std::size_t m_num_scores;
ScoreDataHandle m_score_data;
FeatureDataHandle m_feature_data;
SparseVector m_sparse_weights;
public:
explicit Data(Scorer* scorer, const std::string& sparseweightsfile="");
2012-03-10 12:47:01 +04:00
void clear() {
m_score_data->clear();
m_feature_data->clear();
}
2013-05-29 21:16:15 +04:00
ScoreDataHandle getScoreData() {
return m_score_data;
}
2013-05-29 21:16:15 +04:00
FeatureDataHandle getFeatureData() {
return m_feature_data;
}
2013-05-29 21:16:15 +04:00
Scorer* getScorer() {
return m_scorer;
}
std::size_t NumberOfFeatures() const {
return m_feature_data->NumberOfFeatures();
}
2013-05-29 21:16:15 +04:00
std::string Features() const {
return m_feature_data->Features();
}
void Features(const std::string &f) {
m_feature_data->Features(f);
}
void loadNBest(const std::string &file, bool oneBest=false);
2012-03-10 12:47:01 +04:00
void load(const std::string &featfile, const std::string &scorefile);
2012-03-10 12:47:01 +04:00
void save(const std::string &featfile, const std::string &scorefile, bool bin=false);
2012-03-10 12:47:01 +04:00
//ADDED BY TS
void removeDuplicates();
//END_ADDED
inline bool existsFeatureNames() const {
return m_feature_data->existsFeatureNames();
}
inline std::string getFeatureName(std::size_t idx) const {
return m_feature_data->getFeatureName(idx);
}
inline std::size_t getFeatureIndex(const std::string& name) const {
return m_feature_data->getFeatureIndex(name);
}
/**
* Create shard_count shards. If shard_size == 0, then the shards are non-overlapping
* and exhaust the data. If 0 < shard_size <= 1, then shards are chosen by sampling
* the data (with replacement) and shard_size is interpreted as the proportion
* of the total size.
*/
void createShards(std::size_t shard_count, float shard_size, const std::string& scorerconfig,
2011-11-12 04:24:19 +04:00
std::vector<Data>& shards);
// Helper functions for loadnbest();
void InitFeatureMap(const std::string& str);
void AddFeatures(const std::string& str,
int sentence_index);
};
}
#endif // MERT_DATA_H_