mirror of
https://github.com/moses-smt/mosesdecoder.git
synced 2024-12-25 12:52:29 +03:00
170 lines
5.4 KiB
C++
170 lines
5.4 KiB
C++
// vim:tabstop=2
|
|
|
|
/***********************************************************************
|
|
Moses - factored phrase-based language decoder
|
|
Copyright (C) 2006 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
|
|
***********************************************************************/
|
|
|
|
#include "moses/TranslationModel/PhraseDictionary.h"
|
|
#include "moses/StaticData.h"
|
|
#include "moses/InputType.h"
|
|
#include "moses/TranslationOption.h"
|
|
#include "moses/UserMessage.h"
|
|
#include "moses/InputPath.h"
|
|
#include "util/exception.hh"
|
|
|
|
using namespace std;
|
|
|
|
namespace Moses
|
|
{
|
|
|
|
PhraseDictionary::PhraseDictionary(const std::string &description, const std::string &line)
|
|
:DecodeFeature(description, line)
|
|
,m_tableLimit(20) // default
|
|
,m_maxCacheSize(DEFAULT_MAX_TRANS_OPT_CACHE_SIZE)
|
|
{
|
|
}
|
|
|
|
const TargetPhraseCollection *PhraseDictionary::GetTargetPhraseCollection(const Phrase& src) const
|
|
{
|
|
const TargetPhraseCollection *ret;
|
|
if (m_maxCacheSize) {
|
|
size_t hash = hash_value(src);
|
|
|
|
std::map<size_t, std::pair<const TargetPhraseCollection*, clock_t> >::iterator iter;
|
|
|
|
{
|
|
// scope of read lock
|
|
#ifdef WITH_THREADS
|
|
boost::shared_lock<boost::shared_mutex> read_lock(m_accessLock);
|
|
#endif
|
|
iter = m_cache.find(hash);
|
|
}
|
|
|
|
if (iter == m_cache.end()) {
|
|
// not in cache, need to look up from phrase table
|
|
ret = GetTargetPhraseCollectionNonCache(src);
|
|
if (ret) {
|
|
ret = new TargetPhraseCollection(*ret);
|
|
}
|
|
|
|
std::pair<const TargetPhraseCollection*, clock_t> value(ret, clock());
|
|
#ifdef WITH_THREADS
|
|
boost::unique_lock<boost::shared_mutex> lock(m_accessLock);
|
|
#endif
|
|
m_cache[hash] = value;
|
|
} else {
|
|
// in cache. just use it
|
|
std::pair<const TargetPhraseCollection*, clock_t> &value = iter->second;
|
|
value.second = clock();
|
|
|
|
ret = value.first;
|
|
}
|
|
} else {
|
|
// don't use cache. look up from phrase table
|
|
ret = GetTargetPhraseCollectionNonCache(src);
|
|
}
|
|
|
|
return ret;
|
|
}
|
|
|
|
const TargetPhraseCollection *PhraseDictionary::GetTargetPhraseCollectionNonCache(const Phrase& src) const
|
|
{
|
|
UTIL_THROW(util::Exception, "Legacy method not implemented");
|
|
}
|
|
|
|
|
|
const TargetPhraseCollectionWithSourcePhrase* PhraseDictionary::
|
|
GetTargetPhraseCollectionLegacy(InputType const& src,WordsRange const& range) const
|
|
{
|
|
UTIL_THROW(util::Exception, "Legacy method not implemented");
|
|
//Phrase phrase = src.GetSubString(range);
|
|
//return GetTargetPhraseCollection(phrase);
|
|
}
|
|
|
|
void PhraseDictionary::SetParameter(const std::string& key, const std::string& value)
|
|
{
|
|
if (key == "cache-size") {
|
|
m_maxCacheSize = Scan<size_t>(value);
|
|
} else if (key == "path") {
|
|
m_filePath = value;
|
|
} else if (key == "table-limit") {
|
|
m_tableLimit = Scan<size_t>(value);
|
|
} else {
|
|
DecodeFeature::SetParameter(key, value);
|
|
}
|
|
}
|
|
|
|
void PhraseDictionary::SetFeaturesToApply()
|
|
{
|
|
// find out which feature function can be applied in this decode step
|
|
const std::vector<FeatureFunction*> &allFeatures = FeatureFunction::GetFeatureFunctions();
|
|
for (size_t i = 0; i < allFeatures.size(); ++i) {
|
|
FeatureFunction *feature = allFeatures[i];
|
|
if (feature->IsUseable(m_outputFactors)) {
|
|
m_featuresToApply.push_back(feature);
|
|
}
|
|
}
|
|
}
|
|
|
|
void PhraseDictionary::GetTargetPhraseCollectionBatch(const InputPathList &phraseDictionaryQueue) const
|
|
{
|
|
InputPathList::const_iterator iter;
|
|
for (iter = phraseDictionaryQueue.begin(); iter != phraseDictionaryQueue.end(); ++iter) {
|
|
InputPath &node = **iter;
|
|
|
|
const Phrase &phrase = node.GetPhrase();
|
|
const TargetPhraseCollection *targetPhrases = this->GetTargetPhraseCollection(phrase);
|
|
node.SetTargetPhrases(*this, targetPhrases, NULL);
|
|
}
|
|
}
|
|
|
|
void PhraseDictionary::ReduceCache() const
|
|
{
|
|
if (m_cache.size() <= m_maxCacheSize) return; // not full
|
|
clock_t t = clock();
|
|
|
|
// find cutoff for last used time
|
|
priority_queue< clock_t > lastUsedTimes;
|
|
std::map<size_t, std::pair<const TargetPhraseCollection*,clock_t> >::iterator iter;
|
|
iter = m_cache.begin();
|
|
while( iter != m_cache.end() ) {
|
|
lastUsedTimes.push( iter->second.second );
|
|
iter++;
|
|
}
|
|
for( size_t i=0; i < lastUsedTimes.size()-m_maxCacheSize/2; i++ )
|
|
lastUsedTimes.pop();
|
|
clock_t cutoffLastUsedTime = lastUsedTimes.top();
|
|
|
|
// remove all old entries
|
|
#ifdef WITH_THREADS
|
|
boost::unique_lock<boost::shared_mutex> lock(m_accessLock);
|
|
#endif
|
|
|
|
iter = m_cache.begin();
|
|
while( iter != m_cache.end() ) {
|
|
if (iter->second.second < cutoffLastUsedTime) {
|
|
std::map<size_t, std::pair<const TargetPhraseCollection*,clock_t> >::iterator iterRemove = iter++;
|
|
delete iterRemove->second.first;
|
|
m_cache.erase(iterRemove);
|
|
} else iter++;
|
|
}
|
|
VERBOSE(2,"Reduced persistent translation option cache in " << ((clock()-t)/(float)CLOCKS_PER_SEC) << " seconds." << std::endl);
|
|
}
|
|
}
|
|
|