mosesdecoder/moses/HypothesisStackCubePruning.cpp
Ulrich Germann 240b88c683 Passing around AllOptions or references thereto everywhere,
strong them locally where appropriate, so that compontents can
become independent of StaticData once instantiated.
2015-12-09 23:05:00 +00:00

314 lines
9.4 KiB
C++

// $Id$
/***********************************************************************
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 <algorithm>
#include <set>
#include <queue>
#include "HypothesisStackCubePruning.h"
#include "TypeDef.h"
#include "Util.h"
#include "StaticData.h"
#include "Manager.h"
#include "util/exception.hh"
using namespace std;
namespace Moses
{
HypothesisStackCubePruning::HypothesisStackCubePruning(Manager& manager) :
HypothesisStack(manager)
{
m_nBestIsEnabled = manager.options()->nbest.enabled;
m_bestScore = -std::numeric_limits<float>::infinity();
m_worstScore = -std::numeric_limits<float>::infinity();
m_deterministic = manager.options()->cube.deterministic_search;
}
/** remove all hypotheses from the collection */
void HypothesisStackCubePruning::RemoveAll()
{
// delete all bitmap accessors;
_BMType::iterator iter;
for (iter = m_bitmapAccessor.begin(); iter != m_bitmapAccessor.end(); ++iter) {
delete iter->second;
}
}
pair<HypothesisStackCubePruning::iterator, bool> HypothesisStackCubePruning::Add(Hypothesis *hypo)
{
std::pair<iterator, bool> ret = m_hypos.insert(hypo);
if (ret.second) {
// equiv hypo doesn't exists
VERBOSE(3,"added hyp to stack");
// Update best score, if this hypothesis is new best
if (hypo->GetFutureScore() > m_bestScore) {
VERBOSE(3,", best on stack");
m_bestScore = hypo->GetFutureScore();
// this may also affect the worst score
if ( m_bestScore + m_beamWidth > m_worstScore )
m_worstScore = m_bestScore + m_beamWidth;
}
// Prune only if stack is twice as big as needed (lazy pruning)
VERBOSE(3,", now size " << m_hypos.size());
if (m_hypos.size() > 2*m_maxHypoStackSize-1) {
PruneToSize(m_maxHypoStackSize);
} else {
VERBOSE(3,std::endl);
}
}
return ret;
}
bool HypothesisStackCubePruning::AddPrune(Hypothesis *hypo)
{
if (hypo->GetFutureScore() == - std::numeric_limits<float>::infinity()) {
m_manager.GetSentenceStats().AddDiscarded();
VERBOSE(3,"discarded, constraint" << std::endl);
delete hypo;
return false;
}
if (hypo->GetFutureScore() < m_worstScore) {
// too bad for stack. don't bother adding hypo into collection
m_manager.GetSentenceStats().AddDiscarded();
VERBOSE(3,"discarded, too bad for stack" << std::endl);
delete hypo;
return false;
}
// over threshold, try to add to collection
std::pair<iterator, bool> addRet = Add(hypo);
if (addRet.second) {
// nothing found. add to collection
return true;
}
// equiv hypo exists, recombine with other hypo
iterator &iterExisting = addRet.first;
assert(iterExisting != m_hypos.end());
Hypothesis *hypoExisting = *iterExisting;
m_manager.GetSentenceStats().AddRecombination(*hypo, **iterExisting);
// found existing hypo with same target ending.
// keep the best 1
if (hypo->GetFutureScore() > hypoExisting->GetFutureScore()) {
// incoming hypo is better than the one we have
VERBOSE(3,"better than matching hyp " << hypoExisting->GetId() << ", recombining, ");
if (m_nBestIsEnabled) {
hypo->AddArc(hypoExisting);
Detach(iterExisting);
} else {
Remove(iterExisting);
}
bool added = Add(hypo).second;
if (!added) {
iterExisting = m_hypos.find(hypo);
UTIL_THROW(util::Exception, "Should have added hypothesis " << **iterExisting);
}
return false;
} else {
// already storing the best hypo. discard current hypo
VERBOSE(3,"worse than matching hyp " << hypoExisting->GetId() << ", recombining" << std::endl)
if (m_nBestIsEnabled) {
hypoExisting->AddArc(hypo);
} else {
delete hypo;
}
return false;
}
}
void HypothesisStackCubePruning::AddInitial(Hypothesis *hypo)
{
std::pair<iterator, bool> addRet = Add(hypo);
UTIL_THROW_IF2(!addRet.second,
"Should have added hypothesis " << *hypo);
const Bitmap &bitmap = hypo->GetWordsBitmap();
AddBitmapContainer(bitmap, *this);
}
void HypothesisStackCubePruning::PruneToSize(size_t newSize)
{
if ( newSize == 0) return; // no limit
if (m_hypos.size() > newSize) { // ok, if not over the limit
priority_queue<float> bestScores;
// push all scores to a heap
// (but never push scores below m_bestScore+m_beamWidth)
iterator iter = m_hypos.begin();
float score = 0;
while (iter != m_hypos.end()) {
Hypothesis *hypo = *iter;
score = hypo->GetFutureScore();
if (score > m_bestScore+m_beamWidth) {
bestScores.push(score);
}
++iter;
}
// pop the top newSize scores (and ignore them, these are the scores of hyps that will remain)
// ensure to never pop beyond heap size
size_t minNewSizeHeapSize = newSize > bestScores.size() ? bestScores.size() : newSize;
for (size_t i = 1 ; i < minNewSizeHeapSize ; i++)
bestScores.pop();
// and remember the threshold
float scoreThreshold = bestScores.top();
// delete all hypos under score threshold
iter = m_hypos.begin();
while (iter != m_hypos.end()) {
Hypothesis *hypo = *iter;
float score = hypo->GetFutureScore();
if (score < scoreThreshold) {
iterator iterRemove = iter++;
Remove(iterRemove);
m_manager.GetSentenceStats().AddPruning();
} else {
++iter;
}
}
VERBOSE(3,", pruned to size " << size() << endl);
IFVERBOSE(3) {
TRACE_ERR("stack now contains: ");
for(iter = m_hypos.begin(); iter != m_hypos.end(); iter++) {
Hypothesis *hypo = *iter;
TRACE_ERR( hypo->GetId() << " (" << hypo->GetFutureScore() << ") ");
}
TRACE_ERR( endl);
}
// set the worstScore, so that newly generated hypotheses will not be added if worse than the worst in the stack
m_worstScore = scoreThreshold;
}
}
const Hypothesis *HypothesisStackCubePruning::GetBestHypothesis() const
{
if (!m_hypos.empty()) {
const_iterator iter = m_hypos.begin();
Hypothesis *bestHypo = *iter;
while (++iter != m_hypos.end()) {
Hypothesis *hypo = *iter;
if (hypo->GetFutureScore() > bestHypo->GetFutureScore())
bestHypo = hypo;
}
return bestHypo;
}
return NULL;
}
vector<const Hypothesis*> HypothesisStackCubePruning::GetSortedList() const
{
vector<const Hypothesis*> ret;
ret.reserve(m_hypos.size());
std::copy(m_hypos.begin(), m_hypos.end(), std::inserter(ret, ret.end()));
sort(ret.begin(), ret.end(), CompareHypothesisTotalScore());
return ret;
}
void HypothesisStackCubePruning::CleanupArcList()
{
// only necessary if n-best calculations are enabled
if (!m_nBestIsEnabled) return;
iterator iter;
for (iter = m_hypos.begin() ; iter != m_hypos.end() ; ++iter) {
Hypothesis *mainHypo = *iter;
mainHypo->CleanupArcList(this->m_manager.options()->nbest.nbest_size, this->m_manager.options()->NBestDistinct());
}
}
void HypothesisStackCubePruning::SetBitmapAccessor(const Bitmap &newBitmap
, HypothesisStackCubePruning &stack
, const Range &/*range*/
, BitmapContainer &bitmapContainer
, const SquareMatrix &estimatedScores
, const TranslationOptionList &transOptList)
{
BitmapContainer *bmContainer = AddBitmapContainer(newBitmap, stack);
BackwardsEdge *edge = new BackwardsEdge(bitmapContainer
, *bmContainer
, transOptList
, estimatedScores
, m_manager.GetSource()
, m_deterministic);
bmContainer->AddBackwardsEdge(edge);
}
TO_STRING_BODY(HypothesisStackCubePruning);
// friend
std::ostream& operator<<(std::ostream& out, const HypothesisStackCubePruning& hypoColl)
{
HypothesisStackCubePruning::const_iterator iter;
for (iter = hypoColl.begin() ; iter != hypoColl.end() ; ++iter) {
const Hypothesis &hypo = **iter;
out << hypo << endl;
}
return out;
}
void
HypothesisStackCubePruning::AddHypothesesToBitmapContainers()
{
HypothesisStackCubePruning::const_iterator iter;
for (iter = m_hypos.begin() ; iter != m_hypos.end() ; ++iter) {
Hypothesis *h = *iter;
const Bitmap &bitmap = h->GetWordsBitmap();
BitmapContainer *container = m_bitmapAccessor[&bitmap];
container->AddHypothesis(h);
}
}
BitmapContainer *HypothesisStackCubePruning::AddBitmapContainer(const Bitmap &bitmap, HypothesisStackCubePruning &stack)
{
_BMType::iterator iter = m_bitmapAccessor.find(&bitmap);
BitmapContainer *bmContainer;
if (iter == m_bitmapAccessor.end()) {
bmContainer = new BitmapContainer(bitmap, stack, m_deterministic);
m_bitmapAccessor[&bitmap] = bmContainer;
} else {
bmContainer = iter->second;
}
return bmContainer;
}
}