mosesdecoder/moses/FF/LexicalReordering/LexicalReorderingTable.cpp

// -*- c++ -*-

#include "LexicalReorderingTable.h"
#include "moses/InputFileStream.h"
#include "moses/StaticData.h"
#include "moses/TranslationModel/PhraseDictionary.h"
#include "moses/GenerationDictionary.h"
#include "moses/TargetPhrase.h"
#include "moses/TargetPhraseCollection.h"
#include "moses/TranslationTask.h"

#if !defined WIN32 || defined __MINGW32__ || defined HAVE_CMPH
#include "moses/TranslationModel/CompactPT/LexicalReorderingTableCompact.h"
#endif

namespace Moses
{

//cleans str of leading and tailing spaces
std::string auxClearString(const std::string& str)
{
  int i = 0, j = str.size()-1;
  while(i <= j) {
    if(' ' != str[i]) {
      break;
    } else {
      ++i;
    }
  }
  while(j >= i) {
    if(' ' != str[j]) {
      break;
    } else {
      --j;
    }
  }
  return str.substr(i,j-i+1);
}

void auxAppend(IPhrase& head, const IPhrase& tail)
{
  head.reserve(head.size()+tail.size());
  for(size_t i = 0; i < tail.size(); ++i) {
    head.push_back(tail[i]);
  }
}

LexicalReorderingTable*
LexicalReorderingTable::
LoadAvailable(const std::string& filePath,
              const FactorList& f_factors,
              const FactorList& e_factors,
              const FactorList& c_factors)
{
  //decide use Compact or Tree or Memory table
#ifdef HAVE_CMPH
  LexicalReorderingTable *compactLexr = NULL;
  compactLexr = LexicalReorderingTableCompact::CheckAndLoad(filePath + ".minlexr", f_factors, e_factors, c_factors);
  if(compactLexr)
    return compactLexr;
#endif
  LexicalReorderingTable* ret;
  if (FileExists(filePath+".binlexr.idx") )
    ret = new LexicalReorderingTableTree(filePath, f_factors,
                                         e_factors, c_factors);
  else
    ret = new LexicalReorderingTableMemory(filePath, f_factors,
                                           e_factors, c_factors);
  return ret;
}

LexicalReorderingTableMemory::
LexicalReorderingTableMemory(const std::string& filePath,
                             const std::vector<FactorType>& f_factors,
                             const std::vector<FactorType>& e_factors,
                             const std::vector<FactorType>& c_factors)
  : LexicalReorderingTable(f_factors, e_factors, c_factors)
{
  LoadFromFile(filePath);
}

LexicalReorderingTableMemory::
~LexicalReorderingTableMemory() { }

std::vector<float>
LexicalReorderingTableMemory::GetScore(const Phrase& f,
                                       const Phrase& e,
                                       const Phrase& c)
{
  //rather complicated because of const can't use []... as [] might enter new things into std::map
  //also can't have to be careful with words range if c is empty can't use c.GetSize()-1 will underflow and be large
  TableType::const_iterator r;
  std::string key;
  if(0 == c.GetSize()) {
    key = MakeKey(f,e,c);
    r = m_Table.find(key);
    if(m_Table.end() != r) {
      return r->second;
    }
  } else {
    //right try from large to smaller context
    for(size_t i = 0; i <= c.GetSize(); ++i) {
      Phrase sub_c(c.GetSubString(Range(i,c.GetSize()-1)));
      key = MakeKey(f,e,sub_c);
      r = m_Table.find(key);
      if(m_Table.end() != r) {
        return r->second;
      }
    }
  }
  return Scores();
}

void
LexicalReorderingTableMemory::
DbgDump(std::ostream* out) const
{
  TableType::const_iterator i;
  for(i = m_Table.begin(); i != m_Table.end(); ++i) {
    *out << " key: '" << i->first << "' score: ";
    *out << "(num scores: " << (i->second).size() << ")";
    for(size_t j = 0; j < (i->second).size(); ++j)
      *out << (i->second)[j] << " ";

    *out << "\n";
  }
};

std::string
LexicalReorderingTableMemory::MakeKey(const Phrase& f,
                                      const Phrase& e,
                                      const Phrase& c) const
{
  return MakeKey(auxClearString(f.GetStringRep(m_FactorsF)),
                 auxClearString(e.GetStringRep(m_FactorsE)),
                 auxClearString(c.GetStringRep(m_FactorsC)));
}

std::string
LexicalReorderingTableMemory::MakeKey(const std::string& f,
                                      const std::string& e,
                                      const std::string& c) const
{
  std::string key;
  if(!f.empty()) key += f;
  if(!m_FactorsE.empty()) {
    if(!key.empty()) {
      key += "|||";
    }
    key += e;
  }
  if(!m_FactorsC.empty()) {
    if(!key.empty()) {
      key += "|||";
    }
    key += c;
  }
  return key;
}

void
LexicalReorderingTableMemory::
LoadFromFile(const std::string& filePath)
{
  std::string fileName = filePath;
  if(!FileExists(fileName) && FileExists(fileName+".gz"))
    fileName += ".gz";

  InputFileStream file(fileName);
  std::string line(""), key("");
  int numScores = -1;
  std::cerr << "Loading table into memory...";
  while(!getline(file, line).eof()) {
    std::vector<std::string> tokens = TokenizeMultiCharSeparator(line, "|||");
    int t = 0 ;
    std::string f(""),e(""),c("");

    if(!m_FactorsF.empty()) {
      //there should be something for f
      f = auxClearString(tokens.at(t));
      ++t;
    }
    if(!m_FactorsE.empty()) {
      //there should be something for e
      e = auxClearString(tokens.at(t));
      ++t;
    }
    if(!m_FactorsC.empty()) {
      //there should be something for c
      c = auxClearString(tokens.at(t));
      ++t;
    }
    //last token are the probs
    std::vector<float> p = Scan<float>(Tokenize(tokens.at(t)));
    //sanity check: all lines must have equall number of probs
    if(-1 == numScores) {
      numScores = (int)p.size(); //set in first line
    }
    if((int)p.size() != numScores) {
      TRACE_ERR( "found inconsistent number of probabilities... found "
                 << p.size() << " expected " << numScores << std::endl);
      exit(0);
    }
    std::transform(p.begin(),p.end(),p.begin(),TransformScore);
    std::transform(p.begin(),p.end(),p.begin(),FloorScore);
    //save it all into our map
    m_Table[MakeKey(f,e,c)] = p;
  }
  std::cerr << "done.\n";
}

LexicalReorderingTableTree::
LexicalReorderingTableTree(const std::string& filePath,
                           const std::vector<FactorType>& f_factors,
                           const std::vector<FactorType>& e_factors,
                           const std::vector<FactorType>& c_factors)
  : LexicalReorderingTable(f_factors, e_factors, c_factors)
  , m_UseCache(false)
  , m_FilePath(filePath)
{
  m_Table.reset(new PrefixTreeMap());
  m_Table->Read(m_FilePath+".binlexr");
}

LexicalReorderingTableTree::
~LexicalReorderingTableTree()
{ }

Scores
LexicalReorderingTableTree::
GetScore(const Phrase& f, const Phrase& e, const Phrase& c)
{
  if((!m_FactorsF.empty() && 0 == f.GetSize())
      || (!m_FactorsE.empty() && 0 == e.GetSize())) {
    //NOTE: no check for c as c might be empty, e.g. start of sentence
    //not a proper key
    // phi: commented out, since e may be empty (drop-unknown)
    //std::cerr << "Not a proper key!\n";
    return Scores();
  }

  CacheType::iterator i;

  if(m_UseCache) {
    std::pair<CacheType::iterator, bool> r;
    r = m_Cache.insert(std::make_pair(MakeCacheKey(f,e),Candidates()));
    if(!r.second) return auxFindScoreForContext((r.first)->second, c);
    i = r.first;
  } else if((i = m_Cache.find(MakeCacheKey(f,e))) != m_Cache.end())
    // although we might not be caching now, cache might be none empty!
    return auxFindScoreForContext(i->second, c);

  // not in cache => go to file...
  Candidates cands;
  m_Table->GetCandidates(MakeTableKey(f,e), &cands);
  if(cands.empty()) return Scores();
  if(m_UseCache) i->second = cands;

  if(m_FactorsC.empty()) {
    UTIL_THROW_IF2(1 != cands.size(), "Error");
    return cands[0].GetScore(0);
  } else return auxFindScoreForContext(cands, c);
};

Scores
LexicalReorderingTableTree::
auxFindScoreForContext(const Candidates& cands, const Phrase& context)
{
  if(m_FactorsC.empty()) {
    UTIL_THROW_IF2(cands.size() > 1, "Error");
    return (cands.size() == 1) ? cands[0].GetScore(0) : Scores();
  } else {
    std::vector<std::string> cvec;
    for(size_t i = 0; i < context.GetSize(); ++i)
      cvec.push_back(context.GetWord(i).GetString(m_FactorsC, false));

    IPhrase c = m_Table->ConvertPhrase(cvec,TargetVocId);
    IPhrase sub_c;
    IPhrase::iterator start = c.begin();
    for(size_t j = 0; j <= context.GetSize(); ++j, ++start) {
      sub_c.assign(start, c.end());
      for(size_t cand = 0; cand < cands.size(); ++cand) {
        IPhrase p = cands[cand].GetPhrase(0);
        if(cands[cand].GetPhrase(0) == sub_c)
          return cands[cand].GetScore(0);
      }
    }
    return Scores();
  }
}

void
LexicalReorderingTableTree::
InitializeForInput(ttasksptr const& ttask)
{
  const InputType& input = *ttask->GetSource();
  ClearCache();
  if(ConfusionNet const* cn = dynamic_cast<ConfusionNet const*>(&input)) {
    Cache(*cn);
  } else if (dynamic_cast<Sentence const*>(&input)) {
    // Cache(*s); ... this just takes up too much memory, we cache elsewhere
    DisableCache();
  }
  if (!m_Table.get()) {
    //load thread specific table.
    m_Table.reset(new PrefixTreeMap());
    m_Table->Read(m_FilePath+".binlexr");
  }
};

bool
LexicalReorderingTableTree::
Create(std::istream& inFile, const std::string& outFileName)
{
  typedef PrefixTreeSA<LabelId,OFF_T> PSA;

  std::string
  line,
  ofn(outFileName+".binlexr.srctree"),
  oft(outFileName+".binlexr.tgtdata"),
  ofi(outFileName+".binlexr.idx"),
  ofsv(outFileName+".binlexr.voc0"),
  oftv(outFileName+".binlexr.voc1");

  FILE *os = fOpen(ofn.c_str(),"wb");
  FILE *ot = fOpen(oft.c_str(),"wb");

  PSA *psa = new PSA;
  PSA::setDefault(InvalidOffT);
  WordVoc* voc[3];

  LabelId currFirstWord = InvalidLabelId;
  IPhrase currKey;

  Candidates         cands;
  std::vector<OFF_T> vo;
  size_t lnc = 0;
  size_t numTokens    = 0;
  size_t numKeyTokens = 0;
  while(getline(inFile, line)) {
    ++lnc;
    if(0 == lnc % 10000) TRACE_ERR(".");
    IPhrase key;
    Scores  score;

    std::vector<std::string> tokens = TokenizeMultiCharSeparator(line, "|||");
    std::string w;
    if(1 == lnc) {
      //do some init stuff in the first line
      numTokens = tokens.size();
      if(tokens.size() == 2) {
        // f ||| score
        numKeyTokens = 1;
        voc[0] = new WordVoc();
        voc[1] = 0;
      } else if(3 == tokens.size() || 4 == tokens.size()) {
        //either f ||| e ||| score or f ||| e ||| c ||| score
        numKeyTokens = 2;
        voc[0] = new WordVoc(); //f voc
        voc[1] = new WordVoc(); //e voc
        voc[2] = voc[1];        //c & e share voc
      }
    } else {
      //sanity check ALL lines must have same number of tokens
      UTIL_THROW_IF2(numTokens != tokens.size(),
                     "Lines do not have the same number of tokens");
    }
    size_t phrase = 0;
    for(; phrase < numKeyTokens; ++phrase) {
      //conditioned on more than just f... need |||
      if(phrase >=1) key.push_back(PrefixTreeMap::MagicWord);
      std::istringstream is(tokens[phrase]);
      while(is >> w) key.push_back(voc[phrase]->add(w));
    }

    //collect all non key phrases, i.e. c
    std::vector<IPhrase> tgt_phrases;
    tgt_phrases.resize(numTokens - numKeyTokens - 1);
    for(size_t j = 0; j < tgt_phrases.size(); ++j, ++phrase) {
      std::istringstream is(tokens[numKeyTokens + j]);
      while(is >> w) tgt_phrases[j].push_back(voc[phrase]->add(w));
    }

    //last token is score
    std::istringstream is(tokens[numTokens-1]);
    while(is >> w) score.push_back(atof(w.c_str()));

    //transform score now...
    std::transform(score.begin(),score.end(),score.begin(),TransformScore);
    std::transform(score.begin(),score.end(),score.begin(),FloorScore);
    std::vector<Scores> scores;
    scores.push_back(score);

    if(key.empty()) {
      TRACE_ERR("WARNING: empty source phrase in line '"<<line<<"'\n");
      continue;
    }

    //first time inits
    if(currFirstWord == InvalidLabelId) currFirstWord = key[0];
    if(currKey.empty()) {
      currKey = key;
      //insert key into tree
      UTIL_THROW_IF2(psa == NULL, "Object not yet created");
      PSA::Data& d = psa->insert(key);
      if(d == InvalidOffT) d = fTell(ot);
      else {
        TRACE_ERR("ERROR: source phrase already inserted (A)!\nline("
                  << lnc << "): '" << line << "\n");
        return false;
      }
    }

    if(currKey != key) {
      //ok new key
      currKey = key;
      //a) write cands for old key
      cands.writeBin(ot);
      cands.clear();
      //b) check if we need to move on to new tree root
      if(key[0] != currFirstWord) {
        // write key prefix tree to file and clear
        PTF pf;
        if(currFirstWord >= vo.size())
          vo.resize(currFirstWord+1,InvalidOffT);
        vo[currFirstWord] = fTell(os);
        pf.create(*psa, os);
        delete psa;
        psa = new PSA;
        currFirstWord = key[0];
      }

      // c) insert key into tree
      UTIL_THROW_IF2(psa == NULL, "Object not yet created");
      PSA::Data& d = psa->insert(key);
      if(d == InvalidOffT) d = fTell(ot);
      else {
        TRACE_ERR("ERROR: source phrase already inserted (A)!\nline("
                  << lnc << "): '" << line << "\n");
        return false;
      }
    }
    cands.push_back(GenericCandidate(tgt_phrases, scores));
  }
  if (lnc == 0) {
    TRACE_ERR("ERROR: empty lexicalised reordering file\n" << std::endl);
    return false;
  }
  cands.writeBin(ot);
  cands.clear();

  PTF pf;
  if(currFirstWord >= vo.size())
    vo.resize(currFirstWord+1,InvalidOffT);
  vo[currFirstWord] = fTell(os);
  pf.create(*psa,os);
  delete psa;
  psa=0;

  fClose(os);
  fClose(ot);
  FILE *oi = fOpen(ofi.c_str(),"wb");
  fWriteVector(oi,vo);
  fClose(oi);

  if(voc[0]) {
    voc[0]->Write(ofsv);
    delete voc[0];
  }
  if(voc[1]) {
    voc[1]->Write(oftv);
    delete voc[1];
  }
  return true;
}

std::string
LexicalReorderingTableTree::
MakeCacheKey(const Phrase& f, const Phrase& e) const
{
  std::string key;
  if(!m_FactorsF.empty())
    key += auxClearString(f.GetStringRep(m_FactorsF));

  if(!m_FactorsE.empty()) {
    if(!key.empty()) {
      key += "|||";
    }
    key += auxClearString(e.GetStringRep(m_FactorsE));
  }
  return key;
};

IPhrase
LexicalReorderingTableTree::
MakeTableKey(const Phrase& f, const Phrase& e) const
{
  IPhrase key;
  std::vector<std::string> keyPart;
  if(!m_FactorsF.empty()) {
    for(size_t i = 0; i < f.GetSize(); ++i)
      keyPart.push_back(f.GetWord(i).GetString(m_FactorsF, false));
    auxAppend(key, m_Table->ConvertPhrase(keyPart, SourceVocId));
    keyPart.clear();
  }
  if(!m_FactorsE.empty()) {
    if(!key.empty()) key.push_back(PrefixTreeMap::MagicWord);
    for(size_t i = 0; i < e.GetSize(); ++i)
      keyPart.push_back(e.GetWord(i).GetString(m_FactorsE, false));
    auxAppend(key, m_Table->ConvertPhrase(keyPart,TargetVocId));
  }
  return key;
};


struct State {
  State(PPimp* t, const std::string& p)
    : pos(t), path(p) { }

  PPimp*      pos;
  std::string path;
};

void
LexicalReorderingTableTree::
auxCacheForSrcPhrase(const Phrase& f)
{
  if(m_FactorsE.empty()) {
    //f is all of key...
    Candidates cands;
    m_Table->GetCandidates(MakeTableKey(f,Phrase(ARRAY_SIZE_INCR)),&cands);
    m_Cache[MakeCacheKey(f,Phrase(ARRAY_SIZE_INCR))] = cands;
  } else {
    ObjectPool<PPimp> pool;
    PPimp* pPos  = m_Table->GetRoot();

    // 1) goto subtree for f
    for(size_t i = 0; i < f.GetSize() && 0 != pPos && pPos->isValid(); ++i)
      pPos = m_Table->Extend(pPos, f.GetWord(i).GetString(m_FactorsF, false), SourceVocId);

    if(pPos && pPos->isValid())
      pPos = m_Table->Extend(pPos, PrefixTreeMap::MagicWord);

    if(!pPos || !pPos->isValid())
      return;

    //2) explore whole subtree depth first & cache
    std::string cache_key = auxClearString(f.GetStringRep(m_FactorsF)) + "|||";

    std::vector<State> stack;
    stack.push_back(State(pool.get(PPimp(pPos->ptr()->getPtr(pPos->idx),0,0)),""));
    Candidates cands;
    while(!stack.empty()) {
      if(stack.back().pos->isValid()) {
        LabelId w = stack.back().pos->ptr()->getKey(stack.back().pos->idx);
        std::string next_path = stack.back().path + " " + m_Table->ConvertWord(w,TargetVocId);
        //cache this
        m_Table->GetCandidates(*stack.back().pos,&cands);
        if(!cands.empty()) m_Cache[cache_key + auxClearString(next_path)] = cands;
        cands.clear();
        PPimp* next_pos = pool.get(PPimp(stack.back().pos->ptr()->getPtr(stack.back().pos->idx),0,0));
        ++stack.back().pos->idx;
        stack.push_back(State(next_pos,next_path));
      } else stack.pop_back();
    }
  }
}

void
LexicalReorderingTableTree::
Cache(const ConfusionNet& /*input*/)
{
  return;
}

void
LexicalReorderingTableTree::
Cache(const Sentence& input)
{
  //only works with sentences...
  size_t prev_cache_size = m_Cache.size();
  size_t max_phrase_length = input.GetSize();
  for(size_t len = 0; len <= max_phrase_length; ++len) {
    for(size_t start = 0; start+len <= input.GetSize(); ++start) {
      Phrase f    = input.GetSubString(Range(start, start+len));
      auxCacheForSrcPhrase(f);
    }
  }
  std::cerr << "Cached " << m_Cache.size() - prev_cache_size
            << " new primary reordering table keys\n";
}
}