Changed implementation of indocs (to keep track of which documents phrases come from) from vector to map.

6 files changed, 26 insertions, 13 deletions

diff --git a/moses/TranslationModel/UG/mm/ug_bitext_agenda_job.h b/moses/TranslationModel/UG/mm/ug_bitext_agenda_job.h
index 0e0624351..36b9873e0 100644
--- a/moses/TranslationModel/UG/mm/ug_bitext_agenda_job.h
+++ b/moses/TranslationModel/UG/mm/ug_bitext_agenda_job.h
@@ -137,7 +137,10 @@ int Bitext<Token>::agenda::job
 
   float p = (*m_bias)[sid];
   id_type docid = m_bias->GetClass(sid);
-  uint32_t k = docid < stats->indoc.size() ? stats->indoc[docid] : 0;
+  
+  // uint32_t k = docid < stats->indoc.size() ? stats->indoc[docid] : 0;
+  std::map<uint32_t,uint32_t>::const_iterator m = stats->indoc.find(docid);
+  uint32_t k = m != stats->indoc.end() ? m->second : 0 ;
 
   // always consider candidates from dominating documents and
   // from documents that have not been considered at all yet
@@ -159,11 +162,17 @@ int Bitext<Token>::agenda::job
 	e = root->getCorpus()->sntEnd(sid);
       *log << docid << ":" << sid << " " << size_t(k) << "/" << N
 	   << " @" << p << " => " << d << " [";
-      for (size_t i = 0; i < stats->indoc.size(); ++i)
+      for (std::map<uint32_t, uint32_t>::const_iterator m = stats->indoc.begin();
+	   m != stats->indoc.end(); ++m)
 	{
-	  if (i) *log << " ";
-	  *log << stats->indoc[i];
+	  if (m != stats->indoc.begin()) *log << " ";
+	  *log << m->first << ":" << m->second;
 	}
+      // for (size_t i = 0; i < stats->indoc.size(); ++i)
+      // 	{
+      // 	  if (i) *log << " ";
+      // 	  *log << stats->indoc[i];
+      // 	}
       *log << "] ";
       for (; x < e; ++x) *log << (*m_bitext->V1)[x->id()] << " ";
       if (!ret) *log << "SKIP";
diff --git a/moses/TranslationModel/UG/mm/ug_bitext_jstats.cc b/moses/TranslationModel/UG/mm/ug_bitext_jstats.cc
index bcda9ebf3..517caf783 100644
--- a/moses/TranslationModel/UG/mm/ug_bitext_jstats.cc
+++ b/moses/TranslationModel/UG/mm/ug_bitext_jstats.cc
@@ -76,7 +76,7 @@ namespace Moses
       ++obwd[bwd_orient];
       if (docid >= 0)
 	{
-	  while (int(indoc.size()) <= docid) indoc.push_back(0);
+	  // while (int(indoc.size()) <= docid) indoc.push_back(0);
 	  ++indoc[docid];
 	}
     }
diff --git a/moses/TranslationModel/UG/mm/ug_bitext_jstats.h b/moses/TranslationModel/UG/mm/ug_bitext_jstats.h
index dade27649..03b231487 100644
--- a/moses/TranslationModel/UG/mm/ug_bitext_jstats.h
+++ b/moses/TranslationModel/UG/mm/ug_bitext_jstats.h
@@ -27,7 +27,8 @@ namespace Moses
       uint32_t obwd[Moses::LRModel::NONE+1]; // backward distortion type counts
 
     public:
-      vector<uint32_t> indoc; // counts origin of samples (for biased sampling)
+      std::map<uint32_t,uint32_t> indoc;
+      // vector<uint32_t> indoc; // counts origin of samples (for biased sampling)
       jstats();
       jstats(jstats const& other);
       uint32_t rcnt() const; // raw joint counts
diff --git a/moses/TranslationModel/UG/mm/ug_bitext_pstats.cc b/moses/TranslationModel/UG/mm/ug_bitext_pstats.cc
index 580d7669b..8702d9c50 100644
--- a/moses/TranslationModel/UG/mm/ug_bitext_pstats.cc
+++ b/moses/TranslationModel/UG/mm/ug_bitext_pstats.cc
@@ -58,7 +58,7 @@ namespace Moses
       ++obwd[po_bwd];
       if (docid >= 0)
 	{
-	  while (int(indoc.size()) <= docid) indoc.push_back(0);
+	  // while (int(indoc.size()) <= docid) indoc.push_back(0);
 	  ++indoc[docid];
 	}
     }
diff --git a/moses/TranslationModel/UG/mm/ug_bitext_pstats.h b/moses/TranslationModel/UG/mm/ug_bitext_pstats.h
index 9a14e378b..e5cf4ab26 100644
--- a/moses/TranslationModel/UG/mm/ug_bitext_pstats.h
+++ b/moses/TranslationModel/UG/mm/ug_bitext_pstats.h
@@ -33,8 +33,8 @@ namespace Moses
       uint32_t ofwd[Moses::LRModel::NONE+1]; // distribution of fwd phrase orientations
       uint32_t obwd[Moses::LRModel::NONE+1]; // distribution of bwd phrase orientations
 
-      std::vector<uint32_t> indoc; // distribution over where samples came from
-
+      // std::vector<uint32_t> indoc; // distribution over where samples came from
+      std::map<uint32_t,uint32_t> indoc; 
       typedef std::map<uint64_t, jstats> trg_map_t;
       trg_map_t trg;
       pstats();
diff --git a/moses/TranslationModel/UG/mm/ug_phrasepair.h b/moses/TranslationModel/UG/mm/ug_phrasepair.h
index 7e565c2df..7f03d89df 100644
--- a/moses/TranslationModel/UG/mm/ug_phrasepair.h
+++ b/moses/TranslationModel/UG/mm/ug_phrasepair.h
@@ -30,7 +30,8 @@ namespace Moses
       std::vector<uchar> aln;
       float score;
       bool inverse;
-      std::vector<uint32_t> indoc;
+      // std::vector<uint32_t> indoc;
+      std::map<uint32_t,uint32_t> indoc;
       PhrasePair() { };
       PhrasePair(PhrasePair const& o);
 
@@ -306,10 +307,12 @@ namespace Moses
       out << toString (V1, this->start1, this->len1) << " ::: "
 	  << toString (V2, this->start2, this->len2) << " "
 	  << this->joint << " [";
-      for (size_t i = 0; i < this->indoc.size(); ++i)
+      // for (size_t i = 0; i < this->indoc.size(); ++i)
+      for (std::map<uint32_t,uint32_t>::const_iterator m = indoc.begin();
+	   m != indoc.end(); ++m)
 	{
-	  if (i) out << " ";
-	  out << this->indoc[i];
+	  if (m != indoc.begin()) out << " ";
+	  out << m->first << ":" << m->second;
 	}
       out << "] [";
       vector<float> lrscores;