From 102243ac63d40ab004dc55cd758706dd144b5d14 Mon Sep 17 00:00:00 2001
From: Patrick Simianer <simianer@cl.uni-heidelberg.de>
Date: Wed, 13 Jun 2012 14:52:23 +0200
Subject: max pairs arg

---
 dtrain/dtrain.cc      |  2 +-
 dtrain/pairsampling.h | 15 +++++++++++++++
 dtrain/score.cc       |  6 ++++--
 3 files changed, 20 insertions(+), 3 deletions(-)

diff --git a/dtrain/dtrain.cc b/dtrain/dtrain.cc
index 88413a1d..7f44d4cf 100644
--- a/dtrain/dtrain.cc
+++ b/dtrain/dtrain.cc
@@ -261,7 +261,7 @@ main(int argc, char** argv)
       cerr << setw(25) << "l1 reg " << l1_reg << " '" << cfg["l1_reg"].as<string>() << "'" << endl;
     if (rescale)
       cerr << setw(25) << "rescale " << rescale << endl;
-    cerr << "max pairs " << max_pairs << endl; 
+    cerr << setw(25) << "max pairs " << max_pairs << endl;
     cerr << setw(25) << "cdec cfg " << "'" << cfg["decoder_config"].as<string>() << "'" << endl;
     cerr << setw(25) << "input " << "'" << input_fn << "'" << endl;
 #ifdef DTRAIN_LOCAL
diff --git a/dtrain/pairsampling.h b/dtrain/pairsampling.h
index 71c8ae59..84be1efb 100644
--- a/dtrain/pairsampling.h
+++ b/dtrain/pairsampling.h
@@ -23,6 +23,8 @@ all_pairs(vector<ScoredHyp>* s, vector<pair<ScoredHyp,ScoredHyp> >& training, sc
 {
   sort(s->begin(), s->end(), cmp_hyp_by_score_d);
   unsigned sz = s->size();
+  bool b = false;
+  unsigned count = 0;
   for (unsigned i = 0; i < sz-1; i++) {
     for (unsigned j = i+1; j < sz; j++) {
       if (threshold > 0) {
@@ -32,7 +34,12 @@ all_pairs(vector<ScoredHyp>* s, vector<pair<ScoredHyp,ScoredHyp> >& training, sc
         if ((*s)[i].score != (*s)[j].score)
           training.push_back(make_pair((*s)[i], (*s)[j]));
       }
+      if (++count == max) {
+        b = true;
+        break;
+      }
     }
+    if (b) break;
   }
 }
 
@@ -53,6 +60,8 @@ partXYX(vector<ScoredHyp>* s, vector<pair<ScoredHyp,ScoredHyp> >& training, scor
   unsigned sep_hi = sep;
   if (sz > 4) while (sep_hi < sz && (*s)[sep_hi-1].score == (*s)[sep_hi].score) ++sep_hi;
   else sep_hi = 1;
+  bool b = false;
+  unsigned count = 0;
   for (unsigned i = 0; i < sep_hi; i++) {
     for (unsigned j = sep_hi; j < sz; j++) {
 #ifdef DTRAIN_FASTER_PERCEPTRON
@@ -65,10 +74,15 @@ partXYX(vector<ScoredHyp>* s, vector<pair<ScoredHyp,ScoredHyp> >& training, scor
         if ((*s)[i].score != (*s)[j].score)
           training.push_back(make_pair((*s)[i], (*s)[j]));
       }
+      if (++count == max) {
+        b = true;
+        break;
+      }
 #ifdef DTRAIN_FASTER_PERCEPTRON
       }
 #endif
     }
+    if (b) break;
   }
   unsigned sep_lo = sz-sep;
   while (sep_lo > 0 && (*s)[sep_lo-1].score == (*s)[sep_lo].score) --sep_lo;
@@ -84,6 +98,7 @@ partXYX(vector<ScoredHyp>* s, vector<pair<ScoredHyp,ScoredHyp> >& training, scor
         if ((*s)[i].score != (*s)[j].score)
           training.push_back(make_pair((*s)[i], (*s)[j]));
       }
+      if (++count == max) return;
 #ifdef DTRAIN_FASTER_PERCEPTRON
       }
 #endif
diff --git a/dtrain/score.cc b/dtrain/score.cc
index 5c356c0f..5bb0bcaa 100644
--- a/dtrain/score.cc
+++ b/dtrain/score.cc
@@ -128,6 +128,7 @@ SmoothSingleBleuScorer::Score(vector<WordID>& hyp, vector<WordID>& ref,
   return brevity_penalty(hyp_len, ref_len) * sum;
 }
 
+// TODO single variants!
 
 /*
  * approx. bleu
@@ -136,7 +137,8 @@ SmoothSingleBleuScorer::Score(vector<WordID>& hyp, vector<WordID>& ref,
  *        and Structural Translation Features"
  * (Chiang et al. '08)
  *
- * NOTE: needs some more code in dtrain.cc
+ * NOTE: Needs some more code in dtrain.cc .
+ *       No scaling by src len.
  */
 score_t
 ApproxBleuScorer::Score(vector<WordID>& hyp, vector<WordID>& ref,
@@ -158,7 +160,7 @@ ApproxBleuScorer::Score(vector<WordID>& hyp, vector<WordID>& ref,
     glob_ref_len_ = discount_ * (glob_ref_len_ + ref_len);
     glob_src_len_ = discount_ * (glob_src_len_ + src_len);
   }
-  return (score_t)glob_src_len_ * score;
+  return score;
 }
 
 /*
-- 
cgit v1.2.3