more efficient default recaser training

1 files changed, 14 insertions, 12 deletions

diff --git a/phrase-extract/consolidate-direct-main.cpp b/phrase-extract/consolidate-direct-main.cpp
index 576cdd568..423a3909b 100644
--- a/phrase-extract/consolidate-direct-main.cpp
+++ b/phrase-extract/consolidate-direct-main.cpp
@@ -28,6 +28,8 @@
 
 using namespace std;
 
+std::vector<std::string> tokenize( const char [] );
+
 vector< string > splitLine(const char *line)
 {
   vector< string > item;
@@ -58,7 +60,7 @@ bool getLine( istream &fileP, vector< string > &item )
   string line;
   if (getline(fileP, line)) {
     item = splitLine(line.c_str());
-    return false;
+    return true;
   } else {
     return false;
   }
@@ -107,17 +109,17 @@ int main(int argc, char* argv[])
     if (! getLine(fileDirectP,  itemDirect  ))
       break;
 
-    (*fileConsolidated) << itemDirect[0] << " ||| " << itemDirect[1] << " ||| ";
-
-    // output alignment and probabilities
-    (*fileConsolidated)	<< itemDirect[2]						// prob direct
-                        << " 2.718" // phrase count feature
-                        << " ||| " << itemDirect[3];	// alignment
-
-    // counts
-    (*fileConsolidated) << "||| 0 " << itemDirect[4]; // indirect
-    (*fileConsolidated) << endl;
-
+    vector< string > count = tokenize( itemDirect[4].c_str() );
+    float countEF = atof(count[0].c_str());
+    float countF = atof(count[1].c_str());
+    float prob = countF/countEF;
+
+    (*fileConsolidated) << itemDirect[0] << " ||| "        // source
+                        << itemDirect[1] << " ||| "        // target
+                        << prob << " ||| "                 // prob
+                        << itemDirect[2] << "||| "        // alignment
+                        << itemDirect[4] << " " << countEF // counts
+                        << " ||| " << endl;
   }
 
   fileConsolidated->flush();