Merge branch 'master' of https://github.com/redpony/cdec

5 files changed, 37 insertions, 14 deletions

diff --git a/mteval/mbr_kbest.cc b/mteval/mbr_kbest.cc
index 2519bc01..76d2c7fc 100644
--- a/mteval/mbr_kbest.cc
+++ b/mteval/mbr_kbest.cc
@@ -1,9 +1,14 @@
 #include <iostream>
 #include <vector>
-#include <tr1/unordered_map>
 
 #include <boost/program_options.hpp>
 #include <boost/functional/hash.hpp>
+#ifndef HAVE_OLD_CPP
+# include <unordered_map>
+#else
+# include <tr1/unordered_map>
+namespace std { using std::tr1::unordered_map; }
+#endif
 
 #include "prob.h"
 #include "tdict.h"
@@ -12,7 +17,6 @@
 #include "stringlib.h"
 
 using namespace std;
-using namespace std::tr1;
 
 namespace po = boost::program_options;
 
diff --git a/mteval/ns.cc b/mteval/ns.cc
index b64d4798..c1ea238b 100644
--- a/mteval/ns.cc
+++ b/mteval/ns.cc
@@ -106,7 +106,7 @@ struct BleuSegmentEvaluator : public SegmentEvaluator {
   }
 
   struct NGramCompare {
-    int operator() (const vector<WordID>& a, const vector<WordID>& b) {
+    int operator() (const vector<WordID>& a, const vector<WordID>& b) const {
       const size_t as = a.size();
       const size_t bs = b.size();
       const size_t s = (as < bs ? as : bs);
diff --git a/mteval/ns_ter.cc b/mteval/ns_ter.cc
index 0e1008db..680fb7b4 100644
--- a/mteval/ns_ter.cc
+++ b/mteval/ns_ter.cc
@@ -4,7 +4,12 @@
 #include <cassert>
 #include <iostream>
 #include <limits>
-#include <tr1/unordered_map>
+#ifndef HAVE_OLD_CPP
+# include <unordered_map>
+#else
+# include <tr1/unordered_map>
+namespace std { using std::tr1::unordered_map; }
+#endif
 #include <set>
 #include <boost/functional/hash.hpp>
 #include "tdict.h"
@@ -20,7 +25,6 @@ static const unsigned kREF_WORDCOUNT = 4;
 static const unsigned kDUMMY_LAST_ENTRY = 5;
 
 using namespace std;
-using namespace std::tr1;
 
 bool TERMetric::IsErrorMetric() const {
   return true;
diff --git a/mteval/scorer.cc b/mteval/scorer.cc
index de84e076..fedbca91 100644
--- a/mteval/scorer.cc
+++ b/mteval/scorer.cc
@@ -198,7 +198,7 @@ class BLEUScorerBase : public SentenceScorer {
   virtual float ComputeRefLength(const vector<WordID>& hyp) const = 0;
  private:
   struct NGramCompare {
-    int operator() (const vector<WordID>& a, const vector<WordID>& b) {
+    int operator() (const vector<WordID>& a, const vector<WordID>& b) const {
       size_t as = a.size();
       size_t bs = b.size();
       const size_t s = (as < bs ? as : bs);
@@ -594,26 +594,29 @@ void DocScorer::Init(
       const ScoreType type,
       const vector<string>& ref_files,
       const string& src_file, bool verbose) {
+  cerr << "!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!\n"
+          "!!! This code is using the deprecated DocScorer interface, please fix !!!\n"
+          "!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!\n";
   scorers_.clear();
-  // TODO stop using valarray, start using ReadFile
+  static const WordID kDIV = TD::Convert("|||");
   cerr << "Loading references (" << ref_files.size() << " files)\n";
   ReadFile srcrf;
   if (type == AER && src_file.size() > 0) {
     cerr << "  (source=" << src_file << ")\n";
     srcrf.Init(src_file);
   }
+  std::vector<WordID> tmp;
   std::vector<ReadFile> ifs(ref_files.begin(),ref_files.end());
   for (int i=0; i < ref_files.size(); ++i) ifs[i].Init(ref_files[i]);
   char buf[64000];
   bool expect_eof = false;
   int line=0;
   while (ifs[0].get()) {
-    vector<vector<WordID> > refs(ref_files.size());
+    vector<vector<WordID> > refs;
     for (int i=0; i < ref_files.size(); ++i) {
       istream &in=ifs[i].get();
       if (in.eof()) break;
       in.getline(buf, 64000);
-      refs[i].clear();
       if (strlen(buf) == 0) {
         if (in.eof()) {
           if (!expect_eof) {
@@ -622,9 +625,17 @@ void DocScorer::Init(
           }
           break;
         }
-      } else {
-        TD::ConvertSentence(buf, &refs[i]);
-        assert(!refs[i].empty());
+      } else { // process reference
+        tmp.clear();
+        TD::ConvertSentence(buf, &tmp);
+        unsigned last = 0;
+        for (unsigned j = 0; j < tmp.size(); ++j) {
+          if (tmp[j] == kDIV) {
+            refs.push_back(vector<WordID>(tmp.begin() + last, tmp.begin() + j));
+            last = j + 1;
+          }
+        }
+        refs.push_back(vector<WordID>(tmp.begin() + last, tmp.end()));
       }
       assert(!expect_eof);
     }
diff --git a/mteval/ter.cc b/mteval/ter.cc
index cacc5b00..19b90bbc 100644
--- a/mteval/ter.cc
+++ b/mteval/ter.cc
@@ -5,7 +5,12 @@
 #include <iostream>
 #include <limits>
 #include <sstream>
-#include <tr1/unordered_map>
+#ifndef HAVE_OLD_CPP
+# include <unordered_map>
+#else
+# include <tr1/unordered_map>
+namespace std { using std::tr1::unordered_map; }
+#endif
 #include <set>
 #include <valarray>
 #include <boost/functional/hash.hpp>
@@ -16,7 +21,6 @@ const bool ter_use_average_ref_len = true;
 const int ter_short_circuit_long_sentences = -1;
 
 using namespace std;
-using namespace std::tr1;
 
 struct COSTS {
   static const float substitution;