vest: combine over-similar search directions, exclude primary directions, skeleton for oracle directions

git-svn-id: https://ws10smt.googlecode.com/svn/trunk@227 ec762483-ff6d-05da-a07a-a48fb63a330f

2 files changed, 28 insertions, 13 deletions

diff --git a/decoder/grammar.cc b/decoder/grammar.cc
index 5eb7887d..499e79fe 100644
--- a/decoder/grammar.cc
+++ b/decoder/grammar.cc
@@ -71,7 +71,7 @@ struct TGImpl {
 };
 
 TextGrammar::TextGrammar() : max_span_(10), pimpl_(new TGImpl) {}
-TextGrammar::TextGrammar(const string& file) : 
+TextGrammar::TextGrammar(const string& file) :
     max_span_(10),
     pimpl_(new TGImpl) {
   ReadFromFile(file);
@@ -104,7 +104,7 @@ void TextGrammar::ReadFromFile(const string& filename) {
   RuleLexer::ReadRules(in.stream(), &AddRuleHelper, this);
 }
 
-bool TextGrammar::HasRuleForSpan(int i, int j, int distance) const {
+bool TextGrammar::HasRuleForSpan(int /* i */, int /* j */, int distance) const {
   return (max_span_ >= distance);
 }
 
@@ -121,8 +121,7 @@ GlueGrammar::GlueGrammar(const string& goal_nt, const string& default_nt) {
   //cerr << "GLUE: " << glue->AsString() << endl;
 }
 
-bool GlueGrammar::HasRuleForSpan(int i, int j, int distance) const {
-  (void) j;
+bool GlueGrammar::HasRuleForSpan(int i, int /* j */, int /* distance */) const {
   return (i == 0);
 }
 
@@ -141,7 +140,7 @@ PassThroughGrammar::PassThroughGrammar(const Lattice& input, const string& cat)
   }
 }
 
-bool PassThroughGrammar::HasRuleForSpan(int i, int j, int distance) const {
+bool PassThroughGrammar::HasRuleForSpan(int i, int j, int /* distance */) const {
   const set<int>& hr = has_rule_[i];
   if (i == j) { return !hr.empty(); }
   return (hr.find(j) != hr.end());
diff --git a/decoder/sparse_vector.h b/decoder/sparse_vector.h
index be91f324..bfdeebcc 100644
--- a/decoder/sparse_vector.h
+++ b/decoder/sparse_vector.h
@@ -58,7 +58,7 @@ public:
     }
 
     int max_index() const {
-        if (values_.empty()) return 0;
+      if (empty()) return 0;
         typename MapType::const_iterator found =values_.end();
         --found;
         return found->first;
@@ -75,6 +75,18 @@ public:
     }
 
     template<typename S>
+    S cosine_sim(const SparseVector<S> &vec) const {
+      return dot(vec)/(l2norm()*vec.l2norm());
+    }
+
+  // if values are binary, gives |A intersect B|/|A union B|
+    template<typename S>
+    S tanimoto_coef(const SparseVector<S> &vec) const {
+      S dp=dot(vec);
+      return dp/(l2norm_sq()*vec.l2norm_sq()-dp);
+    }
+
+    template<typename S>
     S dot(const SparseVector<S> &vec) const {
         S sum = 0;
         for (typename MapType::const_iterator
@@ -119,12 +131,16 @@ public:
       return sum;
     }
 
-    T l2norm() const {
+  T l2norm_sq() const {
       T sum = 0;
       for (typename MapType::const_iterator
               it = values_.begin(); it != values_.end(); ++it)
         sum += it->second * it->second;
-      return sqrt(sum);
+      return sum;
+  }
+
+    T l2norm() const {
+      return sqrt(l2norm_sq());
     }
 
     SparseVector<T> &operator+=(const SparseVector<T> &other) {
@@ -149,14 +165,14 @@ public:
         return *this;
     }
 
-    SparseVector<T> &operator-=(const double &x) {
+    SparseVector<T> &operator-=(T const& x) {
         for (typename MapType::iterator
                 it = values_.begin(); it != values_.end(); ++it)
             it->second -= x;
         return *this;
     }
 
-    SparseVector<T> &operator+=(const double &x) {
+    SparseVector<T> &operator+=(T const& x) {
         for (typename MapType::iterator
                 it = values_.begin(); it != values_.end(); ++it)
             it->second += x;
@@ -177,17 +193,17 @@ public:
         return *this;
     }
 
-    SparseVector<T> operator+(const double &x) const {
+    SparseVector<T> operator+(T const& x) const {
         SparseVector<T> result = *this;
         return result += x;
     }
 
-    SparseVector<T> operator-(const double &x) const {
+    SparseVector<T> operator-(T const& x) const {
         SparseVector<T> result = *this;
         return result -= x;
     }
 
-    SparseVector<T> operator/(const double &x) const {
+    SparseVector<T> operator/(T const& x) const {
         SparseVector<T> result = *this;
         return result /= x;
     }