ENH: implement margin and confidence prediction for CvRTreesWrapper

Modules/Learning/Supervised/src/otbCvRTreesWrapper.cxx

@@ -38,10 +38,55 @@ void CvRTreesWrapper::get_votes(const cv::Mat& sample,
                                 CvRTreesWrapper::VotesVectorType& vote_count) const
 {
 #ifdef OTB_OPENCV_3
-  (void) sample;
+  // missing samples not implemented yet
   (void) missing;
-  (void) vote_count;
-  // TODO
+
+  // Here we have to re-implement a basic "predict_tree()" since the function is
+  // not exposed anymore
+  const std::vector< cv::ml::DTrees::Node > &nodes = m_Impl->getNodes();
+  const std::vector< cv::ml::DTrees::Split > &splits = m_Impl->getSplits();
+  const std::vector<int> &roots = m_Impl->getRoots();
+  int ntrees = roots.size();
+  int nodeIdx, prevNodeIdx;
+  int predictedClass = -1;
+  const float* samplePtr = sample.ptr<float>();
+  std::map<int, unsigned int> votes;
+
+  for (int t=0; t<ntrees ; t++)
+    {
+    nodeIdx = roots[t];
+    prevNodeIdx = nodeIdx;
+    while(1)
+      {
+      prevNodeIdx = nodeIdx;
+      const cv::ml::DTrees::Node &curNode = nodes[nodeIdx];
+      // test if this node is a leaf
+      if (curNode.split < 0)
+        break;
+      const cv::ml::DTrees::Split& split = splits[curNode.split];
+      int varIdx = split.varIdx;
+      float val = samplePtr[varIdx];
+      nodeIdx = val <= split.c ? curNode.left : curNode.right;
+      }
+    predictedClass = nodes[prevNodeIdx].classIdx;
+    votes[predictedClass] += 1;
+    }
+
+  vote_count.resize(votes.size());
+  int pos=0;
+  for (std::map<int, unsigned int>::const_iterator it=votes.begin() ;
+       it != votes.end() ;
+       ++it)
+    {
+    vote_count[pos] = it->second;
+    pos++;
+    }
+
+  if (vote_count.size() == 1)
+    {
+    // give at least 2 classes
+    vote_count.push_back(0);
+    }
 #else
   vote_count.resize(nclasses);
   for( int k = 0; k < ntrees; k++ )
@@ -58,11 +103,8 @@ float CvRTreesWrapper::predict_margin(const cv::Mat& sample,
                                       const cv::Mat& missing) const
 {
 #ifdef OTB_OPENCV_3
-  (void) sample;
-  (void) missing;
-  // TODO
-  return 0.;
-#else
+  int ntrees = m_Impl->getRoots().size();
+#endif
   // Sanity check (division by ntrees later on)
   if(ntrees == 0)
     {
@@ -75,18 +117,14 @@ float CvRTreesWrapper::predict_margin(const cv::Mat& sample,
                    classVotes.end(), std::greater<unsigned int>());
   float margin = static_cast<float>(classVotes[0]-classVotes[1])/ntrees;
   return margin;
-#endif
 }
 
 float CvRTreesWrapper::predict_confidence(const cv::Mat& sample, 
                                   const cv::Mat& missing) const
 {
 #ifdef OTB_OPENCV_3
-  (void) sample;
-  (void) missing;
-  // TODO
-  return 0.;
-#else
+  int ntrees = m_Impl->getRoots().size();
+#endif
   // Sanity check (division by ntrees later on)
   if(ntrees == 0)
     {
@@ -98,7 +136,6 @@ float CvRTreesWrapper::predict_confidence(const cv::Mat& sample,
                                               classVotes.end()));
   float confidence = static_cast<float>(max_votes)/ntrees;
   return confidence;
-#endif
 }
 
 #ifdef OTB_OPENCV_3