gistfile1.txt

diff --git a/src/mlpack/methods/lmnn/lmnn_function.hpp b/src/mlpack/methods/lmnn/lmnn_function.hpp
index fab543e2c..ee2a5144b 100644
--- a/src/mlpack/methods/lmnn/lmnn_function.hpp
+++ b/src/mlpack/methods/lmnn/lmnn_function.hpp
@@ -257,7 +257,7 @@ class LMNNFunction
                         const size_t begin,
                         const size_t batchSize);
   // Recalculate impostors.
-  inline void ReCalculateImpostors(const arma::mat& transformedDataset,
+  inline bool ReCalculateImpostors(const arma::mat& transformedDataset,
                                    double transformationDiff);
 };
 
diff --git a/src/mlpack/methods/lmnn/lmnn_function_impl.hpp b/src/mlpack/methods/lmnn/lmnn_function_impl.hpp
index 21c35f532..00e8bc63c 100644
--- a/src/mlpack/methods/lmnn/lmnn_function_impl.hpp
+++ b/src/mlpack/methods/lmnn/lmnn_function_impl.hpp
@@ -89,7 +89,7 @@ LMNNFunction<MetricType>::LMNNFunction(const arma::mat& dataset,
   }
 
   constraint.TargetNeighbors(targetNeighbors, dataset, labels, norm);
-  constraint.Impostors(impostors, dataset, labels, norm);
+  constraint.Impostors(impostors, distance, dataset, labels, norm);
 
   // Precalculate and save the gradient due to target neighbors.
   Precalculate();
@@ -218,7 +218,7 @@ inline void LMNNFunction<MetricType>::TransDiff(
 
 // Recalculate impostors.
 template<typename MetricType>
-inline void LMNNFunction<MetricType>::ReCalculateImpostors(
+inline bool LMNNFunction<MetricType>::ReCalculateImpostors(
                                           const arma::mat& transformedDataset,
                                           double transformationDiff)
 {
@@ -248,13 +248,17 @@ inline void LMNNFunction<MetricType>::ReCalculateImpostors(
       constraint.Impostors(impostors, distance, transformedDataset, labels,
           norm);
     }
+    return true;
   }
   else if (iteration++ % range == 0)
   {
       // Re-calculate impostors on transformed dataset.
       constraint.Impostors(impostors, distance, transformedDataset, labels,
           norm);
+    return true;
   }
+
+  return false;
 }
 
 //! Evaluate cost over whole dataset.
@@ -273,7 +277,8 @@ double LMNNFunction<MetricType>::Evaluate(const arma::mat& transformation)
     transformationDiff = arma::norm(transformation - transformationOld);
   }
 
-  ReCalculateImpostors(transformedDataset,transformationDiff);
+  bool didRecalculate = ReCalculateImpostors(transformedDataset,
+      transformationDiff);
 
   for (size_t i = 0; i < dataset.n_cols; i++)
   {
@@ -309,7 +314,7 @@ double LMNNFunction<MetricType>::Evaluate(const arma::mat& transformation)
         // Calculate exact eval value.
         if (eval > -1)
         {
-          if (iteration - 1 % range == 0)
+          if (didRecalculate)
           {
             eval = metric.Evaluate(transformedDataset.col(i),
                      transformedDataset.col(targetNeighbors(j, i))) -
@@ -369,6 +374,7 @@ double LMNNFunction<MetricType>::Evaluate(const arma::mat& transformation,
   // Apply metric over dataset.
   transformedDataset = transformation * dataset;
 
+  bool didRecalculate = false;
   if (recalculate)
   {
     // Re-calculate impostors on transformed dataset.
@@ -376,6 +382,7 @@ double LMNNFunction<MetricType>::Evaluate(const arma::mat& transformation,
         norm);
     // Set recalculate to false.
     recalculate = false;
+    didRecalculate = true;
   }
 
   for (size_t i = begin; i < begin + batchSize; i++)
@@ -412,7 +419,7 @@ double LMNNFunction<MetricType>::Evaluate(const arma::mat& transformation,
         // Calculate exact eval value.
         if (eval > -1)
         {
-          if (iteration - 1 % range == 0)
+          if (didRecalculate)
           {
             eval = metric.Evaluate(transformedDataset.col(i),
                      transformedDataset.col(targetNeighbors(j, i))) -
@@ -475,7 +482,8 @@ void LMNNFunction<MetricType>::Gradient(const arma::mat& transformation,
     transformationDiff = arma::norm(transformation - transformationOld);
   }
 
-  ReCalculateImpostors(transformedDataset,transformationDiff);
+  bool didRecalculate = ReCalculateImpostors(transformedDataset,
+      transformationDiff);
 
   gradient.zeros(transformation.n_rows, transformation.n_cols);
 
@@ -510,7 +518,7 @@ void LMNNFunction<MetricType>::Gradient(const arma::mat& transformation,
         // Calculate exact eval value.
         if (eval > -1)
         {
-          if (iteration - 1 % range == 0)
+          if (didRecalculate)
           {
             eval = metric.Evaluate(transformedDataset.col(i),
                      transformedDataset.col(targetNeighbors(j, i))) -
@@ -576,6 +584,7 @@ void LMNNFunction<MetricType>::Gradient(const arma::mat& transformation,
   std::map<size_t, double> transformationDiffs;
   TransDiff(transformationDiffs, transformation, begin, batchSize);
 
+  bool didRecalculate = false;
   if (recalculate)
   {
     // Re-calculate impostors on transformed dataset.
@@ -583,6 +592,7 @@ void LMNNFunction<MetricType>::Gradient(const arma::mat& transformation,
         norm);
     // Set recalculate to false.
     recalculate = false;
+    didRecalculate = true;
   }
 
   gradient.zeros(transformation.n_rows, transformation.n_cols);
@@ -622,7 +632,7 @@ void LMNNFunction<MetricType>::Gradient(const arma::mat& transformation,
         // Calculate exact eval value.
         if (eval > -1)
         {
-          if (iteration - 1 % range == 0)
+          if (didRecalculate)
           {
             eval = metric.Evaluate(transformedDataset.col(i),
                      transformedDataset.col(targetNeighbors(j, i))) -
@@ -694,7 +704,8 @@ double LMNNFunction<MetricType>::EvaluateWithGradient(
     transformationDiff = arma::norm(transformation - transformationOld);
   }
 
-  ReCalculateImpostors(transformedDataset,transformationDiff);
+  bool didRecalculate = ReCalculateImpostors(transformedDataset,
+      transformationDiff);
 
   gradient.zeros(transformation.n_rows, transformation.n_cols);
 
@@ -737,7 +748,7 @@ double LMNNFunction<MetricType>::EvaluateWithGradient(
         // Calculate exact eval value.
         if (eval > -1)
         {
-          if (iteration - 1 % range == 0)
+          if (didRecalculate)
           {
             eval = metric.Evaluate(transformedDataset.col(i),
                      transformedDataset.col(targetNeighbors(j, i))) -
@@ -802,6 +813,7 @@ double LMNNFunction<MetricType>::EvaluateWithGradient(
   // Apply metric over dataset.
   transformedDataset = transformation * dataset;
 
+  bool didRecalculate = false;
   if (recalculate)
   {
     // Re-calculate impostors on transformed dataset.
@@ -809,6 +821,7 @@ double LMNNFunction<MetricType>::EvaluateWithGradient(
         norm);
     // Set recalculate to false.
     recalculate = false;
+    didRecalculate = true;
   }
 
   gradient.zeros(transformation.n_rows, transformation.n_cols);
@@ -853,7 +866,7 @@ double LMNNFunction<MetricType>::EvaluateWithGradient(
         // Calculate exact eval value.
         if (eval > -1)
         {
-          if (iteration - 1 % range == 0)
+          if (didRecalculate)
           {
             eval = metric.Evaluate(transformedDataset.col(i),
                      transformedDataset.col(targetNeighbors(j, i))) -