small update

src/mlpack/methods/lmnn/constraints.hpp

@@ -118,11 +118,6 @@ class Constraints
    */
   void Triplets(arma::Mat<size_t>& outputMatrix);
 
-  //! Get the dataset reference.
-  const arma::mat& Dataset() const { return dataset; }
-  //! Modify the dataset reference.
-  arma::mat& Dataset() { return dataset; }
-
   //! Access the value of k.
   const size_t& K() const { return k; }
   //! Modify the value of k.

src/mlpack/methods/lmnn/constraints_impl.hpp

@@ -23,8 +23,8 @@ Constraints<MetricType>::Constraints(
     const arma::mat& dataset,
     const arma::Row<size_t>& labels,
     size_t k) :
-    dataset(dataset),
-    labels(labels),
+    dataset(math::MakeAlias(const_cast<arma::mat&>(dataset), false)),
+    labels(math::MakeAlias(const_cast<arma::Row<size_t>&>(labels), false)),
     k(k)
 {
   // Ensure a valid k is passed.

src/mlpack/methods/lmnn/lmnn_function.hpp

@@ -15,7 +15,6 @@
 
 #include <mlpack/prereqs.hpp>
 #include <mlpack/core/metrics/lmetric.hpp>
-#include "constraints.hpp"
 
 namespace mlpack {
 namespace lmnn {
@@ -188,8 +187,6 @@ class LMNNFunction
   size_t iteration;
   //! Range after which impostors need to be recalculated.
   size_t range;
-  // Constraints object.
-  Constraints<MetricType> constraint;
   //! Holds pre-calculated cij.
   arma::mat p_cij;
   /**

src/mlpack/methods/lmnn/lmnn_function_impl.hpp

@@ -13,6 +13,7 @@
 #define MLPACK_METHODS_LMNN_FUNCTION_IMPL_HPP
 
 #include "lmnn_function.hpp"
+#include "constraints.hpp"
 
 #include <mlpack/core/math/make_alias.hpp>
 #include <mlpack/core/optimizers/function.hpp>
@@ -32,8 +33,7 @@ LMNNFunction<MetricType>::LMNNFunction(const arma::mat& dataset,
     metric(metric),
     regularization(regularization),
     iteration(0),
-    range(1),
-    constraint(dataset, labels, k)
+    range(1)
 {
   // Initialize the initial learning point.
   initialPoint.eye(dataset.n_rows, dataset.n_rows);
@@ -44,7 +44,7 @@ LMNNFunction<MetricType>::LMNNFunction(const arma::mat& dataset,
   targetNeighbors = arma::Mat<size_t>(k, dataset.n_cols, arma::fill::zeros);
   impostors = arma::Mat<size_t>(k, dataset.n_cols, arma::fill::zeros);
   distance = arma::mat(k, dataset.n_cols, arma::fill::zeros);
-
+  Constraints<MetricType> constraint(dataset, labels, k);
   constraint.TargetNeighbors(targetNeighbors);
   constraint.Impostors(impostors);
 
@@ -68,7 +68,7 @@ void LMNNFunction<MetricType>::Shuffle()
   labels = std::move(newLabels);
 
   // Re-calculate target neighbors as indices changed.
-  constraint.Dataset() = dataset;
+  Constraints<MetricType> constraint(dataset, labels, k);
   constraint.TargetNeighbors(targetNeighbors);
 }
 
@@ -84,7 +84,7 @@ double LMNNFunction<MetricType>::Evaluate(const arma::mat& transformation)
   if (iteration % range == 0)
   {
     // Re-calculate impostors on transformed dataset.
-    constraint.Dataset() = transformedDataset;
+    Constraints<MetricType> constraint(transformedDataset, labels, k);
     constraint.Impostors(impostors, distance);
   }
 
@@ -151,7 +151,7 @@ double LMNNFunction<MetricType>::Evaluate(const arma::mat& transformation,
   if (iteration % range == 0)
   {
     // Re-calculate impostors on transformed dataset.
-    constraint.Dataset() = transformedDataset;
+    Constraints<MetricType> constraint(transformedDataset, labels, k);
     constraint.Impostors(impostors, distance, begin, batchSize);
   }
 
@@ -322,7 +322,7 @@ double LMNNFunction<MetricType>::EvaluateWithGradient(
   if (iteration % range == 0)
   {
     // Re-calculate impostors on transformed dataset.
-    constraint.Dataset() = transformedDataset;
+    Constraints<MetricType> constraint(transformedDataset, labels, k);
     constraint.Impostors(impostors, distance);
   }
 
@@ -409,7 +409,7 @@ double LMNNFunction<MetricType>::EvaluateWithGradient(
   if (iteration % range == 0)
   {
     // Re-calculate impostors on transformed dataset.
-    constraint.Dataset() = transformedDataset;
+    Constraints<MetricType> constraint(transformedDataset, labels, k);
     constraint.Impostors(impostors, distance, begin, batchSize);
   }
 

src/mlpack/methods/lmnn/lmnn_main.cpp

@@ -100,7 +100,7 @@ PROGRAM_INFO("Large Margin Nearest Neighbors (LMNN)",
     "literature on L-BFGS.  In addition, a normalized starting point can be "
     "used by specifying the " + PRINT_PARAM_STRING("normalize") + " parameter."
     "\n\n"
-    "By default, the L-BFGS optimizer is used."
+    "By default, the AMSGrad optimizer is used."
     "\n\n"
     "Example - Let's say we want to learn distance on iris dataset with "
     "number of targets as 3 using BigBatch_SGD optimizer. A simple call for "
@@ -121,7 +121,7 @@ PARAM_INT_IN("num_targets", "Number of target neighbors to use for each "
     "datapoint.", "k", 1);
 PARAM_MATRIX_OUT("output", "Output matrix for learned distance matrix.", "o");
 PARAM_STRING_IN("optimizer", "Optimizer to use; 'amsgrad', 'bbsgd', 'sgd', or "
-    "'lbfgs'.", "O", "lbfgs");
+    "'lbfgs'.", "O", "amsgrad");
 PARAM_DOUBLE_IN("regularization", "Regularization for LMNN objective function ",
     "r", 0.5);
 PARAM_FLAG("normalize", "Use a normalized starting point for optimization. It"

src/mlpack/tests/lmnn_test.cpp

@@ -392,26 +392,27 @@ double KnnAccuracy(const arma::mat& dataset,
   knn.Search(k, neighbors, distances);
 
   // Keep count.
-  double count = 0.0;
+  size_t count = 0.0;
 
   for (size_t i = 0; i < dataset.n_cols; i++)
   {
-    arma::Row<size_t> Map;
+    arma::vec Map;
     Map.zeros(uniqueLabels.n_cols);
 
-    for (size_t j=0; j < k; j++)
-      Map(labels(neighbors(j, i))) += std::exp(1 / (j + 1));
+    for (size_t j = 0; j < k; j++)
+      Map(labels(neighbors(j, i))) +=
+          1 / std::pow(distances(j, i) + 1, 2);
 
-    arma::vec index = arma::conv_to<arma::vec>::from(arma::find(Map
+    size_t index = arma::conv_to<size_t>::from(arma::find(Map
         == arma::max(Map)));
 
     // Increase count if labels match.
-    if (index(0) == labels(i))
+    if (index == labels(i))
         count++;
   }
 
   // return accuracy.
-  return (count / dataset.n_cols) * 100;
+  return ((double) count / dataset.n_cols) * 100;
 }
 
 // Check that final accuracy is greater than initial accuracy on