Gustafson-Kessel clustering algorithm

ksiminski · Sep 18, 2023 · 13cee7f · 13cee7f
1 parent 39c018d
commit 13cee7f
Show file tree

Hide file tree

Showing 2 changed files with 20 additions and 22 deletions.
diff --git a/source/partitions/gk.cpp b/source/partitions/gk.cpp
@@ -5,6 +5,7 @@
 #include "../metrics/metric_mahalanobis.h"
 #include "../partitions/gk.h"
 #include "../auxiliary/matrix.h"
+#include "../auxiliary/vector-operators.h"
 
 std::string ksi::gk::getAbbreviation () const
 {
@@ -71,8 +72,6 @@ double ksi::gk::calculateDistance(const std::vector<double>& x, const std::vecto
    return metrics_for_clusters[cluster].calculateDistance(x, y);
 }
 
-
-
 std::vector<std::vector<double>> ksi::gk::modifyPartitionMatrix(const std::vector<std::vector<double>>& mV, const std::vector<std::vector<double>>& mX)
 {
    try
@@ -106,16 +105,6 @@ std::vector<std::vector<double>> ksi::gk::modifyPartitionMatrix(const std::vecto
             }
          }
 
-//          for (std::size_t c = 0; c < nClusters; c++)
-//          {
-//             for (std::size_t x = 0; x < nX; x++)
-//             {
-//                if (Dm[c][x] == 0)
-//                   Dmzeros[x]++;
-//                Dmsums[x] += Dm[c][x];
-//             }
-//          }
-
          for (std::size_t c = 0; c < nClusters; c++)
          {
             for (std::size_t x = 0; x < nX; x++)
@@ -169,10 +158,15 @@ ksi::partition ksi::gk::doPartition(const ksi::dataset& ds)
          {
             mV = calculateClusterCentres(mU, mX);
             std::vector<Matrix<double>> covariance_matrices = calculateCovarianceMatrices(mX, mU, mV);
-            updateMetrics(covariance_matrices, nAttr);
+            if (ksi::is_valid(covariance_matrices))          
+               updateMetrics(covariance_matrices, nAttr);
 
-            mU = modifyPartitionMatrix (mV, mX);
-            normaliseByColumns(mU);
+            auto mU_new = modifyPartitionMatrix (mV, mX);
+            if (ksi::is_valid(mU_new))
+            {
+               mU = mU_new;
+               normaliseByColumns(mU);
+            }
          }
       }
       else if (_epsilon > 0)
@@ -182,12 +176,17 @@ ksi::partition ksi::gk::doPartition(const ksi::dataset& ds)
          {
             mV = calculateClusterCentres(mU, mX);
             std::vector<Matrix<double>> covariance_matrices = calculateCovarianceMatrices(mX, mU, mV);
-            updateMetrics(covariance_matrices, nAttr);
+            if (ksi::is_valid(covariance_matrices))          
+               updateMetrics(covariance_matrices, nAttr);
 
-            auto mUnew = modifyPartitionMatrix (mV, mX);
-            normaliseByColumns(mUnew);
-            frob = Frobenius_norm_of_difference (mU, mUnew);
-            mU = mUnew;
+            auto mU_new = modifyPartitionMatrix (mV, mX);
+            if (ksi::is_valid(mU_new))
+            {
+               mU = mU_new;
+               normaliseByColumns(mU_new);
+            }
+            frob = Frobenius_norm_of_difference (mU, mU_new);
+            mU = mU_new;
          } while (frob > _epsilon);
       }
 
@@ -216,7 +215,6 @@ std::vector<ksi::Matrix<double>> ksi::gk::get_A_matrices_from_metrics(const std:
    return A_matrices;
 }
 
-
 void ksi::gk::updateMetrics(const std::vector<Matrix<double>>& covariance_matrices, const int nAttr)
 {
    metrics_for_clusters.clear();

diff --git a/source/partitions/gk.h b/source/partitions/gk.h
@@ -20,6 +20,7 @@ namespace ksi
       double _volume_rho = 1.0;  // cluster volume parameter
     public:
       gk ();
+      gk (const int nClusters, const int nIterations);
       gk (const double volume_rho);
       gk (const gk & wzor) = default;
       gk (gk && wzor) = default;
@@ -59,5 +60,4 @@ namespace ksi
    };
 }
 
-
 #endif