PndMvaCluster.h

Go to the documentation of this file.

/* ***************************************
 * Clustering algorithms                 *
 * Author: M.Babai@rug.nl                *
 * Version:                              *
 * LICENSE:                              *
 * ***************************************
 */
//#pragma once
#ifndef PND_MVA_CLUSTER_H
#define PND_MVA_CLUSTER_H

#include <iostream>
#include <cassert>
#include <vector>
#include <set>
#include <limits>

// Local includes
#include "PndMvaUtil.h"

#define PNDMVA_CLUSTER_DEBUG 0

typedef std::vector<std::pair<std::string, std::vector<float> *>> DataPoints;

typedef enum ClusteringType {
  KMEANS_HARD = 0,
  KMEANS_SOFT = 1 // Not Implemented yet.
} ClusteringType;

//---------------- Class definition --------------
class PndMvaCluster {
  //--------------------------------------------
  // -------------- public members -------------
 public:
  explicit PndMvaCluster(DataPoints const &InputData, size_t nCluster);

  explicit PndMvaCluster(DataPoints const &InputData, size_t nCluster, bool const prune, bool const forceLabels);

  virtual ~PndMvaCluster();

  virtual DataPoints *Cluster(ClusteringType const ClType = KMEANS_HARD);

  virtual DataPoints *ClusterAndLabel(ClusteringType const ClType, std::vector<std::string> const &labels);

  //------- Getters
  inline size_t GetNumberOfClusters() const;

  inline size_t GetClusterDimension() const;

  //------- Setters
  inline void SetNumberOfClusters(size_t val);

  inline void Setprune(bool const prune = false);

  inline void SetForceToLabel(bool const forceLabel = false);

  //__________________ DEBUG FUNCTIONS ______________
#if (PNDMVA_CLUSTER_DEBUG > 0)

  void printStructs();
#endif
  //--------------------------------------------
  // protected:
  // -------------- private members ------------
 private:
  PndMvaCluster(const PndMvaCluster &other);
  PndMvaCluster &operator=(const PndMvaCluster &other);

  // Functions & Procedures
  // Performs the actual hard K-Means clustering.
  DataPoints *K_Means();

  // Initialize the centroids before clustering.
  void InitCentroids();

  // Partitions the data points among the current cluster centroids.
  void InitialPartition();

  // Compute (modify) the coordinates of centroids.
  void ComputeCentroids();

  // Set all dimensions to zero
  void ResetCenteroids();

  // Clear the currently used data structures.
  void ClearStructures();

  // Init empty Centroid to the furthest point.
  void ReInitEmptyCenter(size_t centerIdx);

  // Variables
  size_t m_num_Cluster;  
  DataPoints m_PointSet; 
  size_t m_dimension;    

  // Container to hold the centroid.
  DataPoints m_Centroids;

  // Connection of each point to a centroid.
  std::vector<size_t> m_PointsToClusters;

  // Responsibility list of each centroid.
  std::vector<std::set<size_t> *> m_ClustersToPoints;

  bool m_prune;        // If prune the current cluster.
  bool m_forceToLabel; // Force to label the current mean
};

//__________________ Inlines ____________

inline size_t PndMvaCluster::GetNumberOfClusters() const
{
  return m_num_Cluster;
};

inline size_t PndMvaCluster::GetClusterDimension() const
{
  return m_dimension;
};

inline void PndMvaCluster::SetNumberOfClusters(size_t val)
{
  m_num_Cluster = val;
};

inline void PndMvaCluster::Setprune(bool const prune)
{
  m_prune = prune;
};
inline void PndMvaCluster::SetForceToLabel(bool const label)
{
  m_forceToLabel = label;
};
#endif // End interface