RocCurve.hh

Go to the documentation of this file.

#ifndef _ROCCURVEHH_
#define _ROCCURVEHH_

#include <openmodeller/om_defs.hh>
#include <openmodeller/Configuration.hh>

#include <openmodeller/Model.hh>

#include <openmodeller/Occurrences.hh>
#include <openmodeller/Environment.hh>
#include <openmodeller/Sampler.hh>

#include <map>

#define ROC_DEFAULT_RESOLUTION 15
#define ROC_DEFAULT_BACKGROUND_POINTS 10000

class dllexp RocCurve
{
public:
    RocCurve();

  ~RocCurve();

  void initialize( int resolution=ROC_DEFAULT_RESOLUTION );

  void initialize( int resolution, int num_background_points );

  void initialize( int resolution, bool use_absences_as_background );

  void reset();

  void calculate( const Model& model, const SamplerPtr& sampler );
  
  int numPoints() const { return _data.size(); }

  double getX( int point_index ) const { return _data[point_index][0]; }

  double getY( int point_index ) const { return _data[point_index][1]; }

  double getTotalArea();

  double getPartialAreaRatio( double e=1.0 );

  bool ready() const { return _ready; }

  ConfigurationPtr getConfiguration() const;

private:

  struct VectorCompare {

    bool operator () ( const std::vector<Scalar> &a, const std::vector<Scalar> &b ) const {

      if ( a[0] != b[0] ) {

        return a[0] < b[0]; // Compare 1 - specificity.
      }
      else {

        return a[1] < b[1]; // Compare sensitivity (1 - specificity values are equal).
      }
    }
  };

  void _loadPredictions( const Model& model, const SamplerPtr& sampler );
 
  void _calculateGraphPoints(); 

  bool _calculateTotalArea(); 
  
  std::vector<int> _category;      // 0=absence, 1=presence
  std::vector<Scalar> _prediction; // associated probabilities

  std::vector< std::vector<Scalar> > _data; // Main data structure to store all points
  
  int _resolution; // Number of points on the curve

  int _approach; // Approach to be used: 0=undefined, 1=traditional (presence x absence), 2=proportional area

  int _num_background_points; // Number of background points to be generated when there are no absences. Only for proportional area approach.
  bool _use_absences_as_background; // Indicates if absence points should be used as background points. Only for proportional area approach.

  int _true_negatives; // Number of true negatives (binarized)
  int _true_positives; // Number of true positives (binarized)

  double _auc; // Area under the curve. Need to store this to avoid recalculating in serialization.

  std::map<double, double> _ratios; // Ratios calculated via getPartialAreaRatio (max omission <=> ratio). Ratios are stored here to be used during serialization.

  std::vector<Scalar> _thresholds; // Thresholds in ascending order

  std::vector<Scalar> _proportions; // Proportional area for each point

  bool _ready;
};

#endif