File heuristic.hpp

namespace seco
class IHeuristic
#include <heuristic.hpp>

Defines an interface for all heuristics that allows to calculate numerical scores that assess the quality of rules, based on the elements of confusion matrices. Given the elements of a confusion matrix, such a heuristic calculates a numerical score in [0, 1].

All heuristics must be implemented as gain metrics, i.e., rules with a greater numerical score are considered better than those with a smaller numerical score.

All heuristics must treat positive and negative labels equally, i.e., if the ground truth and a rule’s predictions would be inverted, the resulting numerical scores must be the same as before.

Public Functions

inline virtual ~IHeuristic()
virtual float64 evaluateConfusionMatrix(float64 cin, float64 cip, float64 crn, float64 crp, float64 uin, float64 uip, float64 urn, float64 urp) const = 0

Calculates and returns a numerical score in [0, 1] given the elements of a confusion matrix. All elements must be equal to or greater than 0. If a rule does not cover any elements, i.e., if CIN + CIP + CRN + CRP == 0, the worst possible quality 0 must be returned.

According to the notation in http://www.ke.tu-darmstadt.de/bibtex/publications/show/3201, a confusion matrix consists of 8 elements, namely CIN, CIP, CRN, CRP, UIN, UIP, URN and URP. The individual symbols used in this notation have the following meaning:

  • The first symbol denotes whether the corresponding labels are covered (C) or uncovered (U) by the rule.

  • The second symbol denotes relevant (R) or irrelevant (I) labels according to the ground truth.

  • The third symbol denotes labels for which the prediction of a rule is positive (P) or negative (N).

This results in the terminology given in the following table: 

       | ground-   |           |
       | truth     | predicted |
——–&#8212;|——–&#8212;|——–&#8212;|–&#8212; covered | 0 | 0 | CIN | 0 | 1 | CIP | 1 | 0 | CRN | 1 | 1 | CRP ——–&#8212;|——–&#8212;|——–&#8212;|–&#8212; uncovered | 0 | 0 | UIN | 0 | 1 | UIP | 1 | 0 | URN | 1 | 1 | URP

Real numbers may be used for the individual elements, if real-valued weights are assigned to the corresponding labels.

Parameters:

  • cin – The number of covered (C) labels that are irrelevant (I) according to the ground truth and for which the prediction in the rule’s head is negative (N)

  • cip – The number of covered (C) labels that are irrelevant (I) according to the ground truth and for which the prediction in the rule’s head is negative (N)

  • crn – The number of covered (C) labels that are relevant (R) according to the ground truth and for which the prediction in the rule’s head is negative (N)

  • crp – The number of covered (C) labels that are relevant (R) according to the ground truth and for which the prediction in the rule’s head is positive (P)

  • uin – The number of uncovered (U) labels that are irrelevant (I) according to the ground truth and for which the prediction in the rule’s head is negative (N)

  • uip – The number of uncovered (U) labels that are irrelevant (I) according to the ground truth and for which the prediction in the rule’s head is positive (P)

  • urn – The number of uncovered (U) labels that are relevant (R) according to the ground truth and for which the prediction in the rule’s head is negative (N)

  • urp – The number of uncovered (U) labels that are relevant (R) according to the ground truth and for which the prediction in the rule’s head is positive (P)

Returns:

The quality that has been calculated

class IHeuristicFactory
#include <heuristic.hpp>

Defines an interface for all factories that allow to create instances of the type IHeuristic.

Public Functions

inline virtual ~IHeuristicFactory()
virtual std::unique_ptr<IHeuristic> create() const = 0

Creates and returns a new object of type IHeuristic.

Returns:

An unique pointer to an object of type IHeuristic that has been created

class IHeuristicConfig
#include <heuristic.hpp>

Defines an interface for all classes that allow to configure a heuristic.

Subclassed by seco::AccuracyConfig, seco::FMeasureConfig, seco::LaplaceConfig, seco::MEstimateConfig, seco::PrecisionConfig, seco::RecallConfig, seco::WraConfig

Public Functions

inline virtual ~IHeuristicConfig()
virtual std::unique_ptr<IHeuristicFactory> createHeuristicFactory() const = 0

Creates and returns a new object of type IHeuristicFactory according to the specified configuration.

Returns:

An unique pointer to an object of type IHeuristicFactory that has been created