Selector::SelectionMetrics Class Reference

Class to extract metrics from either the agent or the environment. More...

#include <selectionMetrics.h>

Inheritance diagram for Selector::SelectionMetrics:

Public Member Functions
	SelectionMetrics ()=default
	Default constructor.
virtual	~SelectionMetrics ()=default
	Default destructor.
	SelectionMetrics (double score, double utility=0)
	Constructor with score and utility initialization.
virtual double	getScore () const
virtual double	getUtility () const
virtual void	initMetrics (const TPG::TPGVertex *agent, const Learn::LearningEnvironment &learningEnvironment)
	Init the metrics for the agent in the learning environment.
virtual void	extractMetricsStep (const TPG::TPGVertex *agent, std::vector< double > actionValues, const Learn::LearningEnvironment &learningEnvironment)
	Extract metrics from the agent in the learning environment.
virtual void	extractMetricsEpisode (const TPG::TPGVertex *agent, size_t nbStepsExecuted, const Learn::LearningEnvironment &learningEnvironment)
	Extract metrics from the agent in the learning environment.
virtual void	weightedSum (std::shared_ptr< SelectionMetrics > other, size_t nbEvaluation, size_t nbEvaluationOther)
	Perform a weighted sum between this SelectionMetrics and another.

Static Public Member Functions
template<class T >
static T	weightedSum (T value, T valueOther, size_t nbEvaluation, size_t nbEvaluationOther)
	Perform a weighted sum between 2 values.

Protected Attributes
double	score = 0
double	utility = 0

Friends
bool	operator< (const SelectionMetrics &lhs, const SelectionMetrics &rhs)
	Comparison function to enable sorting of SelectionMetrics with STL.

Detailed Description

Class to extract metrics from either the agent or the environment.

This metrics can be used to specify the selection of the selector. This class does not implement any metrics, it need to be override by the different selection methods.

Constructor & Destructor Documentation

SelectionMetrics()

Selector::SelectionMetrics::SelectionMetrics ( double score, double utility = 0 )

inline

Constructor with score and utility initialization.

Parameters

[in]	score	the score obtained by the agent.
[in]	utility	the utility obtained by the agent.

Member Function Documentation

extractMetricsEpisode()

void Selector::SelectionMetrics::extractMetricsEpisode ( const TPG::TPGVertex * agent, size_t nbStepsExecuted, const Learn::LearningEnvironment & learningEnvironment )

virtual

Extract metrics from the agent in the learning environment.

This method is called at the end of every episode of the environment evaluation.

Parameters

[in]	agent	the TPGVertex representing the agent.
[in]	nbStepsExecuted	number of steps executed at the end of the episode.
[in]	learningEnvironment	the learning environment in which the agent is evaluated.

Reimplemented in Selector::ClassificationSelectionMetrics, and Selector::MapElites::MapElitesSelectionMetrics.

extractMetricsStep()

virtual void Selector::SelectionMetrics::extractMetricsStep ( const TPG::TPGVertex * agent, std::vector< double > actionValues, const Learn::LearningEnvironment & learningEnvironment )

inlinevirtual

Extract metrics from the agent in the learning environment.

This method is called at every step of the environment evaluation.

Parameters

[in]	agent	the TPGVertex representing the agent.
[in]	actionValues	the action values taken by the agent.
[in]	learningEnvironment	the learning environment in which the agent is evaluated.

Reimplemented in Selector::TimingSelectionMetrics, and Selector::MapElites::MapElitesSelectionMetrics.

getScore()

double Selector::SelectionMetrics::getScore ( ) const

virtual

Return the score of the agent.

Reimplemented in Selector::TimingSelectionMetrics.

getUtility()

double Selector::SelectionMetrics::getUtility ( ) const

virtual

Return the utility of the agent.

Reimplemented in Selector::TimingSelectionMetrics.

initMetrics()

virtual void Selector::SelectionMetrics::initMetrics ( const TPG::TPGVertex * agent, const Learn::LearningEnvironment & learningEnvironment )

inlinevirtual

Init the metrics for the agent in the learning environment.

This method is called at the beginning of the evaluateJob method.

Parameters

[in]	agent	the TPGVertex representing the agent.
[in]	learningEnvironment	the learning environment in which the agent is evaluated.

Reimplemented in Selector::ClassificationSelectionMetrics, and Selector::MapElites::MapElitesSelectionMetrics.

weightedSum() [1/2]

void Selector::SelectionMetrics::weightedSum ( std::shared_ptr< SelectionMetrics > other, size_t nbEvaluation, size_t nbEvaluationOther )

virtual

Perform a weighted sum between this SelectionMetrics and another.

Parameters

[in]	other	the other SelectionMetrics to combine with this.
[in]	nbEvaluation	the number of evaluation used to obtain this SelectionMetrics.
[in]	nbEvaluationOther	the number of evaluation used to obtain the other SelectionMetrics.

Reimplemented in Selector::ClassificationSelectionMetrics, Selector::MapElites::MapElitesSelectionMetrics, and Selector::TimingSelectionMetrics.

weightedSum() [2/2]

template<class T >

static T Selector::SelectionMetrics::weightedSum ( T value, T valueOther, size_t nbEvaluation, size_t nbEvaluationOther )

inlinestatic

Perform a weighted sum between 2 values.

Parameters

[in]	value	the value to combine.
[in]	valueOther	the other value to combine.
[in]	nbEvaluation	the number of evaluations to obtain value.
[in]	nbEvaluationOther	the number of evaluations to obtain valueOther.

Member Data Documentation

score

double Selector::SelectionMetrics::score = 0

protected

Score obtained by the agent at the end of an episode

utility

double Selector::SelectionMetrics::utility = 0

protected

Double value for the utility. Utility is used only if the LearningEnvironment override the specific method. The utility represent a score needed for logs but not for learning.

It can be used for comparison with different score rules for example.

---

The documentation for this class was generated from the following files:

• D:/a/gegelati/gegelati/gegelatilib/include/selector/selectionMetrics.h
• D:/a/gegelati/gegelati/gegelatilib/src/selector/selectionMetrics.cpp