Specialization of the selection class for classification purposes. More...

#include <classificationSelector.h>

Inheritance diagram for Selector::ClassificationSelector:

Public Member Functions
	ClassificationSelector (std::shared_ptr< TPG::TPGGraph > graph, const Learn::LearningParameters &params, uint64_t nbActions)
	Constructor for ClassificationSelector.

virtual std::shared_ptr< SelectionMetrics >	createSelectionMetrics () override

virtual void	doSelection (std::multimap< std::shared_ptr< Learn::EvaluationResult >, const TPG::TPGVertex * > &results, Mutator::RNG &rng) override
	Specialization of the doSelection method for classification purposes.

Public Member Functions inherited from Selector::Selector
	Selector (std::shared_ptr< TPG::TPGGraph > graph, const Learn::LearningParameters &params)
	Constructor for Selector.

virtual void	launchSelection (std::multimap< std::shared_ptr< Learn::EvaluationResult >, const TPG::TPGVertex * > &results, Mutator::RNG &rng)
	This method execute the doSelection method.

virtual void	keepBestPolicy ()
	This method keeps only the bes tRoot policy in the TPGGraph.

virtual void	updateEvaluationRecords (const std::multimap< std::shared_ptr< Learn::EvaluationResult >, const TPG::TPGVertex * > &results)
	Update the bestRoot and resultsPerRoot attributes.

virtual void	updateResultsPerRoot (const std::multimap< std::shared_ptr< Learn::EvaluationResult >, const TPG::TPGVertex * > &results)
	Update the resultsPerRoot.

virtual void	updateBestRoot (const std::multimap< std::shared_ptr< Learn::EvaluationResult >, const TPG::TPGVertex * > &results)
	Update the bestRoot attribute.

virtual const std::pair< const TPG::TPGVertex *, std::shared_ptr< Learn::EvaluationResult > > &	getBestRoot () const
	Get the best root TPG::Vertex encountered since the last init.

virtual std::shared_ptr< TPG::TPGGraph >	getGraph ()
	Getter for the TPGGraph built by the LearningAgent.

virtual void	forgetPreviousResults ()
	This method resets the previous registered scores per root.

virtual const std::map< const TPG::TPGVertex *, std::shared_ptr< Learn::EvaluationResult > > &	getResultsPerRoot () const
	Return the resultsPerRoot map.

virtual const SelectionContext &	updateContext ()
	Update the SelectionContext structure and return it.

virtual void	updateAfterPopulate ()
	Method to call at the end of TPGMutator::populateTPG.

Protected Attributes
uint64_t	nbActions
	Number of actions in the LearningEnvironment.

Protected Attributes inherited from Selector::Selector
std::shared_ptr< TPG::TPGGraph >	graph
	TPGGraph on which the TPGVertex can be selected or deleted.

const Learn::LearningParameters &	params
	Parameters for the selection.

std::pair< const TPG::TPGVertex *, std::shared_ptr< Learn::EvaluationResult > >	bestRoot {nullptr, nullptr}

std::map< const TPG::TPGVertex *, std::shared_ptr< Learn::EvaluationResult > >	resultsPerRoot
	Map associating root TPG::TPGVertex to their EvaluationResult.

SelectionContext	context
	context used by the TPGMutator to populate the TPGGraph.

Detailed Description

Specialization of the selection class for classification purposes.

Constructor & Destructor Documentation

◆ ClassificationSelector()

Selector::ClassificationSelector::ClassificationSelector	(	std::shared_ptr< TPG::TPGGraph >	graph,
		const Learn::LearningParameters &	params,
		uint64_t	nbActions )

inline

Constructor for ClassificationSelector.

Parameters

[in]	graph	shared pointer of the graph on which the selection is done.
[in]	params	parameters used by the Selector.
[in]	nbActions	number of actions in the LearningEnvironment.

Member Function Documentation

◆ createSelectionMetrics()

std::shared_ptr< Selector::SelectionMetrics > Selector::ClassificationSelector::createSelectionMetrics ( )

overridevirtual

Specialization of createSelectionMetrics

Creates and return an instance of ClassificationSelectionMetrics

Copyright or © or Copr. IETR/INSA - Rennes (2025) :

Quentin Vacher qvach.nosp@m.er@i.nosp@m.nsa-r.nosp@m.enne.nosp@m.s.fr (2025)

GEGELATI is an open-source reinforcement learning framework for training artificial intelligence based on Tangled Program Graphs (TPGs).

This software is governed by the CeCILL-C license under French law and abiding by the rules of distribution of free software. You can use, modify and/ or redistribute the software under the terms of the CeCILL-C license as circulated by CEA, CNRS and INRIA at the following URL "http://www.cecill.info".

As a counterpart to the access to the source code and rights to copy, modify and redistribute granted by the license, users are provided only with a limited warranty and the software's author, the holder of the economic rights, and the successive licensors have only limited liability.

In this respect, the user's attention is drawn to the risks associated with loading, using, modifying and/or developing or reproducing the software by the user in light of its specific status of free software, that may mean that it is complicated to manipulate, and that also therefore means that it is reserved for developers and experienced professionals having in-depth computer knowledge. Users are therefore encouraged to load and test the software's suitability as regards their requirements in conditions enabling the security of their systems and/or data to be ensured and, more generally, to use and operate it in the same conditions as regards security.

The fact that you are presently reading this means that you have had knowledge of the CeCILL-C license and that you accept its terms.

Reimplemented from Selector::Selector.

◆ doSelection()

void Selector::ClassificationSelector::doSelection	(	std::multimap< std::shared_ptr< Learn::EvaluationResult >, const TPG::TPGVertex * > &	results,
		Mutator::RNG &	rng )

overridevirtual

Specialization of the doSelection method for classification purposes.

During the decimation process, roughly half of the roots are kept based on their score for individual class of the ClassificationLearningEnvironment. To do so, for each class of the ClassificationLearningEnvironment, the roots provided the best score are preserved during the decimation process even if their global score over all classes is not among the best.

The remaining half of preserved roots is selected using the general score obtained over all classes.

This per-class preservation is activated only if there is a sufficient number of root vertices in the TPGGraph after decimation to guarantee that all classes are preserved equally. In other word, the same number of root is marked for preservation for each class, which can only be achieved if the number of roots to preserve during the decimation process is superior or equal to twice the number of actions of the ClassificationLearningEnvironment. If an insufficient number of root is preserved during the decimation process, all roots are preserved based on their general score.

The results map is updated by the method to keep only the results of non-decimated roots.