TY - GEN
T1 - On Switching Selection Methods to Increase Parsimony Pressure
AU - de Lima, Allan
AU - Carvalho, Samuel
AU - Dias, Douglas Mota
AU - Sullivan, Joseph P.
AU - Ryan, Conor
N1 - Publisher Copyright:
© 2023 by SCITEPRESS – Science and Technology Publications, Lda.
PY - 2023
Y1 - 2023
N2 - We proposed a novel and simple selection system that alternates between tournament and Lexicase selection to tackle the bloat issue. In this way, we used Lexi2, an implementation of Lexicase with lexicographic parsimony pressure, adopting the number of nodes in our solutions as size measurement. In addition, we increased the parsimony pressure by adding a penalty, also based on the number of nodes, to the aggregated fitness score. We analysed different scenarios, including some without extra parameters, in five benchmark problems: 2-bit Multiplier, 5-bit Parity, Car Evaluation, LED and Heart Disease. We succeeded in all of them in at least one scenario, reducing the size significantly while maintaining fitness. Beyond error and size, we also included results for the average number of fitness cases used in each generation.
AB - We proposed a novel and simple selection system that alternates between tournament and Lexicase selection to tackle the bloat issue. In this way, we used Lexi2, an implementation of Lexicase with lexicographic parsimony pressure, adopting the number of nodes in our solutions as size measurement. In addition, we increased the parsimony pressure by adding a penalty, also based on the number of nodes, to the aggregated fitness score. We analysed different scenarios, including some without extra parameters, in five benchmark problems: 2-bit Multiplier, 5-bit Parity, Car Evaluation, LED and Heart Disease. We succeeded in all of them in at least one scenario, reducing the size significantly while maintaining fitness. Beyond error and size, we also included results for the average number of fitness cases used in each generation.
KW - Bloat
KW - Lexicase
KW - Tournament
UR - http://www.scopus.com/inward/record.url?scp=85188240424&partnerID=8YFLogxK
U2 - 10.5220/0012188900003595
DO - 10.5220/0012188900003595
M3 - Conference contribution
AN - SCOPUS:85188240424
T3 - International Joint Conference on Computational Intelligence
SP - 96
EP - 107
BT - Proceedings of the 15th International Joint Conference on Computational Intelligence, IJCCI 2023
A2 - van Stein, Niki
A2 - Marcelloni, Francesco
A2 - Lam, H. K.
A2 - Cottrell, Marie
A2 - Filipe, Joaquim
PB - Science and Technology Publications, Lda
T2 - 15th International Joint Conference on Computational Intelligence, IJCCI 2023
Y2 - 13 November 2023 through 15 November 2023
ER -