Abstract : Given a collection of algorithms, the Algorithm Selection (AS) problem consists in identifying which of them is the best one for solving a given problem. The selection depends on a set of numerical features that characterize the problem to solve. In this paper we show the impact of feature selection techniques on the performance of the SUNNY algorithm selector, taking as reference the benchmarks of the AS library (ASlib). Results indicate that a handful of features is enough to reach similar, if not better, performance of the original SUNNY approach that uses all the available features. We also present sunny-as: a tool for using SUNNY on a generic ASlib scenario.
https://hal.inria.fr/hal-01227600
Contributor : Amadini Roberto <>
Submitted on : Wednesday, November 11, 2015 - 4:59:53 PM Last modification on : Friday, October 30, 2020 - 12:04:03 PM Long-term archiving on: : Friday, April 28, 2017 - 5:12:26 AM
Roberto Amadini, Fabio Biselli, Maurizio Gabbrielli, Tong Liu, Jacopo Mauro. Feature Selection for SUNNY: a Study on the Algorithm Selection Library. ICTAI, Nov 2015, Vietri sul Mare, Italy. ⟨hal-01227600⟩