Portfolio Algorithms for Combinatorial Optimization
Portfoliové algoritmy pro kombinatorickou optimalizaci
Authors
Supervisors
Reviewers
Editors
Other contributors
Journal Title
Journal ISSN
Volume Title
Publisher
České vysoké učení technické v Praze
Czech Technical University in Prague
Czech Technical University in Prague
Date
Abstract
Tato diplomová práce zkoumá vybrané portfolio algoritmy pro kombinatorickou optimalizaci.
Cílem je zjistit, zda tyto algoritmy poskytují efektivní alternativu k
nejčastěji používaným restartovacím strategiím. Porovnává několik verzí MetaMax
algoritmu, s fixní i neomezenou velikostí portfolia a algoritmus
MultiEA, které používají online plánování. Jsou také navrženy dvě modifikované
varianty MultiEA algoritmu. Do srovnání je zahrnut i zástupce bandit strategií,
algoritmus Epsilon-greedy, jehož plán je předem znám a dvě restartovací
strategie. Všechny tyto algoritmy jsou testovány na skupině úloh kombinatorické
optimalizace a je vyhodnocena jejich rychlost konvergence a také schopnost najít
řešení.
This diploma thesis studies selected portfolio algorithms for combinatorial optimization. The goal is to find out whether these algorithms provide an efficient alternative to the most commonly used restarting strategies. It compares several versions of MetaMax algorithm with fixed and also unbounded portfolio sizes and the MultiEA algorithm, that both use an online schedule creation. Two modified variants of MultiEA algorithm are also proposed. A representative of bandit strategies, Epsilon-greedy algorithm, whose schedule is known in advance, and two different restarting strategies are included in comparison as well. All these algorithms are tested on the set of combinatorial optimization tasks and their convergence rate as well as the ability to find a solution is evaluated.
This diploma thesis studies selected portfolio algorithms for combinatorial optimization. The goal is to find out whether these algorithms provide an efficient alternative to the most commonly used restarting strategies. It compares several versions of MetaMax algorithm with fixed and also unbounded portfolio sizes and the MultiEA algorithm, that both use an online schedule creation. Two modified variants of MultiEA algorithm are also proposed. A representative of bandit strategies, Epsilon-greedy algorithm, whose schedule is known in advance, and two different restarting strategies are included in comparison as well. All these algorithms are tested on the set of combinatorial optimization tasks and their convergence rate as well as the ability to find a solution is evaluated.