A Statistical Approach for Algorithm Selection

Pérez, Joaquín; Pazos, Rodolfo; Frausto, Juan; Rodríguez, G.; Romero, David; Cruz, Laura

doi:10.1007/978-3-540-24838-5_31

Cited by 18 publications

(4 citation statements)

References 8 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…They used five deterministic heuristics and two non-deterministic algorithms with 1D-BPP. Three machine learning methods are compared -Discriminant Analysis (DA) (Pérez et al, 2004), a decision tree to build the selectors and a Self-Organizing Map (SOM) (Haykin et al, 2009) to implement the selection system with feedback. Five features were used as input.…”

Section: Asp With Feature-based Approachesmentioning

confidence: 99%

Automated Algorithm Selection: from Feature-Based to Feature-Free Approaches

Alissa¹,

Sim²,

Hart³

2022

Preprint

View full text Add to dashboard Cite

We propose a novel technique for algorithm-selection, applicable to optimisation domains in which there is implicit sequential information encapsulated in the data, e.g., in online bin-packing. Specifically we train two types of recurrent neural networks to predict a packing heuristic in online binpacking, selecting from four well-known heuristics. As input, the RNN methods only use the sequence of item-sizes. This contrasts to typical approaches to algorithm-selection which require a model to be trained using domain-specific instance features that need to be first derived from the input data. The RNN approaches are shown to be capable of achieving within 5% of the oracle performance on between 80.88% to 97.63% of the instances, depending on the dataset. They are also shown to outperform classical machine learning models trained using derived features. Finally, we hypothesise that the proposed methods perform well when the instances exhibit some implicit structure that results in discriminatory performance with respect to a set of heuristics. We test this hypothesis by generating fourteen new datasets with increasing levels of structure, and show that there is a critical threshold of structure required before algorithm-selection delivers benefit.

show abstract

Section: Asp With Feature-based Approachesmentioning

confidence: 99%

Automated Algorithm Selection: from Feature-Based to Feature-Free Approaches

Alissa¹,

Sim²,

Hart³

2022

Preprint

View full text Add to dashboard Cite

show abstract

“…There are two considered variables for characterizing the problem structure, which are proposed in [ 16 ]. The first b characterizes the proportion of the total size of the objects that can be assigned to one container.…”

Section: Characterizing and Analyzing The Relation Bin-packing Problementioning

confidence: 99%

Characterizing and Analyzing the Relation Between Bin-Packing Problem and Tabu Search Algorithm

Landero¹,

Ríos²,

Pérez

et al. 2020

Computational Science and Its Applications – ICCSA 2020

Self Cite

View full text Add to dashboard Cite

The relation between problem and solution algorithm presents a similar phenomenon in different research problems (optimization, decision, classification, ordering); the algorithm performance is very good in some cases of the problem, and very bad in other. Majority of related works have worked for predicting the most adequate algorithm to solve a new problem instance. However, the relation between problem and algorithm is not understood at all. In this paper a formal characterization of this relation is proposed to facilitate the analysis and understanding of the phenomenon. Case studies for Tabu Search algorithm and One Dimension Bin Packing problem were performed, considering three important sections of algorithm logical structure. Significant variables of problem structure and algorithm searching behavior from past experiments, metrics known by scientific community were considered (Autocorrelation Coefficient and Length) and significant variables of algorithm operative behavior were proposed. The models discovered in the case studies gave guidelines that permits to redesign algorithm logical structure, which outperforms to the original algorithm in an average of 69%. The proposed characterization for the relation problem-algorithm could be a formal procedure for obtaining guidelines that improves the algorithm performance.

show abstract

“…The meta-learning approach is oriented to learning about classification using machine learning methods; three methods are explored to solve an optimization problem: Discriminant Analysis (Pérez, 2004), C4.5 and the Self-Organising Neural Network. The hyper-heuristic approach is oriented to automatically produce an adequate combination of available lowlevel heuristics in order to effectively solve a given instance (Burke et al, 2010); a hyper-heuristic strategy is incorporated in an ant colony algorithm to select the heuristic that best adjust one of its control parameter.…”

Section: Approaches To Building Algorithm Selectorsmentioning

confidence: 99%