Publicação: A meta-learning recommender system for hyperparameter tuning: Predicting when tuning improves SVM classifiers
Carregando...
Data
Orientador
Coorientador
Pós-graduação
Curso de graduação
Título da Revista
ISSN da Revista
Título de Volume
Editor
Elsevier B.V.
Tipo
Artigo
Direito de acesso
Acesso aberto
Resumo
For many machine learning algorithms, predictive performance is critically affected by the hyperparameter values used to train them. However, tuning these hyperparameters can come at a high computational cost, especially on larger datasets, while the tuned settings do not always significantly outperform the default values. This paper proposes a recommender system based on meta-learning to identify exactly when it is better to use default values and when to tune hyperparameters for each new dataset. Besides, an in-depth analysis is performed to understand what they take into account for their decisions, providing useful insights. An extensive analysis of different categories of meta-features, meta-learners, and setups across 156 datasets is performed. Results show that it is possible to accurately predict when tuning will significantly improve the performance of the induced models. The proposed system reduces the time spent on optimization processes, without reducing the predictive performance of the induced models (when compared with the ones obtained using tuned hyperparameters). We also explain the decision-making process of the meta-learners in terms of linear separability-based hypotheses. Although this analysis is focused on the tuning of Support Vector Machines, it can also be applied to other algorithms, as shown in experiments performed with decision trees. (C) 2019 Published by Elsevier Inc.
Descrição
Palavras-chave
Meta-learning, Recommender system, Tuning recommendation, Hyperparameter tuning, Support vector machines
Idioma
Inglês
Como citar
Information Sciences. New York: Elsevier Science Inc, v. 501, p. 193-221, 2019.
