Meta learning for defaults: symbolic defaults

Jan N. van Rijn, Florian Pfisterer, Janek Thomas, Andreas Muller, Bernd Bischl, J. Vanschoren

Research output: Contribution to conferencePaperAcademic

Abstract

In this work we propose to use meta-learning to learn sets of symbolic default hyperparameter configurations that work well across many data sets. A well known example for such a symbolic default is the logarithmic relation between the number of features of a dataset and the available features per split of a Random Forest, as observed by Breiman (2001). Symbolic functions allow for a more rich vocabulary to define defaults on. In the past, symbolic and static default values have been obtained either from hand-crafted heuristics or empirical evaluations of specific
algorithms. We propose to automatically learn such symbolic configurations, i.e., formulas containing meta-features, from a large set of prior evaluations of numeric hyperparameters on multiple data sets via symbolic regression and optimization.

Workshop

WorkshopNeural Information Processing Workshop on Meta-Learning
CountryCanada
CityMontreal
Period8/12/188/12/18
Internet address

Keywords

  • Meta-learning
  • Automatic Machine Learning

Cite this

van Rijn, J. N., Pfisterer, F., Thomas, J., Muller, A., Bischl, B., & Vanschoren, J. (2018). Meta learning for defaults: symbolic defaults. Paper presented at Neural Information Processing Workshop on Meta-Learning, Montreal, Canada.
van Rijn, Jan N. ; Pfisterer, Florian ; Thomas, Janek ; Muller, Andreas ; Bischl, Bernd ; Vanschoren, J./ Meta learning for defaults : symbolic defaults. Paper presented at Neural Information Processing Workshop on Meta-Learning, Montreal, Canada.7 p.
@conference{280dadb5de6f47228a9354fb512cbf8b,
title = "Meta learning for defaults: symbolic defaults",
abstract = "In this work we propose to use meta-learning to learn sets of symbolic default hyperparameter configurations that work well across many data sets. A well known example for such a symbolic default is the logarithmic relation between the number of features of a dataset and the available features per split of a Random Forest, as observed by Breiman (2001). Symbolic functions allow for a more rich vocabulary to define defaults on. In the past, symbolic and static default values have been obtained either from hand-crafted heuristics or empirical evaluations of specificalgorithms. We propose to automatically learn such symbolic configurations, i.e., formulas containing meta-features, from a large set of prior evaluations of numeric hyperparameters on multiple data sets via symbolic regression and optimization.",
keywords = "Meta-learning, Automatic Machine Learning",
author = "{van Rijn}, {Jan N.} and Florian Pfisterer and Janek Thomas and Andreas Muller and Bernd Bischl and J. Vanschoren",
year = "2018",
month = "12",
day = "8",
language = "English",
note = "Neural Information Processing Workshop on Meta-Learning ; Conference date: 08-12-2018 Through 08-12-2018",
url = "http://metalearning.ml/2018/",

}

van Rijn, JN, Pfisterer, F, Thomas, J, Muller, A, Bischl, B & Vanschoren, J 2018, 'Meta learning for defaults: symbolic defaults' Paper presented at Neural Information Processing Workshop on Meta-Learning, Montreal, Canada, 8/12/18 - 8/12/18, .

Meta learning for defaults : symbolic defaults. / van Rijn, Jan N.; Pfisterer, Florian; Thomas, Janek; Muller, Andreas; Bischl, Bernd; Vanschoren, J.

2018. Paper presented at Neural Information Processing Workshop on Meta-Learning, Montreal, Canada.

Research output: Contribution to conferencePaperAcademic

TY - CONF

T1 - Meta learning for defaults

T2 - symbolic defaults

AU - van Rijn,Jan N.

AU - Pfisterer,Florian

AU - Thomas,Janek

AU - Muller,Andreas

AU - Bischl,Bernd

AU - Vanschoren,J.

PY - 2018/12/8

Y1 - 2018/12/8

N2 - In this work we propose to use meta-learning to learn sets of symbolic default hyperparameter configurations that work well across many data sets. A well known example for such a symbolic default is the logarithmic relation between the number of features of a dataset and the available features per split of a Random Forest, as observed by Breiman (2001). Symbolic functions allow for a more rich vocabulary to define defaults on. In the past, symbolic and static default values have been obtained either from hand-crafted heuristics or empirical evaluations of specificalgorithms. We propose to automatically learn such symbolic configurations, i.e., formulas containing meta-features, from a large set of prior evaluations of numeric hyperparameters on multiple data sets via symbolic regression and optimization.

AB - In this work we propose to use meta-learning to learn sets of symbolic default hyperparameter configurations that work well across many data sets. A well known example for such a symbolic default is the logarithmic relation between the number of features of a dataset and the available features per split of a Random Forest, as observed by Breiman (2001). Symbolic functions allow for a more rich vocabulary to define defaults on. In the past, symbolic and static default values have been obtained either from hand-crafted heuristics or empirical evaluations of specificalgorithms. We propose to automatically learn such symbolic configurations, i.e., formulas containing meta-features, from a large set of prior evaluations of numeric hyperparameters on multiple data sets via symbolic regression and optimization.

KW - Meta-learning

KW - Automatic Machine Learning

M3 - Paper

ER -

van Rijn JN, Pfisterer F, Thomas J, Muller A, Bischl B, Vanschoren J. Meta learning for defaults: symbolic defaults. 2018. Paper presented at Neural Information Processing Workshop on Meta-Learning, Montreal, Canada.