PASCAL - Pattern Analysis, Statistical Modelling and Computational Learning

An empirical study of MDL Model Selection with infinite parametric complexity
Steven de Rooij and Peter Grünwald
Journal of Mathematical Psychology Volume 50, Number 2, pp. 180-192, 2006.

Abstract

Parametric complexity is a central concept in MDL model selection. In practice it often turns out to be infinite, even for quite simple models such as the Poisson and Geometric families. In such cases, MDL model selection as based on NML and Bayesian inference based on Jeffreys' prior can not be used. Several ways to resolve this problem have been proposed. We conduct experiments to compare and evaluate their behaviour on small sample sizes. We find interestingly poor behaviour for the plug-in predictive code; a restricted NML model performs quite well but it is questionable if the results validate its theoretical motivation. The Bayesian model with the improper Jeffreys' prior is the most dependable.

EPrint Type:Article
Project Keyword:Project Keyword UNSPECIFIED
Subjects:Computational, Information-Theoretic Learning with Statistics
ID Code:2811
Deposited By:Peter Grünwald
Deposited On:22 November 2006