Bayesian Entropy Estimation for Countable Discrete DistributionsEvan Archer, Il Memming Park, & Jonathan W. Pillow 
Journal of Machine Learning Research
15 (Oct): 28332868, 2014
We consider the problem of estimating Shannon's entropy H from discrete data, in cases where the number of possible symbols is unknown or even countably infinite. The PitmanYor process, a generalization of Dirichlet process, provides a tractable prior distribution over the space of countably infinite discrete distributions, and has found major applications in Bayesian nonparametric statistics and machine learning. Here we show that it provides a natural family of priors for Bayesian entropy estimation, due to the fact that moments of the induced posterior distribution over H can be computed analytically. We derive formulas for the posterior mean (Bayes' least squares estimate) and variance under Dirichlet and PitmanYor process priors. Moreover, we show that a fixed Dirichlet or PitmanYor process prior implies a narrow prior distribution over H, meaning the prior strongly determines the entropy estimate in the undersampled regime. We derive a family of continuous measures for mixing PitmanYor processes to produce an approximately flat prior over H. We show that the resulting "PitmanYor Mixture" (PYM) entropy estimator is consistent for a large class of distributions. Finally, we explore the theoretical properties of the resulting estimator, and show that it performs well both in simulation and in application to real data.

This paper supercedes:

online publications 