Discussion of the article "Bayesian cluster analysis: point estimation and credible balls" by Wade and Ghahramani

TL;DR

This paper discusses Wade and Ghahramani's Bayesian clustering method, emphasizing its contribution to uncertainty assessment of clustering solutions and suggesting extensions for broader applicability.

Contribution

It highlights the methodological gap in Bayesian clustering by providing a way to evaluate uncertainty around the optimal partition.

Findings

Provides a framework for assessing uncertainty in Bayesian cluster analysis

Reveals alternative clustering solutions based on uncertainty characterization

Suggests extensions to improve the method's applicability

Abstract

We present a discussion of the paper "Bayesian cluster analysis: point estimation and credible balls" by Wade and Ghahramani. We believe that this paper contributes substantially to the literature on Bayesian clustering by filling in an important methodological gap, by providing a means to assess the uncertainty around a point estimate of the optimal clustering solution based on a given loss function. In our discussion we reflect on the characterisation of uncertainty around the Bayesian optimal partition, revealing other possible alternatives that may be viable. In addition, we suggest other important extensions of the approach proposed which may lead to wider applicability.