Bayes and empirical Bayes changepoint problems

TL;DR

This paper introduces a Bayesian approach for multiple changepoint detection that efficiently computes posterior probabilities, estimates hyperparameters with Monte Carlo EM, and offers advantages over traditional MAP estimators, demonstrated on simulations and real data.

Contribution

It generalizes Liu and Lawrence's method to unknown number of changepoints, using dynamic programming and Monte Carlo EM for hyperparameter estimation.

Findings

Posterior samples are independent, avoiding MCMC convergence issues.

The method effectively detects multiple changepoints in simulated and real data.

Advantages over traditional MAP estimators include better uncertainty quantification.

Abstract

We generalize the approach of Liu and Lawrence (1999) for multiple changepoint problems where the number of changepoints is unknown. The approach is based on dynamic programming recursion for efficient calculation of the marginal probability of the data with the hidden parameters integrated out. For the estimation of the hyperparameters, we propose to use Monte Carlo EM when training data are available. We argue that there is some advantages of using samples from the posterior which takes into account the uncertainty of the changepoints, compared to the traditional MAP estimator, which is also more expensive to compute in this context. The samples from the posterior obtained by our algorithm are independent, getting rid of the convergence issue associated with the MCMC approach. We illustrate our approach on limited simulations and some real data set.