Efficiently Computed Marginal Likelihoods for Multivariate Mixture Models
Abstract
We propose a new methodology for efficiently computing marginal likelihoods for univariate and multivariate mixture models. The estimator that this methodology proposes computes the marginal likelihood from Markov chain Monte Carlo (MCMC) samples, is consistent, asymptotically normal and of finite variance. In addition, it is invariant to label switching, does not require posterior samples from hidden allocation vectors, and is efficiently approximated, even for an arbitrarily high number of components. Its computational efficiency is based on an asymptotically optimal ordering of the parameter space, which can in turn be used to provide useful visualisations. We test it in simulation settings where the true marginal likelihood is available analytically. It performs well against state-of-the-art competitors, even in multivariate settings with a high number of components. We demonstrate its utility for inference and model selection on univariate and multivariate data sets.
Related
- A Proxy-likelihood Estimator for Multivariate Extremes Models with Intractable Likelihoods
- A multivariate unsupervised lognormal-GPD mixture for dependent losses
- Salient Bayesian Clustering for Proportional Data via a Multivariate Beta Mixture Model
- Amortized ratio-estimation importance sampling and localized simulation-based calibration for intractable likelihoods
- Recovery-Directed Symbolic Distillation of Neural Likelihoods
BibTeX
@misc{metodiev2026efficiently,
title = {{Efficiently Computed Marginal Likelihoods for Multivariate Mixture Models}},
author = {Martin Metodiev and Marie Perrot-Dockès and Pierre Latouche and Adrian E Raftery},
howpublished = {HAL},
year = {2026},
month = oct,
eprint = {2504.21812},
archivePrefix = {arXiv},
url = {https://hal.science/hal-05053852v3},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Efficiently Computed Marginal Likelihoods for Multivariate Mixture Models
AU - Martin Metodiev
AU - Marie Perrot-Dockès
AU - Pierre Latouche
AU - Adrian E Raftery
PY - 2026
DA - 2026/10/15
PB - HAL
UR - https://hal.science/hal-05053852v3
AB - We propose a new methodology for efficiently computing marginal likelihoods for univariate and multivariate mixture models. The estimator that this methodology proposes computes the marginal likelihood from Markov chain Monte Carlo (MCMC) samples, is consistent, asymptotically normal and of finite variance. In addition, it is invariant to label switching, does not require posterior samples from hidden allocation vectors, and is efficiently approximated, even for an arbitrarily high number of components. Its computational efficiency is based on an asymptotically optimal ordering of the parameter space, which can in turn be used to provide useful visualisations. We test it in simulation settings where the true marginal likelihood is available analytically. It performs well against state-of-the-art competitors, even in multivariate settings with a high number of components. We demonstrate its utility for inference and model selection on univariate and multivariate data sets.
ER -