Tractable Bayesian density regression via logit stick-breaking priors

There is a growing interest in learning how the distribution of response variable changes with set observed predictors. Bayesian nonparametric dependent mixture models provide flexible approach to address this goal. However, several formulations require computationally demanding algorithms for posterior inference. Motivated by issue, we study class predictor-dependent infinite models, which relies on simple representation stick-breaking prior via sequential logistic regressions. This formulation maintains same desirable properties popular priors, and leverages recent Pólya-gamma data augmentation facilitate implementation computational methods These routines include Markov chain Monte Carlo Gibbs sampling, expectation–maximization algorithms, mean-field variational Bayes scalable inference, thereby stimulating wider density regression practitioners. The associated these are presented detail tested toxicology study.