2017/03/20 by Richard D. Payne, Nilabja Guha, Payne, Richard D. +5
Computer Science · #Gaussian Processes and Bayesian Inference #Bayesian Methods and Mixture Models
paper · pdf · doi:10.48550/arxiv.1703.06978
Conditional density estimation (density regression) estimates the\ndistribution of a response variable y conditional on covariates x. Utilizing a\npartition model framework, a conditional density estimation method is proposed\nusing logistic Gaussian processes. The partition is created using a Voronoi\ntessellation and is learned from the data using a reversible jump Markov chain\nMonte Carlo algorithm. The Markov chain Monte Carlo algorithm is made possible\nthrough a Laplace approximation on the latent variables of the logistic\nGaussian process model. This approximation marginalizes the parameters in each\npartition element, allowing an efficient search of the posterior distribution\nof the tessellation. The method has desirable consistency properties. In\nsimulation and applications, the model successfully estimates the partition\nstructure and conditional distribution of y.\n