| meshed-package {meshed} | R Documentation |
Methods for fitting models based on Meshed Gaussian Processes (MGPs)
Description
meshed is a flexible package for Bayesian regression analysis on spatial or spatiotemporal datasets. The main function for fitting regression models is spmeshed, which outputs posterior samples obtained from Markov chain Monte Carlo which can be summarised using standard tools. The package also provides a function rmeshedgp for quickly simulating correlated spatial or spatiotemporal data at a very large number of locations.
Details
The functions rmeshedgp and spmeshed are provided for prior and posterior sampling (respectively) of Bayesian spatial or spatiotemporal multivariate regression models based on Meshed Gaussian Processes as introduced by Peruzzi, Banerjee, and Finley (2020). Posterior sampling via spmeshed proceeds by default via GriPS as detailed in Peruzzi, Banerjee, Dunson, and Finley (2021). When at least one outcome is not modeled with Gaussian errors, sampling proceeds taking advantage of Metropolis-adjusted Langevin dynamics as detailed in Peruzzi and Dunson (2022).
Author(s)
Michele Peruzzi
References
Peruzzi, M., Banerjee, S., and Finley, A.O. (2022) Highly Scalable Bayesian Geostatistical Modeling via Meshed Gaussian Processes on Partitioned Domains. Journal of the American Statistical Association, 117(538):969-982. doi:10.1080/01621459.2020.1833889
Peruzzi, M., Banerjee, S., Dunson, D.B., and Finley, A.O. (2021) Grid-Parametrize-Split (GriPS) for Improved Scalable Inference in Spatial Big Data Analysis. https://arxiv.org/abs/2101.03579
Peruzzi, M., Dunson, D.B. (2022) Spatial meshing for general Bayesian multivariate models. https://arxiv.org/abs/2201.10080