Analysis 1 - Implemention of NNGP models using \(N\) x \(N\) matrices, and examples with \(N=6\), \(N=500\), and \(N=2000\) models. Shows agreement between full GP model and NNGP approximation.
Analysis 2 - New NNGP implementation now only uses \(N\) x \(k\) matrices, which significantly improves NNGP efficiency. Compares NNGP and full GP models for \(N=2000\) and \(N=5000\) models, shows the NNGP is 30-50 times more efficient than the full GP.
Analysis 3 - Implementation showing how to fit full GP model using MCMC and simultaneously make predictions at new points. Two examples are shown, first with \(N=p=5\) shows proof of concept, but is not very efficient. Second, with \(N=50\) and \(p=2500\) is implemented as efficiently as I could.
Analysis 4 - Implements conjugate Gibbs updating of \(\beta\) mean regression coefficients for NNGP model.
Analysis 5 - An example of using the non-stationary correlation model.
FNS-M2 Model - Reproduces an analysis of the precipitation data set used by Paciorek and Schervish (2006), here using a GP spatial model with a nonstationary covariance function.
Sparse General Vecchia (SGV) Implementation - Demonstration of using the sparse general Vecchia implementation of NNGP, as described in Katzfuss and Guinness (2017).
Nearest-Neighbor Gaussian Process (NNGP): Joint Density, Sampling, and Prediction - This vignette provides a complete workflow for constructing, fitting, and predicting from a Nearest-Neighbor Gaussian Process (NNGP) model in NIMBLE using BayesNSGP. The example demonstrates neighborhood graph construction, specification of NNGP joint densities within BUGS code, comparison of alternative MCMC sampling strategies, and spatial prediction at unobserved locations.