This package implements the model described in “A Framework of Zero-Inflated Bayesian Negative Binomial Regression Models For Spatiotemporal Data” by Qing He and Hsin-Hsiung Huang (2023). https://doi.org/10.1016/j.jspi.2023.106098.
This package is a work in progress, feel free to create an issue if you have suggestions or notice any problems.
Installation Instructions
Install the devtools R package, then run the following command
install.packages(c("BayesLogit", "LaplacesDemon", "MASS", "Matrix", "msm", "mvtnorm", "spNNGP"))
devtools::install_github("KingJMS1/NNGP_ZINB_R")Example Use
Detailed examples with full code can be found in the vignettes folder, or at the following link under articles: https://kingjms1.github.io/NNGP_ZINB_R/
X Other Predictor variables
y Zero inflated count response
coords Spatial coordinates for GP
Vs Spatially varying predictor variables
(e.g. one-hot indication of which location this is for varying intercept),
wrapped in sparseMatrix from Matrix R package.
Will be multiplied by the spatial random effects for prediction.
Vt Temporal varying predictor variables, wrapped in sparseMatrix from Matrix R package.
Will be multiplied by the temporal random effects for prediction.
Ds Spatial distance matrix, diagonal should be 0,
off diagonal is distance between elements i and j in space, inputs to the spatial GP kernel
Dt Temporal distance matirx, diagonal should be 0,
off diagonal is distance between elements i and j in time, inputs to the temporal GP kernel
M How many neighbors to allow in the spatial GP algorithm, defaults to 10.
nsim How long to run MCMC in total, must be greater than burn.
burn How long to run MCMC before saving samples.
Given all of the above, predictions can then be found via:
API Reference
API reference: https://kingjms1.github.io/NNGP_ZINB_R/reference/index.html