sKrig: Draw posterior predictive samples from a spatial Gaussian...

In bisque: Approximate Bayesian Inference via Sparse Grid Quadrature Evaluation (BISQuE) for Hierarchical Models

Description

Draw posterior predictive samples from a spatial Gaussian process model

Usage

sKrig(x, sFit, coords.krig, coords = sFit$coords, burn = 0, ncores = 1)

Arguments

x	Observation of a spatial Gaussian random field, passed as a vector
sFit	posterior samples of model parameters; output from bisque::sFit
coords.krig	Spatial coordinates at which the field should be interpolated
coords	Spatial coordinates at which observations are available
burn	number of posterior samples to discard from sFit before sampling
ncores	Kriging is done via composition sampling, which may be done in parallel. ncores specifies the number of cores over which sampling is done. If ncores>1, bisque::sKrig assumes that a parallel backend suitable for use with the foreach package is already registered.

Examples

library(fields)

simulate.field = function(n = 100, range = .3, smoothness = .5, phi = 1){
  # Simulates a mean-zero spatial field on the unit square
  #
  # Parameters:
  #  n - number of spatial locations
  #  range, smoothness, phi - parameters for Matern covariance function
  
  coords = matrix(runif(2*n), ncol=2)
  
  Sigma = Matern(d = as.matrix(dist(coords)), 
                 range = range, smoothness = smoothness, phi = phi)
  
  list(coords = coords,
       params = list(n=n, range=range, smoothness=smoothness, phi=phi),
       x = t(chol(Sigma)) %*%  rnorm(n))
}

# simulate data
x = simulate.field()

# configure gibbs sampler  
it = 100

# run sampler using default posteriors
post.samples = sFit(x = x$x, coords = x$coords, nSamples = it)

# build kriging grid
cseq = seq(0, 1, length.out = 10)
coords.krig = expand.grid(x = cseq, y = cseq)

# sample from posterior predictive distribution
burn = 75
samples.krig = sKrig(x$x, post.samples, coords.krig = coords.krig, burn = burn)

bisque documentation built on March 26, 2020, 7:27 p.m.