factory-predict {GeneralizedWendland}

Function factory for prediction generator

Description

A function factory which generates a function for sampling from the predictive distribution

Usage

predictionFactory(y, locs0, locs1, covariance, X0 = list2DF(nrow = nrow(locs0)),
  X1 = list2DF(nrow = nrow(locs1)), ..., cov.args = list(), chol.args = list(),
  use_spam = TRUE)

Arguments

y

Observed measurements.

locs0

Locations at which measurements were obtained.

locs1

Locations at which to predict.

covariance

Covariance function.

X0

Covariates. By default corresponds to an empty data.frame() with nrow(locs0) rows.

X1

Covariates for new locations. By default corresponds to an empty data.frame() with nrow(locs1) rows.

...

Additional arguments. Unused.

cov.args

Additional arguments for covariance function.

chol.args

Additional arguments for function used to compute cholesky decomposition.

use_spam

Logical value. If TRUE, use spam to compute predictions.

Value

Returns a function of the form function(n, param), where n corresponds to the number of simulations to run, and param is a parameter vector as returned by optim.

Author(s)

Thomas Caspar Fischer

Examples

set.seed(23)
n <- 100
res <- c(20, 20)

locs <- data.frame(x = runif(n), y = runif(n))
locs_new <- expand.grid(x = seq(0, 1, length.out = res[1]),
                        y = seq(0, 1, length.out = res[2]))
range <- 0.3
dmat <- as.matrix(dist(locs))
theta <- c(range, 1, 1, 0, 0)
cov.args <- list()
chol.args <- list()
Sigma <- cov.wendland(h = dmat, theta = theta, cov.args = cov.args)
y <- c(spam::rmvnorm(n = 1, Sigma = Sigma))

predictionFun <- predictionFactory(y = y, locs0 = locs, locs1 = locs_new,
  covariance = cov.wendland, cov.args = cov.args, chol.args = chol.args,
  use_spam = FALSE)

predictions <- predictionFun(n=10, param=theta)

image(x = seq(0, 1, length.out=res[1]),
      y = seq(0, 1, length.out=res[2]),
      z = matrix(apply(predictions,2,mean), res[1], res[2]),
      col = hcl.colors(9, "Blue-Red"),
      breaks = qnorm(seq(0.05, 0.95, 0.1)),
      xlab = "", ylab = "", xaxt = "n", yaxt = "n",
      useRaster = TRUE)