GPfit-package

A computationally stable approach of fitting a Gaussian Process (GP) model to a deterministic simulator.

Branch of CRAN package with recommended updates. Changes improve performance by ~2 times.

This package describes a Gaussian Process within an n-dimensional hypercube, each with boundaries [0, 1]. The scale_norm helper function has a range argument to allow padding of individual dimensions for the known extent of the dimension.

library(GPfit)
set.seed(45341)
# row dimension
nn <- nrow(mtcars)
id <- (nn - 1):nn
x1 <- scale_norm(seq_len(nn - 2), range = c(1, nn))
# response
y1 <- mtcars$mpg[-id]
# disp dimension
d1 <- scale_norm(mtcars$disp)[-id]
# matrix of inputs
m1 <- cbind(
    ro = x1,
    am = mtcars$am[-id], 
    ds = d1)
# fit model
gpf1 <- GP_fit(
    X = m1,
    Y = y1)
# goodness of fit
plot(
    x = fitted(gpf1)[, "Y_hat"], 
    y = y1, 
    asp = 1)
abline(coef = c(0, 1), col = "red", lty = 2)
# predict for new records
gpp1 <- predict(gpf1, 
    xnew = cbind(
        ro = scale_norm(id, range = c(1, nn)), 
        am = mtcars$am[id], 
        ds = scale_norm(mtcars$disp[id], range = mtcars$disp)))
# response forecast
res <- cbind(actual = mtcars$mpg[id], predicted = gpp1$Y_hat)
res
points(
    x = res[, "predicted"], 
    y = res[, "actual"], 
    col = "green")