IGP: Interchangeable Gaussian Processes

This is an R package that provides a single interface for many different Gaussian process modeling software options.

This package was formerly called UGP, for Universal Gaussian processes, but universal has a different meaning in kriging so the name was changed for clarity.

Installation

You can install IGP from GitHub with:

# install.packages("devtools")
devtools::install_github("CollinErickson/IGP")

Example

The following shows a simple example using the R package laGP as the GP code.

set.seed(0)
library(IGP)
package = "laGP"
n <- 20
d <- 1
f1 <- function(x) {abs(sin(2*pi*x[1]))}
X1 <- matrix(runif(n*d),n,d)
Z1 <- apply(X1,1,f1) + rnorm(n, 0, 1e-3)
gp <- IGP(package=package,X=X1,Z=Z1)

curve(sapply(x, f1), ylab='y')
curve(gp$predict(matrix(x, ncol=1)) - 2 * gp$predict.se(matrix(x, ncol=1)), col=3, add=T)
curve(gp$predict(matrix(x, ncol=1)) + 2 * gp$predict.se(matrix(x, ncol=1)), col=3, add=T)
curve(gp$predict(matrix(x, ncol=1)), col=2, add=T)
points(X1, Z1, pch=19)

Below is the exact same thing except using the R package GauPro. The predictions made are indistinguishable, meaning that they have fit approximately the same parameter values.

set.seed(0)
package = "GauPro"
gp <- IGP(package=package,X=X1,Z=Z1)

curve(sapply(x, f1), ylab='y')
curve(gp$predict(matrix(x, ncol=1)) - 2 * gp$predict.se(matrix(x, ncol=1)), col=3, add=T)
curve(gp$predict(matrix(x, ncol=1)) + 2 * gp$predict.se(matrix(x, ncol=1)), col=3, add=T)
curve(gp$predict(matrix(x, ncol=1)), col=2, add=T)
points(X1, Z1, pch=19)

Package options

The available packages and the platform they run on are shown below. The R packages should run easily. The MATLAB packages are called using the R.matlab R package and have to open a connection to MATLAB. Thus you need to have MATLAB on your computer, it will be slow, and is likely to have problems. Currently the MATLAB packages are not included in the CRAN version of the package, but they can be found on the GitHub repository. The Python packages are called using the R package Python.In.R. It will open a connection to Python and probably will be slow. In addition to requiring that you already have the package (GPy or sklearn) installed, and must be accessible through your default Python path.