gaussianProcess

R package for Gaussian Process regression with various kernels

Usage

If X is a matrix of training covariates and y a vector of training targets then you create a gaussianProcess and automatically tune the hyper parameters with various options (see doc) with

gp <- gaussianProcess(X,y,options)

To predict the output for a new data matrix X.new run

pred <- predict(gp, X.new)

Here pred is a list with the posterior mean in pred$mean and the posterior covariance in pred$covariance. If in 1 dimension, you can plot the posterior process between x.min and x.max with

plot(gp, x.min, x.max)

Using a mean function

If you define a mean function mean.f that takes in a matrix and returns a vector you can use this as the mean function for the GP with

gp <- gaussianProcess(X,y, meanFunc=mean.f)

For instance you can use glmnet to train a mean function to pass into the GP as follows. Fit with glmnet (e.g. with lambda=1), get the beta and create the mean function:

fit <- glmnet(x, y, lambda=1)
beta <- fit$beta
mean.f <- function(x) x %*% beta

Or you can use the pryr package to do a partial application

fit <- glmnet(x, y, lambda=1)
mean.f <- pryr::partial(predict, object=fit)

Cross Validation

You can Leave One Out CV errors (true - predicted) for all data points with loocv(gp)