Predictive mean, variance and log marginal likelihood of a GP. See "2.3 Varying the Hyperparameters" on page 19 of Rasmussen and Williams' book.
gp.predict(y, K = NULL, Kstar, Kstarstar, U = chol(K))
The targets.
The covariance matrix (kernel) for input points, not needed if U is provided.
The cross covariance matrix (kernel)
The cross covariance matrix (kernel) for test points
Cholesky decomposition of K