Gaussian process regression, sampling and fitting

GpCovariance: kernel matrices (SE/ARD, Matern 3/2 and 5/2, rational quadratic, periodic, linear). GpPredict: exact regression (Rasmussen and Williams 2006, Algorithm 2.1) with the log marginal likelihood. GpSample: prior or posterior sample paths. GpFit: type-II maximum likelihood by Nelder-Mead and kernel selection. GpLoo: closed-form leave-one-out. Identical to the Python arm morie.fn.gproc.

Usage

GpCovariance(X1, X2, kernel = "se", ...)

GpPredict(X, y, Xnew, kernel = "se", noise = 1e-06, mean = 0, ...)

GpSample(
  X,
  kernel = "se",
  nsim = 1,
  seed = 1,
  data = NULL,
  noise = 1e-06,
  jitter = 1e-10,
  ...
)

GpFit(X, y, kernels = "se", mean = 0, init = NULL)

GpLoo(X, y, kernel = "se", noise = 1e-06, ...)

Arguments

X1, X2, X, Xnew: Input matrices.
kernel: Kernel name.
...: Kernel parameters (variance, lengthscale, period, alpha, bias).
y: Targets.
noise: Noise variance.
mean: Constant mean.
nsim: Number of paths.
seed: Philox seed.
data: list(X, y) for posterior sampling.
jitter: Diagonal jitter.
kernels: Candidate kernels.
init: Initial log parameters.

Value

Matrix or list.

References

Rasmussen, C. E. and Williams, C. K. I. (2006). Gaussian Processes for Machine Learning. MIT Press.

Examples

GpCovariance(matrix(0), matrix(c(0, 1)), "matern32")
#>      [,1]      [,2]
#> [1,]    1 0.4833577
GpPredict(matrix(c(0, 1)), c(0, 1), matrix(0.5), noise = 0.01)$mean
#> [1] 0.5459203