FunctionalGPs.jl

About

FunctionalGPs.jl extends AbstractGPs.jl with a composable algebra of linear functionals on Gaussian processes. You can condition a GP on, and predict, any linear transform of the latent function: point evaluations, partial derivatives of any order, integrals over intervals and boxes, and sums, scales, tensor products and stacks of these.

The algebra dispatches into specialised assembly paths, with Toeplitz matrices for stationary kernels on regular grids, sparse matrices for compactly supported kernels, and closed-form antiderivatives where available.

Installation

using Pkg
Pkg.add("FunctionalGPs")