Version: v1.0.2 Latest Latest

This package is not in the latest version of its module.

Go to latest
Published: Jul 18, 2021 License: MIT


Gaussian Process regression in Go

GoGP is a library for probabilistic programming around Gaussian Processes, in Go. A Gaussian process is by itself an Infergo model (elemental, with custom gradient). A GP model can be used within a larger model, for example to handle uncertain inputs, non-Gaussian noise, or impose priors on GP hyperparameters.

Kernels are automatically differentiated by leveraging the reverse-mode automatic differentation of Infergo.


More examples in the tutorial folder.

Bare bones

In the basic case, similar to that supported by many Gaussian process libraries, a GP directly serves as the model for inference on hyperparameters (or the hyperparameters can be just fixed).

The library user specifies the kernel:

type Basic struct{}
func (Basic) Observe(x []float64) float64 {
    return x[0] * kernel.Normal.Observe(x[1:])
func (Basic) NTheta() int { return 2 }

and initializes GP with a kernel instance:

gp := &gp.GP{
    NDim:  1,
    Simil: Basic{},

MLE inference on hyperparameters and prediction can then be performed through library functions.

Priors on hyperparameters

If priors on hyperparameters are to be specified, the library user provides both the kernel and the model. GP is initialized with the kernel, and then GP and the model are combined for inference in a derived model:

type Model struct {
    gp           *gp.GP
    priors       *Priors
    gGrad, pGrad []float64
func (m *Model) Observe(x []float64) float64 {
    var gll, pll float64
    gll, m.gGrad = \
      m.gp.Observe(x), model.Gradient(m.gp)
    pll, m.pGrad = \
    return gll + pll
func (m *Model) Gradient() []float64 {
	for i := range m.pGrad {
		m.gGrad[i] += m.pGrad[i]
	return m.gGrad

In Model, gp holds a GP instance and priors holds an instance of the model expressing beliefs about hyperparameters. A Model instance is used for inference on hyperparameters, a GP instance --- for prediction.


Path Synopsis
Package kernel provides a library of primitive kernels.
Package kernel provides a library of primitive kernels.

Jump to

Keyboard shortcuts

? : This menu
/ : Search site
f or F : Jump to
y or Y : Canonical URL