GP classification, sparse GPs, warping and deep-GP samples

GpClassify: Laplace-approximation binary classification with the logistic likelihood (Rasmussen and Williams 2006, Algorithms 3.1-3.2) and MacKay's probit-averaged probabilities. GpSparse: DTC predictions with the DTC evidence and the Titsias (2009) VFE bound. KumaraswamyWarp: input warping. DeepGpSample: deep-GP prior samples.

Usage

GpClassify(X, y, Xnew, kernel = "se", tol = 1e-10, maxit = 100, ...)

GpSparse(X, y, Z, Xnew, method = "vfe", kernel = "se", noise = 0.1, ...)

KumaraswamyWarp(x, a, b)

DeepGpSample(X, layers, seed = 1, jitter = 1e-09)

Arguments

X, Xnew

Input matrices.

y

Labels in -1, 1 or targets.

kernel

Kernel name.

tol, maxit

Newton controls.

...

Kernel parameters.

Z

Inducing inputs.

method

"vfe" or "dtc".

noise

Noise variance.

x

Values in the unit interval.

a, b

Kumaraswamy shapes.

layers

List of list(kernel, params).

seed

Philox seed.

jitter

Diagonal jitter.

Value

List or vector.

References

Titsias, M. K. (2009). Variational learning of inducing variables in sparse Gaussian processes. Proceedings of AISTATS, 567-574.

Snoek, J., Swersky, K., Zemel, R. and Adams, R. P. (2014). Input warping for Bayesian optimization of non-stationary functions. Proceedings of ICML, 1674-1682.

Damianou, A. and Lawrence, N. D. (2013). Deep Gaussian processes. Proceedings of AISTATS, 207-215.

Examples

GpClassify(matrix(c(-1, -0.5, 0.5, 1)), c(-1, -1, 1, 1), matrix(0.8), variance = 4)$probability
#> [1] 0.7254559
KumaraswamyWarp(c(0, 0.5, 1), 2, 1)
#> [1] 0.00 0.25 1.00