Gaussian Process with Histogram Intersection Kernel
Description
Provides an implementation of a Gaussian process regression with a histogram intersection kernel (HIK) and utilizes approximations to speed up learning and prediction.
In contrast to a squared exponential kernel, an HIK provides advantages such as linear memory and learning time requirements. However, the HIK only provides a piecewise-linear approximation of the function.
Furthermore, the number of estimated eigenvalues is reduced. The eigenvalues and vectors are required for the approximation of the log-likelihood function as well as the approximation of the predicted
variance of new samples. This package provides approximations for a single eigenvalue as well as multiple. Further information of the variance and log-likelihood approximation, as well as the Gaussian
process with HIK, can be found in the paper by Rodner et al. (2016) .