Sparse Gaussian Processes using Pseudo Inputs

Implementation in Python

This is repository contains a python implementation of Sparse Gaussian Processes using Pseudo Inputs published in NIPS 2005. [LINK]

In order to use this package, please import the package as follows -

import spgp

This package contains two functions:

• spgp.utilityfn: Contains the function for likelihood calculation and estimating the mean & variance using learned SPGP hyper-params.
• spgp.minimize: Uses Carl's Rasmussen's implementation for finding a local minimum of a nonlinear multivariate function.

This repository also contains an example implementation for a 2D spatial regression problem. This example uses plotly for generating the outputs and these are saved as an offline plot in the output folder.

To use these codes, please refer the following publications:

1. Rajat Mishra, Mandar Chitre, and Sanjay Swarup. "Online Informative Path Planning using Sparse Gaussian Processes." 2018 OCEANS-MTS/IEEE Kobe Techno-Oceans (OTO). IEEE, 2018.
2. Edward Snelson and Zoubin Ghahramani. "Sparse Gaussian Processes using pseudo-inputs." Advances in neural information processing systems. 2006.