Experimental Design for Gaussian Process Regression in Python
Python
Switch branches/tags
Nothing to show
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Failed to load latest commit information.
gpExp
.gitignore
LICENSE
README.md
demo.py
demo2.py
demo_spacefilling.py

README.md

GPEXP

Experimental Design for Gaussian Process Regression
A python package for performing experimental design for Gaussian Process Regression

###General Information

Author: Alex Gorodetsky

Contact: goroda@mit.edu

Copyright (c) 2013-2015, Massachusetts Institute of Technology
License: GPL2

Software Dependencies

Overview

Experimental design deals with the issue of determining where to obtain new data in order to build accurate models. Gaussian process regression is a useful method to build models of raw data or to build surrogate models for complex computational simulations.

This software is different from most other GP software because it focuses on combining experimental design and Gaussian process regression. It was developed for performing the studies provided in the paper detailing the integrated variance experimental design function (link arxiv paper). The current software contains the following options for kernels and experimental design cost functions (though adding new kernels and cost functions is fairly trivial).

(Multidimensional) Kernels

  • Squared exponential
    • isotropic
    • automatic relevance determination
  • Isotropic Matern kernel
  • Mehler Kernel

Experimental Design Cost Functions

  • Integrated Variance
  • Conditional Entropy
  • Mutual Information