Toolboxes and softwares for learning and optimization

DiceKriging / DiceOptim / DiceView

NKO participant: C. Chevalier, D. Ginsbourger, V. Picheny, Y. Richet

Kriging-based R packages. DiceKriging provides estimation, validation and prediction of kriging models; DiceOptim implements the EGO algorithm and its variants; DiceView allows quick vizualization of kriging models (with 2D/3D section views for high dimensions).

All the packages provide a full documentation with simple examples.

The package DiceOptim does not provide yet algorithms for noisy optimization, but a new version should be released soon. alpha-version of noisy kriging-based optimizers are available on demand (victor “dot” picheny “at” ecp “dot” fr or david “dot” ginsbourger “at” stat “dot” unibe “dot” ch).

The packages are downloadable directly from R or at the following URLs:

DiceKriging: http://cran.r-project.org/web/packages/DiceKriging/index.html

DiceOptim: http://cran.r-project.org/web/packages/DiceOptim/index.html

DiceView: http://cran.r-project.org/web/packages/DiceView/index.html

SURROGATES toolbox

NKO participant: Felipe Viana

SURROGATES Toolbox is a general-purpose MATLAB library of multidimensional function approximation and surrogate-based optimization methods. The current version includes the following capabilities:

Design of experiments: central composite design, full factorial design, Latin hypercube design, D-optimal and maxmin designs.
Surrogates: kriging, polynomial response surface, radial basis neural network, and support vector regression.
Analysis of error and cross validation: leave-one-out and k-fold cross-validation, and classical error analysis (coefficient of determination, standard error; root mean square error; and others).
Surrogate-based optimization: efficient global optimization (EGO) algorithm.
Other capabilities: global sensitivity analysis and conservative surrogates via safety margin.

URL: http://sites.google.com/site/fchegury/surrogatestoolbox

Engineering Design via Surrogate Modelling: A Practical Guide Matlab code

NKO participant: Alexander Forrester

A large set of matlab functions that accompany the book Engineering Design via Surrogate Modelling: A Practical Guide by Dr. Alexander Forrester, Dr. Andras Sobester, Prof. Andy Keane. The code is split into the folders:

Sampling Plans
Constructing a Surrogate
Exploring and Exploiting a Surrogate
Advanced Concepts
Example Problems
Example Scripts

Link to the book description:

http://www.wiley.com//legacy/wileychi/forrester/

The functions are downloadable here:

http://www.personal.soton.ac.uk/aijf197/Website%20Code%20November%2010.zip

SCILAB Krisp toolbox

NKO participants: Janis Janusevskis, Rodolphe Le Riche

This toolbox implements kriging based regression (also known as Gaussian process regression) and optimization of deterministic simulators. The toolbox consists of two main components:

Functions for creation of kriging model for deterministic or noisy data (correlation kernels, hyper-parameter estimation, prediction, cross-validation).
Methods implementing Efficient Global Optimization (EGO) for time consuming deterministic functions.

Full description and source code available here :

http://atoms.scilab.org/toolboxes/krisp/