jeffrey-hokanson/ExperimentDesigns

A collection of best experimental designs

A database of (approximately) optimal experimental designs

Editor: Jeffrey M. Hokanson

Introduction

Space-filling experimental designs pose notoriously difficult optimization problems. Here we collect the best known current designs for the dual purpose of

• providing experimentalists with precomputed near-optimal designs and
• a benchmark for the performance of new algorithms for constructing experimental designs.

Visualizations of these designs can be seen at my website.

Related Work

• Packomania: an extensive resource for sphere packing (analogous to maximin design
• Space-filling designs: a database of designs with many maximin designs

Contributing

I welcome other researchers to submit new experimental designs. To submit a design issue a pull request in the directory corresponding to the type of design constructed. For example, to submit a 40-point minimax design using the l2-metric on the unit-square add a file

designs/minimax/l2/square_0040.json

This file should be in json format with the following fields:

• X Coordinates for the design
• objective the design criteria used, in this case minimax
• metric the distance metric used, in this case l2
• domain the space on which the design is constructed, in this case square
• radius the radius of the discs covering the domain (a.k.a. dispersion, fill-distance)
• author who submitted the design
• notes (optional) notes for reproducing the design

When a pull request is issued, Travis-CI will check the submitted design verifying that design has the right number of points and radius value is correct. In addition, these checks will also check that the new designs are better than the ones that are being replaced.