Welcome

This site has now been superceded by the new Huygens II Server.

The Downloads area has been updated with the latest version of the training client program, and its API documentation. This has minor changes - the most notable being that the constructor includes your login (email address), which will be used in the CEC2006 Competition.

  30 March 2006

• A new version of the huygensWSClient training client which accesses the new server gungurru is now available in the Downloads directory. This server returns 1000 evaluations locally in a blistering 2.5 seconds! Run RandomSearch and let me know how long it takes from your part of the world.

  The older version which accessed the machine temp64 should no longer be used.

• The human interface mentioned below now includes an image of the moon you are evaluating (up to number 20-120).

  13 March 2006

The Huygens Server has moved to a new faster dedicated server, generously provided through a grant from the Apple University Consortium. The direct address is:

http://gungurru.csse.uwa.edu.au/cara/huygens/

  A human interface that allows probing of the training series landscapes is now available in beta form here.

  A prerelease of a new SOAP-based training client that allows arrays of values to be sent in each package is available for trialling. See the bottom of the Downloads page.

Welcome to the Hugens Benchmarking Research Project.

The success of evolutionary and allied algorithms has led to a proliferation of population-based and hybrid search and optimisation algorithms. The aim of this site is to provide an objective way for researchers to compare their algorithms against those of many other researchers, without having to reimplement all of the competing algorithms.

The site currently uses fractally generated landscapes as a domain for comparison. There are many reasons for this, some of which are outlined in a GECCO 2005 paper that can be found under the Documentation menu. New users might like to begin by taking a look at some typical landscapes.

The current focus is on real-valued optimisation. It is anticipated that further qualifying problem domains will be added as limitations of the current domain become better understood. Please send arguments supporting changes or other problem domains to cara@csse.uwa.edu.au.

To start testing your own algorithm begin by following the 3 step Quick Start under the Documentation menu. This will get you connected to the server. You can then plug in your own algorithm. Simply use the API or follow the example Uniform Search provided.

Plug-in stubs are currently provided for algorithms written in Java and Matlab, however the system simply uses http posts, so it is relatively easy to write a stub for your favourite language, using the Java stub as an example. A range of standard stubs for other languages will be provided as time permits. Please send requests (or contribute stubs) to cara@csse.uwa.edu.au.

Finally, this site has recently "gone live" at GECCO'2005 in Washington DC. Some teething problems are possible. We appreciate your patience and ask that you send any bugs or suggestions to cara@csse.uwa.edu.au.