Contribute  :  Advanced Search  :  Site Statistics  :  Directory  :  My Downloads  :  Links  :  Calendar  
SFERES Framework dedicated to evolution and simulation
Welcome to SFERES
Thursday, March 11 2010 @ 12:36 AM CET

SFERES documentation

SFERES (SFERES is a Framework Enabling Research on Evolution and Simulation) is a tool dedicated to students, searchers or others interested in the evolutionnary algorithm experiments. Its goal is to provide a generic tool maximizing code reutilization and thus accelerating the development of new algorithms by concentrating on the new aspects. SFERES gathers an evolution framework with a simulation framework. SFERES is written in C++. Development of new modules requires programming skills, but exploitation of existing modules only requires to modify an XML configuration file.

For details about SFERES' structure, look at the following article (in french): S. Landau, S. Doncieux, A. Drogoul, and J.-A. Meyer. SFERES: un framework pour la conception de systèmes multi-agents adaptatifs. Technique et Science Informatiques, 21(4):427-446, 2002. It is available for download from the publications page.

This document aims at presenting SFERES' philosophy and structure. On SFERES web site (http://sferes.lip6.fr) you can also find a FAQ (http://sferes.lip6.fr/faqman/index.php) and a Wiki for tricks and tips and experiences of other users (http://sferes.lip6.fr/ewiki/index.php). Publications that uses SFERES are listed and downloadable from http://webia.lip6.fr/~sferes/external/sferesbib.php .

Introduction

How to use SFERES ?

Module creation

Some generalities about SFERES inner working:

The main modules that can be changed are:

  • in the evaluation part:
    • the Evaluator module, as SFERES has been conceived in the field of Adaptive Systems, it will be most of the time a simulator, that contains agents that interact between each other and with an environment through sensors and effectors
    • the control architecture that links an agent's sensors to its effectors (in the case of the optimization of an agent behavior)
    • the fitness function

  • in the evolution part:
    • the genome, made up with chromosomes, that contains the information that is generated and optimized by evolution. The genetic operators that have to be used (mutation and cross-over) are part of these modules
    • the population that contains the selection algorithm
    • the statistics. These modules can analyze and keep statistics all along evolution and concerning any one of the modules that evolve