Predictive measures for problem representation and genetic operator design
Dirk Thierens · Utrecht University Repository (Utrecht University) · 2002
The design of a problem representation and of the genetic operators acting on this representation are a very important task when applying genetic algorithms. To assist the user in this design task a number of predictive measures have been proposed in the literature. Here we discuss some issues in the use of these predictive measures. We identify the properties which a predictive measure should possess, and introduce an alternative procedure. The results represent a small step towards a predictive tool assisting the GA designer. The paper's main goal is to stimulate the discussion about the necessary ingredients of such a tool at the GECCO Workshop on Representations. To keep the discussion concrete we compute some predictive measures of different GA problem solving approaches for the Bin Packing problem.