A new hyper-heuristic implementation in HyFlex: a study on generality

Reusability is a desired feature for search and optimisation strategies. Low- level, problem-dependent search mechanisms are far from being used on different prob- lems while keeping them unchanged. Meta-heuristics have been employed for taking the heuristic design process to a higher level and for facilitating reusability. These approaches have been usually conceived as heuristic methods that need to be imple- mented and adapted with regard to the characteristics of the goal problem. Thus, the resulting algorithms are still problem-dependent and hard to apply to other prob- lems. Hyper-heuristics take the search process into the heuristic level and manage the heuristic set instead of directly solving a problem. During this management process, any problem-dependent data exchange between hyper-heuristics and problems is disal- lowed. Although any knowledge about the problems is absent for the hyper-heuristics, the generality of hyper-heuristics has not been examined extensively. In a recently pro- posed high level framework, HyFlex, it is easy to test the generality of hyper-heuristics. HyFlex provides a set of problems with a number of instances as well as a group of low-level heuristics. In this study, the design process of a hyper-heuristic upon HyFlex will be discussed. A performance analysis based on the experimental results will be carried out.