Benchmark generator for TD Mk landscapes

Abstract

We introduce a publicly available benchmark generator for Tree Decomposition (TD) Mk Landscapes. TD Mk Landscapes were introduced by Whitley et al. to get rid of unnecessary restrictions of Adjacent NK Landscapes while still allowing for the calculation of the global optimum in polynomial time. This makes TD Mk Landscapes more lenient while still being as convenient as Adjacent NK Landscapes. Together, these properties make it very suitable for benchmarking blackbox algorithms. Whitley et al., however, introduced a construction algorithm that only constructs Adjacent NK Landscapes. Recently, Thierens et al. introduced an algorithm, CliqueTreeMk, to construct any TD Mk Landscape and find its optimum. In this work, we introduce CliqueTreeMk in more detail, implement it for public use, and show some results for LT-GOMEA on an example TD Mk Landscape problem. The results show that deceptive trap problems with higher overlap do not necessarily decrease performance and effectiveness for LT-GOMEA.