Late Acceptance Hill Climbing for Constrained Covering Arrays

The Late Acceptance Hill-Climbing (LAHC) algorithm is a one-point search meta-heuristic with a single parameter. Like Simulated Annealing (SA) it sometimes accepts worsening moves, however it is far more simple and does not require complex parameter setting. In this paper we study an application of LAHC to the Combinatorial Interaction Testing (CIT) problem. CIT is a cost-effective black-box sampling technique for discovering interaction faults in highly configurable systems. There are several techniques for CIT; one of the most established and well-known is Covering Arrays by Simulated Annealing (CASA). CASA is a layered search framework using SA in its most inner layer. Here we replace SA in CASA with LAHC, proposing a modified framework, Covering Arrays by Late Acceptance (CALA). Our experimental evaluation demonstrates that LAHC yields better or equal quality solutions compared to SA for all but one of the 35 benchmark instances tested.