Cluster-Based Constraint Ordering for Direct Diagnosis

Prediction quality and runtime performance are important performance indicators for diagnosis algorithms. In this paper, we propose a new method, which is called ClusDiag (Cluster-Based Constraint Ordered Direct Diagnosis) which can improve both indicators. ClusDiag has a learning phase to find a constraint ordering heuristic. After the learning phase, a diagnosis is found by applying the direct diagnosis algorithm FastDiag on an inconsistent constraint set where the constraints are reordered with respect to the constraint ordering heuristic. Keywords— Configuration Systems; Diagnosis; Constraint Satisfaction; Variable and Value Ordering Heuristics; Clustering; Performance Optimization