Investigating the correlation between indicators of predictive diagnostic optimisation and search result quality

摘要: Combinatorially complex problems are often optimised with heuristic solvers which generally provide acceptable results but no indication as to how the quality achieved compares best possible. In previous work we have introduced Predictive Diagnostic Optimisation (PDO), a based on local search that provides information about space structure through set of indicators whilst searching for optimal solution. PDO can collect useful process, such variation in number steps needed locally optimise random solution and error between expected actual qualities optimum, known prediction error. Given experimental quadratic assignment problem, it appears high coincides lower vice versa. This confirms this assumption help two additional also shows reliability is challenged by structural properties lead homogeneity optima basins. Conversely, increases an indicator quality.