Algorithm runtime prediction: Methods & evaluation

• Published in: Artificial intelligence Vol. 206; pp. 79 - 111
• Main Authors: Hutter, Frank, Xu, Lin, Hoos, Holger H., Leyton-Brown, Kevin
• Format: Journal Article
• Language: English
• Published: Oxford Elsevier B.V 01.01.2014; Elsevier
• Subjects: Algorithmics. Computability. Computer arithmetics; Applied sciences; Artificial intelligence; Computer science; control theory; systems; Empirical performance models; Exact sciences and technology; Flows in networks. Combinatorial problems; Highly parameterized algorithms; Mathematical programming; Mixed integer programming; Operational research and scientific management; Operational research. Management science; Performance prediction; Propositional satisfiability; Response surface models; Supervised machine learning; Theoretical computing; Travelling salesperson problem; Supervised machine learning; Travelling salesperson problem; Performance prediction; Response surface models; Empirical performance models; Mixed integer programming; Propositional satisfiability; Highly parameterized algorithms; Performance evaluation; Compilation; Automatic selection; Algorithmics; Travelling salesman problem; Constraint satisfaction; Modeling; Innovation; Identifier; Portfolio selection; Response surface; Learning algorithm; Execution time; Probabilistic approach; Empirical method; Empirical model; Propositional logic; Production management; Supervised learning; Learning (artificial intelligence); Portfolio management; Artificial intelligence; Algorithm analysis
• ISSN: 0004-3702; 1872-7921
• DOI: 10.1016/j.artint.2013.10.003

Perhaps surprisingly, it is possible to predict how long an algorithm will take to run on a previously unseen input, using machine learning techniques to build a model of the algorithmʼs runtime as a function of problem-specific instance features. Such models have important applications to algorithm analysis, portfolio-based algorithm selection, and the automatic configuration of parameterized algorithms. Over the past decade, a wide variety of techniques have been studied for building such models. Here, we describe extensions and improvements of existing models, new families of models, and—perhaps most importantly—a much more thorough treatment of algorithm parameters as model inputs. We also comprehensively describe new and existing features for predicting algorithm runtime for propositional satisfiability (SAT), travelling salesperson (TSP) and mixed integer programming (MIP) problems. We evaluate these innovations through the largest empirical analysis of its kind, comparing to a wide range of runtime modelling techniques from the literature. Our experiments consider 11 algorithms and 35 instance distributions; they also span a very wide range of SAT, MIP, and TSP instances, with the least structured having been generated uniformly at random and the most structured having emerged from real industrial applications. Overall, we demonstrate that our new models yield substantially better runtime predictions than previous approaches in terms of their generalization to new problem instances, to new algorithms from a parameterized space, and to both simultaneously.