Extending the “Open-Closed Principle” to Automated Algorithm Configuration

Abstract

Metaheuristics are an effective and diverse class of optimization algorithms: a means of obtaining solutions of acceptable quality for otherwise intractable problems. The selection, construction, and configuration of a metaheuristic for a given problem has historically been a manually intensive process based on experience, experimentation, and reasoning by metaphor. More recently, there has been interest in automating the process of algorithm configuration. In this article, we identify shared state as an inhibitor of progress for such automation. To solve this problem, we introduce the Automated Open-Closed Principle (AOCP), which stipulates design requirements for unintrusive reuse of algorithm frameworks and automated assembly of algorithms from an extensible palette of components. We demonstrate how the AOCP enables a greater degree of automation than previously possible via an example implementation.

Metaheuristics "In the Large"

“…In software engineering, a framework which can be configured from an open-ended palette of components while remaining unchanged is said to conform to the "Open-Closed principle" ('a framework should be closed to modification, but open to extension by new components'). We have extended this principle [106] to incorporate the behavior required to support automated design, yielding the "Automated Open Closed Principle" (AOCP); that paper discusses the issue in greater technical detail and describes experiments with a suitably equipped algorithm configurator. The adoption of the AOCP provides open-ended reuse of components, and thus a systematic approach to the automated exploration of the metaheuristic design space.…”

“…[29,19,12,77]), to the best of our knowledge only MitL and CILib employ this principled approach to open-ended state dependencies. The additional contribution of the MitL initiative in this respect is the use of the monadic approach to explicitly support automated assembly [106]: of particular value in this respect is the fact that a strict type-system can be used to discriminate between stateless and stateful operations and to provide information about which aspects of component behaviour contribute to solution quality, this being vital for the elimination of accidental complexity.…”

“…It uses antprogramming as a generative hyper-heuristic [56] to automatically configure a local search framework with components which have different state dependencies. Further technical specifics of enabling communal research via extensible algorithm templates are described in detail in Swan et al [106].…”

“…• Implementation: despite past efforts in developing software frameworks, there is a tendency to re-implement metaheuristics from scratch, hindering reproducibility and replicability [107,78,106].…”

“…We rely heavily on functional programming constructs to express metaheuristic components in a truly reusable way. In constructing this synthesis, we have exposed gaps in the literature that we are now closing with new contributions: in particular the MitL initiative has introduced a) the "Automated Open Closed Principle" [106], which shows how to express algorithm frameworks as 'closed' design spaces which can nonetheless be configured in an open-ended manner via combinatorial assembly, and b) the removal of the domain barrier from hyper-heuristics [108], essential in raising the level of genericity with respect to problem domains. These address Q3, making easier automation of the space of metaheuristics, and hence progress towards answers to questions Q1 and Q2.…”