Deconstructing GAs into Visual Software Components
Leidy Patricia Garzón-Rodriguez, Henry Alberto Diosa, Sergio Rojas–Galeano · 2015
We envisage Genetic Algorithms (GA) as search-based optimisation techniques encompassing independent bio-inspired operators and representations that are realizable as self-contained deployable computational units. In other words, we think of GAs as a set of software components conforming to a formally-defined evolution-oriented composition model. Furthermore, we imagine such components being assembled on a visual programming-free board, much like prefabricated electronic chips are wired up to build electronic devices. Here we introduce Goldenberry-GA, a toolbox of visual software components complying with these premises that has been built over the Orange framework for data mining. The paper describes at user-level the suite of new released components (GeneticAlgorithm, InitialPopulation, SolutionRepresentation, Selection, Mutation, Crossover), including working examples that demonstrate some advantages of the reuse and extension principles of its underlying Component-based software architecture. It also explains the composition model specification of the toolbox and the software design patterns that were taken into account during its development. A qualitative comparative study with similar Evolutionary Computation frameworks is reported so as to highlight strengths and weaknesses of the toolbox, as well as to point out directions for future work.