Algorithms, pseudo-codes, implementations – some reflections on interdependencies and potential implications

The article investigates different implementations of a Variable Neighborhood Search approach for the Capacitated Vehicle Routing Problem. An important goal of the work presented here is to obtain an understanding of the potential effects of implementation details on the observed performance of the algorithm.

Our work is motivated by our experiences in implementation competitions, in which best-possible algorithms to hard optimization problems must be devised and delivered. In those contexts, we may ask: what determines the success of an algorithm, (i) some key algorithmic ingredient, (ii) coding details, or (iii) other aspects? Here, we try analysing (ii) and (iii). Overall, we are able to identify some important details that are often neglected in scientific articles. Moreover, and in order to allow anyone to re-produce our findings, we make the source-code of our implementation available online with this article [Martin Josef Geiger. Different implementations of a metaheuristic for the capacitated vehicle routing problem. Mendeley Data, 2018. https://dx.doi.org/10.17632/s4h89j4tgp.2].