A. Beham, S. Wagner, M. Affenzeller - Optimization Knowledge Center - Companion Publication of the 2016 Genetic and Evolutionary Computation Conference, GECCO'16 Companion, Denver, Colorado, Vereinigte Staaten von Amerika, 2016, pp. 1331-1338
The task of selecting an appropriate algorithm instance for a given optimization problem instance often requires significant experience. Efficient optimization requires a different set of parameters or an entirely different algorithmic approach for some characteristics of problem instances. Obtaining such experience takes significant amount of time and requires an in-depth analysis of the algorithms' performance. In addition to these difficulties, published results only provide a summary, the obtained raw performance data is often not reused later on. In this work we want to give such data more value and more publicity by storing it in a database and reusing it when solving new problem instances. We describe the information that the data should contain in order to maximize reusability. Furthermore, we discuss three use cases that supports optimization experts in their decisions and allows them to perform a manual exploration of the search space using available algorithm instances and the possibility to decide on the starting solutions and thus bias the search in a certain sub-space of the solution space.