Analysis and Visualization of the Impact of Different Parameter Configurations on the Behavior of Evolutionary Algorithms
- 554 Downloads
Evolutionary algorithms are generic and flexible optimization algorithms which can be applied to many optimization problems in different domains. Depending on the specific type of evolutionary algorithm, they offer several parameters such as population size, mutation probability, crossover and mutation operators, or number of elite solutions. How these parameters are set has a crucial impact on the algorithm’s search behavior and thus affects its performance. Therefore, parameter tuning is an important and challenging task in each application of evolutionary algorithms in order to retrieve satisfying results.
In this paper, we show how software frameworks for evolutionary algorithms can support this task. As an example of such a framework, we describe how HeuristicLab enables automated execution of extensive parameter tests as well as its capabilities to analyze and visualize the obtained results. We also introduce a new chart of HeuristicLab, which can be used to compare the performance of many different parameter configurations and to drill down on different configurations in an interactive way. By this means this new chart helps users to visualize the influence of different parameter values as well as their interdependencies and is therefore a powerful feature in order to gain a deeper understanding of the behavior of evolutionary algorithms.
The work described in this paper is part of the COMET Project #843532 Heuristic Optimization in Production and Logistics (HOPL), funded by the Austrian Research Promotion Agency (FFG) and the Government of Upper Austria.
- 3.Wagner, S., et al.: Architecture and design of the HeuristicLab optimization environment. In: Klempous, R., Nikodem, J., Jacak, W., Chaczko, Z. (eds.) Advanced Methods and Applications in Computational Intelligence. Topics in Intelligent Engineering and Informatics, vol. 6, pp. 197–261. Springer, Heidelberg (2014). https://doi.org/10.1007/978-3-319-01436-4_10 CrossRefGoogle Scholar