Evolutionary Hyperparameter Tuning in ML.NET

Microsoft's ML.NET has proven to be a solid machine learning framework that is designed to reliably tackle common data science problem tasks. Since its release in 2018, this software library has enjoyed regular updates, performance improvements and a constantly expanding range of functionality. A relatively new extension is the Model Builder API (Microsoft AutoML) that strives to automatically train models by utilizing a variety of different hyperparameter tuning algorithms. In contrast to other frameworks for automated machine learning, among the provided mechanisms, evolutionary approaches like genetic algorithms or evolution strategies cannot be found, although they might have the potential to evolve optimal parameter configurations over time. Therefore, the aim of this paper is to extend Microsoft's AutoML API with evolutionary hyperparameter tuners and benchmarking them with the already existing algorithms based on different datasets.