Genens: An AutoML System for Ensemble Optimization Based on Developmental Genetic Programming

Gabriela Suchopárová, Roman Neruda

Abstract: We propose an AutoML system for pipeline optimization based on developmental genetic programming - genens. It is built atop of scikit-learn pipelines, and it focuses on both hyperparameter and architecture optimization. Compared to existing systems, it enables to optimize more complex ensembles, while exploring simpler models at the same time. The system has been evaluated on selected benchmark datasets from the AutoML benchmark, producing competitive results.