Bachelor theses
A Heuristic for Automatic Distribution Selection in Open Source H2O AutoML
Author
Vojtěch Müller
Year
2023
Type
Bachelor thesis
Supervisor
Ing. Tomáš Frýda
Reviewers
Ing. Karel Klouda, Ph.D.
Department
Summary
This thesis aims to create a heuristic for the automatic selection of a hyperparameter that represents a statistical distribution of the target variable in the H2O AutoML framework. Various approaches were tested, providing different results, for instance, using artificially generated data, using datasets from OpenML platform, or different benchmark methods. The proposed heuristic improves the prediction performance in four followed criteria by about 5 %. This one is implemented to the H2O AutoML. The thesis also examines unsuccessful attempts to provide a solid baseline for future improvements.