Gradient Boosting Mapping is and nonlinear dimensionality reduction and feature creation method. This repository contains the Python code for GBMAP and the experiments in the paper:
Patron, A., Savvides, R., Franzon, L., Luu, H.P.H., Puolamäki, K. (2025). Fast and Understandable Nonlinear Supervised Dimensionality Reduction. In: Pedreschi, D., Monreale, A., Guidotti, R., Pellungrini, R., Naretto, F. (eds) Discovery Science. DS 2024. Lecture Notes in Computer Science(), vol 15243. Springer, Cham. https://doi.org/10.1007/978-3-031-78977-9_25
Most of the datasets are from OpenML and will be downloaded as needed, but GeckoQ has to be downloaded separately.
The GeckoQ data can be downloaded from Fairdata repository, you'll only have to download the Dataframe.csv and place it to experiments/data.
Script run.sh contains explicit instructions how to install and run the experiments. The results for the experiments are placed to experiments/results and figures to experiments/figures.
If you use the code, please cite the paper:
@incollection{patron2025Fast,
title = {Fast and {{Understandable Nonlinear Supervised Dimensionality Reduction}}},
booktitle = {Discovery {{Science}}},
author = {Patron, Anri and Savvides, Rafael and Franzon, Lauri and Luu, Hoang Phuc Hau and Puolam{\"a}ki, Kai},
year = {2025},
volume = {15243},
pages = {385--400},
publisher = {Springer Nature Switzerland},
address = {Cham},
doi = {10.1007/978-3-031-78977-9_25},
}