Python code for 2SEGP, which was proposed in:
@inbook{virgolin2021genetic,
author = {Virgolin, Marco},
title = {Genetic Programming is Naturally Suited to Evolve Bagging Ensembles},
year = {2021},
isbn = {9781450383509},
publisher = {Association for Computing Machinery},
address = {New York, NY, USA},
url = {https://doi.org/10.1145/3449639.3459278},
booktitle = {Proceedings of the Genetic and Evolutionary Computation Conference},
pages = {830–839},
numpages = {10}
}
Differently from classic evolution which returns a single model, 2SEGP returns an ensemble. Nice aspects of 2SEGP are that:
- it is a very simple modification of classic GP
- it has only 1 additional hyper-parameter compared to classic GP (ensemble size) and it is rather robust to different choices
- it runs almost as fast as classic GP (see Big-O's in the paper)
- despite being simpler than other existing approaches, it typically performs equal or better than them
- main ideas are portable to other evolution-based algorithms (e.g., neural architecture search)
Run pip install . from inside this folder.
Requires numpy >= 1.16.1, scikit-learn >= 0.20.0, simplegp >= 0.9.9.
Here's how to reproduce one run for a dataset "mydataset.csv"
cd Reproduce
python run.py mydataset
Please find all datasets used in the paper in Reproduce/Datasets.
Of course, perform multiple runs to obtain the same statistics of the paper.