Skip to content

An open source python library for automated feature engineering based on Genetic Programming

License

Notifications You must be signed in to change notification settings

hengzhe-zhang/EvolutionaryForest

Repository files navigation

Evolutionary Forest

Documentation Status Updates

An open source python library for automated feature engineering based on Genetic Programming

Introduction

Feature engineering is a long-standing issue that has plagued machine learning practitioners for many years. Deep learning techniques have significantly reduced the need for manual feature engineering in recent years. However, a critical issue is that the features discovered by deep learning methods are difficult to interpret.

In the domain of interpretable machine learning, genetic programming has demonstrated to be a promising method for automated feature construction, as it can improve the performance of traditional machine learning systems while maintaining similar interpretability. Nonetheless, such a potent method is rarely mentioned by practitioners. We believe that the main reason for this phenomenon is that there is still a lack of a mature package that can automatically build features based on the genetic programming algorithm. As a result, we propose this package with the goal of providing a powerful feature construction tool for enhancing existing state-of-the-art machine learning algorithms, particularly decision-tree based algorithms.

Features

  • A powerful feature construction tool for generating interpretable machine learning features.
  • A reliable machine learning model has powerful performance on the small dataset.

Installation

From PyPI:

pip install -U evolutionary_forest

From GitHub (Latest Code):

pip install git+https://github.com/hengzhe-zhang/EvolutionaryForest.git

Supported Algorithms

Example

An example of usage:

X, y = load_diabetes(return_X_y=True)
x_train, x_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=0)
r = EvolutionaryForestRegressor(max_height=3, normalize=True, select='AutomaticLexicase',
                                gene_num=10, boost_size=100, n_gen=20, n_pop=200, cross_pb=1,
                                base_learner='Random-DT', verbose=True)
r.fit(x_train, y_train)
print(r2_score(y_test, r.predict(x_test)))

An example of improvements brought about by constructed features:

https://raw.githubusercontent.com/zhenlingcn/EvolutionaryForest/master/docs/constructed_features.png

Tutorials

Here are some nodebook examples of using Evolutionary Forest:

Documentation

Tutorial: English Version | 中文版本

Credits

This package was created with Cookiecutter and the audreyr/cookiecutter-pypackage project template.

Citation

Please cite our paper if you find it helpful :)

@article{zhang2021evolutionary,
  title={An Evolutionary Forest for Regression},
  author={Zhang, Hengzhe and Zhou, Aimin and Zhang, Hu},
  journal={IEEE Transactions on Evolutionary Computation},
  volume={26},
  number={4},
  pages={735--749},
  year={2021},
  publisher={IEEE}
}

@article{zhang2023sr,
  title={SR-Forest: A Genetic Programming based Heterogeneous Ensemble Learning Method},
  author={Zhang, Hengzhe and Zhou, Aimin and Chen, Qi and Xue, Bing and Zhang, Mengjie},
  journal={IEEE Transactions on Evolutionary Computation},
  year={2023},
  publisher={IEEE}
}