lsboost is a regression boosting algorithm for multicalibration defined in (Globus-Harris et al. 2023). Multicalibration is a realtively new notion of fair machine learning designed to ensure that identified subgroups of the population in your data do not receive predictions which are far away from their conditional label mean.
This package is not yet available for installation via pip and thus must be downloaed from this repository. You can easily download the repository by clicking on the green code button and following the GitHub instructions listed. The following command can be run in the terminal for example:
git clone https://github.com/Declancharrison/Level-Set-Boosting.gitA notebook titled LSBoost_notebook.ipynb has been provided to give an example for using lsboost on census data from the Folktables package. Further descriptions of hyperparameters and their uses can be found in the init for the class LSBoostingRegressor in LSBoost.py.
If you use lsboost, please cite the paper it originates from:
@misc{globusharris2023multicalibration, title={Multicalibration as Boosting for Regression}, author={Ira Globus-Harris and Declan Harrison and Michael Kearns and Aaron Roth and Jessica Sorrell}, year={2023}, eprint={2301.13767}, archivePrefix={arXiv}, primaryClass={cs.LG} }
There is a known problem with MacOS using the parallel implentation in this package. We are currently working on a fix and hope to have this resolved shortly. If you run into any issues or have questions about the package, feel free to create an issue and we will get back to you as quick as possible!