LogitBoost (No Longer Maintained)

This is a Python implementation of the LogitBoost classification algorithm¹ built on top of scikit-learn. It supports both binary and multiclass classification; see the examples.

This package provides a single class, LogitBoost, which can be used out-of-the-box like any sciki-learn estimator.

Documentation website: https://logitboost.readthedocs.io

Installation

The logitboost package can be installed using the pip utility. For the latest version, install directly from the package's GitHub page:

pip install git+https://github.com/artemmavrin/logitboost.git

Alternatively, install a recent release from the Python Package Index (PyPI):

pip install logitboost

Note. To install the project for development (e.g., to make changes to the source code), clone the project repository from GitHub and run make dev:

git clone https://github.com/artemmavrin/logitboost.git
cd logitboost
# Optional but recommended: create a new Python virtual environment first
make dev

This will additionally install the requirements needed to run tests, check code coverage, and generate documentation.

This project was developed in Python 3.7, and it is tested to also work with Python 3.6 and 3.8.

References

---

1. Jerome Friedman, Trevor Hastie, and Robert Tibshirani. "Additive Logistic Regression: A Statistical View of Boosting". The Annals of Statistics. Volume 28, Number 2 (2000), pp. 337–374. JSTOR. Project Euclid.↩