pylift is an uplift library that provides, primarily:
- Fast uplift modeling implementations.
- Evaluation tools (
UpliftEvalclass).
While other packages and more exact methods exist to model uplift, pylift is designed to be quick, flexible, and effective. pylift heavily leverages the optimizations of other packages -- namely, xgboost, sklearn, pandas, matplotlib, numpy, and scipy. The primary method currently implemented is the Transformed Outcome proxy method (Athey 2015).
This branch is a fork from github.com/wayfair/pylift, and is actively being maintained.
The latest version of pylift can be installed through pypi:
pip install pylift
Licensed under the BSD-2-Clause by the authors.
Yi, R. & Frost, W. (2018). Pylift: A Fast Python Package for Uplift Modeling. Wayfair Tech Blog.
Athey, S., & Imbens, G. W. (2015). Machine learning methods for estimating heterogeneous causal effects. stat, 1050(5).
Gutierrez, P., & Gérardy, J. Y. (2017). Causal Inference and Uplift Modelling: A Review of the Literature. In International Conference on Predictive Applications and APIs (pp. 1-13).
Hitsch, G., & Misra, S. (2018). Heterogeneous Treatment Effects and Optimal Targeting Policy Evaluation. Preprint