UTBoost

UTBoost is a powerful uplift modeling library based on boosting framework over decision trees. It can handle large-scale RCT (randomized controlled trial) datasets and demonstrates superior predictive performance.

Documentations

Quick Start

See the tutorial notebook for details.

# import approaches
from utboost import UTBClassifier, UTBRegressor

# define model (CausalGBM algorithm)
model = UTBClassifier(
    ensemble_type='boosting',
    criterion='gbm',
    iterations=20,
    max_depth=4
)

# fit model
model.fit(X=X_train, ti=ti_train, y=y_train)

# predict outcomes
preds = model.predict(X_test)
# predict uplift
uplift_preds = preds[:, 1] - preds[:, 0]

File Locations

src/* — C++ code that ultimately compiles into a library
include/ — C++ header files
python-package/ — python package

License

This project is open-sourced under the MIT license. You can find the terms of the license here.

Reference Paper

Junjie Gao, Xiangyu Zheng, DongDong Wang, Zhixiang Huang, Bangqi Zheng, Kai Yang. “UTBoost: A Tree-boosting based System for Uplift Modeling“.