kuplift 0.0.15

A User Parameter-free Bayesian Framework for Uplift Modeling

kuplift package

kuplift is a Python package that provides a series of uplift modeling methods based on recent research work. kuplift allows users to easily use the following algorithms:

1. Encoding data using a discretization method for treatment effect (uplift) modeling called UMODL.

2. OptimizedUnivariateEncoding is an optimized version of the umodl algorithm written in C++ for continuous and categorical variables.

3. Variable selection for uplift modeling with UMODL-FS.

4. Learning a Bayesian decision tree model for uplift modeling with UB-DT.

5. Learning a random forest model for uplift modeling with UB-RF.

How to install

$ pip install kuplift

User Guide

import kuplift as kp
import pandas as pd

df = pd.read_csv("data.csv")

# Make sure the dtype of all categorical variables is object
df = df.astype({"some_categorical_variable": object})

variables = list(df.columns[:-2])  # Last two columns are treatment and target columns

# Univariate variable transformation
ue = kp.UnivariateEncoding()
encoded_data = ue.fit_transform(df[variables], df["treatment"], df["target"])

# Univariate variable transformation optimized through the use of the C++ implementation of umodl
ue = kp.OptimizedUnivariateEncoding()
encoded_data = ue.fit_transform(df[variables], df["treatment"], df["target"])

# Feature selection
fs = kp.FeatureSelection()
important_vars = fs.filter(df[variables], df["treatment"], df["target"])

# Uplift Bayesian Decision Tree
tree = kp.BayesianDecisionTree()
tree.fit(df[variables], df["treatment"], df["target"])
preds = tree.predict(df[variables])

# Uplift Bayesian Random Forest
forest = kp.BayesianRandomForest(n_trees=4)
forest.fit(df[variables], df["treatment"], df["target"])
preds = forest.predict(df[variables])

Examples

You can find examples in the examples directory.

Documentation

Refer to the documentation at https://udata-orange.github.io/kuplift/

Credits

kuplift has been developed at Orange Labs.

Current contributors

Mina Rafla

Nicolas Voisine

References

Rafla, M., Voisine, N., Crémilleux, B., & Boullé, M. (2022, September). A non-parametric bayesian approach for uplift discretization and feature selection. ECML PKDD 2022

Rafla, M., Voisine, N., & Crémilleux, B. (2023, May). Parameter-free Bayesian decision trees for uplift modeling. PAKDD 2023