BEAT: A Python Package for ML-Based Heterogeneous Treatment Effects Estimation

BEAT is a Python package for estimating heterogeneous treatment effects from observational data via machine learning:

• All arguments are the same as the original package, but there are two new inputs: target.weight.penalty indicates the penalty assigned to the protected attributes. target.weights is a matrix that includes the protected characteristics. X should not inlcude the protected characteristics.
• See full details about the BEAT method in the original paper: Eliminating unintended bias in personalized policies using bias-eliminating adapted trees (BEAT)
• Forked from https://github.com/Microsoft/EconML

Getting Started

Installation

Install the latest release from PyPI:

pip install BEAT_TEST

Usage Examples

Estimation Methods

from econml.grf import BeatForest
#Setting Training treatment and outcome 
treatment = ['W']
outcome = ['Y']
Y = train[outcome]
T = train[treatment]
#Setting Unprotected variables
unprotected_covariate = ['X.V1', 'X.V2', 'X.V3', 'X.V4', 'X.V5', 'Z.V1', 'Z.V2', 'Z.V3', 'Z.V4']
X1 = train[unprotected_covariate]
#set parameters for BEAT and Fit in training values
BEAT = BeatForest(alpha = 10, demean = 0, n_estimators = 8)                     
BEAT.fit(X1, T, Y) 
#Get prediction from test dataset
prediction = BEAT.predict(X_test)

References

Ascarza, E., & Israeli, A. (2022). Eliminating unintended bias in personalized policies using bias-eliminating adapted trees (beat). Proceedings of the National Academy of Sciences, 119(11). https://doi.org/10.1073/pnas.2115293119