xuplift is a library for explainable uplift modeling. It uses linearized kernel feature maps to estimate treatment effects with both speed and mathematical rigor. Instead of computing a massive $N \times N$ kernel matrix, xuplift selects landmark points to project data into a finite-dimensional feature space.

Supported Models

• Regressor: Kernel-based Ridge regressor for outcome and residual modeling.
• Classifier: Kernel-based Logistic classifier for precise propensity score estimation.

Supported Meta-Learners

• DRClassifier, DRRegressor: Doubly robust estimator combining propensity scores and outcome models.
• GRClassifier, GRRegressor: Generalized R-learner supporting both continuous and binary treatments.
• MRegressor: Modified covariates learner optimized for randomized controlled trials (RCT).
• PWRegressor: Propensity score weighted learner using inverse probability weighting.
• RClassifier, RRegressor: Residual learner minimizing an R-objective via residual-on-residual regression.
• SClassifier, SRegressor: Single learner treating treatment assignment as a standard feature.
• TClassifier, TRegressor: Two learner approach fitting independent models for each group.
• XClassifier, XRegressor: Cross learner optimized for significantly unbalanced treatment groups.

Installation

pip install xuplift