causalml

https://arxiv.org/abs/2009.01561

Causalis - State-of-the-art robust causal inference for experiments and observational data in python

Repository for DoubleML tutorial at UseR!2022, UAI2022 and Tools for Causality 2023

Replication of Simulations in Bach et al. (2024) - DoubleML - An Object-Oriented Implementation of Double Machine Learning in R, https://doi.org/10.18637/jss.v108.i03

Shiny App illustrating the Key Ingredients of the Double Machine Learning Approach

KDD 2024 2nd Workshop on Causal Inference and Machine Learning in Practice

Reproducible benchmark of 7 uplift modeling approaches across 5 public datasets (Hillstrom, Criteo, Lenta, RetailHero, MegaFon) + synthetic DGP, with bootstrap CIs, robustness analysis, and presentation-ready comparison plots.

There are all learning materials that I used or assignments that I completed.

Uplift modeling & retention API — predicting which customers a retention contact can actually save. Live on Render

This is the replication of one R tutorial introduced in Machine Learning and Econometrics tutorial in AEA Annual Meeting 2018.

Uplift-модель для таргетированного маркетинга (промокоды). T-learner, R-learner, Optuna, MLflow, Uplift@30%.

A slide deck and set of jupyter notebooks created during my time with Fellowship.ai in order to decide what method would be best for predicting churn using the most up to date and innovative methods.

Exploratory project using uber's causalml to estimate average treatment effects on IT incident durations.

A single behavioral nudge nearly 5×'d email click-through in a randomized experiment. This project proves the effect causally, scales it to ~200K extra clicks, and uses Causal ML to pinpoint exactly which customers to target.

Replicates Nigeria egg demand-creation study; adds targeting & media-mix optimization.

Causal uplift modeling (X-Learner) on a 3-arm email A/B test, identifies persuadable customers and cuts targeting volume 50% while capturing 61-72% of incremental conversions.

Uplift modeling: choosing whom to offer a promo code to, with causalml, scikit-uplift and CatBoost

Comparison of 6 uplift-modeling approaches on the Lenta dataset

Who should get which email? Bayesian A/B/C test and uplift model on 64k customers; best email doubles conversion vs none

Reproducible benchmark of uplift-modeling approaches (meta-learners, causal forests, DML, IV) on the Criteo Uplift dataset.