robustInfer: Robust and Efficient Statistical Inference

RobustInfer provides implementation of modern robust statistical inference for online experimentation, e.g., regression adjustment, generalized estimating equations, Mann–Whitney U, zero-trimmed U, and Doubly Robust Generalized U.

RobustInfer contains a Python library that is tailored for small to medium-scale analysis, as well as a Scala (Spark) library that is tailored for large-scale analysis.

Theoretical details of the algorithms are described in the paper: https://arxiv.org/abs/2505.08128

Copyright

Copyright 2025 LinkedIn Corporation All Rights Reserved.

Licensed under the BSD 2-Clause License (the "License"). See License in the project root for license information.

Usage

Run Notebooks from Docker

1. Build the Docker Image: docker build -t robustinfer-notebook .
2. Run the Docker Container: docker run -p 8888:8888 -v $(pwd):/app robustinfer-notebook The -v $(pwd):/app mounts the project directory into the container. Access Jupyter Notebook at http://localhost:8888.
3. Example usage can be found in the notebooks folder.

Build

• to build: make build
• to clean: make clean
• to run tests: make test

References

@article{wei2025beyond,
  title={Beyond Basic A/B testing: Improving Statistical Efficiency for Business Growth},
  author={Wei, Changshuai and Nguyen, Phuc and Zelditch, Benjamin and Chen, Joyce},
  journal={arXiv preprint arXiv:2505.08128},
  year={2025}
}