Robust Chauvenet Rejection (RCR)

Advanced, but easy-to-use, statistical outlier rejection — robust enough for datasets that are >85% contaminated.

This repository hosts the multi-language reimplementation effort for RCR, originally published in Maples et al. 2018 and shipped as a C++ library with Python bindings.

The work here is a fresh restructuring around three tracked implementations:

Track	Status	Purpose
cpp/	Reference. Frozen upstream code.	Source of truth for algorithmic behavior.
python/	Planned.	Pure-Python reimplementation — readability, hackability, parity tests.
rust/	Planned.	High-performance reimplementation — safety, portability, optional Python bindings via PyO3.

The Python and Rust ports are intentionally separate trees so each can have its own idiomatic layout (Cargo workspace, pyproject.toml, etc.) without contorting around the legacy C++ tree.

What is RCR?

The naive form of outlier rejection is sigma clipping: drop points more than k standard deviations from the mean. Chauvenet (1863) gave a principled prescription for k as a function of sample size. Iterating this rule is "traditional Chauvenet rejection".

The catch: the mean and standard deviation are themselves polluted by the very outliers they are being used to reject. That makes traditional Chauvenet useless on heavily contaminated samples.

Robust Chauvenet Rejection (RCR) swaps in robust replacements — the median or half-sample mode for the center, and calibrated robust replacements for the spread — and applies the iterative rejection on top. It has been simulated extensively and works on samples with contamination fractions in excess of 90%.

For the full method, see the paper (included) and the short preprint.

Repository layout

robust-chauvenet-rejection/
├── README.md            <- you are here
├── AGENT.md             <- ground rules for AI agents working in this repo
├── assets/
│   └── test/            <- canonical test datasets, shared across implementations
│       ├── README.md
│       ├── data_smoke.csv
│       ├── data_singlevalue.csv
│       ├── data_weighted_singlevalue.csv
│       ├── data_linear.csv
│       └── data_exponential.csv
├── agents/              <- exploratory plans and notes for the reimplementation
│   ├── README.md
│   ├── python_vs_rust_plan.md
│   └── test_data_generation.md
├── cpp/                 <- frozen reference implementation (pybind11 module)
│   ├── src/             <- C++ source + pybind11 glue
│   ├── tests/           <- existing pybind smoke tests
│   ├── testdata/        <- original CSVs (also mirrored to assets/test/)
│   ├── docs/            <- Sphinx site sources
│   └── ...
├── python/              <- (planned) pure-Python implementation
└── rust/                <- (planned) Rust implementation

Quickstart (current state)

The only currently runnable implementation is the upstream C++/pybind11 module under cpp/.

python3 -m pip install rcr        # installs from PyPI, not this checkout
python3 cpp/tests/test.py         # smoke-test the installed module

To build from this checkout instead:

cd cpp
python3 -m pip install pybind11
python3 -m pip install -e .
python3 tests/maintest.py

The Python and Rust ports do not exist yet — see agents/python_vs_rust_plan.md for the plan to bring them up.

Testing data

All implementations are validated against the same CSVs under assets/test/. New implementations should add a thin loader that reads from that directory rather than re-generating the data, so that cross-implementation results stay comparable. See assets/test/README.md for per-file expectations.

Citation

If you use RCR in academic work, cite Maples et al. 2018:

@article{maples2018robust,
    title  = {Robust Chauvenet Outlier Rejection},
    author = {Maples, M.P. and Reichart, D.E. and Konz, N.C. and Berger, T.A.
              and Trotter, A.S. and Martin, J.R. and Dutton, D.A.
              and Paggen, M.L. and Joyner, R.E. and Salemi, C.P.},
    journal = {The Astrophysical Journal Supplement Series},
    volume  = {238}, number = {1}, pages = {2}, year = {2018},
    publisher = {IOP Publishing}
}

License

See cpp/LICENSE. RCR is free for academic and non-commercial use; contact the authors for commercial licensing.

Acknowledgements

Original C++/Python author: Nick C. Konz. Former author: Michael Maples. Originally developed at the Department of Physics and Astronomy, University of North Carolina at Chapel Hill.