pinball-qr 1.0.0

Fast and accurate quantile regression for Python, with an sklearn-compatible interface.

The fastest and most accurate methods for quantile regression, now in python.

• Free software: MIT license

• Documentation: https://mhoward2718.github.io/pinball-qr/

Features

• sklearn-compatible API — QuantileRegressor works with pipelines, cross-validation, and all the usual sklearn tooling.

• Multiple solvers — Barrodale-Roberts simplex (br), Frisch-Newton interior point (fn/fnb), preprocessed Frisch-Newton (pfn), LASSO-penalised (lasso), and ADMM via POGS (pogs).

• Statistical inference — IID, NID, and kernel standard errors, rank inversion confidence intervals, and bootstrap methods.

• Nonparametric quantile regression — optimal quantization via Competitive Learning Vector Quantization (CLVQ).

Credits & Acknowledgements

Pinball-QR is authored by Michael Howard.

This project builds on the work of several researchers and open-source projects:

quantreg (R package)

The Fortran solvers (rqbr, rqfn, rqfnb) are ported from the quantreg R package by Roger Koenker, with contributions from Stephen Portnoy, Pin Tian Ng, Blaise Melly, Achim Zeileis, and others. See Koenker, R. (2005) Quantile Regression, Cambridge University Press.

POGS (Proximal Operator Graph Solver)

The ADMM-based solver uses C++ source from POGS by Christopher Fougner (MIT license). See Fougner & Boyd (2018), Parameter Selection and Pre-Conditioning for a Graph Form Solver, in Emerging Applications of Control and Systems Theory, Springer.

QuantifQuantile (R package)

The nonparametric quantile regression via optimal quantization is based on the QuantifQuantile R package by Isabelle Charlier, Davy Paindaveine, and Jérôme Saracco. See Charlier, Paindaveine & Saracco (2015), Estimation of Conditional Quantiles using Optimal Quantization.