Software for Knockoffs

A versatile interface to the knockoff methodology.

Knockoffs

Knockoffs is the main software package for perfoming controlled variable selection with knockoffs.

Currently, it is available for R and Matlab, and it implements the methods described in the papers below.

The code is publicly available from GitHub: https://github.com/msesia/knockoff-filter.

Reference

“Panning for Gold: ”Model-X“ Knockoffs for High-dimensional Controlled Variable Selection”,
Emmanuel Candès, Yingying Fan, Lucas Janson, and Jinchi Lv. J. R. Stat. Soc. B. (2018). Link to the paper.

Reference

“Controlling the False Discovery Rate via Knockoffs”,
Rina Foygel Barber and Emmanuel Candès. Ann. Statist. 43 (2015). Link to the paper.

License: GPLv3

Main features

Generation of knockoff copies for Fixed-X variables.
Generation of knockoff copies for Model-X Gaussian variables.
Generation of second-order knockoff copies.
Computation of several importance statistics for knockoffs.
Computation of data-adaptive threshold for the importance statistics (the knockoff filter).

Authors

The knockoffs algorithms and the original MATLAB code were created by Rina Foygel Barber and Emmanuel Candès. The MATLAB package available on this site were created by Matteo Sesia, Evan Patterson and Lucas Janson. Currently, the maintainer is Matteo Sesia.

The R package was created by Matteo Sesia and Evan Patterson. Currently, the maintainer is Matteo Sesia.