The IRLS sandbox

A safe place to play with iteratively reweighted least squares (IRLS), and variationally reweighted least squares (VRLS), algorithms for estimating sparse solutions to underdetermined linear systems.

Technical details of the algorithms are in preparation for submission to:

Worley, B., Variationally reweighted least squares for sparse recovery, Journal of Machine Learning Research, 2019.

Introduction

FIXME

Organization

Solver implementations

The source code for all implemented algorithms is stored in src.

Documentation

In order to establish a baseline level of notation for the algorithms implemented in src, a LaTeX writeup was created in notes that provides a sketch of the derivations required to obtain each algorithm.

Task management

This project uses doit to manage tasks, including compiling solver source code, running experiments, and making figures. All task management code is stored in sandbox.

Usage

FIXME The make.py script handles all tasks using doit. To see a listing of the supported tasks, simply run:

./make.py list --all

To see a listing of supported commands, run:

./make.py help

Running make.py without arguments will execute all tasks. Unless you're going on a long coffee break, this is not recommended.

Licensing

The irls-sandbox sources are released under the MIT license. See the LICENSE.md file for the complete license terms.

~ Brad.