RobustRegression

This repository contains the package and the notebooks to reproduce the figures of the paper Asymptotic Characterisation of Robust Empirical Risk Minimisation Performance in the Presence of Outliers.

Code Organization

The package is contained in the folder robust_regression is subdivide into 5 other packages:

• sweeps : contains helpers to create sweeps of some parameters
• aux_functions : contains the definitions of fully vectorized (á la Numpy) function like $f_\mathrm{out}$ and $f_{\mathbf{w}}$.
• fixed_point_equations : contains the definitions of the fixed point form for the problems studied. Also it contains the functions to run a single fixed point iteration or to optimize over some parameters.
• regression_numerics : contains the relevant routines to perform ERM simulations and AMP simulations
• utils : miscellaneus stuff used for numerical integration or root finding.

The workspace contains also some resources that can be looked at when setting up a simulation.

The directory tests_robust_regression contains the unit tests for some functions of the package.

Installation

To install the package in the virtual environment of your choice (venv or condaenv) first activate the virtual environment and then in the folder of this package run

$ pip install -r requirements.txt
$ pip install .

Once installed the package can be imported as:

import robust_regression as rr

Contributing

Fell free to open pull request if you want to add something to the package.