Simple steps are all you need: Frank-Wolfe and generalized self-concordant functions

Repository for the paper "Simple steps are all you need: Frank-Wolfe and generalized self-concordant functions", NeurIPS 2021.

Find the preprint here, cite with the CITATION.bib entry.

The step sizes developed in the paper are available in the FrankWolfe.jl package as MonotonicStepSize and MonotonicGenericStepsize.

Due to their large size, the raw instance data files are not included in the repository but available on the Zenodo archive. Run the get_data_instances.sh bash script (or equivalent on your system) to fetch them. If wget or unzip are not available for you, download data.zip from the URL and place the data folder at the top-level of this repository. plotting/plot_experiment_results.py is used to generate the figures stored in Images/. Repository structure:

├── data # raw instance data obtained from get_data_instances.sh
├── Images # Final images
├── plotting # Plotting script
└── results # result JSON files

All top-level scripts generate data or results.

Requirements

The recommended Julia version is 1.6, the Project.toml and Manifest.toml should be used to instantiate the environment. The Python plotting script was run on Python 3.7.9 with matplotlib 3.3.3.