Acceleration of Frank-Wolfe Algorithms with Open-Loop Step-Sizes

Code for the paper: Wirth, E., Kerdreux, T. and Pokutta, S., 2023, April. Acceleration of Frank-Wolfe Algorithms with Open-Loop Step-Sizes. In International Conference on Artificial Intelligence and Statistics (pp. 77-100). PMLR.

Installation guide

Download the repository and store it in your preferred location, say ~/tmp.

Open your terminal and navigate to ~/tmp.

Run the command:

$ conda env create --file environment.yml

This will create the conda environment open_loop_fw.

Activate the conda environment with:

$ conda activate open_loop_fw

Navigate to ~/tmp

To perform the experiments in the paper:

>>> python3 -m experiments.exterior

>>> python3 -m experiments.polytope

>>> python3 -m experiments.kernel_herding

>>> python3 -m experiments.gisette

>>> python3 -m experiments.movielens

The experiments are then stored in ~/tmp/experiments/figures.