Stochastic Extragradient: General Analysis and Improved Rates

This repository is the official implementation of the paper "Stochastic Extragradient: General Analysis and Improved Rates".

Installation

git clone https://github.com/hugobb/Stochastic-Extragradient.git
cd Stochastic-Extragradient
pip install .

Notebooks

To reproduce the results of the paper simply open and run the notebook gamesopt/experiments/Exp1 AISTATS 2022.ipynb.
To generate the plots you need to have LaTeX installed.

Structure of the code

There is two main class Game and Algorithm. To instantiate a quadratic game and run stochastic extragradient (SEG) with same sample:

game = QuadraticGame(dim, n_samples)
alg = SEG(game, lr=lr, lr_e=lr_e, same_sample=True)
results = alg.run(n_iter)

This will return a dict with different metrics.