wenxu-uoft / hierminimax_icpp_ae

Artifact Evaluation for HierMinimax

HierMinimax_ICPP

Code for Distributed Minimax Fair Optimization over Hierarchical Networks (ICPP 2024).

We present our implementation in Jupyter Notebooks. We strongly recommend you use Google Colab to run these Jupyter Notebooks. Otherwise, you can open and run our Jupyter Notebooks via either Jupyter Notebook or JupyterLab, which can be downloaded from here if you do not have one. You will also need to install all required packages by uncommenting the first code block in each of the files.

This repo contains all code needed to replicate the results.

The code for the convex loss function setting is in HierMinimax_ICPP2024_CVX.ipynb and the code for the non-convex loss function setting is in HierMinimax_ICPP2024_NONCVX.ipynb.

Each Jupyter Notebook contains four code blocks: Installation (Please uncomment the pip comments if not using Google Colab), Utilities (all the utilizities methods and training implementation), Training (the code to run experiments and store results), plotting (the code to generate results the same as the paper)

After running all blocks, all results will be stored in the results folder and all figures for comparison of benchmarks will be plotted and stored in the images folder.