Personalized Federated Learning via Stacking

This repository contains the code for Personalized Federated Learning via Stacking.

Structure

• main.py: Main script for launching a single simulated federation, typically used with scripts in launch_scripts/.
• data.py: Handles dataset fetching and preprocessing.
• partitioning.py: Contains functions for natural, quantity skew, label skew, and vertical partitioning.
• transformers.py: Includes a custom scikit-learn transformer for dealing with vertically partitioned data.
• figs.ipynb: Processes results and generates figures as presented in the paper.

Reproducibility

To reproduce the results and figures from the paper, follow these steps:

1. Install Python package requirements with pip install -r requirements.txt.
2. Obtain a Kaggle API key by following these instructions.
3. Run sh launch_scripts/all.sh.
4. Open and execute figs.ipynb.