Triplets-Oversampling-for-Federated-Datasets

Project for paper "Triplets Oversampling for Federated Datasets" submitted to ECML-PKDD2023

To reprocude the results of the paper, you need to follow the procedure below:

1. Clone or copy the repository to your local machine
2. Open a terminal and go to the folder "Triplets_Oversampling_for_Federated_Datasets"
3. Follow Quick start to generate the results of the paper simply.

Quick start

You can reproduce the results with provided scripts.

Prepare the virtual environment and packages with conda:

source ./prepare_venv.sh

Run the centralized learning experiment:

source ./run_cl.sh

Run the federated learning experiment:

source ./run_fl.sh

Python environment and dependencies:

• uv installed
• Python 3.10
• Required package is listed in the file "requirements.txt"

Synthesis quality comparison

The source code for the synthesis quality comparison in the paper is also published. You run the jupyter notebook to get the comporation figures yourself.

You may also customize the dataset you prefer to check the synthesis quality of selected sampling algorithms.