PAF Prediction Challenge Reproducibility

Reproducibility of Machine Learning Models for Paroxysmal Atrial Fibrillation Onset Prediction

Cédric Gilon¹, Jean-Marie Grégoire^1,2, Jérome Helinckx¹, Stéphane Carlier ², Hugues Bersini¹

1. IRIDIA, Université Libre de Bruxelles, Belgium
2. Département de Cardiologie, Université de Mons, Belgium

Computers in Cardiology 2022, Tampere Finland

Dependancies

The source code is written in Python3 and is using Poetry as virtual environment and package manager. The package is divided between the data and the src folders. The src folder contains the features scripts, the models and finally some utiland visualization tools.

To create the virtual environment and install the dependancies, you can use:

poetry install

Run models

To run the scripts, the src folder should be in the PYTHONPATH. The run each model you can use

python src/models/20xx_model_name/main_model_name.py

The results are stored in csv files and the figures are saved as png files.

Data

The data are already in the repositroy and were downloaded from the PAF Prediction Challenge Database, which is available on the Physionet website. Labels are also still available on the results page.

Licence

GNU General Public License v3.0

References

(1) https://physionet.org/content/afpdb/1.0.0/

(2) https://physionet.org/content/challenge-2001/1.0.0/