Authors¹: Rachid Guerraoui, Nirupam Gupta, Rafaël Pinot, Sébastien Rouault, John Stephan

¹_{alphabetical order, as for all the papers from the Distributed Computing Laboratory (DCL) of EPFL.}

There is no particular dependence in the hardware used.
Any hardware which can run the following dependencies will do.

Besides Python 3.7.3 (or above) and its standard library, this software depends on the following Python packages:

• numpy 1.19.1
• torch 1.6.0
• torchvision 0.7.0
• pandas 1.1.0
• matplotlib 3.0.2
• tqdm 4.40.2
• PIL 7.2.0
• six 1.15.0
• pytz 2020.1
• dateutil 2.8.1
• pyparsing 2.2.0
• cycler 0.10.0
• kiwisolver 1.0.1
• cffi 1.13.2

You may probably use any later version for most of these libraries.

The entirety of our results can be reproduced in one command.
From the directory containing reproduce.py, please run:

$ python3 reproduce.py

This script will automatically download the dataset, run the experiments and produce the final graphs.