Back to Reproducibility… in the context of a training session

Abstract

Within the context of the EJCP, we are in charge of a training day on Reproducibility in SE

• What is a day of training at the EJCP? 3h of course and 3-4h of training sessions/lab to be scheduled as we want.
• When? The week of July 3 to 7 in Valence, France.
• Audience? PhD students in the GDRGPL field.
• The ambition of the course To make the students aware of the reproducibility challenge with regard to non-functional properties related to the execution environment of the experiment:
  • challenges related to the stability of the performance;
  • challenges related to the execution environment’s heterogeneity;
  • challenges related to the scalability of the execution platforms and the software layer;
  • challenges related to the configuration of software layers (eg hyperparameters in ML, compilation/build options, default values);
  • … (See examples)

But also challenges related to the volatility of data in the context of the study of open-source projects.

In this context, I would like to introduce them to Software Heritage as an open archive full of valuable data in the field of software engineering.

I would like to take advantage of the coffee for a kind of pedagogical meeting on which training session topic could allow us to raise awareness on this challenge and illustrate which tools could mitigate some of these issues…

Consequently, the next coffee will have a short presentation and we will work in a small group to define different labs’ scenario. Come with some ideas ;)