Conférence de Frédérik Paradis, étudiant au doctorat en informatique, membre du  Groupe de recherche en apprentissage automatique de l’Université Laval (GRAAL)

Poutyne is a Python framework aiming to simplify research and development of neural networks with the automatic differentiation library PyTorch. Poutyne handles much of the boilerplate code needed to train neural networks. It provides a simple Model class that integrates the training and validation loops in addition to the computation of metrics. Furthermore, a powerful mechanism of callbacks is provided allowing the developer to personalize the training and testing. Many pre-defined callbacks are present in the library. Finally, Poutyne provides an Experiment class integrating many callbacks so that training can be stopped and resumed at will.

This class also logs everything in a single directory. Poutyne is an open source library and is available through the Python package manager pip.