Oríon : Experiment Version Control for Efficient Hyperparameter Optimization

Christos Tsirigotis, Xavier Bouthillier, François Corneau-Tremblay, Peter Henderson, Reyhane Askari, Samuel Lavoie-Marchildon, Tristan Deleu, Dendi Suhubdy, Michael Noukhovitch, Frédéric Bastien, Pascal Lamblin

Abstract: We present Oríon, a new black-box optimization tool currently in development that is designed to adapt to the workflow of machine learning researchers for minimal obstruction. We propose a new version control system for experiments, which can significantly improve the organization of research projects in machine learning as well as the efficiency of hyperparameter optimization. The entire tool is built with the goals of promoting reproducibility, fair benchmarking of different machine learning models, and providing a platform for the research of black-box optimization algorithms.

TL;DR: We present Oríon, a black-box optimization tool designed to adapt to the workflow of machine learning researchers, promoting reproducibility, fair benchmarking, and providing a platform for the research of black-box optimization algorithms.

Keywords: Black-box optimization, experiment management, version control