Scikit-learn some perspectives Lundi 17 septembre 2018 Lancement - PowerPoint PPT Presentation

Lundi 17 septembre 2018 Lancement de l’initjatjve scikit-learn @ Fondatjon Inria Scikit-learn some perspectives

Lundi 17 septembre 2018 Lancement de l’initjatjve scikit-learn @ Fondatjon Inria Development dynamics & prospects

Lundi 17 septembre 2018 Lancement de l’initjatjve scikit-learn @ Fondatjon Inria Development dynamics

Scikit-learn : the vision • Democratjzing « machine learning » Mathematjcal building blocks of Accessible beyond AI mathematjcians and computer scientjsts • A tool also for productjon Avoid oversimplifjcatjon : we Quality sofuware engineering target a technical audience • Open source

10 years 2009 2017 v0.1, released by ODSC price for best tool Open Data Science Conference researchers at Inria 2011 1 st internatjonal sprint 2010 2011 2012 2013 2014 2015 2016 2017 2018 v0.1 v0.8 v0.10 v0.13 v0.15 v0.17 v0.18 v0.19 v0.20 SVM Données catégorielles kNN Gradient Boosting Données sparses Parallélisme Clustering Sélection de modèle Calcul distribué Détection d’anomalie Modèles linéaires Vitesse Random Forest

10 years Monthly website traffjc

A large impact >500 000 actjve users 12 000 academic citatjons 3 20 Windows 34 Academic Mac Industrial 50 Linux Other 63 30

Community-driven development Monthly actjve contributeurs

Community-driven development Contributor’s actjvity A few very actjve Actjvity people Many occasional contributors Contributor

A professional core Public Paid to work on the project (2018) : research • Inria : • Columbia : money Actjvity  O. Grisel  A. Mueller (50%) J. du Boisberranger   stagiaires • Sydney university :  G. Lemaître (50%) J. van den Bossche (50%)   J. Nohman (50%) Contributor

Scikit-learn @ fondation Inria Sponsoring Mix community-driven development with industrial interest • Goals : stjmulate the development of scikit-learn • Mixte governance between community and sponsors  Keep the energy and the confjdence of the community  Take strategic recommendatjons from the industry

Lundi 17 septembre 2018 Lancement de l’initjatjve scikit-learn @ Fondatjon Inria Development prospects

Positioning Computational performance Polyvalence Model complexity Large- Databases scale

Scikit-learn, new ambitions Data integratjon Interpretability & understanding • Missing data • Model interpretatjon • Categorical values • Confjdence ? Scaling up ? • Distributed computjng ? Technical committee • Vendor acceleratjons More frequent versions

Thanks to our partners

Alexandre Gramfort Alexander Fabisch Alexandre Passos Andreas Mueller Arnaud Joly Brian Holt Bertrand Thirion David Cournapeau David Warde-Farley Fabian Pedregosa Gael Varoquaux Guillaume Lemaitre Gilles Louppe Jake Vanderplas Jaques Grobler Jan Hendrik Metzen Jacob Schreiber Joel Nothman Joris Van den Bossche Kyle Kastner Lars Buitjnck Loïc Estève Shiqiao Du Mathieu Blondel Manoj Kumar Noel Dawe Nelle Varoquaux Olivier Grisel Paolo Losi Peter Pretuenhofer Hanmin Qin Raghav Rajagopalan Robert Layton Ron Weiss Roman Yurchak Tom Dupré la Tour Vlad Niculae Vincent Michel Wei Li And many others