P. Cotte (Advestis) & B. Gaston (CRIANN), JCAD 2022. Maison des Sciences de l’Homme, Dijon, 10-12 Octobre 2022.
Advestis et le CRIANN étaient présents aux "Journées CAlcul et Données - JCAD" 2022 avec deux communications sur le High Performance Computing (HPC) :
Un poster intitulé "Acceleration of a learning algorithm in Python on a HPC cluster using mpi4py", présenté par notre Lead Python Engineer Philippe Cotte, PhD.
Un talk dont le titre est "Execution of a feature engineering algorithm in Python on a supercomputer with Dask", animé par Benoist Gaston et Philippe Cotte, PhD.
Abstract Poster: Advestis' algorithm based on systematic exploration produces rules for various use cases, including investment recommendations. Advestis collaborated with the CRIANN to refactor and run the code on HPC infrastructures. The first goal is to see whether a very high number of cores and a better memory management could allow for a deeper exploration of the data compared to a local machine, where the hardware limitations prevent the exploration of all rules, hopefully resulting in better predictions. The first goal is to compare the speed and cost of running the code on a private cloud provider, here Google Cloud Plateform (GCP), with the public cluster "Myria", managed by the CRIANN. We found that an increase in the number of explored rules results in a net increase in the predictive performance of the final ruleset, and that Myria is twice faster and cheaper than GCP.
>> N'hésitez pas à nous contacter pour plus d'informations.