MlOps, a must!

Investment in machine learning projects is increasing every year in large companies. According to a Gartner report, the number of companies that have adopted ia tools has increased by 270% between 2015 and 2019.?

In France in particular, the government has made artificial intelligence a priority and has invested €1.5 billion over 5 years to encourage the development of artificial intelligence.

But it should not be forgotten that a large proportion of artificial intelligence projects end in failure.

Indeed, it is very difficult to complete this kind of project, because the tools made available to data scientists are not always the most suitable.?

Through this conference we will see the different tools that are part of the MlOps stack, and allow the different people who work on machine learning projects to carry out a project until production and to keep them up to date throughout the life cycle of the algorithms.?

We will go over the different stages that are essential in any machine learning project, and for each of these stages we will discover which tools will allow us to facilitate the work of the data scientist, but not only.?

The first tool that will be presented is the notebook. It allows data scientists to facilitate the exploratory part of the data, and thus, to start any project on a good basis.?

The second tool that will be presented is the Feature Store, which allows to store the different steps of data cleaning, as well as the different calculations for the creation of features.?

We will then come back to data versioning and source control, two tools that will allow you to go back and see the evolution of the data and the code throughout the project.?

Finally, we will talk about three tools that will allow us to work on the machine learning part, firstly the autoMl that we already presented at the Big Data exhibition last year, the notion of experimentation, which can be implemented with libraries such as MLFlow and finally the model registry which is a kind of model database.?

As for the tools for the production part, we will talk about the use of containers which allow us to have a stable and controlled environment, but also about the monitoring model, a crucial step to follow the evolution of our model in production and never lose sight of our objective.?

Have a good viewing!

Thibault Malherbe, Data Scientist?INETUM