Experimenting with a model or comparing a model's performance with other solutions is often a tricky task. Problems can include absence of an intuitive graphical interface and lack of an SDK that allows this work to be done neatly.
MLflow is a perfect solution. MLflow is an open-source project that covers all the aspects of an ML workflow, from model exploration to production and retirement. Easily integrated into any ML library, it can be run on the cloud and it allows an easy cross-team collaboration. For these reasons, MLflow has been widely used across industries, from Booking.com to Microsoft and Facebook.
In this talk, Stefano Bosisio will explore MLflow Tracking. Tracking allows data scientists to have an intuitive graphical interface to understand and immediately visualise their models' experiments performance and metrics. We will go through a Python SDK implementation of MLflow, to show not only how MLflow is easy to be integrated with custom-based products but also to give data scientists an easy interface to immediately experiment with their models.
Stefano Bosisio is a Machine Learning Engineer at Trustpilot in Edinburgh - https://www.linkedin.com/in/stefano-bosisio1
At Trustpilot, he looks for new solutions to help data scientists better scale up their models and land easily in production.
Before joining Trustpilot, Stefano worked for Sainsbury's Bank. His role involved bringing the bank to a more data-centric position, by creating new models for financial products and paving the road for a new ML infrastructure for model development and deployment.
Stefano holds a PhD in Chemistry from the University of Edinburgh. When he’s not working, Stefano enjoys playing the piano, painting, crocheting (lockdown skills), as well as travelling and getting better and better at baking.
You can read Stefano’s write up on the topic below.
Scale-up your models development with MLflow | by Stefano Bosisio | Nov, 2021 | Towards Data Science