MLOps – what you need to know
Kickstart your Machine Learning Operation (MLOps) journey. Hear about best practices for deployment of ML systems and customer journey for MLOps adoption.
The Continuous delivery Foundation defines MLOps as “the extension of the DevOps methodology to include Machine Learning and Data Science assets as first class citizens within the DevOps ecology”.
In this event, we want to give you a kickstart in your MLOps journey. In the first talk, we have Ed Shee from Seldon, covering the best practices for deployment of ML systems. Then we have Hamza Tahir from ZenML, showcasing his framework to adopt best practices for MLOps and get you out of your jupyter notebook. Finally, we have Mia Ryan, from Redpill Linpro, taking us through a customer journey for MLOps adoption.
- From Model to Micro-service - ML Deployments At Scale
Until recently, the data science / machine learning field has been pretty immature in it's adoption of DevOps tools and processes. That's now changing rapidly as engineering teams realise that, in order to gain any value from their ML models, they need to get them into production. In this talk, Ed will introduce the open source Seldon Core library, build a model using popular machine learning tools and deploy it to Kubernetes to handle production traffic. You will learn how to turn an ML model into a production microservice that handles REST/gRPC traffic, how to use complex model deployment techniques and how to monitor both the infrastructure and the models themselves, spotting drift and outliers as they take place.
Ed Shee, Seldon
- Completing the MLOps Picture with production-grade pipelines
The transition between the experimentation and production phases of the ML lifecycle is key to get right in a production MLOps setting. Teams must be fully aware that having a smooth hand-over from the nascent stages of a ML projects on local notebooks to a remote cloud based MLOps stack is key to success. In this talk, Hamza will showcase how to create a complete MLOps pipeline from scratch using the open-source ZenML framework. You will learn basics of continuous training, data validation, and metadata tracking for reproducibility. By the end, you will have a complete picture of how you can tie different tools and infrastructure together to create a robust ML pipeline.
Hamza Tahir, Co-Creator of ZenML
- MLOps - from whiteboard to customer success In Redpill Linpro DevOps is at the core of everything we do, both internally and for our customers. Our goal was therefore to give ML processors the same automated setup as in software development. This is our journey from scratch to successfully deploying MLOps for one of our customers, including challenges and experiences with both exploration of tools and processes.
Mia Ryan, Redpill Linpro