Machine Learning Operations (MLOps)

 

In this one day workshop you will learn how to operationalize Machine Learning models using popular open-source tools, like Kedro and Kubeflow, and deploy it using cloud computing.

During the course we simulate real-world end-to-end scenarios – building a Machine Learning pipeline to train a model, and deploy it on Kubeflow environment. We’ll walk through the practical use cases of MLOps for creating reproducible, scalable and modular data science code. Next, we’ll propose a solution for running pipelines on Google Cloud Platform, leveraging managed and serverless services. All exercises will be done using either a local docker environment, or GCP account.

 

   Target Audience

Data scientists and DevOps who are interested in implementing MLOps best practices, and build Machine Learning pipelines.

 

    Requirements

Some experience coding in Python, basic understanding of cloud computing and machine learning concepts.Install your favorite code editor (i.e. Visual Studio Code or PyCharm). The trainer will demonstrate the exercises using Visual Studio Code.
Download the Python 3.8 if you don’t have it already (https://www.python.org/downloads/release/python-380/). We’ll use it as a base Python for the virtual environment.

 

    Participant’s ROI

• Practical knowledge of building Machine Learning pipelines using Kedro
• Hands-on experience with building Machine Learning platform with Kubeflow Pipelines
• Tips about real world applications and best practices.

 

    Training Materials

All participants will get training materials in the form of PDF files containing slides with theory and exercise manual with the detailed description of all exercises. During the workshops the exercises can be done using either a local docker environment or within your IDE.

 

    Time Box

This is a one-day event (9:00-16:00), there will be some breaks between sessions.

 

    Agenda

Session #1 - Introduction to Machine Learning Operations (MLOps)
Introduction and key concepts
MLOps components
The challenges of deploying and maintaining Machine Learning models on production
The Machine Learning model lifecycle

Session #2 - Kedro - a framework to structure your ML pipeline
Create reproducible, maintainable and modular data science code
Build your Machine Learning pipeline
Hands-on exercises

Session #3 - Kubeflow and Kubeflow Pipelines
Introduction and key concepts
Example of Kubeflow Pipelines (managed) and Vertex AI (serverless) deployments
Hands-on exercises

Session #4 - Building infrastructure for your Machine Learning platform
Overview of MLOps Frameworks landscape, and reference architectures

Session #5 - Summary and wrap-up

 

   Keywords: MLOps, Kubeflow, Kubeflow Pipelines, Kedro, Google Cloud Platform

 

    Session leader: