Feature Stores

This is the additional presentation found in the middle.

This was meant to be a workshop but it ended up working out much better as a talk, hence the description below.

As machine learning-powered applications become more ubiquitous, so do the tools we create and use to optimize development time and collaboration across teams. One such tool that has gained popularity in recent years is the feature store as its adoption helps solve pressing issues at the infrastructure level before and after machine learning takes place within a data science project. Because of the rise of machine learning, and because of the need to store features in better, more efficient ways, this workshop focuses on teaching participants how to get started developing and using feature stores. In particular, participants will learn how to combine feast–an open source feature store–alongside tools from the common data stack such as pandas and NumPy, and workflow orchestration tools such as Dagster, Prefect, and Metaflow. By the end of the workshop, participants will have operationalized analytics data using a feature store while ensuring consistency across training and serving data for machine learning projects.

The workshop is divided into 4 sections of equal lengths each.

1. In the first part, we cover what feature stores are as we deconstruct each component of an already built solution in a top-bottom approach.
2. The second stage covers the interaction between a model and the online and offline sides of a feature store. More specifically, we walk through making predictions while using an online store built with feast.
3. The third part is a walk-through of the lower level details of feast and the steps required to set up and manage a feature store locally before going to the cloud.
4. Lastly, we cover different deployment options and design patterns, and discuss the limitations of open source feature stores when compared to proprietary solutions. In addition, we will also discuss the ways in which data science and machine learning teams across different industries could adopt feature stores.