Happy Thanksgiving week! Thank you for signing up for our free Sunday news digest. As a token of appreciation and to celebrate the holiday, we offer our Premium subscription to TheSequence Edge with 20% off forever. The offer ends on November 30.

Subscribe with 20% OFF

This holiday week, let’s have some time to catch up with what we’ve covered before. As part of our current MLOps series, we offer you the recap of articles dedicated to feature stores. Why did they become the crucial part of MLOps stack?

🏗🏪 What is a Feature Store?

Feature stores have rapidly become a key component of MLOps infrastructures. This is not surprising if we consider that many challenges in the lifecycle of ML models revolve around data and features. In any large ML team, data scientists spend most of their time extracting, selecting, and transforming data into features and then figuring out how to incorporate those features into production-ready ML models. From an ML architecture standpoint, a feature store can be seen as the missing link between feature engineering and feature serving.

Arguably, the concept of a feature store can be traced back to the original Uber’s Michelangelo architecture. In that setting, the Michelangelo team introduced a feature store as a key but relatively minor component of its architecture. From there, feature stores have evolved all the way to become a booming standalone market in the ML space.

Despite the rise in popularity, the adoption of feature stores in real-world ML applications remains relatively low. In the early phases of ML projects, feature stores can be seen as overkill and are often ignored. However, as machine learning infrastructures become larger and more complex, it becomes advantageous to incorporate feature store capabilities.   

We have covered the topic of feature stores in detail in several issues of The Sequence. Here are some of our favorites:

• Dive into Edge#77 (no subscription is needed) to learn more about what a Feature Store is; hear the story of how Uber Michelangelo began the Feature Store movement; and explore the feature store market.

• Feature Stores as standalone companies came from the efforts of a few powerhouses trying to optimize deploying machine learning at scale. This Edge covers some of the reference architectures used by Google, Facebook, LinkedIn, and Uber to enable their machine learning pipelines.

• As feature store enables a centralized catalog of features in machine learning pipelines, here we explain the difference between feature extraction and feature selection.

• In this Edge, we discuss what makes some feature representations better than others; explore Uber’s architecture to discover optimal features; and overview three architectures powering feature stores at Airbnb, Pinterest, and DoorDash.

• We’d also like to recommend a few deep dives in which we explore the Tecton platform, lightweight open-sourced Feast, and the feature extraction and storage platform of Molecula.

Feature stores movement continues to grow, with a few big launches coming from the leading players on the market. As big fans of the concept, we will keep covering the most advanced efforts in the future.

Next week we will continue with the MLOps series. Stay tuned!