maquette - noun ma·quette /maˈkɛt/
a usually small preliminary model (as of a sculpture or a building)
There’s so many feature store and machine learning system frameworks out there. But no real “easy way” to serve a system without relying on managing Spark clusters, Kubernetes farms and the like.
In this series we’ll try to create one from scratch - without fancy infrastructure. We’ll go through all the pitfalls and problems on the way, so that hopefully you don’t have to.
How can refactoring a feature store using the strategy pattern help us?
[Read More]
Parse Don't Normalize
What's the difference between personal and enterprise level feature stores?
[Read More]
Building a Feature Store the Unix Way
Can we build a feature store using simple files?
[Read More]
Taking a glimpse at Tecton.ai
Tecton.ai is a new enterprise feature store - what can we learn from publically available videos of their API?
[Read More]
The Challenges with Slowly Changing Dimensions
What issues can arise with Slowly Changing Dimensions?
[Read More]
Dealing with Slowly Changing Dimensions
More often than not, we're dealing with slowly changing dimension tables - how do we deal with that?
[Read More]
Event Driven Analytical Records
So we've built an event table - now what?
[Read More]
Data Structures for Calculating Deltas
What data structures are easier for calculating deltas?
[Read More]
Revisiting the Design of Vowpal Wabbit
What can we learn from serving features and building machine learning pipelines using Vowpal Wabbit?
[Read More]
The Design of Denormalized Tables
Denoramlized tables are at the heart of the Feature Store. What are the design considerations here?
[Read More]
Reflecting on feast.dev
What currently exists in the open source world of feature stores?
[Read More]
Integrating services - Take One
What kind of considerations should we make when working with multiple layers in the architecture?
[Read More]
Designing a Feature Store - Reflecting on UBER's Michelangelo
The goals of Michelangelo are numerous - but let's first breakdown the choice of technology and ask; "what kind of design decisions should we make?"
[Read More]