Are Feature Stores The Next Big Thing In Machine Learning?

I’m anticipating 2021 to be the 12 months of the characteristic retailer

Mike Del Balso, CEO and co-founder of Tecton

According to a Gartner study, 85 p.c of AI tasks will flatline by 2022. Even probably the most diligent machine studying fashions could not meet expectations when deployed in an enterprise setting, primarily attributable to two causes — insufficient knowledge infrastructure and expertise shortage.

In the machine studying pipeline, seek for acceptable knowledge and dataset preparation are among the many most time-consuming processes. An information scientist spends round 80 percent of his/her time in managing and making ready knowledge for evaluation. The demand-supply hole for certified data scientists is one other urgent problem.

Enter, feature retailer. 

What Are Feature Stores?

A characteristic retailer permits options (measure items of information) to be registered, found, and used for the machine studying pipelines and on-line functions for mannequin inferencing. They can retailer giant volumes of characteristic knowledge and supply low latency entry to options for on-line functions. A characteristic retailer automates the enter, tracks, and governs knowledge into machine studying fashions. Enterprise AI can profit immensely from such a centralised and reproducible framework to handle machine studying fashions.



In 2017, Uber modified the sport with the introduction of Michelangelo, an ML platform for knowledge administration. Michelangelo supplied a characteristic retailer. In 2019, Feast venture, in collaboration with Google Cloud, introduced a characteristic retailer.

The newest to hitch the bandwagon is Amazon’s AWS SageMaker Feature Store — a completely managed and purpose-built repository. Airbnb, Twitter, Facebook, and Netflix are different main gamers with characteristic shops.

Feature shops (by taking on probably the most mundane but time-intensive knowledge duties) enable knowledge scientists to give attention to important duties similar to mannequin constructing and experimentation fairly than spending time on cleansing and managing knowledge.

See Also

Kaggle journey

Feature shops handle knowledge pipelines that rework uncooked knowledge to characteristic values. These pipelines may be both the scheduled pipelines that mixture a considerable amount of knowledge (petabytes) or real-time pipelines triggered by occasions. Feature shops include the ‘freshest’ characteristic values to machine learning fashions.

Feature retailer exposes APIs and UIs to the info scientist to indicate the at present accessible options, pipelines and different coaching datasets accessible or are below improvement. Data scientists can select the options required for his or her use instances and incorporate them into their fashions.

Feature shops provide the next benefits:

  • One of the principle challenges in implementing a machine studying mannequin in an enterprise atmosphere is that the options used for coaching the mannequin will not be the identical within the manufacturing serving layer. A characteristic retailer supplies a constant characteristic set, enabling a smoother deployment course of.
  • The characteristic retailer retains metadata along with the precise options. This helps knowledge scientists in deciding on explicit options that carried out effectively on current fashions.
  • Unlike conventional strategies the place options are developed in silos, characteristic shops enable sharing options and their metadata with friends. This helps in collaboration and avoids duplication.
  • In essential companies similar to finance, healthcare, and safety, it turns into important to trace the lineage of algorithms being developed. To obtain this, scientists require visibility into the end-to-end movement of the mannequin. A characteristic retailer provides a peek into the info lineage of a characteristic, capturing how a characteristic was developed, offering insights and stories for regulatory compliance.

Wrapping Up

As talked about earlier, bigger tech corporations that extensively take care of AI have constructed their very own characteristic shops. The trade must standardise and automate the core of characteristic engineering. Moreover, characteristic shops are slated to turn into a prerequisite within the machine studying pipeline.


Subscribe to our Newsletter

Get the most recent updates and related gives by sharing your e-mail.


Join Our Telegram Group. Be a part of an interesting on-line neighborhood. Join Here.
Shraddha Goled

Shraddha Goled

I’m a journalist with a postgraduate diploma in pc community engineering. When not studying or writing, one can discover me doodling away to my coronary heart’s content material.

LEAVE A REPLY

Please enter your comment!
Please enter your name here