LinkedIn’s open-source Feathr feature store for machine learning joins the LF AI & Data Foundation

feathr linkedin feature store


  • On 12th September 2022, LinkedIn declared its open source store, named Feather.
  • Feathr robotizes and normalizes communication with the information type, which is utilized in both the preparation and deduction phases of AI.
  • Feathr is presently a sandbox project at LF man-made intelligence + Information.

On 12th September 2022, LinkedIn declared that its open source store, named Feathr, is joining LF man-made intelligence and Information, the Linux Establishment’s umbrella starting point for enormous information and simulated intelligence projects.

Feathr was initially evolved at LinkedIn to help oversee and serve highlights utilized in its AI applications. Rather than physically working with highlights as a component of a singular information pipeline, Feathr robotizes and normalizes the communication with the information type, which is utilized in both the preparation and deduction phases of AI.

What precisely is “Feathr”?

Feathr is a device that was created by LinkedIn and initially set into creation back in 2017. It’s intended to make highlight serving in AI simple, quicker, and more versatile, particularly for ongoing man-made reasoning applications. LinkedIn engineers Hangfei Lin and Jinghui Mo said in a blog entry that the organization’s computer-based intelligence groups use Feathr to store, change, serve and offer elements with low dormancy and high throughput.

All the more explicitly, Feathr fills in as a deliberation layer between the crude information and AI models that assists with normalizing and improving on highlight definition, change, serving, stockpiling and access inside AI work processes and applications. Designers can then zero in more on highlight designing, passing on Feathr to deal with information serialization designs by associating with different data sets. It likewise gives execution streamlining and qualifications to the board capacities.


The catalyst in making Feathr was giving more noteworthy consistency, precision, and execution in its AI programs. By characterizing the information highlights utilized in ML programs once in a typical element namespace, clients can now pull them up “by the name” from inside ML work processes. This permits similar highlights to be utilized in various ML programs, further developing efficiency and exactness. Highlight stores likewise give a more repeatable strategy to changing source information into highlights (which is something not tracked down in all component stores), and lift the presentation of ML serving at the derivation stage by concentrating the capacity and serving of elements.


Since sending off Feathr inside in 2017, utilization of the product has developed. As indicated by LinkedIn, the element store is presently being utilized to follow a great many highlights utilized by the online entertainment monster.

“It has decreased the designing time expected for adding and exploring different avenues regarding new highlights from weeks to days,” LinkedIn information framework engineer Hangfei Lin writes in a blog entry today.

LinkedIn delivered the code behind Feathr under an Apache 2.0 permit this April, permitting the overall population to utilize the open source including stores interestingly. From that point forward, the undertaking “has accomplished significant ubiquity among the AI tasks (MLOps) people group” and is being taken on by organizations across various enterprises, Lin composes.


By giving Feathr to The Linux Establishment’s LF computer-based intelligence and Information bunch, LinkedIn is setting up extra administration around the open source project, which ought to assist with drawing in additional clients and more supporters of the undertaking.

“We’re eager to invite Feathr to LF artificial intelligence and Information and for it to be important for our specialized venture portfolio (41 undertakings and developing) with over 17K engineers,” Dr. Ibrahim Haddad, the senior supervisor of LF simulated intelligence and Information, said in a press proclamation in words-  “We anticipate the task’s proceeded with development and accomplishment as a component of LF computer-based intelligence and Information.”

Microsoft is likewise important for the Feathr story (LinkedIn is claimed by Microsoft). As indicated by Lin, LinkedIn engineers have worked with their Microsoft Purplish blue partners to guarantee Feathr runs well on Purplish blue and is incorporated with other Purplish blue items and tasks.

As indicated by this Purplish blue blog entry, Feathr presently works with Apache Flash, Juypter, Sky blue Mass Stockpiling, HDFS, Snowflake, Databricks Delta Tables, and SQL Waiter. Microsoft was likewise associated with the publicly releasing of Feathr back in April.

Feathr is presently a sandbox project at LF man-made intelligence + Information.