Only three days after previewing a slew of upcoming enhancements for Apache Kafka, LinkedIn Inc. is adding another item to its open-source analytics repertoire with the release of a likewise internally-developed tool that promises to simplify machine learning initiatives. And in particular, coordination across the different stages of the project lifecycle.

Work on large-scale recommendation systems like the kind LinkedIn uses to refer members to groups and job postings that match their professional interests is typically divided between two teams. One is responsible for creating the mathematical models into which raw data is fed for analysis, while the other takes care of testing and the various other chores involved in operationalizing those models.

The problem is that even a minor change in an algorithm can have a significant impact on the work of the latter team, a reality that has traditionally required the data scientists to a make major time investment in incorporating the modifications. That slows down the rate at which new capabilities can be created and tested by a great deal, which is what Feature Fu aims to address.

The tool employs an open-source Java library called Expr to significantly cut the amount of work involved in translating changes made during the modelling process to the subsequent stages of the project. Its parser can handle updates to the formulas that dictate which parts of a given dataset are fed into an algorithm and how without requiring engineers to completely rework the underlying implementation.

Feature Fu has already been adopted by LinkedIn’s Job Relevant Team to help optimize recommendations and will now be able to find use beyond the social networking giant in other applications. Given that its other open-source projects is currently employed at over 100 organizations, including the likes of Yahoo! Inc. and The Goldman Sachs Group, Inc, the tool probably won’t have any trouble attracting users.

Photo via Geralt