Alteryx buys Feature Labs to automate ML feature engineering

Alteryx is beefing up its assisted machine learning modeling capability with an acquisition that automates the selection of model features.
Written by Tony Baer (dbInsight), Contributor

Alteryx has just announced the acquisition of Feature Labs, a tiny three year-old Cambridge, Mass.-based startup that will add capabilities for automating feature engineering to machine learning (ML) models. This was a pure technology acquisition that complements Alteryx's recently unveiled assisted modeling capability that guides users through the process of building ML models.

Best known for its self-service data preparation capabilities, Alteryx has always been a hard company to pin down. Are they just a self-service data prep and visualization tool or broader-based analytics offering? Depending on who you speak to and when, it's one or the other, or both. With the Feature Labs acquisition, Alteryx is doubling down on analytics and machine learning modeling portion of its portfolio. It's clearly where the aspirations of their user base – primarily business analysts with aspirations for becoming citizen data scientists.

In a company blog, Alteryx cited a Kaggle survey showing feature engineering – where you select the variables for ML models – was ranked by respondents as the most important parameter impacting ML model outcomes. The rationale is, as long as you're going to provide a guided experience for building ML models, may as well get developers and analysts over what they may perceive as the most critical hump.

Feature Labs, founded in 2015, was an outgrowth of MIT research that until now has existed just below the radar, having drawn a modest $3 million in funding to date. The tiny company has three products where the common thread is using ML to help developers build ML models.

They include Featuretools, for extracting predictive features from data sets. Tempo is a hosted service in the cloud that automates three steps in the development of ML models: prediction engineering, provided a guided experience for problem definition; automated feature engineering, the core family jewel; and machine learning, providing an automated workflow for identifying the best model. The portfolio is rounded out with MLApps, which offers a library of prebuilt app templates for problems such as predict next purchase, anti-money laundering, predictive maintenance, and credit scoring.

The secret sauce is what the company terms Deep Feature synthesis, a routine conceived at MIT that was adapted by FeatureLabs cofounders Kalyan Veeramachaneni and Max Kanter to automatically build predictive models for complex, multi-table datasets. They claim that with this automated function, they beat two thirds of the 900 human teams participating in an online data science competition. Given that Alteryx does not currently have a hosted SaaS offering, it would be interesting if the Featurelabs Tempo cloud service could become the tail that wags the dog.

Given the modest size of the company, Alteryx didn't disclose the purchase price and was able to close the deal immediately. 

Editorial standards