FinML — Feature Stores with Kevin Stumpf

In this episode of FinML, we talked with Kevin Stumpf, founder of Tecton, about Feature Stores and the problems they solve.

Key takeaways from the talk:

• One of the challenges a Feature Store solves is the unification of batch and streaming data, enabling easy computation of features combining both data sources using a lambda-type architecture.
• Another challenge Feature Stores solve is that they ensure the feature values served in production and used for training are identical and hence there is no skew introduced when in production.
• It’s easy for data scientists to create data sets that leak information from the future and thus artificially improve model performance. Feature Stores solve this problem by providing a framework that requires people to create features to be created by themselves.
• While features are usually implemented by the data engineering team, Feature Stores allow the data scientists to implement the features, thus decreasing delivery times.
• Lack of meta-data management for features can mean that feature design work is duplicated across the organisation or relevant data sets are not found to begin with. Feature stores solve these issues by having a dedicated meta-data layer.
• Data quality issues are a significant problem for many data pipelines in production. Feature Stores allow for easy monitoring of data pipelines since all of them use a common infrastructure.

Presentation as pdf

About FinML

FinML is a meetup group dedicated to applications of Machine Learning in finance. We are a group that is dedicated to discuss the economic and statistical concepts behind running Machine Learning in the real world. We strongly believe in discourse, which is why our sessions are 30 min presentation and 30 min open discussion. Sign up here to be invited to all of our meetups and contact myself in case you are interested in speaking at an event!