Zipline - A Declarative Feature Engineering Library

Explore the architecture and algorithms behind Zipline, Airbnb's declarative feature engineering library for machine learning, in this 34-minute conference talk from Strange Loop. Discover how Zipline significantly reduces the time ML practitioners spend on data collection and transformation tasks. Learn about the system's ability to provide point-in-time correct features for both offline model training and online inference. Delve into the innovative algorithm that makes efficient point-in-time correct feature generation tractable. Gain insights into supervised machine learning in industry, the challenges faced by ML engineers, and the solutions Zipline offers. Examine concepts such as aggregation, reversibility, time complexity, and temporal joins. Understand the feature serving stack, data architecture, and query topology implemented in Zipline. Presented by Nikhil Simha, a Software Engineer on Airbnb's Machine Learning infrastructure team, this talk offers valuable knowledge for data scientists and ML practitioners looking to streamline their feature engineering processes.

Introduction
Supervised Machine Learning
Machine Learning in Industry
Machine Learning Engineers
Goal
Key
Why is it so hard
Example
Aggregation
Reversibility
Time Complexity
Tile Problem
Change Data
Temporal Join
TriMerge
Query Topology
Feature Serving Stack
Architecture
Storage
Data Architecture
Data Classification
Query Log
Questions