Buzzsaw: A System for High Speed Feature Engineering

Feature Engineering is a critical part of machine learning systems. Most feature engineering solutions focus on a single language or computational abstraction for their pipelines, forcing practitioners to either embrace a restricted ecosystem or incur operational and orchestrational complexity. In this paper, we describe Buzzsaw: an independent system for constructing and evaluating complex feature pipelines, designed for embedded usage across multiple languages and platforms. We describe its core interfaces, data types, and algorithms for implementing the system and compare it to three other common feature engineering frameworks.