DataFlow: A Framework for High-Performance Streaming ML

This paper introduces DataFlow, a framework designed to bridge the gap between batch and streaming machine learning, addressing issues like causality violations and reproducibility problems. It emphasizes a unified execution model based on DAGs with point-in-time idempotency, ensuring consistent behavior across different environments. The framework's ability to handle time-series data, support online learning, and integrate with the Python data science stack makes it a valuable contribution to the field.