Explore the intricacies of deploying production models in this 46-minute conference talk from GOTO Chicago 2018. Gain insights from Juliet Hougland, a Data Platform & ML Engineer at Stitchfix, as she delves into the various aspects of model deployment. Learn about different deployment types, including data, API, and code artifact deployments, and how they fit into model development workflows. Discover the challenges of transmitting model outputs to other systems reliably and the importance of feature stores. Understand the concept of AB testing and its role in model evaluation. Examine the engineering requirements for machine learning pipelines, including lambda architecture and model throughput considerations. Explore the dynamics between data scientists and software engineers, and learn strategies for effective communication and handoffs in model deployment. Gain valuable knowledge on topics such as serialization, PMML, and its limitations. This comprehensive talk covers team structures, QA processes, and addresses general questions about production model deployment, providing a holistic view of the subject for data scientists and engineers alike. Read more