MLOps for Time Series in Production

In this session, Eddie Mattia from Outerbounds demonstrates how to build a complete MLOps platform for time series data. The talk covers periodic model retraining and batch inference pipelines, showcasing how to deploy a time series forecasting machine in the cloud. The presentation delves into mapping modeling code to workflow tasks, deploying training and inference workflows, and stitching them together in a production system using Outerbounds platform and the open-source framework Metaflow. Key topics include operationalizing models from experimentation to production, using XGBoost for time series with its interpretability and scalability, and setting up automated execution triggered by new data.