Featurization & Feature Stores: A Crash Course In The ML Lifecycle & MLOps

DataOps, MLOps, Data Engineering.... what's the big difference? Squint at the job descriptions and they'd seem to be the same person, especially around featurization. Can't DataOps tools be used for MLOps? Why is 80% of a data scientist's time stuck with data? Isn't a feature store just an expensive, overly specialized database where machine learning features get parked (only to be forgotten until a pipeline breaks)? Much like how humans share 70% of their DNA with slugs (and 50% with bananas)* the differences, while minute, are significant. My goal in this session is to help illuminate the challenges and vagaries of developing ML models from scratch (for production) and in the process answer the following questions: - What are the main problems MlOps tries to solve? - What does the process look like for developing a model from scratch? And why is feature engineering tricky to automate? - What is a Feature Store? What are the pain-points a feature store is meant to solve? - What are the different types of feature store or platforms that exist and which archetypes are seeing the most adoption? And why?