This repo is a companion to the guest lecture at UC Berkeley for the DATASCI 290 class. It includes the slides as well as Iguazio implementations for each of the design patterns.
Automatically package up a piece of code or model and run on a cluster at scale.
Use CI/CD and infrastructure as code to orchestrate workloads. Use the Git repo as the source of truth.
Use a feature store for more than ingesting and retrieving features - integrate with experiment tracking, model serving, and model monitoring.
Monitor feature drift and automate the re-training process to prevent model decay over time.