Machine Learning with Databricks and MLflow

О курсе

This course teaches you to build, track, and deploy machine learning models on the Databricks platform using MLflow. You start with the reproducibility crisis in ML — understanding why untracked experiments, scattered notebooks, and missing version control create production failures — and learn how MLflow solves these problems with structured experiment tracking, model versioning, and artifact management. You then explore MLflow's architecture in depth: the Tracking layer for logging parameters, metrics, and artifacts; the Model Registry for governance and stage gates; and the Projects layer for reproducible environments. The course covers Feature Store architecture for eliminating training/serving skew, where features are computed once and served two ways — batch for training and real-time for inference. You progress through the ML algorithm spectrum from manual implementations to AutoML, learning when to choose transparency over automation for regulated industries. The second module focuses on production deployment: the MLOps maturity staircase (L0 through L3), inference patterns for batch and real-time serving, and the infrastructure decisions that separate prototype ML from production ML. Hands-on labs on Databricks reinforce every concept.

Навыки, которые вы освоите

Учитесь у экспертов

Noah Gift

Преподаватель курса