Machine Learning on Databricks
From a first model to one running in production — features, evaluation, tracking, and the failures that only appear after deployment.
The modelling part of machine learning is the smallest part. This course gives it proper attention and then spends longer on everything around it, which is where real projects fail.
Do the Python course first if you are not comfortable loading and cleaning a dataset. Everything is taught on Databricks, which is the platform this lab builds on and partners with.
Every lesson is free and needs no sign-up.
Syllabus
What a model is doing
Fitting, generalising, and overfitting — with the arithmetic, not the metaphor.
Features
Building them, reusing them, and the leakage that makes a model look brilliant until it ships.
Training and tuning
Baselines first, then the models worth the extra complexity.
Evaluating honestly
Choosing a metric that matches the decision, and holding out data properly.
Tracking experiments with MLflow
Runs, parameters, artefacts, and being able to say why this version is the one deployed.
In production
Serving, monitoring, drift, retraining, and deciding when a model should be switched off.