Azure’s AutoML: A Quick Look
Walkthrough of Azure’s Automated Machine Learning: New Visual Builders with traditional Py Notebook and API approach.
Published in
6 min readNov 5, 2019
AutoML is a relatively new thing (2–3yrs) — all the major vendors have an offering — for some like Data Robot it is their main thing. Azure has been taunting me to try this out (with its banner ads) .. so I answered the call. AutoML features:
- With one click, auto-design, train, score your experiment
- Most have auto-feature selection and feature engineering
- Most will tune hyper-params like step size, iterations, tree depth etc
- Most will run thru a dozen+ models at once and score & rank them (something Doug Foo has been doing manually)
None of this is magic, it is mostly brute force — they run it on their clusters in parallel — doing a lot of iterative work we humans were doing before.
Walkthrough of Visual AutoML
First create a Azure ML Workspace, then jump into this thing that looks like an advertising banner (Behold — the Azure ML Studio — or whatever its called since the names keep changing)
