Azure ML — 6 ways to structure your code

This blog is a single place where we are talking about various approaches to develop Azure ML code: notebook only, modularizing using ScriptRunConfig, ML pipeline with PythonScriptStep, hyperparameter tuning using HyperDrive, ML pipeline with HyperDriveStep & finally incorporating AutoML step into an ML pipeline.

Azure ML provides functionalities to build, deploy and monitor various machine learning models. There are plenty of samples available in Azure code repositories. However, ML developers often get confused about the best way to follow.

In this blog, from a single notebook approach to finally an AutoML approach — 6 options we’ve shown for doing the same tasks i.e. from downloading the cervical cancer research dataset to using a classification algorithm for creating a model to predict the cervical cancer probability.

Approach 1 — notebook only

This is the most basic way to develop a code in Azure ML. We write all the steps, from data download/data read, data transformations, training and model performance in a single notebook or script. This approach is generally used…