Introduction
Building a good model involves many choices: which algorithm, which preprocessing, which hyperparameter values. Trying them by hand is slow and tedious. AutoML (automated machine learning) hands much of this search to software.
Think of a chef who tries many recipe variations automatically — different oven temperatures, baking times and ingredient amounts — tasting each result and focusing on the most promising ones.
AutoML can automate:
- hyperparameter tuning — the most common use;
- model selection — choosing between algorithms;
- preprocessing and feature engineering choices;
- neural architecture search (NAS) — designing network structures;
- ensembling the best models found.