PyCaret AutoML

If you have ever wanted to know which of fifteen algorithms suits your table before writing fifteen scripts, PyCaret is built for exactly that moment. It is an open-source, low-code Python library that wraps scikit-learn, XGBoost, LightGBM and CatBoost behind a handful of one-line commands, aimed squarely at notebook work.

The workflow
Everything starts with setup(), which inspects the dataframe and assembles a preprocessing pipeline — type inference, imputation, encoding, scaling, the train/test split, the cross-validation strategy — from keyword arguments rather than code. From there:

• compare_models() trains every applicable estimator under cross-validation and returns them ranked on your chosen metric.
• tune_model() searches hyperparameters for a chosen candidate, with Optuna or Hyperopt available as the backend.
• blend_models() and stack_models() combine the leaders into an ensemble; finalize_model() refits on the full dataset and save_model() writes out the fitted pipeline, preprocessing included.

Separate modules cover classification, regression, clustering, anomaly detection and time series, and runs can be logged to MLflow automatically.

Why it matters in AutoML
PyCaret occupies the low-code middle ground rather than the fully automatic end. It does not hide the search behind a single fit call: you see the leaderboard and decide what to tune, ensemble and ship. That makes it both a teaching tool and a fast baseline generator — the quickest honest answer to "is this problem learnable at all?" — while keeping the analyst in the loop.