// the find
ClimbsRocks/auto_ml
[UNMAINTAINED] Automated machine learning for analytics & production
auto_ml is a Python AutoML library from circa 2017-2021 that wraps scikit-learn, XGBoost, LightGBM, and Keras behind a single Predictor API. It handles feature engineering, model selection, and hyperparameter tuning with minimal setup. It's been unmaintained since early 2021 and should not be used for new projects.
The single-dictionary prediction path (claiming ~1ms latency) was a genuinely practical design decision for production inference — most AutoML tools of that era were batch-only. The feature learning trick (using a neural net's penultimate layer as feature input to gradient boosting) is an honest ensemble technique that predates most 'stacking' tutorials. The categorical ensembling API is a clean abstraction for the common 'one model per store/customer' pattern that teams usually solve with messy loops. Test coverage is reasonable with backwards-compatibility tests that include a serialized model artifact.
Abandoned in 2021 — dependencies like TensorFlow 1.x/2.x, older XGBoost, and the dill serialization format have all moved on, so installation will likely fail or produce silent incompatibilities on modern Python. The repo explicitly says UNMAINTAINED in the title, which rules it out for anything production-facing. AutoML has advanced dramatically since then; FLAML, AutoGluon, and optuna-based pipelines all outperform this on both accuracy and ecosystem support. Uses dill for model serialization instead of joblib or ONNX, which means loading a trained model in a different Python/library version environment is a gamble.