auto_ml

Why we included this project

auto_ml is an early AutoML library built on scikit-learn, and it is still worth reading even though development stopped in 2021. You hand it a pandas DataFrame plus a short description of which columns are categorical and which is the prediction target, and it searches across regressors and classifiers to fit a usable model without much hand-tuning. What set it apart was the production focus: trained models serialize to disk, a prediction on a single dictionary row takes about a millisecond, and the categorical ensembling feature trains one model per store or customer behind one API rather than forcing you to keep track of thousands of artifacts. Data scientists curious about how AutoML pipelines worked before modern frameworks, or looking for a compact example of wiring XGBoost, LightGBM, CatBoost, and Keras into one training loop, will get real value from the code. Treat it as a learning resource and design reference, not something to build a current service on.