honestml

honestml is a general tabular AutoML library for binary/multiclass classification and regression. It is built around a clean, extensible core: pluggable models, metrics, cross-validation, feature engineering/selection, tuning, ensembling, serving and tracking.

The differentiator is the honestly best model: out-of-fold selection with a bootstrap equivalence band, leakage-controlled feature engineering and selection, an optional untouched outer holdout, and a tracker-independent run report — the score you see is the score you can expect in production.

pip install honestml

• Quickstart — fit, presets, reports, artifacts (examples run in CI).
• Guide — every capability with copy-paste examples: data input, CV and honest selection, presets and budget, features, HPO and ensembling, reports and tracking, artifacts and ONNX.
• API reference — the pinned public surface.
• Correctness guide — why the scores are honest.
• Plugin contract — ship your own model via entry points.
• Versioning policy — what stays compatible across releases.

Maintainers: Releasing.