Ensemble learning's utility belt for sklearn and beyond

What it does

combo is a Python library for combining machine learning models and their scores. It wraps methods from scikit-learn, XGBoost, and LightGBM into a single API that handles classification, clustering, and anomaly detection ensembles. The library implements everything from basic majority voting and averaging to specialized techniques like Dynamic Classifier Selection, Stacking, and Locally Selective Combination for outlier detection.

The interesting bit

Instead of treating ensemble learning as a classification-only game, combo applies the same unifying philosophy across disparate tasks—clustering and anomaly detection included. It also JIT-compiles hot paths with numba and parallelizes with joblib, which is more effort than you typically see in a research demo library.

Key highlights

• Implements algorithms spanning general-purpose averaging, classification (DCS, DES, Stacking), clustering (EAC, Clusterer Ensemble), and anomaly detection (LSCP, XGBOD, AOM, MOA).
• Claims unified APIs with detailed documentation and interactive examples.
• Optimizes performance with numba JIT and joblib parallelization where possible.
• Published in AAAI 2020 and integrates with mainstream ML libraries like scikit-learn, XGBoost, and LightGBM.

Caveats

• The README explicitly lists Python 3.5–3.7 as supported versions, which may suggest the documentation has not kept pace with newer Python releases.

Verdict

Worth a look if you regularly ensemble models across different tasks and want a consistent interface. Skip it if you only need basic voting for classification and prefer staying within native scikit-learn utilities.