Time series ML, maintained by the researchers who wrote the papers

What it does

aeon is a scikit-learn compatible Python library for learning from time series. It covers classification, regression, clustering, forecasting, anomaly detection, distances, segmentation, similarity search, transformations, and benchmarking. Many implementations are contributed or maintained by the researchers who originally published the methods, including deep learning approaches.

The interesting bit

The same team behind aeon runs the benchmarks the field uses to evaluate new methods, so the library functions as both a toolkit and a living meta-analysis. New methods often arrive alongside their papers, and the recommended defaults are drawn from published comparative studies rather than intuition.

Key highlights

• Broad task coverage: classification, regression, clustering, forecasting, anomaly detection, segmentation, similarity search, and transformations.
• Author-maintained algorithms including InceptionTime, Hydra-MultiRocket, SETAR-Tree, DrCIF, and CLASP.
• Scikit-learn compatible API.
• Peer-reviewed: published in the Journal of Machine Learning Research (2024).
• NumFOCUS affiliated project, with support from INRIA and UKRI/EPSRC.

Caveats

• The README says many algorithms are author-maintained, not all of them.
• The library spans a wide range of tasks; the sources do not clarify whether every subdomain is equally mature or documented.

Verdict

Researchers and practitioners who want a broad, scikit-learn compatible time series library tied to current literature should start here. Those looking for a minimal, single-purpose tool may find the scope larger than necessary.