aeon-toolkit/aeon

A toolkit for time series machine learning and deep learning

aeon – A scikit‑learn‑compatible toolkit for time‑series machine learning

What it is – aeon is a Python library that brings the familiar scikit‑learn API to the whole spectrum of time‑series problems: classification, regression, clustering, forecasting, anomaly detection, segmentation, similarity search, transformations, distance/kernels and benchmarking. The code is written and maintained by the original authors of many of the algorithms, and the project is backed by NumFOCUS, INRIA and UKRI.

Key features

  • Broad algorithm coverage – from classic statistical models (ARIMA) to state‑of‑the‑art deep‑learning nets (InceptionTime, H‑InceptionTime, LITE, ResNet) and specialised methods such as Hydra‑MultiRocket, HIVE‑COTEv2.0, SETAR‑Tree, KASBA, CLASP, DrCIF, TDE, etc.
  • Scikit‑learn‑style interface – all estimators implement fit, predict, score, transform, etc., so they can be used in pipelines and with existing tooling.
  • Benchmarks & reproducibility – the library ships with utilities for reproducible experimental evaluation and mirrors the benchmark suites that the time‑series community relies on.
  • Deep‑learning support – optional TensorFlow/Keras back‑ends (aeon[all_extras]) provide GPU‑accelerated models with the same API as the classical estimators.
  • Extensive documentation & examples – a full website, API reference, and a gallery of notebooks (also runnable on Binder) cover every task.

Typical usage (classification on a UCR dataset)

from aeon.classification.convolution_based import RocketClassifier
from aeon.datasets import load_gunpoint

X_train, y_train = load_gunpoint(split="train")
X_test,  y_test  = load_gunpoint(split="test")

clf = RocketClassifier()
clf.fit(X_train, y_train)
print("Accuracy:", clf.score(X_test, y_test))

The same pattern works for regression, clustering, forecasting, etc.

Installation

pip install aeon               # core package
pip install aeon[all_extras]   # includes deep‑learning deps

Python 3.10+ is required.

Community & governance – active Discord, GitHub Discussions, a code‑of‑conduct, and a clear contribution guide. The project is published in JMLR (2024) and cited via a ready‑to‑use BibTeX entry.

Who should use it?

  • Researchers needing faithful implementations of the latest time‑series methods.
  • Practitioners who want a single, consistent API for many time‑series tasks.
  • Anyone building pipelines that combine classical feature‑based models with deep‑learning nets.

All information above is taken directly from the repository’s README; no additional claims have been added.