sktime

sktime is a Python library for developers and data scientists who want to compare time series models without rebuilding their workflow for each algorithm. It puts classical statistical methods, machine learning, deep learning and foundation models behind a consistent interface. You can run it on your own macOS, Linux or Windows machine.

Its forecasting options include AutoARIMA and Amazon Chronos, with probabilistic predictions that express uncertainty alongside expected values. That matters for work such as retail demand planning or energy load forecasting, where a single estimate doesn't capture the risk of shortages or excess capacity.

The library also handles time series classification, regression and clustering, plus anomaly and change detection. Feature extraction tools turn sequences into inputs for models, while time-to-event modeling estimates when something such as equipment failure may occur.

For model development, sktime supports pipelines, ensembles, tuning and validation. Its scikit-learn-compatible tools let you use regressors for forecasting, and interfaces connect it with statsmodels, tsfresh and PyOD. You can add your own estimators, transformers or adapters, including components from a private codebase, through the same API.

The library is open source under BSD-3-Clause.