pyts

Why we included this project

Most time series work comes down to turning raw ordered sequences into features a classifier can use, and pyts collects the common transforms for that job in one place. It includes BOSS variants, shapelet learning, time series forests, and preprocessing utilities such as SAX, PAA, and discrete Fourier transforms. Since the estimators follow the familiar scikit-learn fit/transform/predict pattern, the package drops into an existing sklearn workflow without much friction, and it ships helpers for pulling benchmark datasets from the UCR/UEA repositories. That combination makes it handy for prototyping classifiers on sensor streams or machine logs, and for researchers who want a working reference implementation of a published algorithm to reproduce results quickly.