skaters

Fast online univariate distributional forecasting. Its forecaster returns a full predictive distribution at each step rather than a point, and is evaluated by held-out log-likelihood on thousands of economic time series against standard methods (ARIMA, ETS, GARCH and others). Ports in Python, JavaScript, R, Julia and Rust match the reference to within 1e-6.

About 21,000 all-time downloads from PyPI.