augurs is a time series analysis library for Rust with bindings for JavaScript. It provides a set of tools for analyzing time series data, including clustering, outlier detection, forecasting, and changepoint detection.
Visit the documentation for more information. Alternatively, give augurs a star on GitHub!
augurs can be used to identify groups of series that behave similarly through time. This can be slow for large number of series but can be parallelized!
A similar but easier problem is to identify series that are outliers: those which behave differently to the majority of series. There are several algorithms; this demo shows the DBSCAN algorithm.
augurs contains several forecasting algorithms, including MSTL, which is a seasonal-trend decomposition procedure for modelling multiple seasonalities, based on LOESS.
augurs also contains an implementation of the Prophet forecasting algorithm, which is a decomposable time series model with three main components: trend, seasonality, and holidays.
augurs also exposes functionality from the excellent
changepoint
crate, which provides a Bayesian approach to detecting
changepoints in time series data. Here, each changepoint is
highlighted in the plot.