tidyverts

tidyverts is a collection of R packages designed for tidy time series analysis and forecasting, built to integrate with the tidyverse. It provides a cohesive framework for representing, transforming, modeling, and forecasting time-indexed data in a consistent, pipe-friendly workflow.

The central data structure in the tidyverts ecosystem is the tsibble, a tidy time series tibble that includes a time index and optional keys to represent multiple related series. This structure enables clear handling of temporal relationships, irregular time steps, and hierarchical or grouped time series within a single data frame. The tsibble concept is complemented by tsibbledata, which supplies ready-to-use time series datasets for examples and teaching.

Key packages in the tidyverts family include fable for modeling and forecasting, feasts for features and statistics related to time series, and fabletools to support the modeling toolkit. Together with tsibble and tsibbledata, these packages provide a unified interface for specifying models, generating forecasts, and assessing accuracy. The fable ecosystem supports a range of models (for example ARIMA, ETS, and regression-based approaches) and can be extended with additional packages for Prophet or other methods.

A defining strength of tidyverts is its emphasis on a tidy, consistent API: you define models with model(), produce forecasts with forecast(), and evaluate performance with accuracy(). The ecosystem also supports cross-validation and forecasting across grouped or hierarchical structures, making it suitable for applied forecasting tasks across multiple time series.

The tidyverts suite is maintained by Rob J Hyndman and collaborators and is available on CRAN and via its GitHub repositories. It also includes tsibbledata for example data and aims to make time series work accessible within the tidyverse paradigm.