Skip to contents

foresightr 0.1.1

  • The installation no longer runs a Rust binary to regenerate the R wrappers, which ship with the package: that step failed on CRAN’s Windows pretest.
  • Tables are tibbles: the ranking, forecast, cumulative, accuracy and bands of a backtest, total_forecast(), find_outliers(), fourier_terms() and seasonal_dummies(). as.data.frame() on a backtest still gives a plain data frame. The time column of forecasts and outliers is a date for monthly and quarterly series.

foresightr 0.1.0

First release: R interface to the Rust crate foresight 0.7.3.

  • Models as specifications (model_*()): benchmarks, Theta, Holt-Winters, log-linear regression, seasonal ARIMA with regressors and automatic orders, exponential smoothing with automatic choice, a Prophet-style trend with events, TBATS, Croston with its variants, forecasts of the seasonally adjusted series, ensembles and Box-Cox scales.
  • backtest(): rolling origin on several threads (foresight_threads()), errors by horizon, the choice by out-of-sample error and empirical intervals by horizon and for totals (total_forecast()).
  • decompose_stl(), decompose_mstl(), fill_gaps(), find_outliers(), clean_series(), tests of stationarity and seasonality, accuracy measures.
  • Charts with ggplot2: autoplot() and theme_foresight().