LoginSign Up
← Back to all sparks
F

forecast

ANALYTICS
Velocity0.0

Automatic forecasting of univariate time series in R

github.com/robjhyndman/forecast ↗

After years of pure maintenance, forecast 9.0.0 reopens the package

forecastingtime-seriesr-statsmajor-releaseapi-redesignmodel-classes
◆Current state
forecast is the long-established R forecasting package that fable was meant to succeed. For several years its releases were RNG fixes, base-R compatibility and documentation. Then 9.0.0 arrived with a batch of new model constructors, wider prediction-interval support and a rewritten accuracy() built on S3 methods.
◆Where it's heading
The major version reframes forecast around explicit *_model() constructors — mean, random walk, spline, theta, Croston — rather than the older function-per-method style, and the 9.0.x patches since have been performance and argument-handling cleanups on top. That is an active maintenance line, not a package winding down in favour of fable.
◆Prediction
Expect continued 9.0.x patches consolidating the new constructors and their forecast methods, with the older interfaces kept working alongside them.

◆Recent moves

  1. 4mo ago

    Fixes for checkresiduals() and mstl() lambda handling

    Two regression fixes on the 9.0 line. Housekeeping after a large major release.

    View source ↗
  2. 5mo ago

    Faster ARFIMA search and forecast.mlm() argument handling

    Speeds up the ARFIMA model search and fixes newdata resolution when forecast.mlm() is called from another function. The kind of consolidation expected in the first patches after 9.0.0.

    View source ↗
  3. 7mo ago

    forecast 9.0.0 adds five model constructors and rewrites accuracy()

    ⚡ SPARK

    The release that ends the maintenance-only stretch. It adds a family of explicit model constructors, extends simulated and bootstrapped intervals across more models, and rebuilds accuracy() on S3 methods.

    View source ↗
  4. 1y ago

    Documentation and bug-fix release

    A contentless maintenance release from the pre-9.0 quiet stretch, listing only documentation improvements and bug fixes.

    View source ↗
  5. 2y ago

    RNG state and base-R head/tail compatibility

    Stops the package from altering RNG state on attach and defers head.ts/tail.ts to base R from 4.5.0. Correctness housekeeping.

    View source ↗
  6. 2y ago

    Much faster hfitted() for ARIMA and ETS models

    Large speedups to historical fitted-value computation for the two most-used model families, with ETS fitted values now derived from initial states. A real performance gain in the middle of an otherwise quiet period.

    View source ↗