LoginSign Up
← Back to all sparks
F

fasster

ANALYTICS
Velocity0.0

Forecasting with Additive Switching of Seasonality, Trend and Exogenous Regressors

github.com/tidyverts/fasster ↗

fasster arrives as a fable-compatible state space model for switching seasonality.

time series forecastingstate space modelsfablemultiple seasonalityr package
◆Current state
fasster implements FASSTER, a state space model with a switching component in the measurement equation, aimed at series carrying several seasonal patterns and abrupt structural change. Version 0.2.0 is the first substantive release: a formula interface with trend(), season(), fourier(), ARMA() and xreg() plus the %S% switching and %?% conditional operators, and the full fable method set. Parameters come from a filtering-and-smoothing heuristic rather than full optimisation.
◆Where it's heading
The package sat at a 2018 development version for over seven years, so the news is that it exists as a usable model at all. Implementing the whole fable contract — forecast(), refit(), stream(), interpolate(), components() — means it slots into an existing forecasting workflow instead of asking for its own. The heuristic estimator is the open question these entries leave unanswered.
◆Prediction
The obvious next step is supplementing the heuristic parameter estimates with proper optimisation, though the two entries here give no direct signal on timing.

◆Recent moves

  1. 6mo ago

    FASSTER lands as a complete fable model

    ⚡ SPARK

    After seven years at a development version, fasster ships as a full fable model with a switching-operator formula interface. The heuristic estimator is stated up front rather than left implicit.

    View source ↗
  2. 7y ago

    Early development build

    A 2018 development snapshot noting only that the response name is passed to forecast construction — the package's sole public marker for the following seven years.

    View source ↗