fasster 0.2.0

Fast Additive Switching of Seasonality, Trend and Exogenous Regressors (FASSTER) is a state space model designed for forecasting time series with multiple seasonal patterns. The model extends traditional state space models by introducing a switching component to the measurement equation, enabling flexible modeling of complex seasonal patterns, and time series dynamics with rapid structural changes.

• FASSTER model implementation:

• Model specification: Flexible formula interface supporting:

  • trend() for polynomial trends
  • season() for seasonal factors
  • fourier() for trigonometric seasonal terms
  • ARMA() for autoregressive moving average components
  • xreg() for exogenous regressors
  • %S% switching operator for group-specific model structures
  • %?% conditional operator for time-varying components

• Model methods: Full integration with the fable framework:

  • fitted() and residuals() for model diagnostics
  • augment() for augmenting data with model estimates
  • tidy() for extracting coefficients (initial state estimates)
  • glance() for model summary statistics (AIC, BIC, log-likelihood)
  • report() for displaying estimated state and observation variances
  • components() for decomposing fitted values into trend and seasonal components
  • forecast() for generating predictions
  • interpolate() for filling missing values
  • refit() for applying a fitted model to new data with optional re-estimation
  • stream() for extending models with new observations

• Heuristic estimation: Model parameters are estimated using a heuristic approach based on filtering and smoothing to obtain initial state parameters and variance estimates.