LoginSign Up
← Back to all sparks
F

fable

ANALYTICS
Velocity0.0

Tidy time series forecasting models that work with tsibble data

github.com/tidyverts/fable ↗

fable keeps widening its model shelf, one econometric class at a time

forecastingtime-seriesr-statsmodel-classeseconometricstidyverts
◆Current state
fable is the tidyverts forecasting engine, and its releases are almost entirely about which model families it can express. The 0.4.x line added the vector-error-correction and VARIMA classes plus impulse-response methods; 0.5.0 adds fractional differencing via ARFIMA. Between those, the releases are CRAN-check patches and documentation passes.
◆Where it's heading
The package is closing the gap with the older forecast package's model coverage while keeping the tidy model-specification grammar. Each substantive release is a new model class plus the generate()/IRF() plumbing to make it behave like the existing ones. Maintenance releases cluster around CRAN policy and ggplot2/tsibble compatibility rather than internal rewrites.
◆Prediction
Expect the next substantive release to add another model class or extend generate()/IRF() coverage to the classes that still lack them, rather than change the modelling interface.

◆Recent moves

  1. 6mo ago

    fable adds ARFIMA and fractional differencing

    ⚡ SPARK

    Continues the pattern of the 0.4.x releases: a new model class arrives with its supporting utilities, this time long-memory processes. ETS() also picks up missing-value tolerance, closing a gap that forced users to impute before modelling.

    View source ↗
  2. 1y ago

    Indexing and generate() fixes for VECM models

    A two-line patch cleaning up the VECM support shipped one release earlier. Nothing new for users who were not already hitting the exogenous-regressor indexing error.

    View source ↗
  3. 1y ago

    VECM and VARIMA models land, plus IRF for VAR and ARIMA

    ⚡ SPARK

    The release that set the current arc: two multivariate model classes plus the impulse-response and simulation methods that make them usable end to end.

    View source ↗
  4. 2y ago

    Patch for C++ R header changes

    A compatibility patch for upstream R header changes, plus documentation. No user-facing behaviour change.

    View source ↗
  5. 2y ago

    CRAN check patch with generate() fixes

    A CRAN-check housekeeping release with two narrow generate() fixes. Typical of the maintenance cadence between model-class additions.

    View source ↗
  6. 3y ago

    TSLM forecasts gain Student's t intervals

    Gives users an opt-in path to the statistically more appropriate Student's t forecast distribution for TSLM, while keeping the Normal approximation as the default for model combination. Also loosens ETS() smoothing-parameter constraints when values are set explicitly.

    View source ↗