pymc_forecast.features#

Seasonal feature builders: Fourier design matrices and periodic tiling.

pymc_forecast.features.fourier_features(time, *, period, num_terms)[source]#

Build a labeled Fourier seasonality design matrix.

Parameters:

  • time – Either an integer duration (phases 0..duration-1) or an array of numeric time positions (e.g. np.arange(len(index)) or fractional day-of-week positions). The positions set the phase; the returned "time" coord carries them.

  • period – Seasonal period, in the same units as time.

  • num_terms – Number of harmonics; the output has 2 * num_terms columns.

Returns:

xarray.DataArray – Dims ("time", "fourier") with coords labeling each harmonic (sin_1 .. sin_k, cos_1 .. cos_k).

pymc_forecast.features.periodic_repeat(pattern, duration, *, axis=0, period=None)[source]#

Tile a seasonal pattern to cover duration time steps.

Works on numpy arrays and on PyTensor variables (e.g. a sampled seasonal latent inside a model). For symbolic tensors whose length along axis is not statically known, pass period explicitly.

Parameters:

  • pattern – The seasonal pattern; its length along axis is the period.

  • duration – Target length along axis.

  • axis – Axis to repeat along (defaults to 0, the package’s time axis).

  • period – Explicit period, required only when it cannot be read from pattern’s static shape.