EvaluationWindows.time_index

EvaluationWindows.time_index(dtype)

Build the positions along the fitted axis that each window covers.

HSGP-based latent variables such as media_temporal_latent_multiplier evaluate their basis functions at these positions, so a window has to say where on the axis it sits rather than counting from zero.

Parameters:

dtypestr

NumPy dtype for the output.

Returns:

np.ndarray

Shape (n_periods, max_window). Indices may run past n_dates where a short window is padded out.