NonlinearSSM.unscented_smoother

NonlinearSSM.unscented_smoother(y, *, alpha=1e-1, beta=2.0, kappa=0.0)

The unscented Rauch-Tung-Striebel smoother.

Runs the unscented filter, then Rauch’s backward recursion with the filtered-to-predicted cross-covariance rebuilt from sigma points at each filtered moment pair – the same deterministic points the forward pass used, so the two passes share one approximation rather than stacking two different ones. Exact through linear maps, like its filter.

Parameters:

• y (ArrayLike) – Data, (n, p); numpy.nan rows are missing.

• alpha (float) – Sigma-point spread; must match the intended filter’s.

• beta (float) – Prior-distribution weighting.

• kappa (float) – Secondary scaling.

Return type:

_RtsSmootherResult

Returns:

The _RtsSmootherResult with method="unscented".

Raises:

• SpecificationError – If the model carries custom hooks or the spread is not positive.

• NumericalError – If either pass degenerates.