NonlinearSSM.unscented_filter

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

The unscented Kalman filter: deterministic sigma points, no Jacobians.

Exact through linear maps for any parameter setting, and accurate to third order in the Taylor sense through smooth nonlinear ones (Julier-Uhlmann); the standard additive-noise form.

Parameters:

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

• alpha (float) – Sigma-point spread; small values keep the points near the mean.

• beta (float) – Prior-distribution weighting; 2 is optimal for a Gaussian.

• kappa (float) – Secondary scaling; 0 is the standard default.

Return type:

_KalmanFilterResult

Returns:

The standard filter record; the log-likelihood is the unscented approximation.

Raises:

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

• NumericalError – If the filter diverges.