Variational Kalman Filter

class kalman.vkf.VBKalmanFilter(*args: Any, **kwargs: Any)

Variational Bayesian Adaptive Kalman Filter (VB-AKF)

forward(observations: torch.Tensor) → Tuple[torch.Tensor, torch.Tensor]

Process full sequence (T, B, obs_dim)

get_measurement_covariance() → torch.Tensor

Returns current estimate of measurement covariance

predict(state: GaussianState, process_matrix: torch.Tensor | None = None) → GaussianState

Prediction step with covariance dynamics

update(state: GaussianState, measurement: torch.Tensor) → GaussianState

Iterative variational update step