The code uses
self.S = P_inv + dot(H_T, R_inv).dot(H)
which is consistent with Dan Simon's book, but I am dubious. In general the shape of covariance in state space and measurement space is different, but this requires them to be identical.
Until this is checked and fixed I am removing the computation of mahalanobis, since it cannot be computed given that the shapes of S and y are not compatible.
The code uses
which is consistent with Dan Simon's book, but I am dubious. In general the shape of covariance in state space and measurement space is different, but this requires them to be identical.
Until this is checked and fixed I am removing the computation of mahalanobis, since it cannot be computed given that the shapes of S and y are not compatible.