Existing implementation of the polynomial filter is fragile. It was using SVD to overcome singular template covariance matrix and applied an arbitrary rcond cut to small singular values. When small singular values were encountered, the results were compromised.
The new implementation accepts singular values down to machine precision and does away with the squaring of the template matrix. The latter greatly improved the condition number of the system.
Existing implementation of the polynomial filter is fragile. It was using SVD to overcome singular template covariance matrix and applied an arbitrary rcond cut to small singular values. When small singular values were encountered, the results were compromised.
The new implementation accepts singular values down to machine precision and does away with the squaring of the template matrix. The latter greatly improved the condition number of the system.