When fitting with GLS, the noise covariance matrix could be cached for the duration of the fit, but gets called several times during each iteration of the fit instead. Larger noise bases lead to significantly slow fitting and this could help reduce that bottleneck.
When fitting with GLS, the noise covariance matrix could be cached for the duration of the fit, but gets called several times during each iteration of the fit instead. Larger noise bases lead to significantly slow fitting and this could help reduce that bottleneck.