This PR factors out the computation of the covariance from normal estimation such that alternative methods can be plugged in.
I implemented one alternative based on the minimum covariance determinant (see Hubert, Debruyne, 2009 and Nurunnabi, 2015). Unfortunately, the method is rather slow as you can see in the added example.
There are still some stability issues, so it's still a draft. I'll mark it for review when I've resolved the issues.
This PR factors out the computation of the covariance from normal estimation such that alternative methods can be plugged in. I implemented one alternative based on the minimum covariance determinant (see Hubert, Debruyne, 2009 and Nurunnabi, 2015). Unfortunately, the method is rather slow as you can see in the added example.
There are still some stability issues, so it's still a draft. I'll mark it for review when I've resolved the issues.