e(sv_coef) is incorrectly sorted
I think I found a bug:
The column sv
is output here:
https://github.com/kousu/statasvm/blob/055a7ffcb8ef4586908eab855bfc8fcb7686e61d/src/_svm_model2stata.ado#L109-L110 https://github.com/kousu/statasvm/blob/master/src/_svmachines.c#L488
to mark which rows are support vectors. But there's also e(sv_coef) output here
https://github.com/kousu/statasvm/blob/master/src/_svmachines.c#L369
which has the relative strengths(I think?) of each support vector.
But: there's this sort() call:
which ..as far as I can tell...unlinks the order of sv_coef from ->sv_indices. So does that make it totally wrong??
I wish we had some obvious test cases (with graphical output!!) that could demonstrate this.