After some testing I figured the previous bug has to do with recurrence of unique values in the data. I took care of it by using
alpha=alpha+0.00001*(1:numel(alpha)), but this is obviously a workaround which isn't satisfactory for a self-respecting algorithm.
At any rate, I forgot to mention how great this toolbox is. It has been of great help, and saved me a lot of time and work.