[andreasvc]

That looks interesting. I wonder if nowadays that result could be replicated with readily available algorithms from scikit-learn.