Comment by paulluuk

Interesting! To be fair, I only tried it once, I did not look at a distribution over several attempts, so I may have just gotten (un)lucky. I'd expect this example to be training case #1 though.