i will work on trying to speed up the candidate node training. there are a few spots where it seems to calculate the same things twice or more (CalcVBar, CalcEoBar, CalcSigmaO…).
currently training the spiral problem with todays changes. file is cascnet22.nb.
