>>/7809/
you see what I mean about the smoothness of the line though
I suppose an exponentially weighted moving average will show more variability for polarized training...