So the turnover function in numerai_tools just returns an unbounded sum (divided by 2), but in any case a number that might be in the hundreds or even thousands. To get turnover percentage, am I then simply dividing this number by the number of post-filtered/sorted weights that I'm comparing (i.e. length of either s1 or s2)? Or by the sum of all the weights? What's the official final calculation?
And does n top/bottom ever come into it? (As far as official scoring.)