Grid search just meant that I took the whole dataset (that I had at the time), and tried out a bunch of different numbers for those parameters until I got a set which was the most predictive of all the games overall.
1 Like
Oh, also relevant, I suppose, is that I’m using 1135.77 as the divisor, in order to get a particular win chance implication for a given skill gap (I don’t remember what has I was targeting…), And that also had implications on what K should be.
Here’s the ipython notebook where I minimized there error: yomi-skill/Yomi Elo.ipynb at master · cpennington/yomi-skill · GitHub
It is a number of days greater than zero, yes.
(Congrats!)
1 Like