@justaz said in #10:
> I picked Ordo because it is standard in computer chess, and I assumed those guys know what they're doing. Check out
www.talkchess.com/forum3/viewtopic.php?f=2&t=44180>
>
> Cool! I'll look into this
>
>
> The key factor in Ordo being slow is that it's doing gradient descent on a single cpu thread. If I use my brain maybe I'll manage to put it on the GPU with Torch or something, and get a 1000x speedup. As for the period size, I look into it when I compute the whole backlog of ratings, maybe 3 months is perfect, maybe 3 weeks, we will see. Keep an eye out for my next rating release when December database drops!
Hey
@justaz , I'm a mathematical physicist more than a statistician by training, so it would be great if in this or future blogs you include a reference to how Ordo works, as it will be easier for casual/new readers like me to jump into the content faster :)
From a logical perspective, I would have a lot of the same comments as
@MyPoorRook around hand-wavy statements, but I am happy to see your open and measured responses :) One additional comment from a presentation of information perspective, I would have is that, at least on my lichess screen, is the first graph with 1+ million points is hard to read and get something from. I understand that no graphing program is going to enjoy trying to represent all of the points in that case, so would it be possible to present representative envelopes for points, or a trend line against lichess ratings to show that the general clump isn't well-modelled by slope 1 or something?
Interesting topic and thanks for posting.