BGonline.org Forums

VRR - Challenge

Posted By: Maik Stiebler
Date: Saturday, 31 January 2015, at 8:10 a.m.

In Response To: VRR - Challenge (Rick Janowski)

I don't think there is an article and can't remember a posting, but I can make one.

This is how you get the variance reduced rating change:

- Remind yourself of how the FIBS formula works.

- As per the FIBS formula, "Let D = abs(r1-r2) [rating difference]. Let P_upset = 1/(10^(D*sqrt(n)/2000)+1) [probability that underdog wins]"

- Now let Ru equal the luck-adjusted result of the underdog as given by the bots. For GnuBG, you will have to add 50% to the output number, as it is normalized to live in a typical range of [-50%, 50%] as opposed to the typical range of [0,100%] that XG uses.

- Let P = Ru - P_upset.

- For each player, let K = max ( 1 , -experience/100+5 )

- The rating change for the underdog is 4*K*sqrt(n)*P, for the favorite it is -4*K*sqrt(n)*P

- I already mentioned that you can increase the unramped game value 4*sqrt(n) when using VR. Maybe you will find that after increasing it enough, no ramping will be needed anymore.

Post Response

Subject:
Message: