[ View Thread ] [ Post Response ] [ Return to Index ] [ Read Prev Msg ] [ Read Next Msg ]

BGonline.org Forums

Suggested Idea for Performance Grading

Posted By: Bob Koca
Date: Sunday, 5 October 2014, at 5:51 a.m.

In Response To: Suggested Idea for Performance Grading (Dmitriy Obukhov)

http://www.bkgm.com/articles/GOL/Feb00/var.htm

Essentially each roll it uses its estimate of the ensuing equities to grade how lucky or how unlucky the roll was and keeps a running count. These estimates are unbiased and since bots are really good now will be strongly positively correlated with actual luck. This luck adjustment is completely independent of the actual play made so a huge advantage of this approach compared to PR approach is that one should be making the play that gives one the best chance against the particular opponent.

This can be incorporated into an Elo system as in the following example: Suppose that the program rates the winner A as being +40% lucky then A gets a .6 win and B gets a .4 win. Suppose that the ELo difference predicted that A had a 58% chance to win the match. A exceeded that by not by much and A's rating increase by a very small amount.

In terms of cutting down variability consider the following results. I had XG advanced level play 1 point matches against itself. Instead of being a bunch of 0s and 1s by just looking at winner or loser the VRR technique gave instead the following: .54, .53, .44, .49, .61, .59, .44

Messages In This Thread

 

Post Response
Your Name:
Your E-Mail Address:
Subject:
Message:

If necessary, enter your password below:

Password:
 

 

[ View Thread ] [ Post Response ] [ Return to Index ] [ Read Prev Msg ] [ Read Next Msg ]

BGonline.org Forums is maintained by Stick with WebBBS 5.12.