
BGonline.org Forums
Using Noisy Evaluations to Measure Performance
Posted By: Chris Yep
Date: Friday, 1 April 2016, at 3:30 a.m.
In Response To: Using Noisy Evaluations to Measure Performance (Bob Koca)
Am I understanding correctly?:
Choosing C in this manner would mean that on average PRs would be the same as under the previous method. Your theory is that counting decisions by your method would reduce the variability from match to match.
Yes, this is correct. The goal is to choose C so that the average PR (for a large pool of representative players) is the same under XG's current PR formula and the proposed new method. My hope is that this new method reduces the variability of PR from match to match. (My same comments apply to GNU's ER formula, which is similar to XG's PR formula.)
Note also that I mainly refer to "decisions" so that the formula (algorithm) is analogous to XG's current PR formula. If I had wanted to I could have defined "complexity units" as "decisions" divided by 500. Then new PR = (Total [magnitude of] errors in normalized units) / (Complexity Units). Talking about "decisions" (hopefully) makes this new PR slightly more accessible to XG users.
I also think that this new method may reduce the PRs (compared to XG's current PR) of players who actively seek out complex positions. Under XG's current PR method, these players are "punished" with higher PRs. Conversely, I think this new method may increase the PRs (compared to XG's current PR) of players who actively seek out simple positions. See my response to Timothy later tonight.

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