
BGonline.org Forums
Using Noisy Evaluations to Measure Performance
Posted By: Chris Yep
Date: Saturday, 9 April 2016, at 5:35 p.m.
In Response To: Using Noisy Evaluations to Measure Performance (Maik Stiebler)
If complexity for both sides is generally strongly correlated, I see that there is a valid reason to account for high complexity. But I don't know that. And I think it might be more transparent to account for making your opponents' life more difficult by another number that is not the PR. For example, the statistics could tell that your PR is 5 and your opponents' average PR in their matches against you is higher by 2 than their normal average.
I think it's usually assumed that complexity for both sides is strongly correlated. For example, an opening 21: 13/11 6/5 probably increases complexity for both sides (compared to 24/23 13/11). Similarly, earlygame innerboard hitting battles probably increase complexity for both sides on average. But, I agree that we don't know this for certain. I would like to see a study on complexity in the future (examining a large sample of human matches).
Adjusting one's PR in a match by (opponent's normal PR  opponent's PR) is an interesting idea, although in many cases the opponent's normal PR is unavailable, especially for live opponents (online opponents' normal PRs can be estimated based on their ratings), so the proper adjustment isn't known.

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