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

BGonline.org Forums

Disagree with Stick's Book Review

Posted By: Daniel Murphy
Date: Saturday, 11 December 2010, at 10:41 p.m.

In Response To: Disagree with Stick's Book Review (Timothy Chow)

Of course not. There can't possibly be such a demonstration. It's all marketing, not science.

Thanks for your response, Tim. I beg to differ with that one statement, but I think I should first clarify that when I said I could not find an evaluative comparison of the XG and Gnubg formulas, I did not mean the **millipoint divisors multipliers, the similarity (or not) of XG PR and Gnubg ER to Snowie ER (Gnubg's being approximately twice Snowie's for checker error and total error and four times Snowie's for cube error, and XG's being approximately 1.1 times Snowie's), or the scale and nomenclature (what is "world class," etc.) of the PR/ER.

I meant the difference(s) in the formulas themselves, namely:

(1) For checker play, Snowie divides your total checker play error by the number of all moves by both players. Gnubg divides your total checker play error by the number of your unforced moves. I believe that like Gnubg, XG does not count forced moves, but also does not count unforced moves when no error is impossible (e.g., you get gammoned no matter what move you choose). Is that so, and are there any other differences?

(2) For cube action, Snowie divides your total cube error by the number of all moves by both players. Gnubg divides your total cube error by the number of your close or actual cube decisions, where "close" means within 0.250 (but all "too good" decisions are counted). Is the XG method identical to Gnubg's?

(3) For total ER, I believe Snowie's formula is (your checker error plus your cube error) divided by (all plays) times 1000, and Gnubg's is (your checker play error + your cube error) divided by (your unforced checker plays plus your close or actual cube decisions) times 1,000. Other than the multiplier, is XG the same as Gnubg here?

So ... I think that an argument can be made, that if the bots calculate (1) (2) and (3) differently, one method is better than the other(s), and it's not "all marketing." For example, I think most everyone here will agree that not counting forced moves is better than counting them. For another, the method of counting checker and cube errors, and adding them together, might make the total error rate more intuitively reflective of ability.

**I think multiplying millipoints by 500 is inferior to multiplying by 1,000, and to multiply by 500 in order to get a Snowie-like number is a bit ironic, considering that Snowie ER is "inherently flawed" and XG PR is the "new standard," but that's another problem, which I addressed separately --and see below:

On the XG website: "Our take on the Snowie Rating is that it is inherently flawed. Snowie rating is the amount of equity (normalized) lost multiply by 1000 divided by the number for roll (for both players). The main problem of it is that it does not take into account forced moves or moves that are completely unimportant." Also: "We are committed to make eXtreme the standard in Backgammon and bringing back old system that we don't believe in is not something we want to do. Since Version 1.10 eXtreme Gammon has the ability to display the Performance Rating. It is the equity per decision multiplied by 500. So the number are in the range of Snowie ER, but still take into account that force move are not to be counted.

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.