| |
BGonline.org Forums
Missed doubles and PR
Posted By: Dmitriy Obukhov In Response To: Missed doubles and PR (Florin Popa)
Date: Friday, 16 September 2016, at 6:43 p.m.
Okay, I conceed. Tom's explained it simply and elegantly. It is so simple, I am surprised I missed it.
To clarify, my proposed method was not based on math, but on my feel of needed reduction. Many ways of reduction could be suggesteid, and I just suggested the first one that popped in my head. If someone is of opinion that there should be no reduction, obviously he won't like any proposed method.
@ Tim: obviously, it is very hard, if even possible, to find a real backgammon position that satisfies all criteria I outlined before. But why is it important? I could come up with, let's say, 0.5 error if you throw gammons into the mix. Or, smaller error could be repeated more times. Or something else.
Still, after this discussion I am somewhat unsatisfied with the way XG currently handles these situations. Okay, it works out in the long run and supposedly should be right in any run, but hear me out. Imaginary situation:
A student takes a history test and answers 8/10 questions. A teacher gives him F grade and marks it 8/18. Naturally, the student asks for an explanation. The teacher explains that after test is checked, he randomly picks 1 question and multiplies it by a random number from 1 to 10. It turned out that one of 2 wrong answers was chosen and a weight of 9 was applied to it. The teacher asures the student that he has no beef with the student and he can check random draw record on the Web and see it was done by independent party. The teacher also explains that he could just as likely benefit from this innovative method and in the long run it all evens out anyway.
I don't know what the he'll the teacher was trying to test in this fashion but the knowledge ain't it. All he acheived is failing couple good students and passing couple bad ones. When you look at the class average grade you see that it didn't change, but this twisted test made many individual grades unrepresentive.
PR is supposed to be a test of skill in backgammon and these repeated errors, or right decisions, have a chance of having enough effect on PR to even make it useless in rare cases. Reoccuring decisions are not new decisions to a human. If he blundered on previous roll he will probably do it on the next one and won't think about this position all over. So, isn't counting errors twice or trice is similar to applying a weight to a history test questions? It surely seems so to me.
Obviosly, if a reduction, or some other method, is applied to repeated errors it should be applied to repeated correct decisions and luck calculations sgould be adjusted accordingly. The point is to reduce varriance and not artificially lower PR.
| |
BGonline.org Forums is maintained by Stick with WebBBS 5.12.