Normalizing averages disregarding extremes whencalculating averages
#1
Posted 2011-November-24, 23:48
#2
Posted 2011-November-25, 02:02
#3
Posted 2011-November-25, 02:33
George Carlin
#4
Posted 2011-November-25, 05:30
The metho to obtain the result is very similar as how matchpoints are calculated, replacing the 1/0.5/0 result by the difference according to IMP table and (optionally; BBO does it) divide by the number of comparisons..
A recent thread went about the advantages of this vs butler.
IMO, the only advantage of butler is needing fewer calculations to obtain a result, and when you use a computer that point is moot.
#5
Posted 2011-November-25, 07:43
Gerardo, on 2011-November-25, 05:30, said:
The metho to obtain the result is very similar as how matchpoints are calculated, replacing the 1/0.5/0 result by the difference according to IMP table and (optionally; BBO does it) divide by the number of comparisons..
A recent thread went about the advantages of this vs butler.
IMO, the only advantage of butler is needing fewer calculations to obtain a result, and when you use a computer that point is moot.
Sixteen does seem an unreasonably low number of comparisons, though. In the days when I was an OKBridge member, with a MUCH lower total membership than BBO, they still managed to use fifty plays per board.
Going back to the OP's point, I'm not so much bothered about exchanging a couple of IMPs for cold games as I am about the morons who give (and accept) impossible results, just because one side landed in a ridiculous contract. THAT is really frustrating, particularly when the board has had only a few plays when you get to it. Yes, I know the scores will even out as the board has further plays, but it still skews the running score. Over nearly 40 years of playing bridge I've screwed up hands in almost every way imaginable, but I've yet to find any way to avoid making at least four tricks when I have AKQJ of trumps in my hand.
#6
Posted 2011-November-25, 08:12
While those ridiculous scores can be annoying, how many of the boards you play in a session actually have them? If you play a couple dozen boards, and one of them had a joker on it, this shouldn't have a big effect on your total.
#7
Posted 2011-November-25, 09:09
George Carlin
#8
Posted 2011-November-25, 09:24
barmar, on 2011-November-25, 08:12, said:
While those ridiculous scores can be annoying, how many of the boards you play in a session actually have them? If you play a couple dozen boards, and one of them had a joker on it, this shouldn't have a big effect on your total.
I realise OKB had the additional motivation for the larger number of comparisons, but it still seems to me that sixteen is a bit low.
And sure, the ridiculous scores don't occur that often. I'd put it a little higher than you do, not the downright impossible scores but the ludicrous contracts (e.g. one member of a partnership is playing strong twos, the other weak, and in amongst all the slams and games you have a 2H+4).
I can see the case for winding up the number of plays and discarding the top and bottom, but then I don't have to implement it.
#9
Posted 2011-November-25, 09:36
For the 3 imp/game problem, you have to do something different. Some ideas:
1) Play a team match
2) Maybe BBO could implement "compare this table vs 4 GIB's" for people who've paid for GIB for that day/week? Can this already be done? This may be better, but would have its own annoyances too.
3) Maybe BBO could implement "compare this table vs par"? This would be interesting, but would again have its own annoyances such as when the other table bids and makes a slam requiring three finesses and a squeeze.
#10
Posted 2011-November-25, 10:08
Never tell the same lie twice. - Elim Garek on the real moral of "The boy who cried wolf"
#11
Posted 2011-November-25, 10:26
gwnn, on 2011-November-25, 02:33, said:
It's probably anti-BBO-philosophy, but one way to combat this might be to create an Advanced Bridge Club, that players have to qualify (in some way, shape or form) to play in.
#12
Posted 2011-November-25, 18:03
The limit was hard coded into the Windows-Client and and that limit propagated into the server software.
One day the support of the Windows-Client might end and after that the 16 score limit will hopefully be history.
#13
Posted 2011-November-25, 23:14
hotShot, on 2011-November-25, 18:03, said:
The limit was hard coded into the Windows-Client and and that limit propagated into the server software.
One day the support of the Windows-Client might end and after that the 16 score limit will hopefully be history.
As a user of the Windows client, albeit under Linux, I certainly hope that day is a long way away. I've tried the web client and I detest the interface.
As a (retired) database programmer myself, I would be surprised to learn that something like that had been hard-coded in such a way that increasing it would require significant amounts of work. Yes, there are limits that you can't get round, e.g. the self-contained database code with which I was most familiar limited you to 1,000,000 records per table and a record size of 32,000 bytes, but I've never heard of a toolbox that would be responsible for a limit of 16 plays.
#14
Posted 2011-November-27, 20:56
For those who do not like Butler socring here are few clarifications:
1) Nothing gets "thrown out", extremes are just not counted when calculating the average, ALL results are then compared to that "normalized" average.
2) Disregarding 3 results from each side would help a lot to reduce that 3 IMPs per board "declarer advantage". If you examine typical cold game score sheet "funny" results are almost never equally distributed at the both ends, usually you have about 3-5 on one side (missed/failed game, tried slam) and 1-3 in the other (contract doubled and made), using middle 10 results would help a lot.
3) Team matches are even more volatile, even if all 8 players are competent, becuase there is only one comparising, playng against "the field" is much better to figure out where is your game
4)"it all evens out anyway" just makes no sense. For start it is quite common that you play set of 20-30 boards that is very onesided result-wise, so yes it all evens out in very long run, but who cares? In the long run we are all dead too.
And one more thing: being involved in lot of software development (yes in Flex/Flash environment too!) I can not shake off the feeling that there is simply not much going on lately at BBO Inc. on that front. If the money is the issue I suggest doing some kind of drive like Wikipedia does, I am the first one who will donate some money if it means hiring an additional programer to help Uday, if it would lead to better user experience.
#15
Posted 2011-November-28, 00:07
zenko, on 2011-November-27, 20:56, said:
We don't compare to an average, we do cross-IMP scoring. Comparing to an average was mainly done in the days of hand scoring, because it required fewer calculations. Cross-IMP is most common with computer scoring.
#16
Posted 2011-November-28, 22:57
barmar, on 2011-November-28, 00:07, said:
No it was not done to reduce number of calculations, it is done to reduce disparity of median and mean results, actually it is even more important to do it with less than statistcally significant number of results to compare, becuse a few freak results will throw the mean completely out of whack.
#17
Posted 2011-November-28, 23:30
zenko, on 2011-November-28, 22:57, said:
It was too.
Comparing to an average value like -235 will yield unbridge-like imp results (results that the imp table was not designed for).
George Carlin
#18
Posted 2011-November-29, 03:09
gwnn, on 2011-November-28, 23:30, said:
Comparing to an average value like -235 will yield unbridge-like imp results (results that the imp table was not designed for).
Having hand-scored more than my fair share of Butlers, all you do is round the calculated average before comparing. What's the problem? One line of code if you're doing it with a program.
#19
Posted 2011-November-29, 03:41
George Carlin
#20
Posted 2011-December-04, 13:52