Author: Robert Hyatt
Date: 20:30:07 02/16/06
Go up one level in this thread
On February 16, 2006 at 14:07:00, Peter Kappler wrote: > >>>>>> >>>>>>In practice this doesn't happen. The real problem is that by round 6, the >>>>>>winner is pretty certain, although three games against weaker opponents are >>>>>>left, and there the luck factor often lurks behind a pawn and jumps out to >>>>>>present a surprise and knock one of the top players off. >>>>>> >>>>> >>>>>I have to believe that you're half-kidding here. Surely you don't believe that >>>>> adding 3 more rounds increases the variance of the final result? >>>>> >>>>>-Peter >>>>> >>>> >>>>Yes, it absolutely does. Simple example: 4 strong programs, 12 weak ones. >>>>after round 1, you have 8 with 1, 8 with 0, the 8 with 1 include the top 4. >>>>After two rounds, the top four have two, the bottom four have 0, and the rest >>>>have 1 (assuming no draws). after 3 more rounds the top four have played. What >>>>now for those other 4 rounds? Other programs? With the luck factor increasing >>>>variance? >>>> >>> >>> >>>What is this "luck factor"??? >> >>It is something seen in _every_ tournament played. A weaker program stumbles >>into a winning position, either being lucky with the book, or just lucky in >>making a move it didn't understand was good, but which turned out to be good >>many moves later. >> >>I can refer you to the game Nuchess - Cray Blitz at the 1984 ACM event. We >>played a horrible-looking Nb8 move. But it later let us trade off all the >>pieces and win a pawn-race because white simply didn't understand unstoppable >>pawns. But had black not played that one ugly move, for the wrong reason, white >>would have won handily. >> >>Every tournament has a luck factor thrown in. Even in my human chess games, I >>occasionally stumble into something I had not forseen, and win because of it. >> >> >>> >>>If one of the strong programs loses or draws against one of the weak programs, >>>that's not luck - it's a signal that the program might not be as strong as you >>>thought. >> >> >>It can be luck plain and simply. You have three moves to choose between, at the >>depth you are searching, all three have _identical_ scores even though one will >>later turn out to be winning, one will turn out to be losing, and one will turn >>out to be pretty "equal". What determines which move your search chooses? Luck >>of move ordering at the root. Since the first move with equal scores will be >>chosen by alpha/beta. I can play a move that is lucky in that it wins even >>though I don't know it at the time, I can play a move that is unlucky in that it >>loses later in the game. >> >>That's luck, since it is based on random choice rather than any sort of chess >>knowledge or skill. >> > >Sure, these things can happen, but it is the exception rather than the rule. >Most chess games are not decided by luck, and this is why playing more rounds in >a tournament definitely reduces the variance of the result. If you look at these extended events, the things are generally done by round 7. 8 and 9 just take up time. It is better to either (a) be close to log2(#players) or else get close to #players-1 rounds (RR). In the middle of that range, things can be less informative, rather than more. I've seen this happen too many times when the age-old cry "too few rounds" causes over-reaction. If all opponents were "pretty close" then more rounds are better, but when there are a few at the top and the rest chasing them, after the top few have played each other, the rest of the rounds are just burning time. > > >> >> >>> >>>It seems obvious to me that increasing the number of rounds always reduces the >>>variance of the final tournament standing, regardless of the distribution of the >>>participants' ratings. >>> >> >>Seems to me that if you look at events starting at round 6 and on, nothing much >>changes since the leaders have all played because of the Swiss pairing rules. >>But lower-rated programs still move up and down and a few collect right below >>the leaders. If all programs were pretty equal, extra rounds would not really >>make that much of a difference. But they are not equal, and strong vs weak in >>late rounds doesn't do a thing for overall result confidence. >> > >Those "extra" rounds have decided the winner in at least a few of the CCT >tournaments. > >-Peter
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.