Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Looks like it might be a 3 way ftie for first

Author: Robert Hyatt

Date: 22:23:45 01/19/03

Go up one level in this thread


On January 19, 2003 at 23:20:42, Ricardo Gibert wrote:

>On January 19, 2003 at 23:00:38, Robert Hyatt wrote:
>
>>On January 19, 2003 at 22:10:53, Ricardo Gibert wrote:
>>
>>>On January 19, 2003 at 21:23:11, Robert Hyatt wrote:
>>>
>>>>On January 19, 2003 at 18:31:55, Russell Reagan wrote:
>>>>
>>>>>On January 19, 2003 at 18:18:00, Sune Larsson wrote:
>>>>>
>>>>>> means Ruffian/Crafty/Yace sharing 1st place with 7/9
>>>>>
>>>>>Crafty beat both Yace and Ruffian. Ruffian beat Yace, and Yace lost to both
>>>>>Crafty and Ruffian. So, Crafty should be the winner. Or are they playing tie
>>>>>breakers?
>>>>
>>>>
>>>>That's not really done in swiss events.  The tie-break for this event
>>>>was (IMHO) not very well thought-out.  IE who wants to use blitz games to
>>>>settle the final standings of a long time-control tournament?
>>>>
>>>>I would suggest the following for future CCT events.
>>>>
>>>>If two are tied for first, have a play-off game, same time control.
>>>>
>>>>If more than two are tied, either use a traditional tie-break or forget
>>>>about it.  IE at the ACM events, we had "co-champions" that were ranked by
>>>>the tie-break scores.  We never had playoffs at any ACM event, nor at any
>>>>WCCC event I attended although I do remember at least one tie-break with
>>>>Belle playing, probably in 1980.
>>>>
>>>>But in any case, blitz games for a standard time control tournament makes
>>>>no sense, no matter how you look at it.  If nobody is happy using the
>>>>normal sum of opponents scores, which is really pretty useless when you have
>>>>too many rounds as we do, then co-champions would be the simplest and most
>>>>accurate outcome.
>>>
>>>
>>>But I like the blitz playoffs! Between humans players they're too random, but
>>>chess engines get plenty of depth to be able to play some decent chess.
>>>
>>
>>However, an engine can be tuned for blitz or tuned for longer games.  Is it
>>fair to have the engine play the main games at a slow time control, then when
>>a couple tie, to use a totally different time control?  Does that _really_ show
>>which of the two should have won the tournament???
>>
>>>
>>>>
>>>>Other points to ponder:
>>>>
>>>>1.  too many rounds.  You really want to have log2(entries) rounds, rounded
>>>>up to a integer.  For 48 players, that would be 6 rounds.  The more rounds
>>>>held _beyond_ that, the more likely there is to have a log-jam at the top
>>>>since the top players can not play each other a second time, and they will
>>>>end up playing lower rated players, giving more a chance to join them.
>>>
>>>
>>>Log2(entries) is the correct formula if all the games are decisive, but the
>>>draws changes things significantly. Even fewer rounds are enough to
>>>differentiate the contestants.
>>>
>>>For example, in all 5 CCT events, there was a contestant in clear first after
>>>just 5 rounds.
>>
>>That's the point.  If you go beyond log2, you run into problems.  It might
>>seem like a good idea, but it introduces a problem with the "logjam" issue.
>
>
>I was assuming you were were rounding log2(entries) up as per your, "For 48
>players, that would be 6 rounds." BTW, that's "beyond log2."

I don't follow.  log2(64)=6, and you have to round up to the next value,
for obvious reasons.  log2(32)=5 so log2(48) is between 5 and 6, and since
I don't know how to do a "partial round" I round up.  :)


>
>To be precise, if all games are decisve, then ceil(log2(entries)) is correct,
>but less works for chess. My first guess would be ceil(log3(entries)), but that
>seems too low to me. Certainly, it is too low for weak events where it is likely
>that almost all the games are decisive. I don't what the real formula should be,
>but I think it should include an estimate of the drawing rate for the event.
>
>

Less works, but you generally have to rely on tie-breaks.  Which means
seeding order is important.

>>
>>>
>>>Also, consider that major swiss events typically have fewer than log2(entries),
>>>so even fewer rounds than log2(entries) works.
>>
>>no doubt at all there.  The ACM events were 4 rounds for years, and the last
>>few years were expanded to 5.  The WCCC events were 5 rounds, at least
>>through 1989, the last one I played in.
>>
>>>
>>>What I don't like about a small number of rounds is luck is too big a factor. To
>>>me, 9 rounds is a kind of minimum. The old WCCC events with just 4 rounds were
>>>ridiculous. You might as well draw names out of a hat.
>>>
>>>This is all why I still like the 5 round double RR idea.
>>>
>>>
>>>>
>>>>2.  More time between rounds.  It makes little sense to end a round and 1 minute
>>>>later start the next.  Hardly anything started on time, suggesting the start
>>>>times were too optimistic.  I'd suggest 2.5 hours per round rather than 2, which
>>>>would make it more relaxed.
>>>>
>>>>3.  If a player is more than 10 minutes late, he forfeits that round, period.
>>>>If he is more than 10 minutes late in two rounds, he is kicked out of the
>>>>event.
>>>>
>>>>It was a fun event, and was well-run, with the mention of the problems
>>>>given above...



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.