Author |
Topic: 2008 World Championship Discussion (Read 5708 times) |
|
Fritzlein
Forum Guru
Arimaa player #706
Gender: 
Posts: 5928
|
 |
Re: 2008 World Championship Discussion
« Reply #15 on: Jan 26th, 2008, 9:24am » |
 Quote  Modify
|
on Jan 21st, 2008, 5:39pm, Fritzlein wrote:| Also Adanac vs. PMertens has been entertaining in the past. It will be hard to call that game an upset no matter who wins. |
|
Just as I predicted, the game was highly entertaining, and the result was not an upset!
|
|
 IP Logged |
|
|
|
mistre
Forum Guru
Gender:
Posts: 553
|
|
Re: 2008 World Championship Discussion
« Reply #16 on: Jan 26th, 2008, 11:50am » |
Quote Modify
|
So is Froody beating Naveed an upset? Froody has a lower rating, but had a P8 rating of 7 higher than Naveed at the time of the match.
|
|
IP Logged |
|
|
|
Janzert
Forum Guru
Arimaa player #247
Gender:
Posts: 1016
|
|
Re: 2008 World Championship Discussion
« Reply #17 on: Jan 26th, 2008, 2:10pm » |
Quote Modify
|
I think the seed ratings should count for upsets. (I need a bikeshed smiley*)
Janzert
*But of course I have to be able to set the color 
|
|
IP Logged |
|
|
|
Fritzlein
Forum Guru
Arimaa player #706
Gender:
Posts: 5928
|
|
Re: 2008 World Championship Discussion
« Reply #18 on: Jan 28th, 2008, 11:19am » |
Quote Modify
|
on Jan 26th, 2008, 11:50am, mistre wrote:| So is Froody beating Naveed an upset? Froody has a lower rating, but had a P8 rating of 7 higher than Naveed at the time of the match. |
|
The best way to tell what is an upset is by who the spectators bet on.  But since the spectator contest hasn't started yet, we might as well go by the seeding. That means two upsets in round three: froody beat naveed; woh beat omar.
Unless round four sees a stunning upset of subs2000 beating naveed, the distribution of records after round four will be 2-6-10-6-2 (i.e. 2 players with no losses, 6 players with one loss, etc.). That would guarantee a distribution of 1-4-8-8-4-1 heading into the final round. Five players would already have four or more wins, and eight players would be 3-2, needing a win stay in the running for the finals.
The top 3-2 player will have to play up against the bottom 4-1 player, and will definitely qualify for the finals with a win. The bottom 3-2 player will have to play down against the top 2-3 player, and will probably not qualify even with a win. The results of these two matches will determine whether there are eight, nine, or ten players with at least four wins by the end of round six.
The middle six 3-2 players will play each other, with the three losers being out for sure, and the three winners probably but not necessarily qualifying, depending on how their opponents from the first five rounds do in the last round to affect tiebreak points.
|
|
IP Logged |
|
|
|
RonWeasley
Forum Guru
Harry's friend (Arimaa player #441)
Gender:
Posts: 882
|
|
Re: 2008 World Championship Discussion
« Reply #19 on: Jan 28th, 2008, 12:36pm » |
Quote Modify
|
This brings up an effect of a player forfeiting all or most matches. The wins the forfeiting player might have had do not contribute to the tie breaker scores of their opponents. So forfiets are annoying in yet another way. This is not as bad as intentional losses to help "teammates", as has been alleged in Swiss system events in chess for decades. I thought a little about whether an intentional loss could be used to obtain easier opponents later in the tournament, to somehow improve the chance of getting a qualifying win-loss record, but I think the expected value of a player's qualifying score is maximized if every game is played at one's highest level.
For the record, here is a proposed rule for next year some of us have discussed in the chat. A player who forfeits is disqualified from the tournament unless that player asks for, and is granted, permission to continue by the tournament director. In general, we expect players who sincerely want to continue, but missed a game due to extenuating circumstances, to ask and be permitted to continue. A player who essentially abandons the tournament need not take any action as it will be assumed that player will not return and no games will be scheduled for that player.
|
|
IP Logged |
|
|
|
Adanac
Forum Guru
Arimaa player #892
Gender:
Posts: 635
|
|
Re: 2008 World Championship Discussion
« Reply #20 on: Jan 28th, 2008, 1:51pm » |
Quote Modify
|
on Jan 28th, 2008, 12:36pm, RonWeasley wrote:This brings up an effect of a player forfeiting all or most matches. The wins the forfeiting player might have had do not contribute to the tie breaker scores of their opponents. So forfiets are annoying in yet another way. This is not as bad as intentional losses to help "teammates", as has been alleged in Swiss system events in chess for decades. I thought a little about whether an intentional loss could be used to obtain easier opponents later in the tournament, to somehow improve the chance of getting a qualifying win-loss record, but I think the expected value of a player's qualifying score is maximized if every game is played at one's highest level.
For the record, here is a proposed rule for next year some of us have discussed in the chat. A player who forfeits is disqualified from the tournament unless that player asks for, and is granted, permission to continue by the tournament director. In general, we expect players who sincerely want to continue, but missed a game due to extenuating circumstances, to ask and be permitted to continue. A player who essentially abandons the tournament need not take any action as it will be assumed that player will not return and no games will be scheduled for that player. |
|
I really liked the strength of schedule system when I read the rules pre-January 8th. It's a big improvement over giving tie-breakers based upon ratings. Unfortunately, it creates 2 problems that now seem all to obvious:
(1) Players that forfeit their games unfairly penalize their opponent(s), even if (especially if?) they're booted out after 1 round.
(2) Players with meaningless late round games can lose intentionally to help their friends that need tie-breaker help.
Is there an easy way to fix these problems? We could brain-storm new ideas for next year like re-calculating the strength of schedule for the top players before the final round, using only the scores of the players still in contention. However, I wouldn't be surprised if that fixes one set of problems while creating new ones. It's too bad, because the 2008 WC has a great system in theory.
|
| « Last Edit: Jan 28th, 2008, 1:52pm by Adanac » |
IP Logged |
|
|
|
The_Jeh
Forum Guru
Arimaa player #634
Gender:
Posts: 460
|
|
Re: 2008 World Championship Discussion
« Reply #21 on: Jan 28th, 2008, 2:45pm » |
Quote Modify
|
Just an offer for next time - I could break any ties by inputting tournament games into the Bradley-Terry model.
|
| « Last Edit: Jan 28th, 2008, 5:07pm by The_Jeh » |
IP Logged |
|
|
|
mistre
Forum Guru
Gender:
Posts: 553
|
|
Re: 2008 World Championship Discussion
« Reply #22 on: Jan 28th, 2008, 2:46pm » |
Quote Modify
|
Couldn't Head-to-Head, if applicable, be the first tie-breaker before looking at Strength of schedule?
Otherwise, I think Strength of schedule is the best method.
Karl, can you explain your SOS column on your standings table?
|
|
IP Logged |
|
|
|
Fritzlein
Forum Guru
Arimaa player #706
Gender:
Posts: 5928
|
|
Re: 2008 World Championship Discussion
« Reply #23 on: Jan 28th, 2008, 5:06pm » |
Quote Modify
|
on Jan 28th, 2008, 2:46pm, mistre wrote:| Karl, can you explain your SOS column on your standings table? |
|
At the moment, SoS is the sum of the wins of your first three opponents. When I get a chance I will add in the wins of everyone's present (fourth round) opponent. As the round progresses, whenever I enter the result of a game, I increment the SoS of all the opponents of the winner. This helps me keep track, but makes SoS difficult to compare mid-round.
Quote:| Couldn't Head-to-Head, if applicable, be the first tie-breaker before looking at Strength of schedule? |
|
I have a strong dislike of Head-to-Head as a method of breaking ties. If two players have identical 4-2 records, the player with the better performance is clearly the one who player tougher opponents overall. Focusing on the result of the one game the two played against each other is a way to base the decision on as little information information as possible, rather than on as much information as possible. The present strength of schedule system is somewhat random, but head-to-head would be more random because it uses less information.
Other ways to use as less information than the entire strength of schedule include: Whoever beat the toughest opponent is best ("quality win" theory); Whoever was defeated by the weaker opponent is worst ("embarrassing loss" theory); Whoever lost later in the tournament is worse ("peaking late" theory). All of these methods and other crazy fixations are used by humans to rank college football teams, because it is frankly information overload for humans to make a decision on the basis of an entire season. We seek out simple rules of thumb, no matter how arbitrary or unfair, because we can't stand the thought that it would take a pocket calculator to be accurate and fair in our judgments.
Head-to-head, however, has the additional problem that it is a headache to apply unless the there are only two tied players. In our case we will have three to five players tied at 4-2 records, which means there can easily be circularities within the group (i.e. A beat B beat C beat A). Even if you implement measures to resolve circularities, their bare existence undermines the whole notion that X beating Y shows that X is better regardless of all the other game results.
|
| « Last Edit: Jan 28th, 2008, 5:28pm by Fritzlein » |
IP Logged |
|
|
|
Fritzlein
Forum Guru
Arimaa player #706
Gender:
Posts: 5928
|
|
Re: 2008 World Championship Discussion
« Reply #24 on: Jan 28th, 2008, 5:23pm » |
Quote Modify
|
In the old discussions of tournament format, I argued the whole way for pure elimination tournaments, because that is the only way to rule out collusion. Not only is collusion ineffective in elimination format, everyone who is still playing is still in contention for the title, and thus has the original incentive to play to win.
Swiss and round-robin tournaments will always be vulnerable to collusion. Moreover, the temptation to collusion is exacerbated among the people who are out of contention. Since everyone plays every round, players who are eliminated from contention have nothing to lose by losing on purpose to help a friend. Less sinister but more common in non-elimination tournament, people who are out of contention lose motivation to even play, and fail to show up for their games, distorting the standings.
The current strength of schedule tiebreaker is manifestly unfair. PMertens, for example, has been hurt, through no fault of his own, that his first-round opponent hasn't showed up for any later games. That said, the strength of schedule is not entirely out of the players' control. By winning early rounds, players earn tougher pairings in later rounds, which earn them more tie-break points. If I win my first four and then lose my last two (a genuine possibility) I will probably have more tiebreaker points than some whose record was WLWLWW. In round five when he was playing someone 2-2, I would have been playing someone who was 4-0.
As for changing the format for next year, I am of two minds. On the one hand, the Swiss tiebreaker isn't totally unfair, and it will only affect folks in the range #7 - #10 in measured performance, and those folks probably aren't likely to win the World Championship even if they make the finals. So why not have a friendlier tournament server as the qualifier for the cut-throat finals?
On the other hand, maybe we shouldn't overload the World Championship tournament with too many objectives. If we want a friendly tournament, let's wait for summer and play six rounds of the level tournament we're talking about in another thread, not worrying so much about fairness and winning as fun. Then we can make the World Championship into a triple-elimination, take-no-prisoners, we-don't-care-if-we-hurt-someone's-feelings type of event.
|
| « Last Edit: Jan 28th, 2008, 5:26pm by Fritzlein » |
IP Logged |
|
|
|
Fritzlein
Forum Guru
Arimaa player #706
Gender:
Posts: 5928
|
|
Re: 2008 World Championship Discussion
« Reply #25 on: Jan 28th, 2008, 5:35pm » |
Quote Modify
|
on Jan 28th, 2008, 2:45pm, The_Jeh wrote:| Just an offer for next time - I could break any ties by inputting tournament games into the Bradley-Terry model. |
|
That's an interesting offer. When this tournament is over, I will be very curious to see how closely the ranking produced by your model agrees with the ranking produced by the tournament rules. If we all look at the two rankings and decide that your list is more reasonable than the official one, then that might be the way to go for next year.
|
|
IP Logged |
|
|
|
The_Jeh
Forum Guru
Arimaa player #634
Gender:
Posts: 460
|
|
Re: 2008 World Championship Discussion
« Reply #26 on: Jan 28th, 2008, 6:14pm » |
Quote Modify
|
Thank you, Fritzlein.
By the way, there's one sure way to avoid being a victim of collusion: don't lose.
|
|
IP Logged |
|
|
|
jdb
Forum Guru
Arimaa player #214
Gender:
Posts: 682
|
|
Re: 2008 World Championship Discussion
« Reply #27 on: Jan 28th, 2008, 6:18pm » |
Quote Modify
|
The following link talks about tiebreakers:
http://home.snafu.de/jasiek/egftsr.html
If two players are tied, head to head (assuming they played) is a better tie breaker than SoS. SoS uses different information than head to head. SoS uses more information, but the information used by head to head is of a much greater quality.
|
|
IP Logged |
|
|
|
Fritzlein
Forum Guru
Arimaa player #706
Gender:
Posts: 5928
|
|
Re: 2008 World Championship Discussion
« Reply #28 on: Jan 28th, 2008, 7:48pm » |
Quote Modify
|
Hmm, JDB, the link you provided says only the following about head-to-head tiebreaker: "Direct Comparison: Provided it can be applied at all, it is very meaningful because it might be interpreted as an already performed knockout playoff among the tied players. So, for the final results, generally it should be the first or even the only tiebreaker."
I find this argument entirely unpersuasive. Suppose for example that I beat Adanac in round four but lose to chessandgo and 99of9 in the two following rounds. Suppose meanwhile that Adanac comes back to beat chessango in the final round so that Adanac and chessandgo each finish 5-1. Which of them should get the top seed into the final?
Applying the argument of your linked page, Adanac's victory over chessandgo can be viewed as an "already performed knockout playoff" between the two players, so Adanac is #1. But if one loss is enough to "knockout" chessandgo, why wasn't Adanac's loss to me two rounds earlier enough to knock him out?
Why is Adanac's victory over chessandgo more meaningful than Adanac's loss to me, and chessandgo's victory over me? There is a circularity of dominance, but it is not considered because I finished 4-2 and am ignored by the tiebreak.
Consider the absurdity that head-to-head picks Adanac over chessandgo, unless I happen to win my final game. If I win, then I am also 5-1. Suddenly we have a three-way tie for head-to-head as the tied players are all 1-1 against each other. How did losing my game in the one scenario prove that Adanac was better than chessandgo, while while my winning in the other scenario evened it up? And they talk about SoS having a luck component!
In the scenarios I describe, I was an opponent of both tied players, so how I do in my other games should reflect identically on both of them. If I win my other games it should make them both look better, and if I lose my other games it should make them both look worse. I don't see how you can determine which of the two performed better overall without looking at their other four opponents, and seeing which of them played tougher competition apart from me, their common opponent.
|
|
IP Logged |
|
|
|
The_Jeh
Forum Guru
Arimaa player #634
Gender:
Posts: 460
|
|
Re: 2008 World Championship Discussion
« Reply #29 on: Jan 28th, 2008, 8:02pm » |
Quote Modify
|
on Jan 28th, 2008, 7:48pm, Fritzlein wrote: I don't see how you can determine which of the two performed better overall without looking at their other four opponents, and seeing which of them played tougher competition apart from me, their common opponent.
|
|
...which is exactly why the Bradley-Terry model is so brilliant. All games are interconnected. It considers your record, your opponents' records, your opponents' opponents' records, your opponents' opponents' opponents' records, etc. etc., out to an infinite degree. So rather than looking at head-to-head matchups or common opponents, we are breaking the tie by considering all paths of comparison, each weighted naturally, no matter how remote.
|
|
IP Logged |
|
|
|
|
Home
Help
Search
Members
Login
Register
Reply
Notify of replies
Send Topic
Print