Author |
Topic: 2011 World Championship (Read 21181 times) |
|
Adanac
Forum Guru
Arimaa player #892
Gender:
Posts: 635
|
|
Re: 2011 World Championship
« Reply #150 on: Feb 1st, 2011, 7:22am » |
Quote Modify
|
Omar, did the pairings get altered sometime between Sunday night and Monday afternoon? I'm now seeing about 8 inconsistencies between the Wiki and the Gameroom. It's mostly gold/silver assignments that are different but I'm also seeing some new pairings. For example, the Wiki shows the original pairings (based upon current rank): #23 naveed vs. #27 beancrisp #24 ginrunner vs. #28 oali #25 722cassi vs. #29 Belteshazzar Now the Gameroom is showing: #23 naveed vs. #30 b599 #24 ginrunner vs. #27 beancrisp #25 722cassi vs. #28 oali etc. The original pairing made more sense to me than the 2nd one.
|
|
IP Logged |
|
|
|
omar
Forum Guru
Arimaa player #2
Gender:
Posts: 1003
|
|
Re: 2011 World Championship
« Reply #151 on: Feb 1st, 2011, 10:49am » |
Quote Modify
|
on Jan 30th, 2011, 7:53pm, Fritzlein wrote:For simplicity I recommend that the value from Round 1 should be used for all rounds. It sort of makes sense, too, given that the ghosts of dropped-out players are still getting draws every round. |
| Maybe I should have two versions of this Swiss based on whether players can join later or not.
|
|
IP Logged |
|
|
|
omar
Forum Guru
Arimaa player #2
Gender:
Posts: 1003
|
|
Re: 2011 World Championship
« Reply #152 on: Feb 1st, 2011, 11:29am » |
Quote Modify
|
on Feb 1st, 2011, 7:22am, Adanac wrote:Omar, did the pairings get altered sometime between Sunday night and Monday afternoon? I'm now seeing about 8 inconsistencies between the Wiki and the Gameroom. It's mostly gold/silver assignments that are different but I'm also seeing some new pairings. For example, the Wiki shows the original pairings (based upon current rank): #23 naveed vs. #27 beancrisp #24 ginrunner vs. #28 oali #25 722cassi vs. #29 Belteshazzar Now the Gameroom is showing: #23 naveed vs. #30 b599 #24 ginrunner vs. #27 beancrisp #25 722cassi vs. #28 oali etc. The original pairing made more sense to me than the 2nd one. |
| Greg, I ran the pairing again after changing the SoS to be computed using the players in the first round. I think that's when some of the pairings changed from what they were before. I thought we had the same F-factor now, so I'm not sure why the pairing changed. We still seem to have different SoS numbers going into round 5: http://arimaa.com/arimaa/events/showGames.cgi?e=2011wc The pairing algorithm does not assign the colors. After the pairing is done, another program does the color assignment. In pairs where it cannot decide, it randomly assigns the colors, so that caused the color assignments to change when I ran the pairing program again. I have already done the scheduling and emailed the players their opponents and game times. If it is critical I can cancel the games this week, try to get the difference sorted out and move round 5 to next week. Let me know ASAP if you think we should do this.
|
|
IP Logged |
|
|
|
Adanac
Forum Guru
Arimaa player #892
Gender:
Posts: 635
|
|
Re: 2011 World Championship
« Reply #153 on: Feb 1st, 2011, 11:54am » |
Quote Modify
|
on Feb 1st, 2011, 11:29am, omar wrote: Greg, I ran the pairing again after changing the SoS to be computed using the players in the first round. I think that's when some of the pairings changed from what they were before. I thought we had the same F-factor now, so I'm not sure why the pairing changed. We still seem to have different SoS numbers going into round 5: http://arimaa.com/arimaa/events/showGames.cgi?e=2011wc The pairing algorithm does not assign the colors. After the pairing is done, another program does the color assignment. In pairs where it cannot decide, it randomly assigns the colors, so that caused the color assignments to change when I ran the pairing program again. I have already done the scheduling and emailed the players their opponents and game times. If it is critical I can cancel the games this week, try to get the difference sorted out and move round 5 to next week. Let me know ASAP if you think we should do this. |
| Omar, I think everything is OK. It seems that the new formula has moved naveed below ginrunner and that is why the pairings are different. I think we should keep the pairings the way they are in the gameroom and I'll redo the Wiki tonight. Our SoS calculations seem to be consistent it's just the presentation that is different. Your table takes the SoS at the End of Round 4 and uses it in Round 5. If you compare your Round 5 the Wiki Round 4 SoS, they match: http://arimaa.com/arimaa/mwiki/index.php/2011_Open_Classic_Round_4 My Round 5 SoS has already added in the strength of the round 5 opponent, which is no big deal because the pairings have already been created.
|
« Last Edit: Feb 1st, 2011, 11:55am by Adanac » |
IP Logged |
|
|
|
omar
Forum Guru
Arimaa player #2
Gender:
Posts: 1003
|
|
Re: 2011 World Championship
« Reply #154 on: Feb 1st, 2011, 6:19pm » |
Quote Modify
|
Great. We can move forward with round 5 then. Glad to see that our SoS are matching.
|
|
IP Logged |
|
|
|
Adanac
Forum Guru
Arimaa player #892
Gender:
Posts: 635
|
|
Re: 2011 World Championship
« Reply #155 on: Feb 2nd, 2011, 8:09am » |
Quote Modify
|
on Feb 1st, 2011, 6:19pm, omar wrote:Great. We can move forward with round 5 then. Glad to see that our SoS are matching. |
| I finally figured out the final mystery. It's obvious in retrospect, but because I thought it was due to my error I looked for a non-existent mistake it took a bit longer to figure it out. Some F-factor values had Naveed ahead of Ginrunner and 722caasi and other ones have Naveed behind both of them. After 4 rounds Naveed's opponents had this many wins: 4-0-3-2 After 4 rounds both ginrunner's and 722caasi's opponents had this many wins: 2-1-2-2 Even though Naveed's opponents had 9 total wins, the new formula gave him a lower SoS after 4 rounds than ginrunner & 722caasi. This is why the pairings for 1-3 players changed. It's all correct and it matches both Omar's calculations and my own. The strange thing is that when I posted the Round 5 standings it switched the order back to the original schedule, with Naveed ahead of ginrunner & 722caasi. I thought I made a mistake somewhere but it's just that the opponent win totals are now: Naveed 4-0-3-2-1 GR/722: 2-1-2-2-1 Even though there haven't yet been any changes in the number of opponent wins, updating the round from 4 to 5 has altered the relative values of each round's opponent and now Naveed is back in front. So my first instinct was to expect the original pairings whereas Omar was correct to use the ones in the gameroom. If that makes sense.
|
« Last Edit: Feb 2nd, 2011, 9:20am by Adanac » |
IP Logged |
|
|
|
Fritzlein
Forum Guru
Arimaa player #706
Gender:
Posts: 5928
|
|
Re: 2011 World Championship
« Reply #156 on: Feb 2nd, 2011, 9:40am » |
Quote Modify
|
on Feb 2nd, 2011, 8:09am, Adanac wrote:Some F-factor values had Naveed ahead of Ginrunner and 722caasi and other ones have Naveed behind both of them. After 4 rounds Naveed's opponents have this many wins: 4-0-3-2 After 4 rounds both ginrunner's and 722caasi's opponents have this many wins: 2-1-2-2 |
| Thanks for explaining. It looks like the new SoS is behaving as intended. Although the old SoS would see nine opponent wins versus seven opponent wins, the new SoS sees that naveed played three stronger and one weaker opponent, whereas ginrunner and 722caasi have each played three stronger and one equal opponent. So the new SoS sees that one win and three losses is just what is expected from naveed, but one win and three losses is better than expected from ginrunner and 722caasi, so the latter are higher in the standings. Quote:Even though there haven't yet been any changes in the number of opponent wins, updating the round from 4 to 5 has changed the relative values of each round's opponent and now Naveed is back in front. |
| It matters most how many stronger, weaker, and equal opponents one has had, but there is still some balance with how much stronger or weaker the opponents were. The formula (correctly) thinks that the difference between a 2-3 player and a 1-4 player is less than the difference between a 2-2 player and a 1-3 player. So changing the round number from 4 to 5 suddenly gives ginrunner and 722caasi less credit for their three stronger opponents, since it thinks those opponents weren't as much stronger. It is enough to shift them behind naveed in the standings. But omar is correct to make the pairings while the round number is still 4, because the opponents are really 2-2 and 1-3, not yet 2-3 and 1-4. Indirectly we have discovered the limits of the sensitivity to the F-factor, because changing N from 4 to 5 (which is equivalent to multiplying F by 5/6) changed the order of players. I am still left with a mystery, though. Omar has the standings at the end of Round 4 being ginrunner 2, 3.037, 1790 722caasi 2, 3.037, 1596 naveed 2, 2.962, 1778 qswanger 2, 2.814, 1797 beancrisp 2, 2.814, 1690 oali 2, 2.814, 1626 b599 2, 2.692, 1480 Belteshazzar 2, 2.692, 1597 whereas you have the standings as being ginrunner 1790 [22] 3.0374 722caasi 1596 [28] 3.0374 naveed 1778 [23] 2.9618 qswanger 1797 [21] 2.8136 beancrisp 1690 [25] 2.8136 oali 1626 [26] 2.8136 Belteshazzar 1597 [27] 2.6916 b599 1480 [30] 2.6916 Note that the two of you have Belteshazzar and b599 flipped. To me it appears that omar's standings are wrong. Belteshazzar and b599 have the same record and the same SoS, so we go to the next tiebreaker, which is seed, a.k.a. pre-tournament rating. Belteshazzar has the higher pre-tournament rating, and therefore should be higher in the standings. In consequence of which, the pairings of b599 vs. naveed qswanger vs. Belteshazzar which were made this round should actually be Belteshazzar vs. naveed qswanger vs. b599 I think we need a ruling from RonWeasley here. On the one hand, the games have already been paired and scheduled, so re-pairing and re-scheduling represents a disruption. Also it's not a big deal since 1-1 players got paired against each other. On the other hand, the rules are the rules, and it appears the currently listed pairings are incorrect according to the rules. On further inspection, I can guess that the source of the mis-ordering is rounding error. Belteshazzar and b599 have the same opponent strengths except that Belteshazzar has a 2-2 and a 0-4 opponent in the place of b599 having two 1-3 opponents. For a 1-3 player, each of these combos is average, and each pair of opponents should contribute 1.00000000 to the strength of schedule. However, Belteshazzar might have only gotten credit for 0.99999999, and thus slipped a notch in the standings.
|
« Last Edit: Feb 2nd, 2011, 10:18am by Fritzlein » |
IP Logged |
|
|
|
RonWeasley
Forum Guru
Harry's friend (Arimaa player #441)
Gender:
Posts: 882
|
|
Re: 2011 World Championship
« Reply #157 on: Feb 2nd, 2011, 11:35am » |
Quote Modify
|
Right now I'm going to accept the rounding error theory and leave the pairings as is. I don't want to have to discover and correct for rounding errors several times in a round, so I'm choosing expediency this time. This means for the rest of the tournament I will accept the automated pairings if we still believe all unexpected pairings can be explained by rounding error.
|
|
IP Logged |
|
|
|
Fritzlein
Forum Guru
Arimaa player #706
Gender:
Posts: 5928
|
|
Re: 2011 World Championship
« Reply #158 on: Feb 2nd, 2011, 11:50am » |
Quote Modify
|
on Feb 2nd, 2011, 11:35am, RonWeasley wrote:Right now I'm going to accept the rounding error theory and leave the pairings as is. I don't want to have to discover and correct for rounding errors several times in a round, so I'm choosing expediency this time. This means for the rest of the tournament I will accept the automated pairings if we still believe all unexpected pairings can be explained by rounding error. |
| That's reasonable for this round, but if Omar is able to discover and fix a rounding error, you would like him to do so prior to pairing Round 6 and prior to determining the final top eight players, right?
|
|
IP Logged |
|
|
|
Adanac
Forum Guru
Arimaa player #892
Gender:
Posts: 635
|
|
Re: 2011 World Championship
« Reply #159 on: Feb 2nd, 2011, 12:46pm » |
Quote Modify
|
on Feb 2nd, 2011, 11:35am, RonWeasley wrote:Right now I'm going to accept the rounding error theory and leave the pairings as is. I don't want to have to discover and correct for rounding errors several times in a round, so I'm choosing expediency this time. This means for the rest of the tournament I will accept the automated pairings if we still believe all unexpected pairings can be explained by rounding error. |
| I like this ruling for round 5. We got really lucky in that this rounding error has virtually no chance whatsoever of affecting who qualifies for the top 8, or in which order. But if there's a rounding error in round 6 (or heaven forbid, at the end of the tournament) we might not be so lucky.
|
« Last Edit: Feb 2nd, 2011, 1:12pm by Adanac » |
IP Logged |
|
|
|
RonWeasley
Forum Guru
Harry's friend (Arimaa player #441)
Gender:
Posts: 882
|
|
Re: 2011 World Championship
« Reply #160 on: Feb 2nd, 2011, 3:39pm » |
Quote Modify
|
on Feb 2nd, 2011, 11:50am, Fritzlein wrote: That's reasonable for this round, but if Omar is able to discover and fix a rounding error, you would like him to do so prior to pairing Round 6 and prior to determining the final top eight players, right? |
| Yes, if we can make it better, we should.
|
|
IP Logged |
|
|
|
RonWeasley
Forum Guru
Harry's friend (Arimaa player #441)
Gender:
Posts: 882
|
|
Re: 2011 World Championship
« Reply #161 on: Feb 2nd, 2011, 3:51pm » |
Quote Modify
|
on Feb 2nd, 2011, 12:46pm, Adanac wrote: But if there's a rounding error in round 6 (or heaven forbid, at the end of the tournament) we might not be so lucky. |
| This is a good point. Rounding error, if not identified and corrected before the next pairings, is tolerable in my view. It is a random process (doesn't favor a certain kind of player) and the error doesn't cause a lot of divergence. The final standings is different and we should double check that they're correct. For the final standings, there's more time to do this without disrupting people's schedules. The mathematically correct standings will take precedence over the automated standings.
|
|
IP Logged |
|
|
|
aaaa
Forum Guru
Arimaa player #958
Posts: 768
|
|
Re: 2011 World Championship
« Reply #162 on: Feb 3rd, 2011, 9:41am » |
Quote Modify
|
I think I can fix the rounding error problem in the script, but I would need to know its interface, i.e. what parameters to supply to it and, for which these are file names, the respective formats.
|
|
IP Logged |
|
|
|
omar
Forum Guru
Arimaa player #2
Gender:
Posts: 1003
|
|
Re: 2011 World Championship
« Reply #163 on: Feb 5th, 2011, 2:21pm » |
Quote Modify
|
I am only display 3 digits after the decimal, but the sorting is using the full number. So the change in player order was due to some very low significant digit of SoS being higher for b599 than for Belteshazzar. I'll change the algorithm to round the SoS to 4 digits after the decimal and use that for sorting as well as displaying.
|
|
IP Logged |
|
|
|
Fritzlein
Forum Guru
Arimaa player #706
Gender:
Posts: 5928
|
|
Re: 2011 World Championship
« Reply #164 on: Feb 6th, 2011, 4:49pm » |
Quote Modify
|
on Feb 5th, 2011, 2:21pm, omar wrote:I'll change the algorithm to round the SoS to 4 digits after the decimal and use that for sorting as well as displaying. |
| Yay! The latest standings don't have any reversals like b599/Belteshazzar at the end of last round. Unfortunately, there is another potential issue with the pairing algorithm. I would like RonWeasley's ruling on whether to use the (IMO inferior) pairing produced by the software, or instead a slightly variant of that algorithm that both Adanac and I were expecting from our experience with Swiss-paired chess tournaments. We all agree on how to pair the top six players. Then it comes to the 3-2 players: 7 hanzack 8 99of9 9 Nevermind 10 The_Jeh 11 Hippo 12 jdb 13 omar 14 woh 15 Sconibulus 16 ArifSyed The "sliding pairing" used by Swiss systems is 7 hanzack 12 jdb 8 99of9 13 omar 9 Nevermind 14 woh 10 The_Jeh 15 Sconibulus 11 Hippo 16 ArifSyed Unfortunately, this pairing can't happen because hanzack has already played jdb and Nevermind has already played woh. A human TD might see that both problems can be solved with minimal disruption by swapping woh with jdb, and call that a good pairing, but let's see what the algorithm does. It tries to pair hanzack with jdb and rejects it as a repeat. Then it tries to pair hanzack with omar, and rejects that as a repeat too. Then it happily pair hanzack with woh. So far so good. 7 hanzack 14 woh Here's where the trouble starts. Both Adanac and I would expect the algorithm to next try to pair 99of9 with jdb, and work on down through the bottom half of the 3-2 players in order, until finding a suitable match. Omar's algorithm, however, skips jdb, and first tries to pair 99of9 with omar, yeilding 7 hanzack 14 woh 8 99of9 13 omar Then it tries to pair Nevermind, but again skips jdb and starts with woh, Nevermind's "ideal" opponent. Since woh is already paired, it drops down and pairs Nevermind with Sconibulus. Proceeding in this fashion yields 7 hanzack 14 woh 8 99of9 13 omar 9 Nevermind 15 Sconibulus 10 The_Jeh 16 ArifSyed 11 Hippo 12 jdb in which jdb is only paired at the end when all the 3-2 players have run out. Comparing this to straight sliding pairing, it doesn't look like we are off too much, except that Hippo has to play a far stonger opponent than he was supposed to. Everybody else got off by one or two seeds from expected opponent strength, but Hippo and jdb got thrown off by four seeds. Meanwhile, it would be perfectly straightforward for the algorithm to not forget about jdb until the end, but instead keep trying to pair him until it succeeds. This would result in pairings of 7 hanzack 14 woh 8 99of9 13 omar 9 Nevermind 12 jdb 10 The_Jeh 15 Sconibulus 11 Hippo 16 ArifSyed The latter pairing is clearly more in line with Swiss sliding pairing. It happens to be exactly what a human TD would have done, but it doesn't require application of human judgment; it only requires rigid application of a less-dumb algorithm. This scrambling of pairings is an unnecessarily large deviation sliding pairing, but it isn't a huge deal, and I wouldn't make a fuss about it if not for the fact that it makes an even greater hash out of the bottom of the pairings. The standings are: 17 megamau 18 Harren 19 ocmiente 20 knarl 21 Heyckie 22 722caasi 23 ChrisB 24 qswanger 25 ginrunner 26 naveed 27 Rad 28 beancrisp 29 b599 30 oali 31 Belteshazzar 32 ddyer The gaps indicate the score groups 2-3, 1-4, and 0-5. There are an odd number of 2-3 players, so Swiss pairing dictates that the bottom 2-3 player should be paired with the top 1-4 player, and similarly the bottom 1-4 player with the 0-5. We can't have everyone play an opponent with the same record, but that's the closest we can come. Thus the "ideal" Swiss pairings are 17 megamau 22 722caasi 18 Harren 23 ChrisB 19 ocmiente 24 qswanger 20 knarl 25 ginrunner 21 Heyckie 26 naveed 27 Rad 28 beancrisp 29 b599 30 oali 31 Belteshazzar 32 ddyer Alas, Harren vs. ChrisB, Rad vs. beancrisp, and Belteshazzar vs. ddyer are all repeats. What a human TD would do is up for debate, as there is more than one reasonable solution, but the algorithm in use by the tournament settles on an unreasonable solution. Just as it did in forgetting jdb until the end of the 3-2 players, the algorithm rejects Harren vs. ChrisB and then forgets ChrisB until all the other 2-3 players have been paired! So the 2-3 player who drops down to a 1-4 opponent is not the bottom 2-3 player or close to the bottom 2-3 player; it is the one-below-middle 2-3 player ChrisB. The algorithm's final pairings are 17 megamau 22 722caasi 18 Harren 24 qswanger 19 ocmiente 25 ginrunner 20 knarl 26 naveed 21 Heyckie 27 Rad 23 ChrisB 30 oali 28 beancrisp 31 Belteshazzar 29 b599 32 ddyer As before, most players are only off a couple from the expect strength of opponent, except that there was a huge swing for ChrisB and oali, the 1-4 player forced to face him. This is even more out of whack than the jdb-Hippo situation, because of the way it crosses score boundaries. Rather than give you my human-judgement pairings, I submit that the less-dumb algorithm that doesn't forget about unpaired players in the bottom half of a score group will mechanically do a better job, yeilding 17 megamau 22 722caasi 18 Harren 24 qswanger 19 ocmiente 23 ChrisB 20 knarl 25 ginrunner 21 Heyckie 26 naveed 27 Rad 29 b599 30 oali 31 Belteshazzar 28 beancrisp 32 ddyer (Note that once only four players remain, there is only one possible pairing to avoid giving ddyer a repeat opponent.) This algorithmic pairing also deviates from the ideal Swiss, but its deviations are less glaring than the actual pairings I feel that the principles of Swiss pairing have been violated enough to warrant replacing the algorithm-generated pairings of 1 Fritzlein 3 chessandgo 2 rabbits 4 Tuks 5 Adanac 6 Nombril 7 hanzack 14 woh 8 99of9 13 omar 9 Nevermind 15 Sconibulus 10 The_Jeh 16 ArifSyed 11 Hippo 12 jdb 17 megamau 22 722caasi 18 Harren 24 qswanger 19 ocmiente 25 ginrunner 20 knarl 26 naveed 21 Heyckie 27 Rad 23 ChrisB 30 oali 28 beancrisp 31 Belteshazzar 29 b599 32 ddyer with the slightly-smarter-algorithm-generated pairings of 1 Fritzlein 3 chessandgo 2 rabbits 4 Tuks 5 Adanac 6 Nombril 7 hanzack 14 woh 8 99of9 13 omar 9 Nevermind 12 jdb 10 The_Jeh 15 Sconibulus 11 Hippo 16 ArifSyed 17 megamau 22 722caasi 18 Harren 24 qswanger 19 ocmiente 23 ChrisB 20 knarl 25 ginrunner 21 Heyckie 26 naveed 27 Rad 29 b599 30 oali 31 Belteshazzar 28 beancrisp 32 ddyer On the other hand, expediency may again dictate simply going forward with what the software has already spit out. Ron, I await your ruling. Thanks in advance.
|
|
IP Logged |
|
|
|
|