From: Jochum Drechsler web de> Date: 15 sep 2004 Subject: Shogi SOS-based pairing at the EC/WOSC This is a multi-part message in MIME format. --------------040604050503060801090401 Content-Type: text/plain; charset=us-ascii; format=flowed Content-Transfer-Encoding: quoted-printable Your free subscription is supported by today's sponsor: ------------------------------------------------------------------- Amazing Diet Patch The fastest - Easiest way to lose weight! Try it now FREE! http://click.topica.com/= caacBBBa2i6YsbnuqMaa/MyDietPatches ------------------------------------------------------------------- Dear shogi friends, thank you Pieter for your comments about the pairing system. I would like to add some remarks: The system is not strange. Most tournaments in Germany used this pairing= =20 system in the past. All major Go-tournaments in Europe also used this=20 method (and the programme). Also: This pairing system has nothing to do with MacMahon. The programme= =20 supports the MacMahon system and this methode can also be used in the=20 MacMahon system, but it is not restricted to it! =20 Let me first examine why it is reasonable to level the SOS (sum of=20 opponents score) for all players. I will use the WOSC 2000 and 2002 as=20 examples (for many tournaments the SOS and other secondary criteria are=20 not available on the web). WOSC 2002 15 Majewski Thomas DE 2D 1633 +39 6 40.5 24 32 26 17- 44+=20 13+ 9- 43+ 6- 33+ 29+ 30+ 16 Pfaffel Thomas AT 2D 1849 -16 5 50 23 38 31 21+ 11-=20 37+ 35+ 25+ 26+ 2- 3- 14- The highest ranked player with five wins (Pfaffel) has 50 SOS points,=20 the lowes ranked player with six wins (Majewski) has 40.5 SOS points. So= =20 Pfaffel was playing a stronger tournament than Majewski. Who performed=20 better in the tournament? =20 WOSC 2000 19 Blackstock, Les ....... ENG 1973 48- 26- 71+ 51+ 35+ 28- 44+ 41+=20 32+ 6 40.5 20 Pfaffel, Thomas ....... AUT 1200 82+ 25- 27- 40+ 55+ 22- 38+ 47+=20 30+ 6 38 21 Uemura, Yoshiyuki ..... FRA 2272 54+ 13+ 32+ 2- 1- 30+ 25+ 4-=20= =20 7- 5 56 22 Nilsson, Carl-Johan ... SWE 1828 28+ 6+ 10+ 3- 12- 20+ 7- 26-=20 56+ 5 53 23 Segers, Hans .......... NED 1932 41+ 3- 39+ 35+ 7- 32+ 13- 40-=20 44+ 5 49 24 Oosterwijk, Jan ....... NED 1959 44+ 48+ 12- 36+ 43+ 18+ 6- 7-=20 11- 5 48.5 25 Dysart, Doug .......... USA 1889 65+ 20+ 1- 46+ 6- 38+ 21- 32-=20 52+ 5 48 26 Maass, Stefan ......... GER 1510 62+ 19+ 6- 30- 33+ 55+ 16- 22+=20 17- 5 47 27 27 Kashiwazaki, Shintaro . GER 1740 49+ 7- 20+ 14- 50+ 31- 42+ 37+=20 10- 5 47 26 28 Drechsler, Jochen ..... GER 1768 22- 64+ 67+ 4- 60+ 19+ 5- 31+=20 16- 5 46 29 Litjens, Mathijs ...... NED 1773 30+ 12- 70+ 6- 76+ 7- 56+ 36+=20 14- 5 45 30 Heeffer, Albrecht ..... BEL 1778 29- 75+ 54+ 26+ 9- 21- 39+ 43+=20 20- 5 43 24 31 Connors, Don .......... USA 1894 75+ 50+ 2- 44- 53+ 27+ 8- 28-=20 48+ 5 43 22 32 Tran, Philippe ........ FRA 1765 67+ 61+ 21- 18- 54+ 23- 51+ 25+=20 19- 5 42.5 33 Lamb, Mike ............ USA 1704 77+ 1- 62+ 16- 26- 70+ 45+ 17-=20 50+ 5 42 34 Kaufman, Ray .......... USA 1886 17- 42- 65+ 45+ 52+ 36- 35+ 46+=20 13- 5 41 35 Suwa, Keiko ........... JPN 1620 74+ 8- 72+ 23- 19- 50+ 34- 51+=20 57+ 5 40 36 Ngaolertloi, Wisit .... THA 1680 50- 71+ 66+ 24- 45+ 34+ 14- 29-=20 43+ 5 39 22 19 37 Takemura, Hiroshi ..... JPN 1860 76+ 66+ 8- 43- 39+ 13- 52+ 27-=20 46+ 5 39 22 18 38 Christoffersen, Terje . NOR 1705 15- 68+ 45- 42+ 62+ 25- 20- 54+=20 59+ 5 39 21 39 Olufsen, Asle ......... NOR 1398 4- 63+ 23- 66+ 37- 62+ 30- 64+=20 49+ 5 38 =20 The highest ranked player with six wins (Uemura) has 56 SOS points, the=20 lowes ranked player with five wins (Pfaffel) has 38 SOS points! So=20 Uemura was playing a much stronger tournament than Pfaffel. Who=20 performed better in the tournament? Can you compare the performance of=20 those players at all?? The next player with the same SOS points than Pfaffel is 19 ranks below=20 (Olufsen)! Is that fair?? =20 The pairing programme is not directly minimizing the SOS points. The pairing is done following predefined rules an parameters. To find=20 the best pairing according to the differently weighed parameters the=20 programme utilizes the maximum-weight-perfect-matching-algorithm (based=20 on Edmonds). It follows three rules. The rules are dominant over the=20 other in this order: 1. No two players will be paired twice. 2. The global number of players paired up or down has to be minimized. 3. The choice of players that are paired up or down. =20 Pairing within a point (or MacMahon) group The players within one group (with the same number of points or macmahon= =20 points) are paired like this: the first player (according to his SOS)=20 against the last; the second against the second last. The players that=20 had a strong tournament so far (shown in their high sos) will get a=20 relatively weaker opponent. It is assumed that the stronger players=20 right so are in this point group whereas for the players at the lower=20 end of the group it is assumed that their score is not justified by=20 their strength. Therefore they get the chance to justify their score (by= =20 a win against a stronger player); or they will be identified by the=20 system as a weaker player (by a loss against a strong player). The power= =20 of the ranking (how well it reflects the real perforemance strength of=20 the players) will be enhanced clearly if the strongest players are=20 paired against the weakest. The stronger players can place themselves=20 above the weak players through a direct win. =20 Pairing players up and down. If a point group (the players with the same number of points before that= =20 round) is uneven, one player has to be paired up or down. If a player=20 has to be paired down, the strongest player according to the secondary=20 criteria (SOS) will be paired down. This is reasonable because the=20 players at the top of that group have had the strongest opponents on=20 average and still reached that score. If a player is paired down, he=20 will get a relatively weak player. It is fair to give that player that=20 had the strongest opponents so far a relatively weak opponent. But by no= =20 means should a player that had previously rather weak opponents get the=20 chance to improve in the ranking with a win over another weak opponent. =20 If this method is used, you do not get so big differences in the SOS of=20 the players. With the SOS-based pairing you do not influence the final=20 ranking (and the pairings) to much by performances that lie long before=20 the actual tournament. =20 It was expected (as Pieter already pointed out) that the interference of= =20 the EC with the WOSC could also be reduced with SOS-based pairing. But=20 of course the knock-out-tournament always interferes with the open=20 tournament in some way. This year we had some other effects having a stronger influence (I=20 think). We had only two strong players in the open tournament from the=20 start. Their pairings in the first two rounds could only be against=20 relatively weak players (Difference of more than 2000 elo-points!). Another problem were the dan players. We had some very strong dan=20 players that lost only among themselves. The few dan players below those= =20 lost against the top, gained many SOS points against them and had to=20 play weaker players afterwards that they could relative easily beat.=20 Nevertheless final the SOS-distribution very good. Only the SOS of Mr.=20 Sera is a bit to low. But it was impossible to increase it more after=20 the first two rounds in the open tournament. =20 This indicates another important aspect for a meaningfull ranking=20 especially for the top. If you have great strength differences (elo=20 differences) and start with all players in one group (as it was in the=20 WOSC), you also get many pairings with great elo differences. These=20 pairings (with more than say 600 elo difference) are very bad for the=20 aim to find a meaninfull (final) ranking of the players. Further more=20 these pairings are not exciting for both players in nearly all cases. FESA decided in this years FESA-meeting that we will change this. But it= =20 was not yet decided what we will do exactly (use separate groups, use=20 MacMahon groups). With the programme we used this year (Gerlachs MacMahon programme) it is= =20 also possible to set other parameters for the lower MacMahon groups (if=20 a MacMahon system is used at all). For the top MacMahon group it is of=20 course important to find a clear winner and a good ranking and to be=20 very strict about the pairing. But for the players in the lower MacMahon= =20 groups the enjoyment is maybe more important. One could avoid parings of= =20 players from the same club (with a predefined reasonable weight) and=20 from the same country (with a lower weight). It would also be possible=20 to minimize (to a certain extend) pairings with too big strength (elo)=20 differences. This was done before in German tournaments with positive=20 feedback by most players. I hope that everyone now understands that the pairing of the WOSC was=20 indeed fair and that it followed some reasonable considerations. With kind regards, Jochum. Your free subscription is supported by today's sponsor: ------------------------------------------------------------------- Save up to 67% on Omaha Steaks + Get 6 FREE Burgers and a=20 FREE Cutlery Set + Cutting Board! http://click.topica.com/= caacBBEa2i6YsbnuqMaf/OmahaSteaks ------------------------------------------------------------------- --^---------------------------------------------------------------- This email was sent to: = shogi-l shogi net EASY UNSUBSCRIBE click here: http://topica.com/u/?a2i6Ys.= bnuqMa.= c2hvZ2kt Or send an email to: shogi-unsubscribe topica com For Topica's complete suite of email marketing solutions visit: http://www.topica.com/?p=3DTEXFOOTER --^---------------------------------------------------------------- --------------040604050503060801090401 Content-Type: text/html; charset=us-ascii Content-Transfer-Encoding: quoted-printable = 
Your free subscription is supported by today's sponsor:
-------------------------------------------------------------------
Amazing Diet Patch
The fastest - Easiest way to lose weight! Try it now FREE!
http://click.topica.com/=
caacBBBa2i6YsbnuqMab/MyDietPatches
-------------------------------------------------------------------

Dear shogi friends,
 
Let me first examine why it is reasonable to level the SOS (sum of opponents score) for all players. I will use the WOSC 2000 and 2002 as examples (for many tournaments the SOS and other secondary criteria are not available on the web).

WOSC 2002
15 Majewski Thomas       &= nbsp;  DE  2D 1633  +39 6 40.5 24 32   26 17- 44+ 13+  9- 43+  6- 33+ 29+ 30+<= /small>
16 Pfaffel Thomas       &n= bsp;   AT  2D 1849  -16 5 50   23 38   31 21+ 11- 37+ 35+ 25+ 26+=   2-  3- 14-

The highest ranked player with five wins (Pfaffel) has 50 SOS points, the lowes ranked player with six wins (Majewski) has 40.5 SOS points. So Pfaffel was playing a stronger tournament than Majewski. Who performed better in the tournament?
 
WOSC 2000
19 Blackstock, Les ....... ENG  1973  48- 26- 71+ 51+ 35+ 28- 44+ 41+ 3= 2+  6  40.5
20 Pfaffel, Thomas ....... AUT  1200  82+ 25- 27- 40+ 55+ 22- 38+ 47+ 3= 0+  6  38<= /b>
21 Uemura, Yoshiyuki ..... FRA  2272  54+ 13+ 32+ = ; 2-  1- 30+ 25+  4-  7-  5&nb= sp; 56
22 Nilsson, Carl-Johan ... SWE  1828  28+  6+ 10+  3- 12- 20+  7- 26- 5= 6+  5  53
23 Segers, Hans .......... NED  1932  41+  3- 39+ 35+  7- 32+ 13- 40- 44+ = 5  49
24 Oosterwijk, Jan ....... NED  1959  44+ 48+ 12- 36+ 43+ 18+  6-  7- 11-  5  48.5
25 Dysart, Doug .......... USA  1889  65+ 20+  1- 46+  6- 38+ 21- 32- 52+ = 5  48
26 Maass, Stefan ......... GER  1510  62+ 19+  6- 30- 33+ 55+ 16- 22+ 17-  5&nb= sp; 47    27
27 Kashiwazaki, Shintaro . GER  1740  49+  7- 20+ 14- 50+ 31- 42+ 37+ 10-  5  47    26
28 Drechsler, Jochen ..... GER  1768  22- 64+ 67+  4- 60+ = 19+  5- 31+ 16-  5  46
29 Litjens, Mathijs ...... NED  1773  30+ 12- 70+  6- 76+<= span style=3D"">  7- 56+ 36+ 14-  5  45
30 Heeffer, Albrecht ..... BEL  1778  29- 75+ 54+ 26+&= nbsp; 9- 21- 39+ 43+ 20-  5  43    24
31 Connors, Don .......... USA  1894  75+ 50+  2- 44- 53+ 27+  8- 28- 48+ = 5  43    22
32 Tran, Philippe ........ FRA  1765  67+ 61+ 21- 18- 54+ 23- 51+ 25+ 1= 9-  5  42.5
33 Lamb, Mike ............ USA  1704  77+  1- 62+ 16- 26- 70+ 45+ 17- 50+  5&nb= sp; 42
34 Kaufman, Ray .......... USA  1886  17- 42- 65+ 45+ 52+ 36- 35+ 46+ 13-  5  41
35 Suwa, Keiko ........... JPN  1620  74+  8- 72+ 23- 19- 50+ 34- 51+ 57+  5  40
36 Ngaolertloi, Wisit .... THA  1680  50- 71+ 66+ 24- 45+ 34+ 14- 29- 4= 3+  5  39    22    19
37 Takemura, Hiroshi ..... JPN  1860  76+ 66+  8- 43- 39+ 13- 52+ 27- 46+  5&nb= sp; 39    22   = 18
38 Christoffersen, Terje . NOR  1705  15- 68+ 45- 42+ 62+ 25- 20- 54+ 5= 9+  5  39    21
39 Olufsen, Asle ......... NOR  1398   4- 63+ 23- 66+ 37- 62+ 30- = 64+ 49+  5  38<= /b>
 
The highest ranked player with six wins (Uemura) has 56 SOS points, the lowes ranked player with five wins (Pfaffel) has 38 SOS points! So Uemura was playing a much stronger tournament than Pfaffel. Who performed better in the tournament? Can you compare the performance of those players at all??
The next player with the same SOS points than Pfaffel is 19 ranks below (Olufsen)!
Is that fair??
 
The pairing programme is not directly minimizing the SOS points.
The pairing is done following predefined rules an parameters. To find the best pairing according to the differently weighed parameters the programme utilizes the maximum-weight-perfect-matching-algorithm (based on Edmonds). It follows three rules. The rules are dominant over the other in this order:
1. No two players will be paired twice.
2. The global number of players paired up or down has to be minimized.<= /span>
3. The choice of players that are paired up or down.
 
Pairing within a point (or MacMahon) group
The players within one group (with the same number of points or macmahon points) are paired like this: the first player (according to his SOS) against the last; the second against the second last. The players that had a strong tournament so far (shown in their high sos) will get a relatively weaker opponent. It is assumed that the stronger players right so are in this point group whereas for the players at the lower end of the group it is assumed that their score is not justified by their strength. Therefore they get the chance to justify their score (by a win against a stronger player); or they will be identified by the system as a weaker player (by a loss against a strong player). The power of the ranking (how well it reflects the real perforemance strength of the players) will be enhanced clearly if the strongest players are paired against the weakest. The stronger players can place themselves above the weak players through a direct win.
 
Pairing players up and down.
If a point group (the players with the same number of points before that round) is uneven, one player has to be paired up or down. If a player has to be paired down, the strongest player according to the secondary criteria (SOS) will be paired down. This is reasonable because the players at the top of that group have had the strongest opponents on average and still reached that score. If a player is paired down, he will get a relatively weak player. It is fair to give that player that had the strongest opponents so far a relatively weak opponent. But by no means should a player that had previously rather weak opponents get the chance to improve in the ranking with a win over another weak opponent.
 
If this method is used, you do not get so big differences in the SOS of the players. With the SOS-based pairing you do not influence the final ranking (and the pairings) to much by performances that lie long before the actual tournament.
 
It was expected (as Pieter already pointed out) that the interference of the EC with the WOSC could also be reduced with SOS-based pairing. But of course the knock-out-tournament always interferes with the open tournament in some way.
This year we had some other effects having a stronger influence (I think). We had only two strong players in the open tournament from the start. Their pairings in the first two rounds could only be against relatively weak players (Difference of more than 2000 elo-points!).
Another problem were the dan players. We had some very strong dan players that lost only among themselves. The few dan players below those lost against the top, gained many SOS points against them and had to play weaker players afterwards that they could relative easily beat. Nevertheless final the SOS-distribution very good. Only the SOS of Mr. Sera is a bit to low. But it was impossible to increase it more after the first two rounds in the open tournament.
 
This indicates another important aspect for a meaningfull ranking especially for the top. If you have great strength differences (elo differences) and start with all players in one group (as it was in the WOSC), you also get many pairings with great elo differences. These pairings (with more than say 600 elo difference) are very bad for the aim to find a meaninfull (final) ranking of the players. Further more these pairings are not exciting for both players in nearly all cases.

FESA decided in this years FESA-meeting that we will change this. But it was not yet decided what we will do exactly (use separate groups, use MacMahon groups).

With the programme we used this year (Gerlachs MacMahon programme) it is also possible to set other parameters for the lower MacMahon groups (if a MacMahon system is used at all). For the top MacMahon group it is of course important to find a clear winner and a good ranking and to be very strict about the pairing. But for the players in the lower MacMahon groups the enjoyment is maybe more important. One could avoid parings of players from the same club (with a predefined reasonable weight) and from the same country (with a lower weight). It would also be possible to minimize (to a certain extend) pairings with too big strength (elo) differences. This was done before in German tournaments with positive feedback by most players.

I hope that everyone now understands that the pairing of the WOSC was indeed fair and that it followed some reasonable considerations.

With kind regards, Jochum.



Your free subscription is supported by today's sponsor:
-------------------------------------------------------------------
Save up to 67% on Omaha Steaks + Get 6 FREE Burgers and a=20
FREE Cutlery Set + Cutting Board!
http://click.topica.com/=
caacBBEa2i6YsbnuqMag/OmahaSteaks
-------------------------------------------------------------------

--^^-------------------------------------------------=
--------------
This email was sent to: =
shogi-l  shogi  net

EASY UNSUBSCRIBE click here: http://topica.com/u/?a2i6Ys.=
bnuqMa.=
c2hvZ2kt
Or send an email to: shogi-unsubscribe  topica  com

For Topica's complete suite of email marketing solutions visit:
http://www.topica.com/?p=
=3DTEXFOOTER
--^^---------------------------------------------------------------
--------------040604050503060801090401--