Proposal for a New Ranking System for Insei League

[haha]

If we consider than teachers are better than top players, whatever the system, only teachers will compete for the prices, so we need a handicap :

• handicap in game => teachers play against top players in simul or in 1 to 1 but with H2 handicap. I feel teachers will argue than players must have hard games to improve and top players will want to play teachers in 1 to 1, challenge more interesting.
• handicap in score => in the topazg system (let's call it "June system" ), teachers can begin with an initial score of -20 or teachers can get less points per win/lose. Here again, simulation and/or experience can adjust the parameter

still the problem come from teachers able to compete for prices. I understand it's a motivation for top players but could teachers be only motivate to have an attractive league with lot of regular members ? this is somewhat their prices.

Feedback of top players and teachers are welcome !

[topazg]

A couple of ideas are:

a) initial negative score as you say. 20 feels like it might not be so big come the end though.
b) smaller bonus, such as 0.3 * SODOS rather than 0.5

For those that want to play with figures, I've gone and put it into an Excel spreadsheet: http://www.topazg.com/insei.xls

All the figures for each division are up to date as of today, and all the key parameters are adjustable at the bottom of each table. The teachers have their own parameters there too, including an initial handicap value.

One further thing of note: All the other divisions record wins against the teachers, but exclude losses. Firstly, this may be unnecessary now as losing is likely to give rewards and won't punish their percentages (ok, not as big a reward as a win, but bigger than not playing at all). If the teachers want to carry on excluding these losses from the records, which is my initial assumption, I've adjusted the spreadsheet worksheets from Div B through to Div E to ignore losses. This means that the reward for simply playing a game in my system (only 1 point per game admittedly) isn't there for the lower division games against the teachers.

And for those interested, current standings in A (with -20 and 0.3 modifier for teachers) are now:

1. danigabi - 138 (13-1)
2. Kalmah - 96.5 (8-5)
3. ha - 69.5 (5-1)
4. Nata - 63 (5-2)
5. Syptryn - 59.5 (5-10)
6. breakfast - 43.7 (8-1)
7. RamenBoya - 29.5 (2-2)
8. OohAah - 27.5 (3-0)
9. fantastigo - 26.5 (1-9)
10. Arlequ1 - 24 (1-15)
11. DRhazar - 20.5 (1-3)
12. Teamrocket - 17 (1-5)
13. roln111 - 11.8 (4-0)
14. YraUkr - 11 (1-4)

[haha]

i think the weight of bonus is too big.
0.2 bonus modifier both for teachers and students seems right.
with this smaller bonus, the initial handicap of -20 for teachers become ok too.

[gaius]

A scoring system using the actual score + bonuses based on a few formulas sounds rather difficult if you ask me... Might I interject with a simple(r) system:

* The players are ranked by their SODOS score.
* In the case of a player playing multiple games against the same opponent, the score against that opponent is multiplied by Sqrt(1/n) where n is the number of games they played against each other.

Let's give an example. Suppose I play three games against somebody with 10 wins; out of those games, I win two and lose one. This would give me Sqrt(1/3)*(10+10+0) = 11.55 points. That's more than if I just play 1-0 against this opponent. It also gives me a much better result than if I would only play lower ranked players - so much better, in fact, that it might become almost impossible to win a division without some wins against strong players. Which is exactly what you want.

With this system:

• Almost all games either increase your score or leave it the same. OK, if you're 1-0 against someone, then losing will cost you a few points, but winning the game will earn you more than losing will cost you so there's still no reason not to play.
• Playing more games against the same opponent is progressively rewarded less, and playing against a new opponent never costs you points.
• It is vital for strong players to play other strong players.
• Most importantly, it is easy to understand and accurate both at the top and at the bottom end of the rankings!

EDIT: I just noticed this is kind of a necropost... hopefully it's still useful to someone anyway...

[topazg]

A scoring system using the actual score + bonuses based on a few formulas sounds rather difficult if you ask me... Might I interject with a simple(r) system:

* The players are ranked by their SODOS score.
* In the case of a player playing multiple games against the same opponent, the score against that opponent is multiplied by Sqrt(1/n) where n is the number of games they played against each other.

Let's give an example. Suppose I play three games against somebody with 10 wins; out of those games, I win two and lose one. This would give me Sqrt(1/3)*(10+10+0) = 11.55 points. That's more than if I just play 1-0 against this opponent. It also gives me a much better result than if I would only play lower ranked players - so much better, in fact, that it might become almost impossible to win a division without some wins against strong players. Which is exactly what you want.

With this system:

• Almost all games either increase your score or leave it the same. OK, if you're 1-0 against someone, then losing will cost you a few points, but winning the game will earn you more than losing will cost you so there's still no reason not to play.
• Playing more games against the same opponent is progressively rewarded less, and playing against a new opponent never costs you points.
• It is vital for strong players to play other strong players.
• Most importantly, it is easy to understand and accurate both at the top and at the bottom end of the rankings!

EDIT: I just noticed this is kind of a necropost... hopefully it's still useful to someone anyway...

It's not yet, I know Alex is still deciding what to do on this issue.

Can you simulate your system on Div A - it looks potentially interesting, and I'd like to see how it compares...

[topazg]

And for what it is worth, latest standings update:

1. Kalmah - 246.5 (15-13)
2. danigabi - 240 (15-3)
3. breakfast - 170 (15-1)
4. ha - 153 (7-7)
5. Syptryn - 142.5 (7-12)
6. Nata - 132.5 (7-4)
7. Teamrocket - 89.5 (4-6)
8. fantastigo - 81 (3-11)
9. Arlequ1 - 81 (3-23)
10. roln111 - 78.4 (8-0)
11. OohAah - 70.5 (5-0)
12. YraUkr - 58.5 (3-5)
13. RamenBoya - 45.5 (2-7)
14. DRhazar - 35 (1-3)

i think the weight of bonus is too big.
0.2 bonus modifier both for teachers and students seems right.
with this smaller bonus, the initial handicap of -20 for teachers become ok too.

With this modification, the standings are:

1. Kalmah - 133.4 (15-13)
2. danigabi - 124.8 (15-3)
3. breakfast - 122 (15-1)
4. ha - 78 (7-7)
5. Syptryn - 76.8 (7-12)
6. Nata - 68 (7-4)
7. roln111 - 53.6 (8-0)
8. Arlequ1 - 51.6 (3-23)
9. Teamrocket - 46.6 (4-6)
10. fantastigo - 44.4 (3-11)
11. OohAah - 37.2 (5-0)
12. YraUkr - 31.8 (3-5)
13. RamenBoya - 26 (2-7)
14. DRhazar - 17.6 (1-3)

I agree, this seems a bit more balanced, and fairer also on the teachers I think.

[gaius]

I calculated the system I described in my previous post through for division A. The result is:

Code: Select all

Name       Wins      Losses    Points total (SODOS system)
ha         5         1         29.07
danigabi   13        1         25.95
breakfast  8         1         22.83
Nata       5         2         14.07
Kalmah     8         5         9.66
Syptryn    5         10        8.45
DRhazar    1         3         8.00
fantastigo 1         9         4.62
RamenBoya  2         2         4.54
OohAah     3         0         3.00
roln111    4         0         2.73
YraUkr     1         4         1.00
Teamrocket 1         5         1.00
Arlequ1    1         15        1.00

Somewhat counter-intuitively, ha, with 5 wins and 1 loss, ends above danigabi with 13 wins. The reason for this is that ha played 1-0 against danigabi. Because danigabi had 13 wins, this earned ha a disproportionate amount of 13 points.

Noticing this, I set so that instead of adding
N(opponent wins) * Sqrt( 1 / N(games against this opponent) )
points for each win, it now adds:
Sqrt( N(opponent wins) + 1 ) * Sqrt( 1 / N(games against this opponent) )

This way, even beating someone with 0 wins will give you a point; also, beating someone with loads of wins gives you a somewhat more balanced number of points.

The result looks excellent:

Code: Select all

Name      Wins      Losses    Points total (new SODOS system)
danigabi   13        1         17.23
breakfast  8         1         13.63
ha         5         1         11.62
Nata       5         2         8.74
Kalmah     8         5         8.33
Syptryn    5         10        6.37
OohAah     3         0         4.24
roln111    4         0         3.86
RamenBoya  2         2         3.15
DRhazar    1         3         3.00
fantastigo 1         9         1.73
YraUkr     1         4         1.41
Teamrocket 1         5         1.41
Arlequ1    1         15        1.41

[topazg]

Yeah, that looks really good - I think it would benefit from a minor handicap to the teachers, but that also looks like a good contender.

I think intuitively I would find it easier to see how many points I'm likely to gain from my system with regards to total points because my mental arithmetic struggles with square roots somewhat, but yours certainly gives a nice feel to the results.

[gaius]

I think intuitively I would find it easier to see how many points I'm likely to gain from my system with regards to total points because my mental arithmetic struggles with square roots somewhat, but yours certainly gives a nice feel to the results.

All you have to do is put a "price" on everyone's head, equal to Sqrt( N(opponent wins) + 1 ) and put that price on the website. The players don't even need to know the exact formula, but they know that a win against this opponent will give them that many points.

[haha]

I didn't read carefully all the details of SODOS system but the last simulation is a bit strange.
3 players with the same score (arlequ1,teamrocket, YraUkr), for a same win (against fantastigo ?) but different number of loses games...
If i remember well the initial idea (3 criterias : merit, number of wins, number of games), with additional games, your score must increase even with loses no ?
so Arlequ1 must be above Teamrocket and Yraukr in this example ?

[topazg]

Final (31st May) standings for Div A of the insei league in my proposed system with bonus modifier down to 0.2 as suggested by haha:

1. Kalmah - 150.8 (16-14)
2. breakfast - 135 (16-1)
3. danigabi - 131.4 (15-3)
4. roln111 - 111 (14-0)
5. ha - 91.2 (8-8)
6. Syptryn - 82.4 (7-13)
7. Nata - 74.8 (7-7)
8. Teamrocket - 66.8 (6-8)
9. Arlequ1 - 54.6 (3-24)
10. fantastigo - 53.8 (4-11)
11. OohAah - 40.4 (5-2)
12. YraUkr - 35.2 (3-6)
13. RamenBoya - 26.4 (2-7)
14. DRHazar - 18.4 (1-3)

[haha]

i'm affraid that with the current parameters, teachers easily finish first and second. Also, player will not be affraid anymore to play them so they will have a lot of games (perhaps even more that every player). So initial handicap must increase.
if we imagine top player playing his eight games against teachers and lose them all. With 30 games, let's say he will have 22/30. teacher will have 28/30, here again simulation is needed to put correct handicap (one who put top player above of teachers).

Meanwhile, let's see which rules breakfast will chose for June.

[Targund]

I want to offer next rating system:

- Each player has "personal rating";
- Each player has "league rating"
- Personal rating changes every game and reflects that players strenght (such as EGR);
- League rating resets to 0 every month and reflects current month league progress (instead of w/l ratio);
- Personal rating is calculated in Elo system;
- League rating is calculated in Elo system, based on Personal rating difference.

[usagi]

1. The current ranking system discourages players with higher win % to play games.
2. The current ranking system encourages people to hunt down weaker members of the group and avoid games vs strong members.

These two problems cause the phenomena where many players do not play games for fear of ruining their ranking. This is quite bad, given that in Group C, for example, this month, half the people didn't player enough games.

Hello, you raise some interesting points, and as I have experience as the founder of the ASR league system on KGS I will weight in. Disclaimer, I have not read the thread, only the OP.

First, in practice, the current insei league system is better than the old ASR league system when I was running it, because players are more free to play games with the same person, or other people. This leads to the weakness you pointed out that "players A, B and C will seek out only weaker players...". However (and I am sure someone else has pointed this out by now) this isn't a weakness at all - if those players are strong enough to go 12-0 against the other group members, then they deserve to promote or to remain at the top spot in the room. It's really that simple. And it will fulfill the purpose of the league, which is to facilitate learning by facilitating games between stronger players and weaker players. So in this instance, the weakness you pointed out is actually a major strength of breakfast's system.

Second point, the first weakness you pointed out. Again in practice this is fine. If a player is 12-0, why would they want to play more games anyways? They should be promoted and play in the next class (or at least take two teaching games). And if they won their teaching games, well, maybe they should quit the league? Yet, again, the existance of the theoretical (or in practice) 12-0 player is only a boon to the goal of getting stronger players to play weaker players for their benefit. This is why the 12 game rule is there, so people play a minimum number of games. That's fine, as long as they play the minimum number of games.

In general I understand the concern of your comments but as an experienced league-runner myself, I would point out that even if players are not seeded properly, in an ABCDE system they are guranteed to find their proper class within four periods. And as I'm aware players are seeded by their rating on KGS (or dgs). So after the first round they will get a good idea of their position in the league. The mythical 12-0 player could only be the top 1-2 spots in A league.. and again that can only be a benefit to the league system.

It's my opinionated view that a player who avoids playing certain people or tries to game the system by stopping at 12-0 will, in actuality, do nothing. Such a player likely misunderstands how insei-style leagues work on a fundamental level -- as in practice you will find yourself very quickly placed into a class where you belong, and the people who do not play find themselves dropping down as the people around them simply become stronger and are promoted.

Also something else just occurred to me. I'm under the impression that breakfast or the other moderators can request that people play, which would even out the system somewhat. Done strategically this is an excellent rule.

A final disclaimer is that I am not in breakfast's league although I have applied for it starting in september. I have very high hopes for it, from what I have read it is everything I had ever hoped the ASR league to be, and yet somehow so much more.

[usagi]

To continue my thoughts from earlier;

There is actually one other solution you might consider.

Rate players using SODOS -- i.e. sum of defeated opponent's scores.

Here's a sample score list:

Code: Select all

 +---+--------------+---------+---------+---------+---------+---------+---------+---------+---------+
 | # | Name         |    1    |    2    |    3    |    4    |    5    |    6    |    7    |    8    |
 +---+--------------+---------+---------+---------+---------+---------+---------+---------+---------+
 | 1 | Arthur 7d    | x-x-x-x |  - - -  |  - - -  |  - - -  |  - - -  | 1-1-1-1 | 1-1-1-1 | 1-1-1-1 |
 | 2 | Bethseda 6d  |  - - -  | x-x-x-x |  - - -  |  - - -  |  - - -  | 1-1-1-1 | 1-1-1-1 | 1-1-1-1 |
 | 3 | Cervantes 5d |  - - -  |  - - -  | x-x-x-x | 1-0-1-0 | 1-0-1-0 | 1-1-1-0 |  - - -  |  - - -  |
 | 4 | Dimitri 4d   |  - - -  |  - - -  | 0-1-0-1 | x-x-x-x | 1-1-1-0 |  - - -  | 1-1-1-0 |  - - -  |
 | 5 | Egon 3d      |  - - -  |  - - -  | 0-1-0-1 | 0-0-0-1 | x-x-x-x |  - - -  |  - - -  | 1-1-0-0 |
 | 6 | Fauntleroy 2d| 0-0-0-0 | 0-0-0-0 | 0-0-0-1 |  - - -  |  - - -  | x-x-x-x |  - - -  |  - - -  |
 | 7 | Gargamel 1d  | 0-0-0-0 | 0-0-0-0 |  - - -  | 0-0-0-1 |  - - -  |  - - -  | x-x-x-x |  - - -  |
 | 8 | Horace 1k    | 0-0-0-0 | 0-0-0-0 |  - - -  |  - - -  | 0-0-1-1 |  - - -  |  - - -  | x-x-x-x |
 +---+--------------+---------+---------+---------+---------+---------+---------+---------+---------+

Now, this is a hypothetical example whereby the strong have preyed on the weak.
If you organize the winners by score, (or winning percentage) it looks like this:

1. Arthur (12 wins)
2. Bethseda (12 wins)
3. Dimitri (8 wins)
4. Cervantes (7 wins)
5. Egon (5 wins)
6. Fauntleroy (1 win)
7. Gargamel (1 win)
8. Horace (1 win)

In the above example, Arthur and Bethesesa were clearly colluding to play only against the weakest players in the group. It works out well if the system uses winning percentages.

However what if instead of simply rating people by their scores, we rated them by the sum of their defeated opponent's scores? For example, Fauntleroy pulled off a lucky win against Cervantes (or maybe it was his skill). Cervantes has 7 wins, so fauntleroy's score is seven. Similarly, Gargamel beat Dimitri, so Gargamel now has a score of 8. And Horace beat Egon twice, so Horace's score is 10, since Egon won five times.

It is a little tricky but I have a spreadsheet that calculates this this automatically and spits out a sortable SODOS (Sum of Defeated Opponent's Scores) table.

If the above league was calculated using SODOS, Arthur and Bethseda would find themselves with scores of 4 each -- in this case, their wins against Fauntleroy, Gargamel and Horace being worth 1+1+2, or only four points. In this case their wins against weaker players were not worth as much to them as a win against a stronger opponent. Their strategy of preying on the weak failed. It would have been better for them to play each other, and a variety of other opponents as well!

In fact, the SODOS system strongly encourages players to play at least one game against every other member of their room. In an open system where one may play multiple games against other opponents (to a maximum of four) it also strongly encourages participation among active players!

It also does not punish losses. If you lose a game, your score doesn't go down. This removes the fear people have of playing games in the league.

Let's peek ahead and see what the league looks like using SODOS:

Promoted:
1. Dimitri 4d (32 pts)
2. Cervantes 5d (29 pts)

Remain in room:
3. Egon 3d (26 pts)
4. Horace 1k (10 pts)
5. Gargamel 1d (8 pts)
6. Fauntleroy 2d (7 pts)

Demoted:
7. Arthur 7d (4 pts)
8. Bethseda 6d (4 pts)

Who were the winners? The people who played a variety of players; some stronger, some weaker. Notably, Horace's two wins against a stronger opponent, and Gargamel 1d's lucky--or skillful--win earn them a right to remain in this interesting and challenging room.

Who gets demoted? The 6d and 7d who tried to game the system.
Who gets promoted? The strongest players who play a variety of opponents.

This system rewards players for demonstrating their skill. You can't do that if you only play weaker opponents. Note also that it does not discourage stronger players from playing weaker players, since weaker players may rack up a lot of wins by playing each other!

Perhaps SODOS is right for the Insei League? Breakfast! If you like I can E-Mail you the SODOS calculator I wrote for the ASR. I will also be using a version of this to run a 100 person league (eight rooms of 12 people) at a local high school. This method works well, that is my experience! We already have 82 sign ups for the highschool league! Well, that's another story. But yeah if you want a SODOS spreadsheet let me know It takes some learning but it helps doing the calculations.