Proposal for a New Ranking System for Insei League

[topazg]

Out of interest usagi, how much have you read the content of the rest of this thread?

I suspect everyone is familiar with SODOS involved in the league, and yeah, it has a lot of merits - have you read why the other proposals were put forward and what was being sought out for an "ideal" system?

[usagi]

Out of interest usagi, how much have you read the content of the rest of this thread?

I suspect everyone is familiar with SODOS involved in the league, and yeah, it has a lot of merits - have you read why the other proposals were put forward and what was being sought out for an "ideal" system?

Well, my idea is that above everything else, you have to ensure that everyone plays the same number of games. Secondly I thought SODOS would be better than winning, because it can remove the ability of stronger players to prey on weaker players. Those 2 ideas seemed to solve all of the problems people were bringing up about the current system which is why I mentioned them.

The other problems could be removed if teaching games are simply not a part of the league. You can comment them and review them, but I don't think it is really proper to include their results in the legaue. They're teaching games. Not really competitive games. I've put my thoughts forward on that with a solution (less prize money and higher fees for teachers) in another thread.

Anyways here's my take on what was said so far re: sodos

It was said that SODOS doesn't punish losing to weaker players. An example was given that someone might win a game against a stronger player but lose against three weaker players. The problem with this idea is that you are assuming the results of the league before the league is finished. If someone wins five games and doesn't lose any, and another person wins 13 and loses only one, you have absolutely no way to compare those two results honestly. That's because the second person may well have been 7-0 or 8-0 before losing a game. Wasn't their result better? So, that is one reason why you need same # of games.. but more than that you cannot presume that anyone in the league is weak or strong until that period of the league is finished. Just because someone is 8d or 4d doesn't mean that you can assume they will or will not win a league either. When the league is finished, then and only then can you know the final result. You cannot decide the results beforehand and then use that to criticize the results of the league. If said player is truly weak, they will demote next period and not re-promote. So in that sense, I do not really understand that criticism being made about SODOS.

A somewhat related criticism is that SODOS doesn't punish people for losing. I am not sure whoever said that really understands how SODOS works. Let's say you have two players, both are 6-5. Assume both of their scores are even.
Both of them play a game; one plays someone with eight wins, the other plays someone with nine wins. The one who plays against the person with more wins has a higher score in the end. But assuming everyone plays the same number of games, against the same people, it doesn't matter if they play these games first or last; the results aren't known until the end. That's because everyone who won against those players will now increase in score as well; which pushes the two players who just lost their games down. That is why SODOS does, in fact, punish people for losing. Although it does it in a way which is somewhat difficult to see at first.

Here is my experience about running leagues. First, the only way you are ever going to be able to run a perfectly fair and proper league is if everyone plays the same number of games. You simply cannot allow one player to play eight games and another, twelve. Or twelve and sixteen. If you do that, it is unfair on one level or another. Look at the results above. ha, with six games, had a score above dangnabi, with fourteen games. What you don't see is that no system can possibly fairly score those results. Ha needs more games, plain and simple. If ha plays more games, he will certainly win against more people, and his score will improve. So he lost a bunch? Are we saying that the result of his win against dangnabi is invalid?

The problem is that there is no way to filter out such results because it is a subjective perception. Ha needs to play a total number of games like dagnabi. Fourteen and no more. Look at roln111's results. 4-0. Just not enough games compared to others. So you have to understand, no matter what you ever do to the math, if half the people play five games and half play fifteen, you will never have an accurately scored league. That's my experience running two large (and I daresay extremely successful) leagues.

My idea isn't the only or best way to run a league. It's just what I like best.

Unfortunately you cannot solve all the problems. I think SODOS + same # of games solves far more problems than it creates.

[topazg]

Well, my idea is that above everything else, you have to ensure that everyone plays the same number of games. Secondly I thought SODOS would be better than winning, because it can remove the ability of stronger players to prey on weaker players. Those 2 ideas seemed to solve all of the problems people were bringing up about the current system which is why I mentioned them.

I don't think this is viable - nothing can make the players keep to the same games as each other, and availability will always make this unachievable in practice.

The other problems could be removed if teaching games are simply not a part of the league. You can comment them and review them, but I don't think it is really proper to include their results in the legaue. They're teaching games. Not really competitive games. I've put my thoughts forward on that with a solution (less prize money and higher fees for teachers) in another thread.

In the top league, they are competitive games, in the others, they are already handled (points awarded for the win, result ignored for a loss)

Here is my experience about running leagues. First, the only way you are ever going to be able to run a perfectly fair and proper league is if everyone plays the same number of games. You simply cannot allow one player to play eight games and another, twelve. Or twelve and sixteen. If you do that, it is unfair on one level or another. Look at the results above. ha, with six games, had a score above dangnabi, with fourteen games. What you don't see is that no system can possibly fairly score those results. Ha needs more games, plain and simple. If ha plays more games, he will certainly win against more people, and his score will improve. So he lost a bunch? Are we saying that the result of his win against dangnabi is invalid?

The problem is that there is no way to filter out such results because it is a subjective perception. Ha needs to play a total number of games like dagnabi. Fourteen and no more. Look at roln111's results. 4-0. Just not enough games compared to others. So you have to understand, no matter what you ever do to the math, if half the people play five games and half play fifteen, you will never have an accurately scored league. That's my experience running two large (and I daresay extremely successful) leagues.

You can only work with what you have. You need to accept that in the insei league people aren't going to play the same number of games, and handle a system that can rank as fairly as possible anyway. The incentive to allow people to play as many games as they can has overriden a fairly ranked league (and, IMHO, I think this is probably the right decision, as the more games that are played the more improvement can be achieved).

All of your proposals are assuming that same # of games will happen - I don't think anyone is wanting this for the league at the moment ..

[usagi]

You can only work with what you have. You need to accept that in the insei league people aren't going to play the same number of games, and handle a system that can rank as fairly as possible anyway. The incentive to allow people to play as many games as they can has overriden a fairly ranked league (and, IMHO, I think this is probably the right decision, as the more games that are played the more improvement can be achieved).

All of your proposals are assuming that same # of games will happen - I don't think anyone is wanting this for the league at the moment ..

Unfortunately you are right, that's the same problem the ASR had. The ASR faced the same problems and used a 2 points win, 1 point loss strategy. It seems to have worked well, active players would sit in the room and get their games done often within the first few days or a week. I saw many other players sit in the room and not play games they could have played. So go figure, incentive/disincentive - people are paying $100 a month here. If that isn't incentive enough I don't know what is. I think same # of games is ideal but maybe it only works out in real life when everyone has a regular meeting time.

I think that all things considered, maybe Breakfast's current system is fine? If it was using SODOS that might be better, since players with only 4 or 5 games couldn't promote anyways. If players can't even get 12 games in the current system I don't think they belong in the league. Secondly, there is some incentive I don't think anyone has mentioned. Someone with a low number of wins has an incentive to play someone else with a low number of wins so one of them can promote. This can threaten people mid way up who may then need to play another game in order to keep their spot. This leads only to the problem of the supergroup. Unfortunately if you then mandate that they play other people, there may be scheduling conflicts there. But what else can you do?

Well whatever solution you come up with, I am sure it will be great I just really hope I can play in the league in September. Personally I don't really care what system they use, I aim to play a wide variety of people and win the majority of my games.

[usagi]

I had a few more ideas I would like to toss out there.

First, I went back and modified the SODOS calculator I wrote for the ASR to work with the insei league. here are the results for june:

Code: Select all

June 2010 w/SODOS
1   supertjc   71 (14-3)
2   flashback   61 (9-4)
3   roln111    52 (13-2)
4   breakfast   46 (11-3)
5   Kalmah     35 (13-8)
6   Nata       34 (7-4)
7   Syptryn    31 (5-21)
8   remake     29 (3-16)
9   ha         20 (3-3)
10   zazie      20 (3-2)
11   minismurf    9 (4-5)
12   artem92     8 (1-1)
13   Teamrocket    2 (1-4)
14   Snowbars    0 (0-11)

What I found most interesting about using plain old SODOS is that if you won games against a small number of strong players (like flashback) you could still do well, and you could also do well by beating a lot of weaker players (like kalmah) or you could still lose a lot but beat some strong players and stay in. But, you couldn't really prey on weaker players. Also, if you also lost a lot of games (again, like syptryn) this ended up pushing you down, as your score got added to the players above you. You can see this in the difference between roln111 and kalmah, not only did roln111 beat a stronger set of opponents, he lost fewer games, pushing down the people he beat versus everyone else (by denying them points).

But then I had an even better idea. What if we invented a system where we used SODOS, but also winning percentage, and yet, we didn't have the problem about incentives or disincentives based on # of games played? Wouldn't that be perfect? for example, SODOS-DOG -- SODOS, but distributed over the number of games you played. This way, playing stronger players (in general) is desirable, but not necessarily playing someone stronger than yourself. Here's the June results using SODOS-DOG (SODOS divided by games played):

Code: Select all

June 2010 w/SODOS-DOG
1   flashback   4.70
2   supertjc   4.18
3   zazie*      4.00
4   artem92*    4.00
5   roln111    3.47
6   ha*         3.33
7   breakfast   3.29
8   Nata*       3.09
9   Kalmah     1.67
10   remake      1.53
11   Syptryn    1.19
12   minismurf*   1.00
13   Teamrocket*   0.40
14   Snowbars   0.00

(and below, the above table subsorted by players* who played less than 12 games)

1   flashback   4.70
2   supertjc   4.18
3   roln111    3.47
4   breakfast   3.29
5   Kalmah     1.67
6   remake      1.53
7   Syptryn    1.19
8   zazie*      4.00
9   artem92*    4.00
10   ha*         3.33
11   Nata*       3.09
12   minismurf*   1.00
13   Teamrocket*   0.40
14   Snowbars*   0.00

A question here is what to do with players like snowbars or artem92. What would happen if snowbars lost a 12th game? Should we count artem's results at all? The problem here is what is the rule regarding minimum number of games. We have to agree on a rule and stick to it. We can't allow someone to have 3-3 and another 1-1 and 3-2 and say this is a league, when other people are playing 12, 15 and 20 games. So, if we stick to a minimum game rule, we are faced with a situation like if Snowbars had won a 12th game he could stay in A-room. But this is not a fault of SODOS or SODOS-DOG. But, in this respect I don't like SODOS-DOG as much as SODOS. There are other problems.

Finally, I had a third idea. Something unique. Please comment on this one! I think I've really done it this time!

You see, I played World of Warcraft for a number of years, and I was in a top raiding guild on a top server. The idea I had is based on a DKP system. To explain quickly, a DKP system is a system where by the players who are present for the event acquire points they can use to spend on valuable in-game items. The idea I had is related to how items are spent. When a player spends an item, they pay the item's cost (going negative if they have to) but the point value of the item is then re-distributed among all the other players who attended the event. In this way players who contribute receive a portion of the reward based on their contribution.

So the way I thought of applying this was to award each player who did not play at least twelve games a loss for his missed game. In this way they "spend" the right not to play a game. Then, for every loss awarded this way, award a fractional win to every /other/ player in the system. This can easily be done by adding to the players wins the number of awarded losses in total (minus their number of awarded losses) divided by 13 (for a 14 person league). Then you calculate the "new winning percentage".

um, like this:

New Losses = MAX (0, 12-games played)
New Wins = Old Wins + ((sum of awarded losses) - (player's own awarded losses)) / 13
New Winning % = (New Wins / New Losses) + 1% for each played game

here are May 2010's results in this system.

Code: Select all

x. roln111   14   0   114.00%
x. breakfast   16   1   111.51%
1. danigabi   15   3   102.39%
2. Kalmah   16   14   85.15%
3. OohAhh   7   2   71.16%
4. ha      8   8   69.53%
5. Nata     7   7   67.99%
6. Syptryn   7   13   58.72%
7. Teamrocket   5   8   56.72%
8. fantastigo   4   11   47.16%
9. Arlequ1   3   24   41.94%
10. YraUkr   3   6   40.89%
11. RamenBoya   2   7   33.32%
12. DRHazar   1   3   20.76%

And here are June's:

Code: Select all

x. roln111   13   2   114.00%
1. supertjc   14   3   111.51%
x. breakfast   11   3   102.39%
2. flashback   9   4   85.15%
3. Kalmah   13   8   71.16%
4. Nata    7   4   69.53%
5. minismurf   4   5   67.99%
6. Syptryn   5   21   58.72%
7. teamrocket   3   4   56.72%
8. remake   3   16   47.16%
9. ha      3   3   41.94%
10. Zazie   3   2   40.89%
11. snowbars   0   11   33.32%
12. artem92   1   1   20.76%

I have, of course, numbered the players by promotion (1-4 promote, 9-12 demote).

Now, if you wanted to apply SODOS to this system, here is my suggestion. When you award a fractional win, the "defeated opponent" is the average SODOS of the room, not including the SODOS of the player for which you are calculating the value of the fractional win.
I.E. Player's SODOS, plus: Fractional Wins * (sum of all SODOS - player's SODOS) / players / *(total of all games played - player's games played)

It's easier if you do it with a spreadsheet.

Here are the results I came up with using my new SODOS-DKP calculator:

Code: Select all

June 2010 with SODOS-"DKP" system:
1   supertjc (76.37)
3   flashback (66.52)
14   roln111 (57.67)
13   breakfast (51.76)
2   Kalmah (40.93)
4   Nata (39.78)
9   Syptryn (36.99)
10   remake (35.02)
5   zazie (24.93)
6   ha (25.10)
8   minismurf (14.79)
7   artem92 (12.53)
11      Teamrocket (7.15)
12   Snowbars (6.28)

In practice, this system is the same as SODOS or SODOS-DOG, but if players A and B both win and lose the same number of games, but one of them lose one more game, it will properly differentiate between them. And if someone wins 2 and loses 5, and another wins 2 and loses 10, the one who lost 10 will tend to be placed below the one who lost 5. Assuming they both played the same people, the one who lost 10 will be rated lower. So I think there are a lot of advantages to this one as well, but the math is very complex.

Maybe because of that the version using winning percentages is better, but this final system seems to me to solve all of the problems mentioned so far in this thread.