SE4 Rating System

[geoschmo]

Point inflation is a problem with the "unequal" system I suggested most recently. Point inflation is bad because you are giving ome players more points then you are taking away from other players. This makes it harder on new players to climb the rankings.

The "round-robin" style on the other hand has point multiplication. It's not inflation in the sense that you aren't generating points out of ether, but the massive point swings generated by such a system are an unfair and potentially damaging factor. If you play two games, lose one, and win one, you should basically be back close to where you started. Not exactly of course, becasue it will be factored by the the relative abilities of the players you played. But if you play a 20 player game and lose, you will have to play and win 20 1x1 games to get back close to where you started. That's just not fair IMHO.

Gpa's recent suggestion, call it "Round-Robin/2" has the same problem, only to a lesser degree because of the artifical manipulation of the results.

My original "fractional" suggestion doesn't have any point inflation or point multiplication. Gpa's point that it doesn't allow for enough rank movement isn't really accurate. The reason there isn't much rank movement isn't because of the point system being used, it's because SE4 games take a long time to finish. It's inherantly slow. Much slower then chess of course.

I am not wedded to the "fractional" system at all. I would welcome an alternative suggestion. I just haven't seen one yet that isn't what I wouldn't call seriously flawed.

[Grandpa Kim]

How about this

Rs = standard rating change
Rm = modified rating change for multiplayer
n = number of players

Rm=Rs(1.5 root [n-1]/[n-1])

For three players

1.59/2 or .794 of a one on one game

For 10 players

4.33/9 or .481

For 20 players

7.12/19 or .375

Just to simplify things I will consider that a victory over a single player will give a victor a 16 point jump in the Ratings.

In a three player game the victor would currently get 32 points. With this system he gets 25.

In a ten player, instead of 144, he gets 69.

In a twenty player instead of 304, he gets 114.

I could live with this. I tried it using square root but the final numbers were just too low for my taste.

I don't know if anyone likes this or if Slynky is willing to use it, but it seems like a good compromise to me. Is a hundred point loss to the first man out in a twenty player game too much? I think it's okay. What does everyone else say?

Edit: Note this works for one on one games too since any root of 1 is 1. I don't know how your calculating program is set up but you could possibly just insert this into the formula.

[ March 17, 2004, 04:13: Message edited by: Grandpa Kim ]

[Alneyan]

Not only Primitive will lose points (several, depending on the formula), but I will gain more Slynky. If you compute my score for the 2vs2 game with 985 (after my defeat with Primitive) instead of my current 1000, I will win several points in the bargain.

And obviously, you do not want to compute Primitive's game first, lest he takes over your nice seat at the head of the Ratings. (And with Asmala coming, you ought to be wary Slynky, as this seat might become slippery. *Chuckles*)

[primitive]

Originally posted by Slynky:
Games needing rating computation (in this order):

4 player multi-player (Alneyan, Belisarius, Gecko, Slynky)

Alneyan vs Primitive

Joachim vs Warlord Ragnarok

I can compute the Last game (it is dependent on nothing else) but if I compute the second game without doing the first one, Prmitive gets less points (possibly...and probably only 1 point less).

So, should I wait for what appears to be an agreement on rating multi-player and then do the second game or go ahead and go ahead and compute the second game (which, I suspect, is mostly up to Primitive).

Feel free to calculate, this can take a while

[primitive]

Too get the ball rolling, maybe we could put this to a vote:

A: Linear system: Equal point spread between the positions.

B: Some sort of Logarithmic system: Significant difference between 1st and 2nd, Small difference between Last and second to Last.

C: Something completely different.

I’m all for B

[parabolize]

B

[Baron Grazic]

Once Geo & Kim started talking Mathematics, they lost me.

I'll abstain from the Vote, but am if forced will lean towards B. Just no-one ask me to do any calculations.

[Grandpa Kim]

Yeah, you can always tell when the Baron is playing with his socks off: he has 20 ship fleets.

Primitive, this part doesn't matter to me; it's the truncation I'm interested in. (Look that up in your Funk & Wagnall! )

Edit: And possibly more important. How easy will it be for Slynky to implement?

[ March 18, 2004, 02:55: Message edited by: Grandpa Kim ]

[Slynky]

Games needing rating computation (in this order):

4 player multi-player (Alneyan, Belisarius, Gecko, Slynky)

Alneyan vs Primitive

Joachim vs Warlord Ragnarok

Primitive vs RexTorres

(reading all comments on a group rating system...still looking for something that doesn't "skew" the Ratings too much and doesn't result in 20 minutes of computations...I think, if I understand Geo's Last suggestion, that it would result in a lot of manual computation/adjustment. Let's not give up yet.)

[primitive]

Multiplayer Ratings

I've been messing with a spreadsheet and some formulas and are half way to a system with a table. The value from this table should be multiplied with the score from the standard calculation, using loosing players ranking vs the average of the remaining players rankings (or if Slynky is lazy, the average of all players).

There is a lot of factors to play with, but I would like some input before I spend more time (And yes, I have noticed there are less points lost for coming Last in the 20 player game than in the 10 player game )

The table so far gives these values for:
3 players: 1,51 / -0,08 / -1,44
4 players: 2,04 / 0,29 / -0,74 / -1,59
10 players: 3,73 / 1,58 / 0,71 / 0,17 / -0,25 / -0,60 / -0,91 / -1,20 / -1,48 / 1,75
20 players: 5,26 / 2,56 / 1,59 / 1,05 / 0,68 / 0,40 / 0,18 / -0,02 / -0,20 / -0,36 / -0,52 / -0,66 / -0,80 / -0,93 / -1,06 / -1,19 / -1,31 / -1,44 / -1,56 / -1,68

Too high ? Too low ? Stupid idea to start with ?

And if anybody has the time for a crash course in displaying tables, it would be highly apreciated