SE4 Rating System

[Suicide Junkie]

That would only be true if you were going 1/n²

As soon as you hit five players, you get 1 + 1/2 + 1/3 + 1/4.
That's a bit more than 1 + 1/2 + 1/4 + 1/4 = 2

Still OK.
What about the 2nd place player?
1/4 + 1/3 +1/2 - 1 = 0.083
barely any gain at all.

3rd place gets 1/4 + 1/3 - 1 = -0.417
4th gets -0.75
5th gets -1

[ September 03, 2003, 02:39: Message edited by: Suicide Junkie ]

[Baron Grazic]

I like Geoschmo's reasoning on the points scored for the large games, but any way you deciede to do the scoring Slynky, I'm behind you.

I'll post any "Pairs" empire deaths either here (or in the Pairs forum)

I also take it that the Ratings taking in the valuation are the time you calculate the Ratings, not when our games starts, Yes?

[geoschmo]

Originally posted by Suicide Junkie:
Still OK.
What about the 2nd place player?
1/4 + 1/3 +1/2 - 1 = 0.083
barely any gain at all.

Not exactly. That is just the multiplier. You still have to factor in the difference in Ratings going into the match. The total points gained or lost for a 1v1 game can be anywhere from 1 to 32 depending on Slynky's formula. It's impossible to calculate the exact points the 2nd player would get without knowing the Ratings of the players going in. But you are probably correct that it wouldn't be very many. Maybe somehwere in the neighborhood of 0 to 4 points. But that's not bad considering you didn't win. And you won't be losing any points even though you lost at the end.

Originally posted by Baron Grazic:
... but any way you deciede to do the scoring Slynky, I'm behind you.

I second this. However you decide to do it I will support it. I am just making suggestions.

[geoschmo]

Originally posted by geoschmo:
But you are probably correct that it wouldn't be very many. Maybe somehwere in the neighborhood of 0 to 4 points.

Actually it could be much mroe then this. It all depends on the Ratings really. And the 2nd place player could possibly end up negative.

If all four players had the same rating going in it would be pretty easy to calculate.

1st place would get 29 points
2nd place would get -3 points
3rd place would get -11 points
4th place would get -16 points

Change any of the Ratings going in though and things vary wildly. I suppose you could set the range. The most you could possibly get would 116. To do this you'd have to be in a 20 player game and be ranked way below all the other players going into the game.

[ September 03, 2003, 11:44: Message edited by: geoschmo ]

[Slynky]

It's a difficult decision. I certainly see the merit of the discussion on rating multi-player games. Here are some thoughts on this "problem":

(1) As I posted much earlier in this thread (I think), one must be aware of multi-player games and the impact the outcome could have on one's rating. Given a game with 6 players competing for a rating, surely one person will suffer a big (depending on the Ratings of the people in the game) loss in points. Conversely, there are bunches of points to be had...assuming a person can manipulate themselves into 1st or 2nd place. Given 6 players, 2 will suffer some point loss, 2 will turn out about the same and the winning 2 will realize a point gain. I don't think that's too much. By "too much", I'm thinking 4 out of the 6 will NOT have much of a detriment to their rating. But remember, as I said, think long and hard about multi-player games and the chances you take when joining one and having it count as a rating.

(2) As far as making this rating system compare to a tried and true rating system (used for MANY years the world over), it's not far from what is done. Certainly, without much consideration, a one-on-one game comes the closest to simulating a chess match and score. When we get to a multi-player game, things change a bit. BUT, not too much, I think. In a chess tourney, let's say 30 players for 4 rounds, Ratings computations are NOT computed for how one placed in the tourney but the number of wins and losses as computed against the people played against and their Ratings. The actual formula that I use is the same as the formula the USCF (United States Chess Federation) with the exception of this: the formula calculates all the player Ratings that one played against, the number of wins and losses, and figures the score based on one computation. That's not much different than calculating each person's loss or win against each person they won or lost to. I think the USCF does it the way they do to make it easy (and quick) to calculate all the Ratings that they have to process each day from all the results that are pouring in each day.

(3) Lord Chane wrote the program to do the calculation. I hesitated to ask him to write me something to do the computation but he sometimes has the available time and interest to help out. Certainly, he's very busy at work writing programs to help our agency out all the time (currently, he's working on a full-fledged helpdesk program that will be spread over the US in our agency). The program uses the database table of players and their Ratings to compute the new Ratings and post those results in the player table as well as the games table. All I have to do is click on the program, enter the winning player (from a "drop down"), enter the losing player (from a "drop down"), and the date. The logic necessary to incorporate multiple-player games and their respective results would be quite an undertaking, in my opinon (based on my outdated knowledge of programing logic in Clipper and Pascal).

(4) The suggested adjustment to the points calculation deviate a bit from what is practiced in chess matches. It looks good on the surface but I worry it deviates too much so that we don't get the true results that the USCF has been getting for so many years. In other words, I'd hate to tinker with a formula that seems to work for chess even though we calculate multi-player games a bit differently than the chess federation does. But remember, 5 people in a multi-player game is not so much different than a chess tourney of 5 players...someone will win and someone will lose big. And everyone's score in that 5-person tourney will be adjusted just about the same as in our computations.

(5) Finally, not to be obtuse...applying the USCF Ratings formula to our SE4 games will not be perfect. But, it is a verified formula that takes into consideration the expected win probability and calculates points awarded accordingly. THAT formula works very well. I played competitive chess for many years. I played in Germany and in the US. And when I played against an opponent with a certain rating, the results were pretty much on target...in other words, if I was paired against a player with a higher rating than me, I usually lost. And, the inverse was also true. BTW, for anyone interested, my highest chess rating in the USCF was 1777...these days, though, I figure I'd play around a 1300 rating.

Though not directly related to the suggestions for a computation "adjustment", my sincere thought is this: given enough players, given enough time for the Ratings to "smooth out", a person should be fairly confident of their chances for a win or loss by looking at their opponent (or opponents in a multi-player game). I'd like to see 50 or more people in the Ratings system. Heck, I'd like to see everyone there (well, except for the work...hehe). But, people might not join because they don't think they can be rated near the top. I look at it this way: I think chess is the best game in the world! I'm not the best and never will be a Bobby Fischer. BUT, I wanted to see how I DID rate with other players. So, if people love this game as much as I do and others do, they would want to do the same thing...see how they measured up. Perhaps they might have a low rating BUT, if I were in their place, I'd like to see if my rating got higher...to see if I was getting better. That's what I did in chess.

[geoschmo]

That's fine. As I said it was just a suggestion. I think you are going to find the current system will limit the number of people willing to get rated in large games though.

If you think about it, a player with a high rating is taking a large risk for relativly little to gain by getting involved in a game with several other players. The other players will want to gang up on him because the difference in Ratings makes him a more valuable target. And if he loses early he gets the full force of all the losses to several players below in the rankings. By playing the same number of lower ranked people in 1 v 1 games, he has the same potential gain and much better prospects at achieving victories as his enemies wont be able to cooperate against him. That's where the comparison to chess breaks down unfortunatly. Because no matter how many players in a chess tourneament, you still get to face them one at a time.

The myleague ladder that I suggested get's around this problem by not having the loser move down, only the winner move up. So then the good player might not gain anything still, but at least he's not losing big ground. But it's not a perfect system either. It's got other problems.

Geoschmo

[ September 03, 2003, 18:12: Message edited by: geoschmo ]

[Gozra]

We I already mentioned this maybe I was not clear enough.
Have a seperate rating system for Singles, doubles, and multiplayer that way an indivdual gets three scores. I think that will ease your calculation problems. And make an interesting Ratings system.