Forums: Projects Theory and Miscellanea:

I wonder whether anyone here can help with this.

I am attempting to come up with an algorythm for ranking players in a multiplayer game. The ELO system (used in Chess et-al) does not seem to fit as that is geared towards head-to-head games and im dealing with a variable player-number multiplayer game.

I basically have 4 (possibly 5) "statistics" that I can capture about a players history. I may want to weight some of those statistics more than others.

Can anyone start me off on how to proceed?

tell us about the 4/5 statistics and how you might want to weight them?

Hmm, i'll be treading on thin-NDA-ice. I probably am already.

Lets call them:

1 - Play Frequency. I don't want somebody who has played once and won ranked stupidly high, but at the same time I want people rewarded for playing more. I don't want to encourage people to not play to preserve their rating.

2 - Success Rate. This is a delta. Lets say you start with 100 points and you finish the game with 200 points - thats good. It is just as possible to start with 100 and end with zero, or 5 points. This is, as stated a measure of your success in the game. Rating should be elevated by a better delta.

3 - Abort rate. This is the number of times you have chosen not to play. This is NOT a drop-out or disconnect rate. its something different which I cant easily explain. However, the more of these you have the more your rating should shift down.

4 - Speed rate. Put simply this is your response rate in the game. It is affected slightly by slower connections, but the game itself is not a hard core FPS or anything so we're talking marginal. Obviously the faster your response the better your rating.

*5 - Rating. Obviously for each game you play there is already a rating. That rating is already based on the above 4 items (in order of importance). Your overall rating could be affected by the individual ratings of games you've played.

dunno what a 'delta' is? That sounds tricky - I guess you can start a game with something other than 100 points too?

kdice.com/ seems to have quite a complex scoring mechanism that I haven't really figured out yet. You get points for winning (and more points for winning by a bigger margin), and your points drop when you lose.

By giving the players a point score that they can increase by winning more - you encourage them to play more.

It's not really an algorithm, more a way of controlling how many points you give to each player at the end of each game. The ranking is then easy.

Sorry this is a bit thin on details... it's a bit hard what you're doing here!

maybe have a look at my stats on kdice for ideas:

kdice.com/profile/44761168

see how each game gave me, or took away, different amounts of points.

Thanks, you've actually given me some ideas!

I advise you to not play kdice btw - it has sucked hours from my life lately.

oh ive played kdice-a-plenty. just didnt think of looking at the leaderboard for inspiration!

I play as metameat and étapes. poorly.
you guys?

don't think I've seen you online then - I'm on with my real name, DontBogartMe. Although I'm trying to cut down to be honest... damn dice-crack.

true dat.

Looking at my averages I expect to stay hovering between 500 and 0 indefinitely. I honestly don't understand how people get up in the 5k range even on the 500/month buy-in. I know you can kind of game the rank by playing the highest rank boards you can afford as you win way more for performing poorly on higher rank boards so if you average in the top 4 you should still slowly grow but even so...

Serious thanks... kdice cracked it 100%. I have simplified the model massively and its now based on just 2 parameters which give me a really really easily calculatable score.

100 thankyous DBM. I offer you my firstborn* to show my gratitude.


* I dont have any offspring yet.

glad my ramblings helped

Originally posted by: jamiec
I offer you my firstborn* to show my gratitude.

hmm... looks like kdice is increasingly pay to play now. Sucks.

Also lots of interesting new rules, no team play, for instance...