Page 3 of 5
Posted: 01 Mar 2007, 23:52
by manored
Radja wrote:manored wrote:I didnt understood that explanationm from ELO rating very well, but it sounds nice...
Just one thing: since the ELO rating is made for 1vs1 games, does that mean that only 1vs1 games will count for your ELO rating?
no rating system supports multiple players per team nor multiple (+2) teams, we'd need judges for that, and those would need some complicated rules too
clan based rating would be easy, casual play teamA vs teamB would not
-----
simply put, in elo each player has a score, the higher you score compared to your enemy, the easier it is supossed for you to win
lets say your scores' difference means that you should win 3 matches out of 5, but after 5 matches you only win 2, your scores dont represent this and they are changed (you lose some points and the other wins some points)
of course it has quiet some mathematics so that you'll get your new score after each match, also seasoned players dont win/lose points so fast
but again this is for games where both sides are identical (except the first move advantage), there are no mods, no arm / core, etc. so it might need some tweaking before using it here
Why dont we use some simpler system, like a points system that gives you points depending of how many points the guy you just defeated haves and that eliminates a large deal of points from everone (like 50%)from time to time? (to keep people from having too many points)
Posted: 02 Mar 2007, 15:41
by Radja
manored wrote:Why dont we use some simpler system, like a points system that gives you points depending of how many points the guy you just defeated haves and that eliminates a large deal of points from everone (like 50%)from time to time? (to keep people from having too many points)
that would skyrocket higher rating players when they fight each other, and the -50% would destroy the whole idea of keeping track of points
mathematics is not the problem, its all done automaticly by the server and the standar elo system already works (for chess), what we need is to find a decent way to rate casual 4v4 or 2v2 games (maybe just giving small boosts, or no rating them at all), and some other considerations (mods, factions, etc)
Posted: 02 Mar 2007, 15:58
by Betalord
There would be two type of battles: 1on1 and clan-on-clan. This is for practical reasons only ├óÔé¼ÔÇ£ there is no accurate way to rate 2on2 (or n-on-n) games using ELO rating that I know of. So clans would be a single entities with a list of players in it (of which 1 or more would be marked as clan founders, giving them access to add more players to the clan etc.). Clan-on-clan games can then have arbitrary number of players (2on2, 3on3, etc.) and the ELO points would be assigned to the clan entity and not to the players themselves.
Posted: 02 Mar 2007, 17:44
by manored
Radja wrote:manored wrote:Why dont we use some simpler system, like a points system that gives you points depending of how many points the guy you just defeated haves and that eliminates a large deal of points from everone (like 50%)from time to time? (to keep people from having too many points)
that would skyrocket higher rating players when they fight each other, and the -50% would destroy the whole idea of keeping track of points
mathematics is not the problem, its all done automaticly by the server and the standar elo system already works (for chess), what we need is to find a decent way to rate casual 4v4 or 2v2 games (maybe just giving small boosts, or no rating them at all), and some other considerations (mods, factions, etc)
Keeping track of how many resources they produced and stored and how costly was the damage they caused to the enemy maybe?
I really think that there should be something to rate higher that 1vs1 games.
Posted: 02 Mar 2007, 20:10
by Radja
damage/income ratio would need a lot of tracking by the game that im not sure its feasible
i cant really tell who is the most useful /skilled player in a +2v2 game, the rusher? the one behind going econ and t2-3? the one who did a lot of fighting instead of teching to give time to others time to tech up?
is a horrible player in the winning team better than the one who struggled the most but lost in the end? who unskilled is wasting resources? unit spam is worse than commando style raids?
in 1vs1 (or clan vs clan) you have victories and losses, if one wins he is considered better (or luckier) than the other, but in big games with casual players you just cant say that belonging to the winning team makes you a better player than the loser team
Posted: 02 Mar 2007, 21:55
by 10053r
A friend of mine did a large paper on OTA for college back around 7 years ago. It was a statistics paper, and he found that the highest correlation between wins and any other statistic (in a single game) was metal spent. Although mods have come and gone since then, I think it is still safe to say that metal spent (either by building something, or by giving it to your teammates) is the single largest predictor of skill.
So you take metal spent, normalize for the length of the game (keeping in mind that metal spent grows exponentially, so you must normalize exponentially), normalize for the amount of metal present on the map (so playing metal maps doesn't help you more than green fields), and you have a great stat about how good a player is.
I always look at the statistics at the end of every game I play. It is EXTREMELY rare for the winning team to have spent less metal than the losing team, and even more rare for the "helpful" players to have spent less metal than the "freeloaders".
Posted: 02 Mar 2007, 23:53
by LordMatt
In a com continues game, that would likely always hold true, however in com ends one can get a win in the teeth of a loss, with some fancy tricks.
Posted: 03 Mar 2007, 00:12
by manored
Also, the winner team always wasting more metal may have something to see with the fact that they continue to waste metal after anniquilating the enemys ability to do so and starting the hunt for the last alive units.
Posted: 03 Mar 2007, 00:14
by Radja
statistcs are the mathematics of the devil ><
victory points should be awarded by working toward / fullfilling victory condition
Posted: 03 Mar 2007, 00:27
by manored
Radja wrote:statistcs are the mathematics of the devil ><
victory points should be awarded by working toward / fullfilling victory condition
The problem is that this is extremely hard to define. You would need to think in almost all possible existent ways of helping your team, give em all a value and then balance em all.
Posted: 03 Mar 2007, 00:34
by Radja
thats my point, you'd need human judges, and theyd discuss to no end about it
maybe individual skills should be rated individually
(or in ffa!)
Posted: 03 Mar 2007, 00:46
by manored
Radja wrote:thats my point, you'd need human judges, and theyd discuss to no end about it
maybe individual skills should be rated individually
(or in ffa!)
games with more of 2 teams involve too much luck to rate someone's ability.
Posted: 03 Mar 2007, 03:36
by Ishach
The whole topic of a new ranking system is to show a player is even better than the average Silver Star.
If you don't feel confident playing 1v1, maybe you aren't ready to be anything higher than an average Silver Star
Posted: 03 Mar 2007, 09:50
by Lolsquad_Steven
Then why not just give
those few better than
average
Posted: 03 Mar 2007, 12:26
by quantum
If you want to get rated on team games, find a couple of buddies and form a clan. If plenty of clans are created as a result of the ranking system, it won't be hard to find opponents.
You'll also discover that clan games are a lot more fun than random team games :D
Posted: 03 Mar 2007, 17:50
by Licho
AF wrote:This is technically impossible atm since spring doesnt report win/lose stats, hopefully itll kick start the process again.
And how do you think springies report stats?
Spring does report game result..
http://unknown-files.net/stats/games.php
Posted: 03 Mar 2007, 21:02
by Dragon45
1v1 is the best way to determine skill. Period. team clan games are just tasty gravy :)
In any case, wouldnt it be possible to reduce NvNvN games (Multi-clan orgies essentially) to a matrix of NvN games, and then assign points accordingly? Surely von Neumann can be of some use here :D
Posted: 03 Mar 2007, 21:17
by Neddie
Skill is relative, subjective, multidimensional and something which cannot be accurately measured. I disagree with the assertion that any particular category of play is the best way to determine skill.
Posted: 03 Mar 2007, 22:03
by manored
neddiedrow wrote:Skill is relative, subjective, multidimensional and something which cannot be accurately measured. I disagree with the assertion that any particular category of play is the best way to determine skill.
Lets then jugde the results of skill... (resources stored (wasted resources are lack of skill), amount of damage caused to the enemy, etc)
Posted: 03 Mar 2007, 22:29
by Radja
we could define skill (in a game) as how easily you win, and that can be measured (win/loss, elo, someother way, etc)
of course we could rate teching skill, rushing skill, microing skill, etc. but i think that winning a game (or being the best out of 2, 3, 5, e^n matches) is a good enough indicator of how skilled a player is (compared to the other one)