|
Post by Canucksoldier on Nov 3, 2007 16:09:34 GMT -5
Thanks Scott, to clarify a teamer is treated just like a duel as far as the system is concerned with the points split across the members of the team in the last step.
CS
|
|
|
Post by thegreatsatan on Nov 5, 2007 10:01:53 GMT -5
I think this is a great idea. You should contact Geforced and we should combine all ideas and set up 2 new ladders, one based on BTS which is more geared towards playing against an AI / other people who like to test the skills they learned in Single player vs other people in the world and another ladder based on the balanced MP mod we are making. When people first start playing Civ 4 they will be able to log into GameSpy and get adjusted to the game. They can be a part of your guys ladder where they will grow into ok players who understand the general concepts of the game and how to micro-manage. The players who arn't competitive and don't mind the original version with its strategic loopholes will stay in your ladder. The players who prefer a more competitive version of the game that is balanced with minimum - no strategic loopholes will transfer over to our ladder. Those competitive players will be able to partake in serious competition with cash prizes and not have to worry about random newbs or flaws in the strategic depth of the game ruining their gaming experience. What you guys think? a ladder made by a POS like u mgt would be crap. 1st off, you'll never make a 100% fair version. People like you will just find new problems to throw the same little cry baby fits. The little cry baby fits will never go away because they're built into the POS's that have them.(like mgt) But if it takes all the crying,whining loosers away. I'm all for it. A league made up of POS's like u would just be a POS. By the way, u never reported that cton I killed u in. u died like a b1Tch and cryed to everyone. remeber? P.S. Get over yourself. Did I mention you're a total jackhole/POS. Your just another AVERAGE civ player at best. I've seen u die and have killed u myself many times.
|
|
|
Post by NumberOneMercury on Nov 5, 2007 12:56:55 GMT -5
Not sure the teamer system works 100% right--I think there will be some situations where losing teamers could increase a persons skill rating. Here's a simple example of how I understand the system works, correct me if I'm wrong:
For a 2v2 your distribution is 40-30-20-10. So the winning team gets 70% of the ante, 35 to each team member, the losing team gets 30%, 15 to each team member.
Now assume that the winning team is much better with say skill ratings of 2000 each, so that each antes 10% = 200, and the losing team has a member with skill rating of 2000 and the other has a rating of 1000, so that the antes are 200 and 100 respectively. The total ante is 200+200+200+100=700.
Assuming team two loses the skill rating of the 1000 becomes 1000 - 100 (ante) + 15% * 700 (total ante) = 1000 - 100 + 105 = 1005 (!!)
The issues here are (a) people shouldn't gain for losing teamers with good teammates and (b) do we really need more people clammoring to get into elitist teamers to improve their ratings simply by being there?
Maybe the people with the higher skill ratings should get a larger part of the winnings to compensate for anteing a larger amount?
I think distributing the winnings/reimbursement based upon fraction of teams ante (rather than evenly) would cure the problem entirely and remove the negative incentives I noted above.
In the above example that would mean that the 1000 guy would be 1000 - 100 (ante) + (100/(100+200)) * 30% * 700 = 1000 - 100 + 10% * 700 = 1000 - 100 + 70 = 970.
(Also I think 10% ante is too high.)
|
|
Slaughter
Settler
God save the screen!
Posts: 48
|
Post by Slaughter on Nov 5, 2007 13:56:04 GMT -5
for a teamer the distribution would always be 67-33 since both teams are treated like 1 player. with the total ante split like that, each teams part is then split among all team players equally.
but still you're right that it can happen that with extreme values high ranked players playing together in a team might lose skill even if they win and the other way round would low skilled players gain skill for a lost game.
we had a longer discussion about that too and we also had the idea to spread it like you suggested. but after going through some more examples it showed that redistributing it based upon the given fraction would favor the better players when winning and penalize the worse players when losing just too much in the long run. in your example losing vs 2 players with double your skill the 1000 player would lose 30 skill, that's really a lot for such an unbalanced game. and it's even worse if the difference is larger and that team wins, the highest skilled player in the team would get the most skill because he anted the most. but that would mean that the player will gain too much skill for playing vs players that are each lower skilled than him.
we don't want to have players camping at the top playing games with the lowest skilled player against other low skilled players and still gain lots of skill even if they would easy win it 1v2 anyway. so we decided to favor that the lower skilled players lose less skill than higher ranked if they lose and in case of a win they will win more skill.
so we added a global rule that a losing team NEVER gains skill and a winning team NEVER loses skill for a won match. with extreme values this will still give good results for players not affected by this rule and just cut off negative values but does work better in the long run and works even great if people of equal skill play each other.
|
|
|
Post by NumberOneMercury on Nov 5, 2007 14:09:05 GMT -5
for a teamer the distribution would always be 67-33 since both teams are treated like 1 player. with the total ante split like that, each teams part is then split among all team players equally. but still you're right that it can happen that with extreme values high ranked players playing together in a team might lose skill even if they win and the other way round would low skilled players gain skill for a lost game. we had a longer discussion about that too and we also had the idea to spread it like you suggested. but after going through some more examples it showed that redistributing it based upon the given fraction would favor the better players when winning and penalize the worse players when losing just too much in the long run. in your example losing vs 2 players with double your skill the 1000 player would lose 30 skill, that's really a lot for such an unbalanced game. and it's even worse if the difference is larger and that team wins, the highest skilled player in the team would get the most skill because he anted the most. but that would mean that the player will gain too much skill for playing vs players that are each lower skilled than him. we don't want to have players camping at the top playing games with the lowest skilled player against other low skilled players and still gain lots of skill even if they would easy win it 1v2 anyway. so we decided to favor that the lower skilled players lose less skill than higher ranked if they lose and in case of a win they will win more skill. so we added a global rule that a losing team NEVER gains skill and a winning team NEVER loses skill for a won match. with extreme values this will still give good results for players not affected by this rule and just cut off negative values but does work better in the long run and works even great if people of equal skill play each other. Okay, but if 30 points is alot to lose for such an imbalanced game, how can you justify that the 2000 would be at around 2000 - 200 + 105 = 1905...95 points lower for the imbalanced games... The really ironic part is that if the other player was stronger the 2000 loses less skill!!!! say the other loser was a 2000 then it's around 2000 - 200 + 15% * 800 = 2000 - 200 + 120 = 1920!!!! LOL! Who is going to play with worse players then? Not me. Another problem IMO is that the ante is too high, this isn't a question of someone camping out at top, because losing any one game will cause a person to plummet. Also, FYI the way that Case's deals with the issue of playing weak players to gain rank is by building in a probability of win based upon the skill rating (you can see it in one of the help pages) so people don't gain much skill from beating weaker players and lose alot for losing to weaker players--you may want to incorporate something like that into your mechanism.
|
|
Slaughter
Settler
God save the screen!
Posts: 48
|
Post by Slaughter on Nov 5, 2007 15:57:55 GMT -5
Okay, but if 30 points is alot to lose for such an imbalanced game, how can you justify that the 2000 would be at around 2000 - 200 + 105 = 1905...95 points lower for the imbalanced games... 95 is 4,75% of the initial skill of 2000 and 30 is 3% of the 1000 ... that's not really far away form each other taking into account that the more skill you have, the more ante you bring in. The really ironic part is that if the other player was stronger the 2000 loses less skill!!!! say the other loser was a 2000 then it's around 2000 - 200 + 15% * 800 = 2000 - 200 + 120 = 1920!!!! LOL! Who is going to play with worse players then? Not me. well, if you play with same skilled players your teammates pay their due and it's not mostly your part anteing skill. you just forget that if you play with higher skilled players you will have the advantage on your side ... that's how it works, with high skill you have more to lose. but then if other players bypass you it will be in your favor the next games . also if you win such a game, your opponents will lose even more skill than you would, that's worth the risk the problem itself is not new though, the way it's right now at cases is similar: losing a 6v6 teamer with a low skilled player who dies to a warrior makes up for 6 reports against you (with the difference that it might not affect your rank at all, but that's one thing all players were mad about so we try to change it - you can't have it all, sorry ;D) Another problem IMO is that the ante is too high, this isn't a question of someone camping out at top, because losing any one game will cause a person to plummet. it's just a question of where the average and top skill levels will be, maybe it's not 2000+ skill like it's at cases atm, you start with 1000 skill, so maybe 1700-1800 is a top skilled player already. with a higher ante the top and average skills will just be lower but that doesn't harm the overall system, with generally lower skills the problems you described will alos have a lower effect. Also, FYI the way that Case's deals with the issue of playing weak players to gain rank is by building in a probability of win based upon the skill rating (you can see it in one of the help pages) so people don't gain much skill from beating weaker players and lose alot for losing to weaker players--you may want to incorporate something like that into your mechanism. hmm, the ante system works just like that (at least with duels just like at cases), you gain more skill beating higher skill players than winning vs lower skilled and you lose more when losing vs lower skilled than losing vs better players. i don't see a problem there. after all it's really hard to get all game styles together in a single rating system - the ante system just does a very good overall job.
|
|
|
Post by NumberOneMercury on Nov 5, 2007 16:26:27 GMT -5
95 is 4,75% of the initial skill of 2000 and 30 is 3% of the 1000 ... that's not really far away form each other taking into account that the more skill you have, the more ante you bring in. My point is that I think we all agree that if the result is expected based upon the current skill ratings, then skill ratings should not change much EDIT: relative to results unexpected based upon skill ratings. Your system does not work like this. If we don't agree on this then we must have different ideals of how the system should work. The Case's and your system both work like this to some extent for duels (I think Case's skill rating system actually handles this better IMO by building in an exponential function to calculate probability of win). Neither your system or Case's works like this for teamers. We accept this for Case's bc we know Case's was made for duels and not teamers. But if we are starting from scratch why not make a theoretically sound model that works with our ideals? I was a math major in college and would be happy to make a proposed revised model, with a full explanation of how each component works, which you can tweak as you see fit if you would like. it's just a question of where the average and top skill levels will be, maybe it's not 2000+ skill like it's at cases atm, you start with 1000 skill, so maybe 1700-1800 is a top skilled player already. with a higher ante the top and average skills will just be lower but that doesn't harm the overall system, with generally lower skills the problems you described will alos have a lower effect. This is not correct because the change in skill level is based upon a % of the players rank--what you are thinking of is that the starting rating for each player does not matter, which simply affects the average and top levels. The % change greatly affects the volatility of the rating. The way it is structured now volatility will be MUCH greater than ranks on Case's currently. after all it's really hard to get all game styles together in a single rating system - the ante system just does a very good overall job. Yes it does do a very good job, but my point is I think it can be better for teamers without affecting the other styles much at all.
|
|
|
Post by Bantams on Nov 5, 2007 17:20:13 GMT -5
the withdraw button needs hiding in the edit my profile section and the Gold account cost should be same as it is now maybe make any profits go towards end of year CCC Trophey for the Clan with the most wins from the past 12 months oh and PayPal is easy to use but in all Great Job everyone ps hope this aint the last time my rank is 15 ;D
|
|
Slaughter
Settler
God save the screen!
Posts: 48
|
Post by Slaughter on Nov 5, 2007 17:31:35 GMT -5
My point is that I think we all agree that if the result is expected based upon the current skill ratings, then skill ratings should not change much. Your system does not work like this. If we don't agree on this then we must have different ideals of how the system should work. The Case's and your system both work like this to some extent for duels (I think Case's skill rating system actually handles this better IMO by building in an exponential function to calculate probability of win). Neither your system or Case's works like this for teamers. We accept this for Case's bc we know Case's was made for duels and not teamers. But if we are starting from scratch why not make a theoretically sound model that works with our ideals? I was a math major in college and would be happy to make a proposed revised model, with a full explanation of how each component works, which you can tweak as you see fit if you would like. that would be really great, if you can make a model that works for duels, teamers and ctons alike i would give it a shot for sure. although i wasn't really bad at math, probability theory and statistics were not my favorites ;D ... so a small explanation of your model would be nice would be really cool if you can make it within 2 weeks, after then it will be much harder for me to change the skill calculations when finishing the tournament reporting system - i'd really like to implement it if it's suitable before that. This is not correct because the change in skill level is based upon a % of the players rank--what you are thinking of is that the starting rating for each player does not matter, which simply affects the average and top levels. The % change greatly affects the volatility of the rating. The way it is structured now volatility will be MUCH greater than ranks on Case's currently. hmm, not sure if you can generalize it for the whole range of skills. right now a player ranked 300 winning vs rank #10 would jump to 155, i think that is really volatile and worse than the ante system which is kinda stable within a standard deviation from the mode value - at the edges it might have a greater volatility though (not sure if it's really a bad effect or not). but i was referring to the spread not the volatility, i guess with the ante system it will be a 'steeper' gaussian curve which minimizes the effects of extreme values in calculation. Yes it does do a very good job, but my point is I think it can be better for teamers without affecting the other styles much at all. like i said, that would be great ... i was and will never be good at theoretical modelling, but if you can provide one i'll be able to implement it - that's what i'm good at
|
|
|
Post by rokkitlauncher on Nov 5, 2007 18:48:39 GMT -5
only the 1v1s and the ctons should be ranked, the teamers rarely depend on skill anyway.
either that or not every ladder game has to be ranked, you could have to include [Ranked] at the end of the game name if you want it to be super serious then you can leave it out if you want to play for a bit of fun, but also just want to play with some friends
|
|
|
Post by Ellestar on Nov 6, 2007 3:42:31 GMT -5
[/li][li]Skill gain in team games will be based on your teammates' skills too and not only on your own skill. It's more hard to gain skill with stacked top players on one team and you'll gain more skill if you beat a overall 'good' skilled team with mostly 'bad' skilled teammates on your own team. Of course it also works in the opposite direction, if you lose with your team against a team of all lower skilled players, you (and all your teammates) will lose significantly more skill than you would by losing vs a nearly equal or better skilled team.[/quote] Something tells me it's not exactly so. Let's see... Say, more experienced players have skill H and less experienced ones have skill L. So, for a standart 5v5 teamer with 1 experienced and 4 unexperienced vs 5 experienced it will be: pool: (4 * L + 6 *H) * 10% = 2 / 5 * L + 3 / 5 * H If unexperienced team wins then gain of the only experienced player in a team is (2 / 3) / 5 * ( 2 / 5 * L + 3 / 5 * H) - H / 10 = = 1 / 150 * ( 8 * L + 12 * H - 15 * H) = ( 8 * L - 3 * H ) / 150 With only experienced players in a teamer, the same player gains (2 / 3 ) / 5 * ( H ) - H / 10 = ( 20 * H - 15 * H) / 150 = (5 * H) / 150 So, let's check when skill gain is bigger. 5 * H > 8 * L - 3 * H 8 * H > 8 * L (H, L >0) Since we agreed that H > L, 5 * H is always bigger than 8 * L - 3 * H In other words, experienced player playing with unexperienced players against more experienced team gets less skill from a win compared to a win in a game where all players are experienced. So, your original statement isn't true Well, i sensed that something is fishy with that system. So it was a good idea to check it
|
|
|
Post by 9iron9 on Nov 6, 2007 3:52:51 GMT -5
yeah i was doinmg the same equations on my ladybird "my first calculater" countdown edition.
|
|
|
Post by Ellestar on Nov 6, 2007 4:30:30 GMT -5
we had a longer discussion about that too and we also had the idea to spread it like you suggested. but after going through some more examples it showed that redistributing it based upon the given fraction would favor the better players when winning and penalize the worse players when losing just too much in the long run. in your example losing vs 2 players with double your skill the 1000 player would lose 30 skill, that's really a lot for such an unbalanced game. and it's even worse if the difference is larger and that team wins, the highest skilled player in the team would get the most skill because he anted the most. but that would mean that the player will gain too much skill for playing vs players that are each lower skilled than him. ROFLMAO ;D With whom you're discussed a lot? They should go back to school ;D Since in your system ante is proportional and redestribution is equal, lower skill players will gain more skill after a match compared to higher skill players in the same team. I'm not related to any work concerning mathematics (that is, i'm not a specialist and not even an amateur) but even for me it's obvious. Another obvious conclusion from that is that lower skill players will gain skill by just playing teamers with higer-skill players, given about equal win/losses - obviously, at the expense of the higher-skilled players, even if lower-skill players don't contribute to wins. Just a frequent participation is enough. If you ask me, it will be more balanced if skill gains in case of a loss should be proportional. Since ante is proportional, total skill loss because of a match will be proportional too (some % of a total skill at the start of a match). And in case of a win, winning pool should be used to compensate ante first (proportional part) and after that you can distribute extra equally. Then total skill gain will be equal for all players no matter the skill. Of course, it still has the same problems but at least it's not that extreme. so we added a global rule that a losing team NEVER gains skill and a winning team NEVER loses skill for a won match. with extreme values this will still give good results for players not affected by this rule and just cut off negative values but does work better in the long run and works even great if people of equal skill play each other. Team or player? Even if a team doesn't lose skill because of a win, that obviously doesn't mean that all individual players will not lose a skill because of a win. So IMHO that rule doesn't make sense. Well, i guess it's because you don't understand the fact mentioned above. that would be really great, if you can make a model that works for duels, teamers and ctons alike i would give it a shot for sure. although i wasn't really bad at math, probability theory and statistics were not my favorites ;D ... so a small explanation of your model would be nice would be really cool if you can make it within 2 weeks, after then it will be much harder for me to change the skill calculations when finishing the tournament reporting system - i'd really like to implement it if it's suitable before that. I'm suggesting you to keep all match details in the database so skills can be recalculated retroactively if you'll want to change the system (as a bonus, you'll be able to test the settings like ante % on a real data and choose the best ones). I bet that your first version will not work as you expected.
|
|
Slaughter
Settler
God save the screen!
Posts: 48
|
Post by Slaughter on Nov 6, 2007 5:58:19 GMT -5
Something tells me it's not exactly so. Let's see... Say, more experienced players have skill H and less experienced ones have skill L. So, for a standart 5v5 teamer with 1 experienced and 4 unexperienced vs 5 experienced it will be: pool: (4 * L + 6 *H) * 10% = 2 / 5 * L + 3 / 5 * H If unexperienced team wins then gain of the only experienced player in a team is (2 / 3) / 5 * ( 2 / 5 * L + 3 / 5 * H) - H / 10 = = 1 / 150 * ( 8 * L + 12 * H - 15 * H) = ( 8 * L - 3 * H ) / 150 With only experienced players in a teamer, the same player gains (2 / 3 ) / 5 * ( H ) - H / 10 = ( 20 * H - 15 * H) / 150 = (5 * H) / 150 So, let's check when skill gain is bigger. 5 * H > 8 * L - 3 * H 8 * H > 8 * L (H, L >0) Since we agreed that H > L, 5 * H is always bigger than 8 * L - 3 * H In other words, experienced player playing with unexperienced players against more experienced team gets less skill from a win compared to a win in a game where all players are experienced. So, your original statement isn't true Well, i sensed that something is fishy with that system. So it was a good idea to check it yea, the explanation was a bit wishy-washy, from the point of a high skilled player with all lower skilled teammates it is not really correct since it depends on the overall anted skill which can't be higher if you have alot low skilled players in game. With whom you're discussed a lot? They should go back to school ;D Since in your system ante is proportional and redestribution is equal, lower skill players will gain more skill after a match compared to higher skill players in the same team. I'm not related to any work concerning mathematics (that is, i'm not a specialist and not even an amateur) but even for me it's obvious. umm, yea? that's exactly how it was supposed to work, lower skilled players gain more skill from the same game than higher skilled teammates if they win (and lose less if they lost). Another obvious conclusion from that is that lower skill players will gain skill by just playing teamers with higer-skill players, given about equal win/losses - obviously, at the expense of the higher-skilled players, even if lower-skill players don't contribute to wins. Just a frequent participation is enough. isn't it always this way? it's the same with cases right now and it will be the same with every other skill system too, if you team up with a low skilled player and your team wins, he will have the most gain from it if he didn't contribute to the win. it's always at the higher skilled player's risk to play with lower skilled player in the same team. what you are right with is that the skill gain should ideally be greater for the high skilled player when winning with a low skilled team than winning with a high skilled team. but with the ante system the total pot is lower with more low skilled players participating. If you ask me, it will be more balanced if skill gains in case of a loss should be proportional. Since ante is proportional, total skill loss because of a match will be proportional too (some % of a total skill at the start of a match). And in case of a win, winning pool should be used to compensate ante first (proportional part) and after that you can distribute extra equally. Then total skill gain will be equal for all players no matter the skill. Of course, it still has the same problems but at least it's not that extreme. the idea of compensating the ante 1st before redistributing equally sounds good, this will favor the higher skilled players a bit more but is still better for the lower skilled players. i will give it some test runs and see how it works. Team or player? Even if a team doesn't lose skill because of a win, that obviously doesn't mean that all individual players will not lose a skill because of a win. So IMHO that rule doesn't make sense. Well, i guess it's because you don't understand the fact mentioned above. the rule is applied to each individual player after the redistribution of the ante, so it only affects players that actually have negative gain for a win or the other way round. only with really unbalanced teams it will happen to all players on one team, but mostly it will affect only the highest (or lowest skilled) players on a team. I'm suggesting you to keep all match details in the database so skills can be recalculated retroactively if you'll want to change the system (as a bonus, you'll be able to test the settings like ante % on a real data and choose the best ones). I bet that your first version will not work as you expected. let's see what nom comes up with, hope his theoretical model will work in practice as expected. i will have both skill systems run on the same random set of pairings after and see how they behave in the long run, working on such a test program right now. i will also change the function for skill calculation and make it interchangeable so whenever someone comes up with a good model it can be put in easily. another look around for other rating systems would not hurt either, but there really isn't a good system incorporating duels, team and ffa games. if anyone knows of a good system please speak up ... something like the MS True Skill System might be a good, if someone could explain it to me ... (not enough math skills for that )
|
|
|
Post by Ellestar on Nov 6, 2007 9:15:51 GMT -5
umm, yea? that's exactly how it was supposed to work, lower skilled players gain more skill from the same game than higher skilled teammates if they win (and lose less if they lost). Ok, but then it contradicts your statement "and you'll gain more skill if you beat a overall 'good' skilled team with mostly 'bad' skilled teammates on your own team". That is, from my point of view because i think i'll have a higher than average skill. isn't it always this way? it's the same with cases right now and it will be the same with every other skill system too, if you team up with a low skilled player and your team wins, he will have the most gain from it if he didn't contribute to the win. it's always at the higher skilled player's risk to play with lower skilled player in the same team. Of course it isn't. AFAIK in every such skill ranking system you don't lose much score if you're expected to lose anyway. Cases system will work perfectly for 1v1 and maybe for ctons too but it shouldn't be used for teamers the way we use it. Say, that Microsoft TrueSkill system you mentioned works that way. That's why i said that we'll be better using separate ladders for 1v1, ctons and teamers, at least until we'll find a better solution. If we don't, all that "skill" thing doesn't make sense at all because you don't use it properly. In your system it's a very bad idea to play with weak players against a strong players, even if you lose 67% of the usual loss, your chances to win are near zero. Maybe a stacked team's gains will be lower, but they'll mostly be at the expense of a stronger players in a weaker team. And i don't think that denying new players to teamers is a good idea, it happens too much as it is (because it's not fun to play such games). Obviously, your system makes it even worse. At least, in Case's system you don't lose your own place when you're a part of a much weaker team. Also, skills will represent a real player skills only if there will be no weird jumps in a skill if your team is supposed to lose (according to a skill comparision). That's the whole point of the ranking like ELO etc. If a stronger players will lose relatively the same score when playing with weaker players against a much stronger team, you'll never have a fair ranking. what you are right with is that the skill gain should ideally be greater for the high skilled player when winning with a low skilled team than winning with a high skilled team. but with the ante system the total pot is lower with more low skilled players participating. I just pointed at a mistake in your advertisement let's see what nom comes up with, hope his theoretical model will work in practice as expected. i will have both skill systems run on the same random set of pairings after and see how they behave in the long run, working on such a test program right now. i will also change the function for skill calculation and make it interchangeable so whenever someone comes up with a good model it can be put in easily. another look around for other rating systems would not hurt either, but there really isn't a good system incorporating duels, team and ffa games. if anyone knows of a good system please speak up ... something like the MS True Skill System might be a good, if someone could explain it to me ... (not enough math skills for that ) It's incomplete, looks like a full description is here ftp.research.microsoft.com/pub/tr/TR-2006-80.pdf. Good thing is that it already includes the way to calculate skills based on team games. And it was made by a professionals. So if you really want to have a ladder that incorporates team games, we should use this skill rating system. However, i just read their descriptions so far, i didn't check the math. Also, they have an example dataset here research.microsoft.com/mlp/apg/downloads.aspx
|
|
|
Post by SweViking on Nov 6, 2007 13:47:16 GMT -5
Who cares about rank anyway? But ithink you guys doing a great job and its cool you sacrifice alot of free time to try to make a better ladder. Another tip: Maybe we can implement es combat calculator in the site ;D ;D ;D ;D ;D
|
|
marr
Worker
Posts: 169
|
Post by marr on Nov 6, 2007 14:32:05 GMT -5
One other problem commonly found with a ranking system that cases did somehow manage to address with its simplistic system is multiple accounts. I happen to agree that multiple accounts are neccesary in a ranked system as someone with a high ranked nic, will inivitably only want to play with another high ranked nic as it will increase rank and lead to less of a drop in rank if you lose.
High rank plays with low rank = less gains for a win = higer losses for a loss.
One possible idea to promote this and to encourage ranked play is to allow each registered user up to three nics each viewable under thier user profile. A registered user could keep his high nic, play high ranked games and allow him the oppurtunity to play low ranked games without sacrificing his high end nic.
Perhaps a gold membership could help facilitate this in allowing a gold memeber either a limited number of nics on his account, or as many as he wants.
|
|
|
Post by Canucksoldier on Nov 6, 2007 17:24:51 GMT -5
umm, yea? that's exactly how it was supposed to work, lower skilled players gain more skill from the same game than higher skilled teammates if they win (and lose less if they lost). Ok, but then it contradicts your statement "and you'll gain more skill if you beat a overall 'good' skilled team with mostly 'bad' skilled teammates on your own team". That is, from my point of view because i think i'll have a higher than average skill. Of course it isn't. AFAIK in every such skill ranking system you don't lose much score if you're expected to lose anyway. Cases system will work perfectly for 1v1 and maybe for ctons too but it shouldn't be used for teamers the way we use it. Say, that Microsoft TrueSkill system you mentioned works that way. That's why i said that we'll be better using separate ladders for 1v1, ctons and teamers, at least until we'll find a better solution. If we don't, all that "skill" thing doesn't make sense at all because you don't use it properly. In your system it's a very bad idea to play with weak players against a strong players, even if you lose 67% of the usual loss, your chances to win are near zero. Maybe a stacked team's gains will be lower, but they'll mostly be at the expense of a stronger players in a weaker team. And i don't think that denying new players to teamers is a good idea, it happens too much as it is (because it's not fun to play such games). Obviously, your system makes it even worse. At least, in Case's system you don't lose your own place when you're a part of a much weaker team. Also, skills will represent a real player skills only if there will be no weird jumps in a skill if your team is supposed to lose (according to a skill comparision). That's the whole point of the ranking like ELO etc. If a stronger players will lose relatively the same score when playing with weaker players against a much stronger team, you'll never have a fair ranking. I just pointed at a mistake in your advertisement let's see what nom comes up with, hope his theoretical model will work in practice as expected. i will have both skill systems run on the same random set of pairings after and see how they behave in the long run, working on such a test program right now. i will also change the function for skill calculation and make it interchangeable so whenever someone comes up with a good model it can be put in easily. another look around for other rating systems would not hurt either, but there really isn't a good system incorporating duels, team and ffa games. if anyone knows of a good system please speak up ... something like the MS True Skill System might be a good, if someone could explain it to me ... (not enough math skills for that ) It's incomplete, looks like a full description is here ftp.research.microsoft.com/pub/tr/TR-2006-80.pdf. Good thing is that it already includes the way to calculate skills based on team games. And it was made by a professionals. So if you really want to have a ladder that incorporates team games, we should use this skill rating system. However, i just read their descriptions so far, i didn't check the math. Also, they have an example dataset here research.microsoft.com/mlp/apg/downloads.aspxYes we looked at the TrueSkill system, but unfortunately MS only gives you have the info for free, the real details are not public intellectual property. But I am sure that we can tweak our ante system to ensure that teamers are as fairly treated as ctons and 1v1's, but we also have to ensure that while we create a system that ensures the league is competitive across the board we also don't want to create a huge rift between new players and the top players. We have enough people screaming "elitism" as it is now. If we do this system right it should create a normal bell curve of player skill values over a few month's of reports. CS CS
|
|
|
Post by TheBadSeed on Nov 6, 2007 23:25:15 GMT -5
Just wanted to clarify a few things here. First, how exactly would the skill depreciation work when someone is off line? Does skill depreciation begin at 24 hours? 48, a week? My personal opinion is that it should take at least a few days for skill depreciation to begin, and then be very mild. One of the things I dont like about the current ranking system is that its much more based on how often you play than how skilled you are. I think that should be switched. Currently anyone who plays regularly in ladder teamers can get into the top 20 nearly every night, but then be absent for a couple days, and your rank jumps into the hundreds. I like the idea that duels and ctons are going to be given the same creedence as teamers with the new system, I think this will expand the top 20 to include a lot of good players who otherwise never show up, and encourage people to try different sorts of games as well. Second, how are non-standard teamers going to be worked out? For example, if someone is playing a 3v3v3 or a 2v2v2v2, how will those be tracked, is there a way to implement a non-standard teamer into the new ladder? Also, I agree that the possibility of losing skill while winning a game, or gaining skill while losing a game should be eliminated. You might gain very little for beating someone much lower ranked than you, or lose very little for losing to someone much higher skilled than you, but the possibility of actually losing skill while winning or gaining skill while losing will deter a lot of games, and create a situation where people will even more so weed who they are willing to play with. Not an ideal situation.
|
|
|
Post by Canucksoldier on Nov 7, 2007 1:23:47 GMT -5
Just wanted to clarify a few things here. First, how exactly would the skill depreciation work when someone is off line? Does skill depreciation begin at 24 hours? 48, a week? My personal opinion is that it should take at least a few days for skill depreciation to begin, and then be very mild. One of the things I dont like about the current ranking system is that its much more based on how often you play than how skilled you are. I think that should be switched. Currently anyone who plays regularly in ladder teamers can get into the top 20 nearly every night, but then be absent for a couple days, and your rank jumps into the hundreds. I like the idea that duels and ctons are going to be given the same creedence as teamers with the new system, I think this will expand the top 20 to include a lot of good players who otherwise never show up, and encourage people to try different sorts of games as well. Second, how are non-standard teamers going to be worked out? For example, if someone is playing a 3v3v3 or a 2v2v2v2, how will those be tracked, is there a way to implement a non-standard teamer into the new ladder? Also, I agree that the possibility of losing skill while winning a game, or gaining skill while losing a game should be eliminated. You might gain very little for beating someone much lower ranked than you, or lose very little for losing to someone much higher skilled than you, but the possibility of actually losing skill while winning or gaining skill while losing will deter a lot of games, and create a situation where people will even more so weed who they are willing to play with. Not an ideal situation. The skill depreciation system has yet to be formulated, but we do plan to balance it so that neither will there be huge swings in rank as there is in Cases, but also that we won't have the top ranks stagnant with absent players. Multi-team games are not difficult, they will be simply calculated like a cton, with the teams splitting the points at the end. CS
|
|