On March 06 2013 04:39 krzych113 wrote: why isn't iccup ranking a ranking of elo ?
I'd think that would be a good question for ICCup staff, no? Not sure they have ever explained why they use the methods they do.
I don't answer on that stuff anymore, it's pointless. No matter what you write, people will start to argue against you. It's like being the messenger and being nuked.
Other than this, I like this topic quite a lot.
Ah, that's a shame (that people attack you either way). I've never actually heard the reasoning though, but from what you say it sounds like ICCup has it reasons.
Makes me wonder if their could ever be a system like Birdie's used for match-making, and then the main ICCup ranking for determing ladder ranks and champions?
The MOTW System is a copy of the PGTour system, a lot of the senior admins worked on both portals. It was the in-thing the new thing really, compared to what WGTour and Neo Game-i used. One was more or less similar to ICCup, the other was ELO. Why they changed it, dunno. The advantage is that you can easily assign MOTWs, which in theory should people get to play different maps. That the majority plays Python and Fighting Spirit regardless of what we set up is a different topic. It would, theoretically, be possible to add MOTWs for an ELO based system (or Glicko), using different weights - however, that'd be a delicate issue in my eyes. Or you could remove and add maps to the pool per week, but that'd erase the option to play on non-MOTWs for casuals, which again is quite bad.
Then there's the issue - even if the super admins and the head admins would agree on an ELO or Glicko based system - we have no access to the source code or the server. The devs are busy to fix their shit on DotA and WoW, we're glad to get the most important fixes in a short time frame. Note, only very few tech people are left over and these are already old and thus have limited time. Also, for some reason, YelloAnt turns down many non-dev people. Up to some point I get that, other times I don't. You have to earn his trust first, apparently. This slows down a lot of things and we're basically always in the middle between the users and the higher ups.
I mean, I don't want to blame the guys that still keep up the server, despite it being very, very small compared to all the other games on the portal. We are only about 2% of the entire user basis. So I have a hard time to blame the devs too much. I can't force the new people to care for our game, even though I often tried to talk to them. The only upside is that Ant, Unk and x64 really won't shut down the portal until everyone leaves. "weeeee"
MOTW wasn't some "new in-thing" at all developed by PGT. Cloud/Cloria Ladder, which existed around the time WGTour did (while WGT was still French), used MOTW as well. The only difference is that those only used 1 or 2 (to ensure that there was a map for both 1v1 and 2v2).
I think we should go back to one MOTW map, you can only play on it, get ELO system and get TheShimmy to choose MOTW.
While this is interesting, I still believe ICCup's ranking system works better for BW.
ICCup ranks are quite stable and when someone reaches their max rank, they'll be stuck on it as long as they don't actually improve their game. For example, when someone tells me their rank max rank is B- I can quite accurately asses their BW skill and what to expect from their game. And when you take into consideration that there is big difference even between B- and B+, you'd notice that there are many levels of BW skill and they need to be represented accordingly.
For example, I think Fish's ELO ranking is horrible for BW as it's too dynamic. Ask any top foreigner what is their average ranking on Fish. I doubt they'll be able to say accurately as it sometimes spikes to 1400-1500 and sometimes they're stuck on 1100. It's because even few win/loss streaks change your ranking considerably. This doesn't happen with ICCup's ranking. On ICCup you're almost guaranteed to ladder up to your max rank and then play against opponents of the same level (too bad this is not accurate anymore due to so few people playing above C+ rank ).
Now I'm not sure how much Glicko's ranking system is different from ELO ranking, but I seriously don't see what's so inaccurate about ICCup's ranking system. Especially since it has proven itself to work for BW over many years.
Cool stuff. You might be interested in isotropic Dominion's leaderboard (http://dominion.isotropic.org/faq/#leaderboard), which uses an implementation of TrueSkill that has been open sourced on GitHub.
To answer 2Pacalypse's questions above, iCCup's rating system arguably takes too long to converge to be effective for match-making, especially since ratings are reset every season. I guess it may work well enough for players that play a ton of games per season, but it's not great for playing occasionally.
Systems like Glicko and TrueSkill are better for stability than ELO, because if you play a lot, then your rating uncertainty becomes low. Then your rating won't move much with each game. There are also some tweakable parameters involved, so Fish may simply have misconfigured their rating system.
On March 06 2013 04:53 ninini wrote: This ranking will keep all data and rankings across new iccup seasons? If yes, awesome!
It can do; I can also reset it independently. I haven't actually set up the code to automatically handle new iCCup seasons; it's pretty simple to do though.
On March 06 2013 05:07 LML wrote: please fix the font or the background image.. but it's unreadable if you scroll down ;/ edit: this occures around rank 960 or so, the main frame just stops there as if it has a max-height
What browser are you using? It works fine for me on Google Chrome.
On March 06 2013 05:18 fabiano wrote: Is the ranking calculated per season? Or does it take into consideration previous seasons?
That's up to me At the moment it only has data from the new season.
On March 06 2013 05:56 dRaW wrote: Cool but this is still really inaccurate. It's a ladder system so you can really use elo systems with so many unknowns.
It's very accurate after players have played enough games; it reaches a high level of certainty much quicker than ELO, for example. It's not a ladder system, it's a points ranking with the points representing skill.
On March 06 2013 08:21 2Pacalypse- wrote: While this is interesting, I still believe ICCup's ranking system works better for BW.
ICCup ranks are quite stable and when someone reaches their max rank, they'll be stuck on it as long as they don't actually improve their game. For example, when someone tells me their rank max rank is B- I can quite accurately asses their BW skill and what to expect from their game. And when you take into consideration that there is big difference even between B- and B+, you'd notice that there are many levels of BW skill and they need to be represented accordingly.
For example, I think Fish's ELO ranking is horrible for BW as it's too dynamic. Ask any top foreigner what is their average ranking on Fish. I doubt they'll be able to say accurately as it sometimes spikes to 1400-1500 and sometimes they're stuck on 1100. It's because even few win/loss streaks change your ranking considerably. This doesn't happen with ICCup's ranking. On ICCup you're almost guaranteed to ladder up to your max rank and then play against opponents of the same level (too bad this is not accurate anymore due to so few people playing above C+ rank ).
Now I'm not sure how much Glicko's ranking system is different from ELO ranking, but I seriously don't see what's so inaccurate about ICCup's ranking system. Especially since it has proven itself to work for BW over many years.
Glicko2 is also quite stable. Fish's system is, as you said, unstable/dynamic, because it doesn't have a confidence rating. With Glicko2, the more games played, the more confident the system is in the player's skill, and the slower the points change. When the system is 99% confident that your skill is 1980, then if you lose 10 games to a 1500 player you might only drop 80 points or so. If the system is 50% confident your skill is 1980 and you lose the same 10 games you might drop 400 points (EXAMPLES ONLY, I didn't use actual values). So the first few games you play, your ranking will jump all over the place until the system has enough games stored that it can be more confident about your skill.
blueblimp explained this somewhat too. Blueblimp, thanks for your link, I may consider it but I already have a PHP implementation of the Glicko2 system and am too lazy to learn Python just yet/convert the TrueSkill one to PHP. The +/- display of the deviation is good, I'll try implement that so people can see how confident the system is.
On March 06 2013 05:56 dRaW wrote: Cool but this is still really inaccurate. It's a ladder system so you can really use elo systems with so many unknowns.
Well, there are known knowns and there are known unknowns and then there are unknown unknowns; things we don't know that we don't know.
On March 06 2013 13:53 Savant wrote: This great thanks. It would also be great if you could make the rankings sortable by category. Shouldn't be too hard right?
What do you mean, by category? You mean by matches played, matches won, and so on?
On March 06 2013 14:38 quirinus wrote: Seems nice, how often does it update?
Hourly, starts the update at about 50 minutes past the hour and takes a couple minutes to update. So it should be completely updated every hour on the hour.
On March 06 2013 04:39 krzych113 wrote: why isn't iccup ranking a ranking of elo ?
I'd think that would be a good question for ICCup staff, no? Not sure they have ever explained why they use the methods they do.
I don't answer on that stuff anymore, it's pointless. No matter what you write, people will start to argue against you. It's like being the messenger and being nuked.
Other than this, I like this topic quite a lot.
Ah, that's a shame (that people attack you either way). I've never actually heard the reasoning though, but from what you say it sounds like ICCup has it reasons.
Makes me wonder if their could ever be a system like Birdie's used for match-making, and then the main ICCup ranking for determing ladder ranks and champions?
The MOTW System is a copy of the PGTour system, a lot of the senior admins worked on both portals. It was the in-thing the new thing really, compared to what WGTour and Neo Game-i used. One was more or less similar to ICCup, the other was ELO. Why they changed it, dunno. The advantage is that you can easily assign MOTWs, which in theory should people get to play different maps. That the majority plays Python and Fighting Spirit regardless of what we set up is a different topic. It would, theoretically, be possible to add MOTWs for an ELO based system (or Glicko), using different weights - however, that'd be a delicate issue in my eyes. Or you could remove and add maps to the pool per week, but that'd erase the option to play on non-MOTWs for casuals, which again is quite bad.
Then there's the issue - even if the super admins and the head admins would agree on an ELO or Glicko based system - we have no access to the source code or the server. The devs are busy to fix their shit on DotA and WoW, we're glad to get the most important fixes in a short time frame. Note, only very few tech people are left over and these are already old and thus have limited time. Also, for some reason, YelloAnt turns down many non-dev people. Up to some point I get that, other times I don't. You have to earn his trust first, apparently. This slows down a lot of things and we're basically always in the middle between the users and the higher ups.
I mean, I don't want to blame the guys that still keep up the server, despite it being very, very small compared to all the other games on the portal. We are only about 2% of the entire user basis. So I have a hard time to blame the devs too much. I can't force the new people to care for our game, even though I often tried to talk to them. The only upside is that Ant, Unk and x64 really won't shut down the portal until everyone leaves. "weeeee"
MOTW wasn't some "new in-thing" at all developed by PGT. Cloud/Cloria Ladder, which existed around the time WGTour did (while WGT was still French), used MOTW as well. The only difference is that those only used 1 or 2 (to ensure that there was a map for both 1v1 and 2v2).
I meant more like popular not in terms of revolutionary. There's nothing new under the sun.
On March 06 2013 08:21 2Pacalypse- wrote: ...
The point with ICCup system and Elo system is that it goes up and down with your user base. Both have big problems, whenever you look at casual players, new accounts naturally cause a bias, etc. Especially Elo isn't used like it was originally inteded, the real Elo Number, from what I read at least, judges a number of games, not a single one. E.g. Fish calculates "in real time" - you lose immediately points after a game, while the Wiki page of Elo reads (I might be wrong) that in Chess the entire series is calculated and you hence lose fewer points or win fewer points if you play more than one game against the same opponent, depending on the outcome. Glicko is really "better", because it wasn't made for a game that is played in real life, has no ties (again not sure) and has more or less the thoughts of ladders implemented.
On March 06 2013 09:50 blueblimp wrote: To answer 2Pacalypse's questions above, iCCup's rating system arguably takes too long to converge to be effective for match-making, especially since ratings are reset every season. I guess it may work well enough for players that play a ton of games per season, but it's not great for playing occasionally.
Systems like Glicko and TrueSkill are better for stability than ELO, because if you play a lot, then your rating uncertainty becomes low. Then your rating won't move much with each game. There are also some tweakable parameters involved, so Fish may simply have misconfigured their rating system.
Yes, I agree that it might take too long for ICCup's ranking to converge in the beginning, but that's what they should implement the system they use on their Dota ladder. If you were at 6000-9000 (B- to B+) points last season, you start new season with 2000 points and if you were 9000+ (A- and above) you start new season with 3000 points.
But alas, I'll agree that ICCup's system might not be ideal anymore since there are so few people playing it.
On March 06 2013 08:21 2Pacalypse- wrote: While this is interesting, I still believe ICCup's ranking system works better for BW.
ICCup ranks are quite stable and when someone reaches their max rank, they'll be stuck on it as long as they don't actually improve their game. For example, when someone tells me their rank max rank is B- I can quite accurately asses their BW skill and what to expect from their game. And when you take into consideration that there is big difference even between B- and B+, you'd notice that there are many levels of BW skill and they need to be represented accordingly.
For example, I think Fish's ELO ranking is horrible for BW as it's too dynamic. Ask any top foreigner what is their average ranking on Fish. I doubt they'll be able to say accurately as it sometimes spikes to 1400-1500 and sometimes they're stuck on 1100. It's because even few win/loss streaks change your ranking considerably. This doesn't happen with ICCup's ranking. On ICCup you're almost guaranteed to ladder up to your max rank and then play against opponents of the same level (too bad this is not accurate anymore due to so few people playing above C+ rank ).
Now I'm not sure how much Glicko's ranking system is different from ELO ranking, but I seriously don't see what's so inaccurate about ICCup's ranking system. Especially since it has proven itself to work for BW over many years.
Glicko2 is also quite stable. Fish's system is, as you said, unstable/dynamic, because it doesn't have a confidence rating. With Glicko2, the more games played, the more confident the system is in the player's skill, and the slower the points change. When the system is 99% confident that your skill is 1980, then if you lose 10 games to a 1500 player you might only drop 80 points or so. If the system is 50% confident your skill is 1980 and you lose the same 10 games you might drop 400 points (EXAMPLES ONLY, I didn't use actual values). So the first few games you play, your ranking will jump all over the place until the system has enough games stored that it can be more confident about your skill.
Ah, I see. Then indeed, I'll agree that Glicko2 system is pretty good. Though I still prefer seeing those colored grade icons instead of numbers, so you could maybe make your own representation of rating in Glicko2 system by using letters as well
edit: also, you should make some pagination on the site
On March 06 2013 16:55 Sayle wrote: Could you make the player names into links to their ICCup profiles?
Done.
On March 06 2013 17:23 2Pacalypse- wrote: edit: also, you should make some pagination on the site
Will do, if and when I add in search capability. I'll add in iCCup-style letter rankings once there are more matches ranked so that it will seem roughly similar to the actual iCCup rankings.
I'm pretty confident there is a somewhat major bug in there somewhere, which I'm looking for. Basically there are matches being added which are NOT 1v1s on official maps, so I'll have to find out where those matches are coming from and why my scraper is adding them when it shouldn't be.
EPT
At the moment it only has data from the new season. 
