Loading...
  OR  Zero-K Name:    Password:   

Post edit history

Design a Human-Friendly Ladder

To display differences between versions, select one or more edits in the list using checkboxes and click "diff selected"
Post edit history
Date Editor Before After
3/7/2019 1:09:21 PMCHrankAdminDeinFreund before revert after revert
Before After
1 There are quite a few requirements that have been set for the rating, so let me just quickly give an overview: 1 There are quite a few requirements that have been set for the rating, so let me just quickly give an overview:
2 \n 2 \n
3 == Game balance and prediction == 3 == Game balance and prediction ==
4 The rating system should be designed such that it can make the best possible predictions and balance. Players are often subject to daily variations, which is what the rating tries to adapt to, especially with the latest update. Keep in mind that the goal of this rating number is solely to predict one future match accurately, not represent past achievements or trends. 4 The rating system should be designed such that it can make the best possible predictions and balance. Players are often subject to daily variations, which is what the rating tries to adapt to, especially with the latest update. Keep in mind that the goal of this rating number is solely to predict one future match accurately, not represent past achievements or trends.
5 \n 5 \n
6 == Ladder stability == 6 == Ladder stability ==
7 The ladders should have a stable rating that represents long-time progress. As the ladders have a monthly cut-off, it'd make sense to use a monthly aggregation. I've described that in more detail above, and as I've heard until now the simple average would be the preferred solution. 7 The ladders should have a stable rating that represents long-time progress. As the ladders have a monthly cut-off, it'd make sense to use a monthly aggregation. I've described that in more detail above, and as I've heard until now the simple average would be the preferred solution.
8 \n 8 \n
9 == Ranks/Leagues == 9 == Ranks/Leagues ==
10 Ranks are used to give the player an easily comprehensible representation of their skill progress. I think this is where GoogleFrog's points should apply. Ranks are designed such that they can only be gained on a victory and lost on a defeat. They also have overlapping ranges, meaning that the threshold to gain a rank is higher than the one to lose it. This overlap is dynamically calculated from the rating variance of the player, meaning it'll be harder to both gain and lose ranks if your rating is actively changing. This is such that, independent of the configuration of WHR, ranks should remain stable. 10 Ranks are used to give the player an easily comprehensible representation of their skill progress. I think this is where GoogleFrog's points should apply. Ranks are designed such that they can only be gained on a victory and lost on a defeat. They also have overlapping ranges, meaning that the threshold to gain a rank is higher than the one to lose it. This overlap is dynamically calculated from the rating variance of the player, meaning it'll be harder to both gain and lose ranks if your rating is actively changing. This is such that, independent of the configuration of WHR, ranks should remain stable.
11 \n 11 \n
12 == Rating categories == 12 == Rating categories ==
13 There are often requests to separate the rating of different game types, with the current compromise being one rating for MM and one for non-MM (casual) games. As there is a separation in the player base with some players only playing one of these modes, both modes come with their own ranks. 13 There are often requests to separate the rating of different game types, with the current compromise being one rating for MM and one for non-MM (casual) games. As there is a separation in the player base with some players only playing one of these modes, both modes come with their own ranks.
14 \n 14 \n
15 This is currently implemented as giving out ranks to a certain percentage of the active players in each mode, which often leads to confusion as some players gain their rank in one mode where they wouldn't deserve that rank in the other mode. (E.g. Sparkles=blue, Freund=purple). A possible change would be to give out the same number of ranks in both modes, meaning that even the lowest rated MM player would already gain one of the higher ranks. 15 This is currently implemented as giving out ranks to a certain percentage of the active players in each mode, which often leads to confusion as some players gain their rank in one mode where they wouldn't deserve that rank in the other mode. (E.g. Sparkles=blue, Freund=purple). A possible change would be to give out the same number of ranks in both modes, meaning that even the lowest rated MM player would already gain one of the higher ranks.
16 \n 16 \n
17 == Decayed and True rating == 17 == Decayed and True rating ==
18 There are two different ratings in use, a true rating and a decayed rating. The decayed rating will decrease whenever somebody doesn't play. It also starts at maximum decay for newbies, giving them an effective -400 elo offset. The goal of the decayed rating is to give (unbalanced) easier matches to teams with newbies and inactive players on them. The decayed rating is implemented in WHR as a lower bound, minimum confidence interval. This means that it'll not only give a lower rating to inactive players, but also to inconsistent players. 18 There are two different ratings in use, a true rating and a decayed rating. The decayed rating will decrease whenever somebody doesn't play. It also starts at maximum decay for newbies, giving them an effective -400 elo offset. The goal of the decayed rating is to give (unbalanced) easier matches to teams with newbies and inactive players on them. The decayed rating is implemented in WHR as a lower bound, minimum confidence interval. This means that it'll not only give a lower rating to inactive players, but also to inconsistent players.
19 \n 19 \n
20 The decayed rating is currently used for balance, matchmaking and the ladders. The true rating is currently used for game predictions. 20 The decayed rating is currently used for balance, matchmaking and the ladders. The true rating is currently used for game predictions.
21 \n 21 \n
22 == Issues == 22 == Issues ==
23 * The first and the last point seem to contradict each other. I'm not too fond of intentionally unbalancing games. 23 * The first and the last point seem to contradict each other. I'm not too fond of intentionally unbalancing games.
24 * Ladders use the decayed rating, which is meant to balance future games, not represent achievements. 24 * Ladders use the decayed rating, which is meant to balance future games, not represent achievements.
25 * Ranks are possibly not yet fulfilling their requirements, some look at elo over ranks(?) 25 * Ranks are possibly not yet fulfilling their requirements, some look at elo over ranks(?)
26 * The use of four different rating numbers (decayed vs true, competitive vs casual) is confusing. 26 * The use of four different rating numbers (decayed vs true, competitive vs casual) is confusing.
27 \n
28 == Planned changes ==
29 * Smooth out the ladder rankings by averaging skill over the last month
30 * Simplify the ladder cut-off by ranking everyone who has played in the last month.
31 \n
32 @GoogleFrog I'm not too sure about the progress indicator. With the wildly different gamemodes and whack MM balance it could take one or a hundred wins to rank up. I don't think this can be displayed in a simple way. Some days one can get only matched against Godde where every win would be a rank up, on other days all matches are against newcomers where wins don't mean anything. Unless we have reliable "in-league" matchmaking, rank changes are going to vary wildly.
33 \n
34 @malric Everyone should get a rank with the change. If you want to see who's nearby, check the [url=http://zero-k.info/Ladders/Full]full ladders[/url]. I don't think the ladders should be the go-to place to check your rough rank though, that's what the rank icons are for. Especially at the more densely populated ranks, ladder position can vary wildly.