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

Forum index  > News   >

Ladder rework

17 posts, 2495 views
Post comment
Filter:    Player:  
sort

4 years ago
Ladder ratings have been modified to be more intuitive. Now your ladder rating will always increase when you win a game and decrease when a game is lost. The rank brackets are now global and equal for all users. Once you reach a 20% threshold in the next bracket, you will rank up. Once you go more than 20% below your current bracket, you will rank down. Ratings will change less between games.

The way this is implemented is with two major changes. Firstly, ladder rating is only updated playing a game. So there will be no rating changes when you're not playing and it will never go opposite to whether you won/lost. Additionally, the rating changes use a moving average to smooth out changes. Secondly, the target rating for the ladder rating is now a monthly average. This means that instead of just looking at your most recent rating, your rating over the whole past month will be averaged, weighted by the number of games played each day. This should make for more stable monthly ladders.

There's also been a fix for people who haven't played within the last 30 days still being displayed on the ladders. This means that there will be less rated players, and thus less purple ranks (top 1%).
+8 / -0

4 years ago
PS: I'm sorry for all the rank disruption yesterday, ranks should work as intended now. Please report further issues.
+2 / -0

4 years ago
Only 30 days?
Ahhh I seee...
So that you could become top 3. I got you bro
+3 / -0
Maybe it would be a good idea to put the above explanations in https://zero-k.info/Charts/Ratings ?

I also assume that until everybody plays a game, the colors/next/previous level are borked, is that correct?
+0 / -0

4 years ago
Yes, for those who had their ranks changed it will require some games to converge to their correct rank. You'll see either "Rank up next game" or 0% rank progress on your user page if there's an impending change.
+0 / -0


4 years ago
Nicely done.
+1 / -0
Wait, so if I were to become active on the MM ladder again (assuming my MM rating didn't change significantly in the games played to regain active status) right now I'd get purple rank? That... feels wrong...
+0 / -0

4 years ago
No, you probably would uprank to light blue/silver though.
+0 / -0
4 years ago
i want to drop to blue!
+0 / -0

4 years ago
This makes no sense. Now two people in the top 15 have blue rank badges when everyone else at that level is purple. That's directly contrary to the stated nature of the colors, and also the opposite of intuitive.
+0 / -0

4 years ago
That mostly stems from the lack of 1v1s played. People will only de-rank after playing a game, and the color is also not dependant on your absolute ladder position on only one of the ladders. AUrankAdminGoogleFrog who is currently a single place below me on the 1v1 ladder appparently hasn't played a multiplayer game in quite some time, where as I've jammed out many teamgames yesterday. Looking at the teams ladder, the only purples that exist below me haven't logged on since a day or 2 before the update, so as soon as they play againn their colors will change. Within 30 days of people dropping from the ladders due to inactivity, or playing a game under the new system, the colors will even out(I assume). I guess there may be strange things, like a player may only have a high 1v1 rank, but positioned #45 in teams due to infrequent play long ago. that may have a rank 45 team player being purple. viceversa, it may be the case the other way. Often, though, players have similar ladder positions in both ladders, so it's incredibly extreem to say that this color shift will ever stabalize more than a couple rank positions.

Unless....
quote:
CHrankAdminDeinFreund,
The rank brackets are now global and equal for all users.

Does this mean that 1v1 and casual now use the same WHR? I don't know what this means.
+0 / -0

4 years ago
That's kind of ridiculous. Either you can only rank down after playing a game, or you drop ranks with inactivity, in which case you could just take the lower bound of variance and get the same effect. But you can't have both. Also why sacrifice WHR's accuracy for.. what was the benefit of this again?

If you want to make ratings more human readable then just rescale the visible ratings to a 0-1000 range where 0 is the worst player and 1000 is the best. Then rating simply corresponds to a 3-digit percentile standing, with actual (top player) rank being calculated by the raw, non-human-readable WHR rating. A 3-digit percentile rating also corresponds directly to badge colors.

By contrast neither WHR nor ELO really have much significance to the average person, and you can't even tell how high either one goes without looking at the top ranked player. With a percentile-based visible rating you always know that if your rating is ~500 then you're average skill, and you get a precise measure of where you stand in the (active) community with a glance.

Another cool thing about WHR is that it would allow us to have seasons very easily, by rolling previous seasons into unranked when they end. It's no problem when you're recalculating from scratch periodically. I really think we should wipe awards with every season though. Awards become especially meaningless once you have 9001 of them.
+0 / -0
CHrankAdminDeinFreund thank you for your efforts.

quote:
rank disruption yesterday
you cannot get online ONE SINGLE DAY and thats when the crazy shit is going down. what did i miss?
+2 / -0
USrankaeonios This problem with colors remains as long as we value matchmaking and casual ladders equally. To be in the top 1% of MM you have to be in the top 2, for casual in the top 14. Thus the majority of players are going to get their purple badge in casual. If we wanted the MM ladder to be as smooth as the casual we'd have to give it easier percentiles.

Rank or rating decay during inactivity is something I explicitly avoided, as it would add the dreaded rating change without playing right back in. Instead I'd rather just remove inactive players from the ladders.

As for that 3 digit rating you're suggesting. The percentiles are already calculated, so we could just display them instead of Elo numbers.
+0 / -0
quote:
This problem with colors remains as long as we value matchmaking and casual ladders equally. To be in the top 1% of MM you have to be in the top 2, for casual in the top 14. Thus the majority of players are going to get their purple badge in casual. If we wanted the MM ladder to be as smooth as the casual we'd have to give it easier percentiles.

Well you have basically two purposes to ratings and badge colors:

1. To be able to tell how strong a player is when you end up in a game against them or on their team.
2. Bragging rights.

Honestly I don't think it should be "MM vs Casual" but rather "MM vs All". In other words, MM games count towards "All" but not the other way around. That way you have at least one rating that's as accurate as possible across all games played on featured maps and modes (maybe even supported, as long as that doesn't include crap like speed metal), while MM is all about bragging rights. In that case bragging rights would need its own badge thingy, or as an extra decoration to the normal badge so that it doesn't take up more space in the user list.

You could also use somewhat different scaling for MM, like purple: top 10, blue: top 50, and maybe the other colors could be the same using percentiles.

quote:
Rank or rating decay during inactivity is something I explicitly avoided, as it would add the dreaded rating change without playing right back in. Instead I'd rather just remove inactive players from the ladders.

Well no argument there, as long as it's consistent.

I still think the problem with ratings changing even if you don't play is just that your rating changes from one mostly-meaningless number to another. Even when you look at the graph on the ratings page it's scaled to your own ratings changes so you can only compare to yourself. If the scale were fixed at 0-1000, with lines showing each badge color threshhold, it'd be a lot clearer wrt what you're looking at and what it means. As it is you can't just go "oh my rating changed because the community is getting stronger/weaker", it's just "oh my ratings changed and wtf does that even mean?".

Using a percentile-based rating the offline changes also ought to be smaller in general, and thus less alarming. It would feel a lot more like "oh it's just making small corrections" rather than "wtf why did I drop 100 points?".

quote:
As for that 3 digit rating you're suggesting. The percentiles are already calculated, so we could just display them instead of Elo numbers.

You'd still have to make a small change to ensure that the top player (and only the top player) gets a rating of 1000, and vice versa for the worst player and 0. Either way it doesn't sound like it'd be a lot of work.

How do inactive players count wrt the percentile calculation though? IMO if you get dropped from the ladder you shouldn't be counted except maybe for the non-visible WHR rating.
+0 / -0

4 years ago
Percentile is only calculated among the active players (played in last 30 days). Unranked players don't get a percentile.

quote:
"MM vs All"

That's exactly what I started out with, but there was protest to separate the ratings. Some, like RUrankizirayd even ask for FFA ladders.

If MM was to receive easier ranks than casual I'd just go with equal amounts of players per rank for each ladder. So if there are 14 purples from casual there'd also be 14 competitive purples. This would mean that the lowest ranked competitive player would already get somewhere around a yellow rank. This would be equivalent to setting all players that don't play competitive to 0 competitive rating. Needless to say I don't think this would improve the situation. In the end there should be just one rating that counts.
+0 / -0
quote:
Percentile is only calculated among the active players (played in last 30 days). Unranked players don't get a percentile.

Well that's even less work then. :P

quote:
That's exactly what I started out with, but there was protest to separate the ratings. Some, like RUrankizirayd even ask for FFA ladders.

Meh, I guess that was an issue since MM started. Treating MM as special only causes problems compared to how it used to work where any qualifying game counted towards rating. Splitting the ladders organically according to game type (1v1, Teams, FFA) makes a lot more sense than MM vs everything else, and you can still have an overall rating that counts all qualifying games of all types.

Rating is supposed to be a measure of skill, and that breaks down when you make arbitrary restrictions on what counts vs what doesn't. I mean why should a high rating 1v1 on public hosts count differently than a high rating 1v1 under MM? MM was supposed to be a convenient feature for finding games rather than some dictatorial arena mode.

quote:
If MM was to receive easier ranks than casual I'd just go with equal amounts of players per rank for each ladder. So if there are 14 purples from casual there'd also be 14 competitive purples. This would mean that the lowest ranked competitive player would already get somewhere around a yellow rank.


That's lame. Colors should have meaning, whether it's "xth percentile+" or "top x players". What you're suggesting is more like a participation award.

I mean we have awards for that. Like you could get a "duelist award" for winning competitive 1v1s, a "lobster pot award" for winning competitive teams, and maybe a "chaos general" award for FFA. Better than "here have a free gold star".
+0 / -0