Posted on 22 August 2017 - 11:07 PM

I've been thinking about this post for a while before making it, and I'm glad I waited. Initially I was going to talk about how the most important thing about a rating system is how enjoyable it is, and how MMR is a great theoretical rating system, but it isn't enjoyable.

But let's be honest here - MMR is an awful theoretical rating system for minecraft 1v1s, (I can back that up in another post if people are interested) and it's easy to predict how enjoyable a rating system will be using its theory.

Also, before I go any further, Ny_ has a very informative thread on how elo works for those of you who aren't familiar: https://www.badlion.net/forum/thread/184602
To summarize, the k value determines how "swingy" the rating system is. A higher k value means you lose and gain more elo on average per match, while a lower k value means the reverse. The lower the k value, the more "grindy" a rating system is, the higher the k value, the more consistency is rewarded and inconsistency is punished.

For reference, old Badlion elo has a k value of 32, new Badlion elo has a k value of 16, and Ultra's k value appears to be somewhere around 50.

Badlion's big issue is that it's trying to have its cake and eat it too when it comes to switching back to elo, and in the process it seems to be forgetting that it has ladders other than BuildUHC. Let me back that up a bit. This rating system is better than MMR in that it restores the single number, rather than rank, way of denoting rating, and it no longer has the promotion/demotion match system which made little to no sense. It also restores the 1:1 rating tradeoff, which helps reduce inflation. Besides that, though, the system is very similar to MMR in terms of how quickly you gain rating.

Badlion's new elo system is perfect for a ladder like BuildUHC - you never have to wait long to get a match, matches don't take that long, and it's not all that easy to be incredibly consistent. However, for a ladder like Iron, where it takes a year and a day to get a match, getting half as much rating from a match is a huge negative. Similarly, ladders like Debuff where it both takes a long time to find a match AND matches take a long time this rating system is a nightmare. Who wants to wait 5 minutes for a 2 minute debuff match only to gain 8 elo?

So what am I getting at here? This rating system further skews the rating system to popular ladders, like Build. Build works superbly with a k value of 16, but a ladder like debuff would do better with a k value of 40. There will never be a perfect rating system that works well for all of Badlion's ladders, but the old k value of 32 was a nice compromise (unless there was a different rating system for each ladder, which causes a whole host of new problems).

And finally, the low k value doesn't punish inconsistency as harshly as a higher k value, so the highest ranked players on Badlion this season could very easily become the players who play the most, similar to how some #1 players last season (10k masters archer guy, no offense) weren't necessarily the best player but had a huge amount of time to play. To put it simply, the k value is how many matches you need to win per match you lose to never stop gaining elo, and while a comparatively large amount of people win 16 matches per one they lose, very few win 32 per match they lose.

This was a copy-paste from a twitlonger I made (with the ultra bit cropped out because it's not relevant here).

I hope that the owners do consider going back to the old elo system.

Last edited on 23 August 2017 - 02:00 AM by xSah

+2000 anyone who likes the new elo just couldn't do well with the other elo or are just another mmr peasent xdd

Posted on 23 August 2017 - 02:04 AM

It's also a lot more fun to gain higher numbers, gaining +8 just isn't fun :(

Last edited on 23 August 2017 - 02:06 AM by Genocado

I liked s12 elo better, s15 is more of a grind. but s12 in my opinion is more risky when your higher up

Someone be ballsy and tag archy :v

Posted on 23 August 2017 - 02:15 AM

I agree, however, 32k value is too high as well for high rated players in unplayed ladders, loosing a match with the 32k rating means that you have to win 31 matches in a row, which can take around 2-3 hours (if you constantly fight low elo players). The 16k system does make more sense for higher rated players but not for lower rated players because gaining a small amount of points once you just start playing a ladder isn't motivating, you do not get the satisfaction of draining a higher rated plater. The 23k value would perform better for both low and high rated players in my opinion

Posted on 23 August 2017 - 04:08 AM

It is definetly harder to rank up in less played ladders but I prefer the current system to waiting in the q for 5 mins playing a 2 min match and loosing 31 elo to an alt

Posted on 23 August 2017 - 11:05 AM

@Archybot

Posted on 23 August 2017 - 11:47 AM

Having to win 31 matches in a row per loss is exactly what makes this elo so good - it forces top level players to be ruthlessly consistent. Still, a 24k system would be a decent compromise.

Posted on 23 August 2017 - 11:50 AM

I dunno, engineering your rating system on the basis of "what minimizes the damage from elo sniping most" seems like a bad idea when Badlion could nearly eliminate sniping with some rules changes. I do see your point though.

Posted on 23 August 2017 - 11:51 AM

@MasterGberry as well, sorry for the tags but I think I make good points here so I'd be interested in your response.

Posted on 23 August 2017 - 12:42 PM

I'm not getting involved here. The community has already voted here: https://twitter.com/archyb0t/status/900077539938496512

Last edited on 23 August 2017 - 05:57 PM by ThatOneCombo

i guess we r going back to 31 now

Posted on 23 August 2017 - 08:07 PM

how about 12k?

Posted on 23 August 2017 - 09:24 PM

I assume you mean 24k - 12k would be ~6 elo per match and even grindier than right now.

Last edited on 23 August 2017 - 11:43 PM by Archybot

Having to win 31 matches in a row also is exactly why most top players stop playing once they reach their goal instead of pushing on, especially in ladders that have higher hacker to legit ratio, like nodebuff, or ladders that take long matches, such as gapple.

Posted on 24 August 2017 - 01:07 AM

I agree, if you compare season 12 matches played vs season 14 matches played, you can see that the top ranked players on each ladder played 5-6x more matches to maintain the top positions on the leaderboards, this was because you gained 13 elo each win and only lost 214 max which is equivalent to 16 elo anyways 14 x 8 = 224 so basically season 14's gain/loss ratio is very similar to this seasons elo system, the only difference is that you dont get promotion matches/demotion matches. Losing a lot of elo is when you are highly ranked is very demotivating

Posted on 24 August 2017 - 08:58 AM

Initially I thought I'd favor the old 32 k-value elo, but after playing a few hundred ranked iron matches I had a change of heart. Because bow ladders are relatively difficult to be consistent at, I realized that the 16 k-value is quite beneficial for bow based players like myself. Just recently I lost to a 1450 elo player 4 consecutive times, then beat him 7 times in a row (not sure of the exact numbers). Through all this I only lost around 35-40 elo, which is not that much considering I lost 4 straight times to someone I should be beating with consistency.

Posted on 24 August 2017 - 11:20 AM

Well there's also the fact that this season was long.

Last edited on 24 August 2017 - 12:32 PM by Icehawk

Didn't stop players like @ravenatorman, @iHackLiTE, or @Antonathor in S14.

Edit: Yes, it may have been a rating system where you could have infinitely gained points off of anyone. However, players like raven and antonather (LongRangeRandom) still kept going even after a point in the Masters Ranking that it was obnoxious. It might work out if Archy DOES change it back to the original ELO system.

Posted on 24 August 2017 - 02:43 PM

How about 0 elo gained and 0 elo lost.