Last edited on 19 August 2017 - 05:25 PM by Ny_

My thread from last week exploded today, and I decided to make a thread revisiting the issue. I talked for a while with Archy about how ELO works, and he's convinced me that the S15 system is reasonable. Both have their pros and cons though, and my goal is to present them in an unbiased way so that we as a community can discuss it reasonably.

How Elo Works

The main component of an Elo system is the k-value. While there's a bunch of math that goes into it, at the end of the day, the k-value is the max amount of Elo you can gain/lose. In Season 12 and before, we used a k-value of 32 (written 32k Elo). This is because in a match with a >719 difference, the top player could either win +0 or lose -32, and vice versa for the bottom player. These matches were blocked in the later seasons because no one wants to play matches for 0 Elo.

So, this season uses 16k Elo - in the most unbalanced match, you could gain +0 or lose -16 (this is blocked on Badlion so it's actually +1/-15). Now that you guys know the terminology, I'll explain the differences.

Season 12 vs Season 15

Obviously it's more than just the extremes that are different. When 2 relatively equal players vs each other, they usually gain/lose 8ish Elo. Before it was 16. This means spreading out takes about twice as long to happen. A player of around 2k level skill may be only 1.6k now, a week into the season, whereas a week into season 12 they would be 1.8k. Here's an example:


It takes longer to get to the same skill level. This will show up a lot in the first month or so of a new season, as players separate out from 1400. The S12 32k Elo system still has some of this, but a lot less as Elo is gained and lost much much faster.

This effect evens out, and after a couple of months you won't notice a huge difference.

—

So that explains the effect on normal players. What about for the top 1% of players? Well, they experience a different effect. More competition.


This shows the top 20 build players from S12 and from the S15 beta. Notice how much closer together the top 20 are this season. This effect stays in place, even after a large number of matches. We can expect many more players at the tail end of the curve, leading to more competition for the top spot, rather than one person running away with it. A big reason for this is because you can only lose 15 Elo at a time in the 16k system, so you only need a 16W/L max to keep gaining. Losing to another top player won't set you back 16 anymore, only 8ish. And you gain 1 from people up to 1000 Elo away from you.

Say for a minute, that S12 Danteh was playing in the new system.

Danteh (2800) vs RandomIdiot (1800) in S12 can't even happen.
Danteh (2800) vs RandomIdiot (1800) in S15 will result in Danteh winning +1 or losing -15.

Danteh (2800) vs RandomMoron (2300) in S12 will result in Danteh winning +2 or losing -30.
Danteh (2800) vs RandomMoron (2300) in S15 will result in Danteh winning +2 or losing -14.

See how this Elo is friendlier to top players? Top players will be able to go higher, and compete with each other better. No longer will you need a ridiculous amount of consistency to gain in the top 10 - you won't have to win all of your games, just most of them.

All this means that the Elo curve will end up looking like this:


—

Pros and Cons

Let's go through the pros and cons of these two systems one last time.

Season 12 / 32k Elo
Pros:
-Fast to gain early in season
-Rewards consistency

Cons:
-Very aggressive (1 mistake costs -31)
-Little competition at the top

Season 15 / 16k Elo
Pros:
-More balanced (mistakes not punished as harshly)
-Top of ladder is very competitive

Cons:
-Slow to gain early in season
-Doesn't punish inconsistency as hard

—

Thanks a ton to Archy for talking this over with me and helping me understand the main differences and how they effect Badlion. I hope we can look at the two systems now as a community and have an informed discussion on which we'd rather use.

Posted on 19 August 2017 - 05:33 PM

This was incredibly useful for someone like me who hasn't looked too far into how ELO systems work and only have a rough idea on the difference of each system. Thanks a ton for making this.

Last edited on 19 August 2017 - 05:35 PM by officialarrow

cool

Posted on 19 August 2017 - 06:01 PM

This is an amazing thread and I completely agree with your opinion. I see why the elo was changed and there are indeed positive aspects about it which you have listed and they may or may not even outweight the negative ones, but after all the new elo system is yet another symptom of Badlion's favourite type of mistake, which is changing something that doesn't require a change.

I've already stressed this multiple times in my thread concerning the S15 arena ladders: in a nutshell the idea of changing something for the good is frequently misunderstood by the Badlion owners - despite their good intentions of course - and these experiments mostly led them to losing players. The best example for that is probably the elo to MMR change.

S1-12 elo had their flaws which are fixed in S15 elo and vice versa. But people wanted to have elo back and with "having elo back" they solely want the S1-12 elo system back, not an altered version. In this case, I believe the vast majority of the players are very clear about their opinion and may even refuse to think twice about the advantages of a changed system. It may suck from the owners' perspective, but you can't blame them. If that is their wish, it should be granted. I honestly don't feel like currently is the right time for experiments.

tl;dr old elo >≠ new elo

Oh and sorry for grammar mistakes, I'm quite tired.

Posted on 19 August 2017 - 06:06 PM

How long did this take you.

Last edited on 19 August 2017 - 06:08 PM by GlueEater22

I agree, my problem with this system is that it rewards time played not actual skill. You can way too much for killing people lower than you, and don't lose enough if they beat you. I"m 1600 elo, but I definitely don't deserve it. I only fight 1300's and considering I wouldn't lose a lot of elo if I lost to them, it's very easy to gain elo from people lower than you.

Badlion also needs to stop changing things that really don't need a change. I've seen it happen before. A server would change things the players didn't really want changed, and that would completely DESTROY their playerbase.

Posted on 19 August 2017 - 06:08 PM

Agreed, good thread.

Last edited on 19 August 2017 - 10:03 PM by Archybot

There is another major thing you did not add to your post @Ny_

Our biggest mistake was using 32k value early on (it was just a random value we picked from an old game we used to play, and these games would take anywhere from 10 minutes to 2-3 hours). There is documented evidence inside of chess, the game that ELO is made for, that bigger k values are anti-competitive.

The way K-values work in chess are as following (please note that an even game is +/- kvalue/2 - so 32 is +/- 16.):
For all people their first 30 games, regardless of age/rating, is 40 kvalue.
For all people ages 17- is kvalue 40 unless they are above 2300.
For all people ages 18+ after their first 30 games, but still under 2400, they have a kvalue of 20 (this includes age 17- and between 2300 and 2400).
For all people above 2400, regardless of anything else, is a kvalue of 10. AN EVEN GAME FOR A GRANDMASTER IS +- 5.
Ratings are done independently of each other, so when your rating changes you use your own kvalue, and your opp uses his.

This is how chess works, but chess is a game that takes anywhere from a few minutes to hours for a single game. 1v1 Arena matches do not last that long, so we should look into what chess does for rapid/blitz games (games where each player has 5 minutes or less to play total). The official kvalue for blitz/rapid games is 20kvalue, at all levels/ages/skills.

32k value is not recommended, or even good, for a game that is highly volatile and end in less then 5 minutes. Even chess has different kvalues for skill levels, but that would make inflation even worse in a game where you can grind out 100s of matches in a single day. We chose 16 over 20 because 16 is half of 32 and would be the easiest to understand for our current community that is used to 32 (as it is a perfect half). This is already proven by the fact that in the top 20 in S12 and S15 it is much closer inside of S15. Something you do not see in these screenshots is the fact that the amount of games played are basically the same. Lower kvalue encourages more competition at the top, and reduces ladder anxiety for those afraid to lose points.

Unrelated to theorycrafting is blatant cheaters cannot climb as fast in a system that is slower, which makes it a higher chance a moderator/gcheat gets to them before they break into high level gameplay and messing it up there.

Regardless, we will most likely have a end of season vote for people to determine whether they want 32k ELO or 16k ELO, but it will most likely be decided only by people who have played inside of S15 for x amount of games.

Posted on 19 August 2017 - 10:18 PM

Automatically Deleted

Posted on 20 August 2017 - 02:22 AM

cool, i agree

Last edited on 20 August 2017 - 03:11 AM by Ryankoo1

Probably the best argument why s15 elo is better, harder for cheaters to mess things up. I still think the challenge of beating Danteh's 2800 elo record legitimately would promote more competition, but we'll see

Posted on 20 August 2017 - 03:28 AM

In fact the elo record could be a great subject for marketing, if there was a cash prize for the first person to beat Danteh's elo record it would almost certainly create more competition than first global

Posted on 20 August 2017 - 07:34 AM

I like 16k elo. This is something the community can wrap their heads around, which I think is important.

Posted on 20 August 2017 - 07:55 AM

Perfect thread with a lot of high quality information. Stuff like this only makes me more hyped for s15's official release :)

Posted on 20 August 2017 - 08:48 AM

Finally someone made a good thread on this. Really good points and I totally agree that S15 elo is better

Posted on 20 August 2017 - 11:12 AM

+1

Posted on 20 August 2017 - 01:27 PM

Amazing and insightful thread, I now see why the 16k system is a lot better. Thanks.

Posted on 20 August 2017 - 03:30 PM

Automatically Deleted

Posted on 20 August 2017 - 03:40 PM

A very well-written thread which makes it very easy to understand. Thank you!

Posted on 20 August 2017 - 07:43 PM

esketit