Tournament ELO

[mungo__]

Hey guys!

The project aoe-elo.com is online. We already have 80+ tournaments and 4000+ games in the database.

What it is

It's basically a Elo ranking (like Voobly) of all players of major tournaments, based only on tournament games. Motivation:

• A meaningful and up-to-date ranking of pro players
• A database of all major tournaments and all games
• Graphs and Stats! For example Elo development over a player's whole career.
• Interactivity: Scroll back in time, compare player legacies, ...

Using only tournament games means:

• No choice of opponents, like on Voobly
• No smurfing, no noob bashing
• Only important games (tournament games are not rated on Voobly!)
• No aliases or second accounts

Current Progress

We are currently working on:

• Connecting Voobly Accounts (API request pending...)
• More readable tournaments in graphs

On the waiting list:

• Customizing the ladder options (included tournaments, time frame, ...)
• Better explanation of the algorithm
• API for fetching data

Spoiler: Done

• Tournament labels in Elo development graphs
• Manually sortable tables
• Elo decay on inactivity
• Compare players page

Spoiler: Team game Elo

We are thinking about how to do this. Some people suggested to implement something like in Tennis or other sports. The problem is the high amount of games needed for the ranking to be meaningful. These games all have to be manually inserted, as for example Challonge Brackets have no information of the exact players that played the games. The required data for a meaningful estimation is also quite a lot higher than for 1v1 games.

Spoiler: Tournament/stage weighting

Suggested by multiple people.
One problem with this is that introduces a subjective bias to the ranking. Who decides which tournament is important?
Also, weighting different stages is difficult. To give the winner more points, we have to take them away from the loser, this is the principle of Elo. If we weight a finale more than a semi-finale, the losing player of the finale could be even punished for reaching the finale, because he might lose more points than he would have by losing the semi-finale.
Elo is really only about the statistically expected result. Sometimes this can be misleading.
Example: If the Elo is just right for two players Adam and Berta, it should not change much after a series. If Adam has a higher Elo than Berta, the win probability could be for example 66% for Adam. If Adam now wins the series with 2:1 (66% of games for A., 33% games for Berta), the Elo change would be: (win, +4), (win, +4), (lose, -8) = 0
So it does not change, meaning that it was a correct estimate for the skill difference.

Spoiler: Statistical model

Currently a series of 3:4 is rated as 7 individual games, each of which is evaluated separately using the Elo system, just as in Voobly or HD. Some people suggested weighting a 4:0 not double than a 2:0. It is correct that games within a series are correlated (if a player "has a bad day" etc.). However it is difficult to find a transparent algorithm which takes this into account. I will start by a detailed description of the current algorithm and am open to suggestions. Pure Elo has the advantage that it is understood by most visitors immediately.

Spoiler: Different initial Elo

Suggested by @linetyReborn
Currently, players with a lot of games start with 2000 Elo, more unknown players with 1900. This initial value is only an estimage for the Elo of the player at the beginning of his career. Using a lower value (1900) for more unknown players has the purpose of not buff the players too much who win against them (for examle in early tournament stages).
It is true that this is never the ideal estimate, it only gets better after more games. But other methods are difficult: Subjectively rating players manually is never fair. And taking "future success" into account (like letting Viper start at a high rating, as he will win a lot of tournaments later in his career, as we know) also kind of destroys the image of the early stages of the career or a player.
The idea behind Tournament Elo is that we collect enough data so that also the earlier stages of the career are reflected. This way, the Elo evolves with the "true" skill.

--------------------------------------------------------------------------------
Original Post:

One of the most asked questions is probably "who are the top 10 players atm". The voobly ELO is good in general, but contains double accounts, is very fluctuating and for experts is not really precise enough (viper is not #1 for example).

So I wanted to write a website for the ELO only from tournament games (which are unrated in voobly). I am not certain where I take the data from (maybe automatically from this forum or idk). The script for the ELO calculation I could write.

Would you in general be interested in such a ranking? And what features would you like? I thought about something like:

• List of about 50 relevant players (only games among these are relevant)
• Time-sorted list of all recent tounament games
• Separated top ELO list for RM 1v1/ TG / DM
• Interface for manually adding tournament games
• ...?

 

[Huehuecoyotl22]

https://www.aoezone.net/threads/2018-competitions-outcomes.138391/

This will be very useful for you.

 

[_balcanman]

maybe there are not enough games to give plausible values

 

[buhanisson]

DM? How many DM tournaments are there, even. Other than that, sounds good and something that I, too, always wished someone made in a way that people would really care about. I think one additional thing that might be interesting to incorporate could be total prize money won. For TG, this obviously might not make as much sense, but for 1v1 the way I envision the data layout could be like:

player name ; series won ; series lost ; games won - games lost ; series win-% ; games win-% ; tournament wins ; tournament 2nd places ; tournaments played ; money won ; (maybe average money won per tournament) ; tournament rating ; (maybe a word, or a "title" describing the player, either a classification only based on his current tournament rating, or a historical title that, once won, cant be taken away, e.g. a bit chess-like: "world champion" - "super grand master" - "grand master" - "expert player" - "solid performer")

with ability to sort the players based on any variable the user wants.

For ELO calculation, I think you should at least consider making the ratings "start" somewhere other than 1600 (e.g. 2000) because the amount of players and games is relatively low and it might not appeal to some people's esthetic sense if Viper was rated 1725 and daut 1432, or vice versa. Of course, this is only a matter of some very vague idea of how certain numbers are psychologically perceived and doesnt affect the actual mathematics, but anyway, something to consider.

 

[amazing_knight]

Would you in general be interested in such a ranking? And what features would you like?

Yes, I would be

As of now, most places only have a tally of the number of tournaments won by players. But a properly done rating can give us more insights on who've been the most consistent players in tourneys.

Hoping you can take into account the who lost to who in the ELO algorithm. For instance, Jordan lost to TheViper in KotD1 Round2, but I am hoping he still gets more points for taking 1 game from Viper than some others, specially considering the fact that Viper did not lose any other game in the entire tournament till the Finals.

I am also hoping that all tournaments are considered but assigned different weightage depending on the average ELO of the players who participated. So, KotD can be given a higher weightage than, say, NAC or LoTH (Legacy of the Huns). But, all should get counted.

 

[nimanoe]

I like the idea, and a feature I would suggest is to give more weight to the later rounds than to the earlier rounds

 

[mungo__]

Hi guys! I finished a first version of the webiste! Have a look: aoe-elo.com

I would really appreciate if you tell me what you think!

 

[Poxo]

Yellow, thuongthoi and SaymyName are the same player.

Psycho and Mentalist are the same player

 

[kw1k000000]

Great initiaive. For this to be used seriously,

• We need a tournament weighing similar to ATP 500, 1000, slam etc to have results in perspective.
• Whether the game was a knock out or some sort of round robin, double elimination, Swiss system etc. The most is at stake when its a knock out.
• Weighing for winning a set of BO5 3-0 vs 3-2
• Weghing for game/set stage. A Final is more pressure/presitigious than a quarter final. Similarly round robin group stage is far less at stake than a winner take all final.

 

[iViktorius]

3 times in the elo graph for TheViper I see he loses elo after a won match.
- KotD1 final, wins 4-5 vs Liereyy, loses 5 elo
- KO phase ECL, wins 1-2 vs ACCM, loses 2 elo
- QF LotH2, wins 2-1 vs Whoop, loses 5 elo

How does that exactly work?

I checked some others as well and it happens to them as well.

 

[dodageka]

3 times in the elo graph for TheViper I see he loses elo after a won match.
- KotD1 final, wins 4-5 vs Liereyy, loses 5 elo
- KO phase ECL, wins 1-2 vs ACCM, loses 2 elo
- QF LotH2, wins 2-1 vs Whoop, loses 5 elo

How does that exactly work?

I checked some others as well and it happens to them as well.

If you win 5 games and loose 4, than your Elo goes down if it projected you to win more than that.

@kw1k000000 @nimanoe if this is meant to be an Elo, you can’t rate or weigh tournaments differently. Elo is purely about the statistical probability of winning vs another player, independent of the situation

 

[Clemensor]

3 times in the elo graph for TheViper I see he loses elo after a won match.
- KotD1 final, wins 4-5 vs Liereyy, loses 5 elo
- KO phase ECL, wins 1-2 vs ACCM, loses 2 elo
- QF LotH2, wins 2-1 vs Whoop, loses 5 elo

How does that exactly work?

I checked some others as well and it happens to them as well.

This is a guess, but since Viper supposedly has far more elo than his opponents he will lose rating if he doesnt win a certain amount of games in a series, given that every game counts and affects the elo rating. so for example Viper gains like 20 points winning 2 games vs accm while accm gets 22 or smth because he is way lower rated. Similar to rating system on voobly/HD.

 

[kw1k000000]

If you win 5 games and loose 4, than your Elo goes down if it projected you to win more than that.

@kw1k000000 @nimanoe if this is meant to be an Elo, you can’t rate or weigh tournaments differently. Elo is purely about the statistical probability of winning vs another player, independent of the situation

ok let me rephrase. if this rating is to be used in any meaningful way for tournament seeding then in the current form its useless.

 

[buhanisson]

I think the assumption that all games are equally important as long as they are tournament games is fine. It just needs lot more data to round off the effects of those few meaningless games that are bound to happen every now and then. When you reach higher elo, you are implicitly expected to always reach higher and higher rounds in the tournaments to keep improving your elo, anyway.

The one big problem with weighing different rounds differently is the fact that all tournaments dont have same amount of rounds. One 64 or 128 player tournament would give ridiculously more elo than many 8-player invitationals combined. Or the finals of king of realms would be much more important than, say, semifinals of kotd2. In reality this would distort the ratings much more than those few meaningless games where viper might or might not be trolling a bit, but probably actually isnt.

But of course it's not always a simple question as to what actually should count as a tournament. Should showmatches, for example, be included when actual money is on the line?

 

[pumukel]

some food for thought: https://www.aoezone.net/threads/the-aoe2-career-points-project.136483/

 

[c3mbre]

It's cool but I think u should consider only last year. We want to see top players at this moment

 

[King_Marv]

Cool Website!

Some suggestion:

Why not take kotd1 ( its there i missed it) and moa5 into consideration as they were both last year and 1v1 and remove old tournaments from 2015.

Lovecheng = shuianui
Mattheus = melkor

They are two times in the list.

 

[Hellrider_23]

There's a lot of things wrong with this, a lot of smurf accounts, very old experts as well (seems like you used WSVG maldives for the rating).

I would suggest starting with KOTD 1, and even that is super outdated by now, but if you won't use that, you are stuck with two ECL events and maybe a few invitational tournaments+ arena clowns and hunwhore tournaments.

 

[King_Marv]

There's a lot of things wrong with this, a lot of smurf accounts, very old experts as well (seems like you used WSVG maldives for the rating).

I would suggest starting with KOTD 1, and even that is super outdated by now, but if you won't use that, you are stuck with two ECL events and maybe a few invitational tournaments+ arena clowns and hunwhore tournaments.

a lot of smurfs? which one?

 

[buhanisson]

Maybe you could implement some proper decay system to just punish/reset people who dont play enough tournaments in, say, a year or something.

And maybe not display all the accounts with only 1 game or so, to make the list look at least slightly better visually. Or maybe lower the threshold where you give that error sign of not enough data.

 

[Poxo]

wtf how does the point system work? 111



A semifinal is worth 7 times the points of a loser bracket finals? LUL

 

[CounterForce]

wtf how does the point system work? 111

A semifinal is worth 7 times the points of a loser bracket finals? LUL

As mungo said, it's the ELO system applied to the single games in tournaments. You win and lose points just as if this were a rated game. Stage of the tournament is not taken into account.

 

[Poxo]

LUL just rechecked that picture, Riut won points losing vs Yo en the semifinals 111111111

 

[chicklet]

LUL just rechecked that picture, Riut won points losing vs Yo en the semifinals 111111111

Yes, because he won 40% of the games. Slightly higher than the 37% predicted.

 

[i_no_english]

You can gain points by losing... if a 1600 player loses 5-4 to TheViper im sure he will gain points...