Tournament ELO

[mrdmr]

I think you calculated TaToH vs Hera twice ??

 

[_balcanman]

the idea to use only tournament games for the rating is needed, but the idea of different ratings according to rounds is somewhat crazy.^^

 

[Desert_F0X]

If u use Legacy of the Huns 2 there then it certainly would be nice to use Legacy of the Huns 1 as well

 

[Hellrider_23]

a lot of smurfs? which one?

mentalist is also psycho
yinghua is also shuainiu
saymyname is also yellow and also thuongthoi
mattheus is melkor
chart is also riskydice

6 out of 83 might not be ''a lot'' but you get my point

 

[dodageka]

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

If it’s Elo and there is enough data then going back as long as possible doesn’t hurt - even though adding a small decay could make sense in our case as there are probably not enough tournament games. Another challenge that might come up is that there are not enough tournaments where top 20-30 and players below that level mix to ensure that ratings are proportionally (the most effective for that would be open tournaments in Swiss format).

After I’ve taken a closer look at the data early next week I’ll leave more feedback/comments

 

[damdres]

Beautiful website, great job. Congratulations! Go forward!

 

[guitarizt]

The data will always be so far behind and inaccurate to what everyone already knows.

 

[mungo__]

Hey!

Sorry for not responding this week, I am working full time (RL) and had a lot to do this week. But in the meantime I reworked the Elo graphs, in order to make it clearer.
See it for example here: https://aoe-elo.com/team/5/Secret

Thank you for all the feedback! Some points multiple people mentioned:

• Confusing point exchanges (losing points while winning):
  Clemensor and some others said it right, you lose Elo if the result is worse than statistically predicted. You can imagine as if every game within a series was played with the Voobly Elo system. This means
  • Equal opponents will have symmetrical reward for every outcome
  • Viper will lose points if he wins 4:3 agains some pleb (3 games lost are more costly than 4 games won are rewarding)
  • Winning X:0 will always give you points
  • Losing 0:X will always cost you points
• Duplicate Players/Matches
  Thanks a lot! I think we merged them now and everything should be fine. If you spot something else, let us know...
• Weighting Games/Torunaments by stage or prestige/prizemoney
  I really hesitate implementing this: As you see, there is already some controversy about the system as simple as it is now (maybe a bit more explanation from my side would have helped). Initially it is really thought as "Elo rating from tournament games". Every step to make it more complex will make it less interpretable and cause more confusion, unless it is done really precisely/comprehensively. I want to indroduce the least amount of personal bias possible, and who decides which tournament is more important? Or if a final is 2x or 3x more important than a quarterfinal? If we weight a final more, we would maybe punish player more for losing a final more than reward him/her for even reaching it, and so on.
  Right now, Viper has a solid ~100 points advantage over the 2nd place, I think this captures his current dominance enough.
• Missing Tournaments/Games
  We are aware that there is still a lot missing, but we are working on it. I wrote some backend interfaces to allow some community members to enter data manually. Also I will implement automatic duplicate checking (like for the Tatoh-Hera match)

What I am working on now:

• A "explain this match" page, where you can review the calculation for a match. Hopefully the resulting rating changes are more comprehensible then.
• Rating decay, as proposed by multiple of you (buhanisson, dodageka, ..)
• More data & tournaments
• More graphs 11

Let me know what you think about this!

 

[mungo__]

I think you calculated TaToH vs Hera twice ??

Thanks! And OMG, it looks horrible on mobile 11 - I will fix this some day

 

[mungo__]

Hi guys! Thanks for all the feedback. We finished a new version of the website, have a look: aoe-elo.com

• Muuuuch more data (thanks to pumukel and uberrushung)
• Elo Decay: After 3 months inactivity, a player loses 10 Elo per month
• More readable graphs (tournaments are displayed)

Currently, if a player plays again after being inactive, the decay is removed. It is basically only for retired players. Do you think this decay is ok as it is now, or should it be modified?

What would you like to see on the site otherwise?

 

[iViktorius]

quick feedback

"Subscripe to the Newsletter"

That should be Subscribe

 

[mungo__]

quick feedback

"Subscripe to the Newsletter"

That should be Subscribe

fixed, thanks!

 

[archxeon]

This looks really amazing. Standings reflect well to what I've seen most people deem the top 10 to be. So, that's a good start to make a quick judgement if the analysis are meaningful or not. Thanks for the work!

 

[buhanisson]

Currently, if a player plays again after being inactive, the decay is removed. It is basically only for retired players. Do you think this decay is ok as it is now, or should it be modified?

What would you like to see on the site otherwise?

Its very difficult to say what the correct decay is. I think only time will show. So far, I think it seems prety good.

I think I would like to see the teams arranged based on something other than alphabetics. Since the players are in order of rating, I think its a bit confusing if the teams are not. But which statistic to use for ordering is of course difficult to say. By the way, ordering the teams based on win-% doesnt seem to work.

 

[The_Night_Panda]

I love the idea of this! The main change I feel is needed is some kind of weighted difference between different tournaments. It seems wrong that daut got almost 50 points from winning King of the realm while losing only 12 points from losing in first round of kotd2. A possible solution could be to multiply any elo change with a number between 0 and 1 for any tournament based on f.eks. the number of players, format of the tournament (single elimination vs doubble elimination vs round robin) and the length of sets (bo3 vs bo7) so f.eks. tournaments with few players, long formats and longer sets wont lead to potentially much larger elo changes than short format high level tournaments like kotd2 and ecl.

 

[archxeon]

I see it a different way compared to how some others have. I see the "tournament" part differentiating just serious games from non-serious games.
Granted that some players take some tournaments more serious than others, I feel like this approach is just right to exclude the troll games and games against vastly inferior opponents.

 

[Spring_]

well to start you need a ladder that has 1 nick per player to avoid smurfing to save your own rating or smurfing to steal enemy rating. Maybe a separate "tournament 1v1 or tg seeding ladder" that pros are expected to be seeded off of

 

[nimanoe]

Looks great! Small fixes: goeffrey should be geoffrey and Lovecheng AKA yinghua is playing ECL under his ShuaiNiu nick as part of SY. Further xiaoxiong is part of SY and not of Frantic.

well to start you need a ladder that has 1 nick per player to avoid smurfing to save your own rating or smurfing to steal enemy rating. Maybe a separate "tournament 1v1 or tg seeding ladder" that pros are expected to be seeded off of

He's only using tournament games to seed the players and he combines all the different nicks that they use, so this is already fixed :D

 

[pumukel]

Looks great! Small fixes: goeffrey should be geoffrey and Lovecheng AKA yinghua is playing ECL under his ShuaiNiu nick as part of SY. Further xiaoxiong is part of SY and not of Frantic.

you still got that 2k4 feedback rating 11, fixed, thanks.

 

[JoshuaR]

This is amazing. Well done, looks beautiful, and it gives us a great picture of comparable elo in serious tournament games.

I agree with you that there is no need to weight one game higher than another. The point is these are "autopairings" in "serious" games. The players play seriously and with theoretically more "balanced" settings than random map elo. No need to weight differently because then you get into some subjectiveness of whether KOTD is worth more or less than LOTH, or whether a random group stage game from a more "important" tournament is worth more or less than a single elimination from a "less important" tournament.

I'm mixed on the "decay" system. I think it's useful to see who the top active tournament players are, so in that case it's a plus, but it would also be cool to have for posterity the top players "period" with a minimum of X games. Say if you had archival data all the way back to the first Microsoft tournament, you could compare DauT's overall legacy to say Chris'. If you've got the data anyway, you could even have a separate non-decayed rankings. It doesn't make a difference now since all these players are pretty active, but maybe in 2-3 years, if some of these players play less often or whatnot, we can still compare their records.

I'm curious why you chose to adjust ratings based on "matches" rather than individual games. It probably comes out to be essentially the same, but I bet you'd lose a fair amount of the complaints about players losing elo when winning matches when you show their gains/losses for each game.
E.G. instead of Viper 4 - 2 Yo, Viper loses 4 elo pts for underperforming against expected win percentage, you could display:
Yo wins game 1: -4
Yo wins game 2: -4
Viper wins game 3: +1
Viper wins game 4: +1, game 5 +1, game 6 +1
Series Sum: Viper wins 4-2 and loses 4 elo.

 

[mungo__]

you could even have a separate non-decayed rankings. It doesn't make a difference now since all these players are pretty active, but maybe in 2-3 years, if some of these players play less often or whatnot, we can still compare their records.

Good point, also the things you wrote earlier. This decay is only a first version: We fill introduce a non-decayed ranking soon I think. To compare the "legacy", I think I will make a "compare players" function, where the elo graphs are plotted next to each other. This could be pretty :=)

I'm curious why you chose to adjust ratings based on "matches" rather than individual games. It probably comes out to be essentially the same, but I bet you'd lose a fair amount of the complaints about players losing elo when winning matches when you show their gains/losses for each game.

I think you are right. I will work this saturday on a more comprehensive display/explanation, so this gets clearer.

 

[mungo__]

I love the idea of this! The main change I feel is needed is some kind of weighted difference between different tournaments. It seems wrong that daut got almost 50 points from winning King of the realm while losing only 12 points from losing in first round of kotd2. A possible solution could be to multiply any elo change with a number between 0 and 1 for any tournament based on f.eks. the number of players, format of the tournament (single elimination vs doubble elimination vs round robin) and the length of sets (bo3 vs bo7) so f.eks. tournaments with few players, long formats and longer sets wont lead to potentially much larger elo changes than short format high level tournaments like kotd2 and ecl.

Hey, so I am still hesitating on introducing a subjective tournament weight, but the length of sets is already taken into account: Actually, every game within a series is evaluated separately.

I added a small description underneath the Elo change under the matches: https://aoe-elo.com/player/31/TaToH

JoshuaR explained it quite well, and I will make a better description on the website soon.

Make sure to press F5 to refresh the full page and clear the cache (otherwise the browser sometimes only reloads some things).

 

[The_Night_Panda]

Hey, so I am still hesitating on introducing a subjective tournament weight, but the length of sets is already taken into account: Actually, every game within a series is evaluated separately.

I added a small description underneath the Elo change under the matches: https://aoe-elo.com/player/31/TaToH

JoshuaR explained it quite well, and I will make a better description on the website soon.

Make sure to press F5 to refresh the full page and clear the cache (otherwise the browser sometimes only reloads some things).

Im not sure i understand how length of sets are taken into account. It seems like a bo7 with each game played would be weigthed more than twice as much as a bo3 with each game played but maybe i am missing something?

 

[mungo__]

Im not sure i understand how length of sets are taken into account. It seems like a bo7 with each game played would be weigthed more than twice as much as a bo3 with each game played but maybe i am missing something?

Oh I think I misunderstood what you mean. You think that is is not good that lets say a 4:0 win in a BO9 is weighted double as a 2:0 in a BO5? That may be true, as the games within a series are statistically speaking correlated. However I dont see a consistent and transparent method to weight them softer, as this gets complicated and incomprehensive quite fast...

 

[The_Night_Panda]

Oh I think I misunderstood what you mean. You think that is is not good that lets say a 4:0 win in a BO9 is weighted double as a 2:0 in a BO5? That may be true, as the games within a series are statistically speaking correlated. However I dont see a consistent and transparent method to weight them softer, as this gets complicated and incomprehensive quite fast...

You could divide the elo change pr game with the number of games played in the set. (And possibly add a fixed multiplier independent of the format to keep the avarage elo change per set the same.)

Eksample of fixed multiplier: 4

Eksample of two sets:

1
Bo3 ending 2:0 and before that would transfer 10 points each game.
Any elochange from a game gets divided by 2 and multiplied by 4
So each game would transfer 10/2x4=20 elo transfer for each game and 40 elo transfer in total

2
Bo9 ending 5:0 and before each game would transfer 10 points
Now it would be 10/5x4=8 points per game and a total transfer of 40

Before the first set would transfer 20 points and the second 50 points. Now both would be weigthed the same