Tournament ELO

[The_Night_Panda]

The problem with this kind of feature is the heavy calulation overhead. To calculate a player's Elo, we need the Elo of all his opponents in history. The Elo of the opponents is affected by the Elo of their opponents, etc. The current calculation time is not too crazy, only some seconds. But if would have to be done every time someone adjusts the list.

However, as so many people suggested it, I might implement it with a bit of caching...
Possible filters would be:
- Tournaments included/excluded
- Time frame
- any other ideas?

mungo

Maps maybe, but i think those are the major ones

 

[mungo__]

Hello guys! Some news:

• You can now compare the Voobly ratings. They are updated every midnight GMT using Voobly API.
  You can see that they sometimes are quite different from the tournament Elo. (Especially for Viper and some chinese players)
• I removed the "Upcoming events" preview from the start page and replaced them by "Recent Results".
  It was difficult for me to keep them updated, and there are also other, dedicated resources (aoezone calendar, aoe2calendar.com). Maybe I can some day embed them or something.
  Also, the player comparison serves now the purpose of this preview (win chances, Elo history...)
• DemEyesRed Arena Cup is included (thanks @pumukel)

Tell me what you think

mungo

 

[Tarsiz]

I'm honestly very impressed by the quality of your website.
It looks very neat, nice and professionnal, information is easy to access and gives me very quickly what I want to know: performance in a single tournament or the chance that X beats Y knowing their respective elos.

For someone who is interested in the competitive history of our game, your work is essential. Huge props to you! I hope to see this used for seeding in future tournaments.

Something you could add is the name of the organizer and/or a link to the official (or AoEZone) page of the tournaments in the information snippet you have for each tournament.

 

[archxeon]

One thing that bugs me is RiuT being so high. Not that I don't want him that high. I actually wish he'd play more and be deserving of the Top 5. But it doesn't feel right to have him top 5 when he didn't have a single tournament match in 2017 and just three tournament matches in 2018. Maybe introduce some decay (to penalize) or some form factor (to boost) the values?

 

[mungo__]

One thing that bugs me is RiuT being so high. Not that I don't want him that high. I actually wish he'd play more and be deserving of the Top 5. But it doesn't feel right to have him top 5 when he didn't have a single tournament match in 2017 and just three tournament matches in 2018. Maybe introduce some decay (to penalize) or some form factor (to boost) the values?

Yup, the next thing I will do is a soft decay in inactivity phases. Maybe 2 points/Month? The current decay is only for retired players, and it is removed when they play again.

 

[Huehuecoyotl22]

Yup, the next thing I will do is a soft decay in inactivity phases. Maybe 2 points/Month? The current decay is only for retired players, and it is removed when they play again.

How do you decide who is a "retired player"? Does that mean no international tournament participation for x months?
E.g. I wouldn't call fangxiaoqi (aka Win_wu_kong) retired, he just very rarely plays in international tournaments: CoT and WSVG, later only in some Chinese ones.

 

[dodageka]

Instead of a decay, which is always very tricky and potentially controversial, an easier and more straightforward solution could be to filter out players that don't have an entry within the last 2 or 3 months. That way, we would only see players in the list that are reasonably active (and you can always add the option to view the whole list in its current form)

 

[dodageka]

• You can now compare the Voobly ratings. They are updated every midnight GMT using Voobly API.
  You can see that they sometimes are quite different from the tournament Elo. (Especially for Viper and some chinese players)

Especially for Viper they are very similar, it is less than 100 points difference jk, I know what you mean. Good job with the voobly ranks, that is really neat as well

 

[chicklet]

Congratulations @mungo__ on creating such an impressive tool and website... And a huge thank you for providing me with hours of procrastination

I really hope this is used for future tournament seedings.

One question (and apologies if it has already been covered): at present the 'Teams' section shows the average elo of the top 4 players for each team. Would it be possible for the team to have its own elo, independent to those of its members?

 

[dodageka]

A few more comments/suggestions:

• In the team section, the elo change is last 30 days, in the player section it is last 60 days - I think it would be nice to make this consistent
• Is it possible to add a functionality where one can type the name/first letter for the player comparison? At the moment it takes a bit of scrolling etc. to e.g., compage Yo vs. Spring
• On the player comparion site, it would be cool if you could also show the probabilities for a BO3/5/7/9, maybe with a drop-down in addition to the single game probability

Thanks for this amazing website, I love it and seriously hope people will start using it for seeding soon instead of the voobly ladder

 

[mungo__]

Thanks for the suggestions everybody!

Instead of a decay, which is always very tricky and potentially controversial, an easier and more straightforward solution could be to filter out players that don't have an entry within the last 2 or 3 months. That way, we would only see players in the list that are reasonably active (and you can always add the option to view the whole list in its current form)

Good idea. Yes, a Elo decay is really difficult to get right - we dont want to punish players too much, who just happen to play a tournament every one in a while, if their results are good. It wouldnt make much sense... I will definetly think about something like this filter, its a good idea!

add a functionality where one can type the name/first letter for the player comparison

Planned!

On the player comparion site, it would be cool if you could also show the probabilities for a BO3/5/7/9, maybe with a drop-down in addition to the single game probability

Also planned

One question (and apologies if it has already been covered): at present the 'Teams' section shows the average elo of the top 4 players for each team. Would it be possible for the team to have its own elo, independent to those of its members?

Well the question is how meaningful it would be, as every torunament has different player compositions. Also, separate 2v2-4v4 ladders? rating the whole team or the players individually? A lot of data would have to be collected. But it would be a very nice thing for the future (not soon though I think...).

How do you decide who is a "retired player"?

Currently, a player who does not participate at a tournament loses 10points/month, starting after 3 months inactivity: Example

 

[archxeon]

Instead of a decay, which is always very tricky and potentially controversial, an easier and more straightforward solution could be to filter out players that don't have an entry within the last 2 or 3 months. That way, we would only see players in the list that are reasonably active (and you can always add the option to view the whole list in its current form)

If the goal of ratings is to correctly reflect the strength of players, ratings have to account for long gaps somehow, IMO. Long gaps has shown to decrease skill level in at least a few of the top players. But yeah, how to properly quantify is difficult.

 

[Huehuecoyotl22]

While I do understand the “inactivity“ rules, the word is not the most fitting one. Whoop for example signed up for KotD2, but his rating was not high enough. Still he plays almost every day.
As there have been many rating-restricted tournaments recently it is more difficult for <2k3 rated players to play in tournaments all the time.
But it is difficult to decide and your way is already very good.
At least with the Aspiring Experts there will be many new data.

 

[_Mr_St4rk_]

There'll never be a fair rule,
A Player might be not playing tourneys, but keeping himself sharp for next Events..
Other might just not be playing at all, and truly Decaying..
But any rule adopted will only fit (or try to) to one Case..
Good thing u guys are considering Tourney Participation only as a Rule...
I Can see tourney-rated to be included in next tourneys rate (we usually do Max rate + current one... maybe Max + Current + Tourney / 3 could be seen...)

 

[robdyrob]

Yup, the next thing I will do is a soft decay in inactivity phases. Maybe 2 points/Month? The current decay is only for retired players, and it is removed when they play again.

In chess, these players are listed as "inactive", but without any elo decay. I guess this is the best way to address the problem. A linear elo decay implies imho that the quality of the games increases continously (so that e.g. 2200 elo in 2014 is the same as 2000 elo in 2019). This is not necessarily true.

Besides adding filters, you could e.g. add the date of the last played match in your list and/or display inactive players in a different color.

Otherwise, your list is really great. As a more aesthetic side note, I would increase the starting elo a bit. In princivple, it does not change anything, bout would make your values more comparable with the voobly ranking. The mean value of your top 20 is around 2050, in voobly it is much higher.

 

[patao]

Just want to say I appreciate that DauT will forever be #1 in this ranking. ID #1 in the database at least.

 

[Psyychoz]

Just want to say I appreciate that DauT will forever be #1 in this ranking. ID #1 in the database at least.

indeed a fine little acknowledgement of his legendary status :DD

 

[mungo__]

Just want to say I appreciate that DauT will forever be #1 in this ranking. ID #1 in the database at least.

Good catch haha! I put all the players in pretty random order at first, based on the teams in ECL. But of course the Lord had to be ID #1!

 

[The_Night_Panda]

While I do understand the “inactivity“ rules, the word is not the most fitting one. Whoop for example signed up for KotD2, but his rating was not high enough. Still he plays almost every day.
As there have been many rating-restricted tournaments recently it is more difficult for <2k3 rated players to play in tournaments all the time.
But it is difficult to decide and your way is already very good.
At least with the Aspiring Experts there will be many new data.

One way to to adress the issue could be to only have elo decay for players with over 2k tournament elo (currently 19 players). These players should mostly be able to sign up for most bigger tourneys and it would also set a natural limit to how much a player could be punished for being inactive.

It could be something like a 10 point penalty for not playing any tourney-games every 3 months for all players over 2k elo. Those points would then be dealt out to the "active" 2k+ players so the total elo sum would remain unchanged.

As an example Riut has been mostly losing points since the end of 2016 but has remained around top 5 until now due to him only having played 5 1v1 matches in that period. While he may have been somewhat active in the period this still leads to an artificially high rating. With the proposed decay he would end up being 70 points lower. He would still be around top 11 so its not like the change would be out of proportion.

 

[patao]

I forgot what the latest thought on showmatches was, but I'd love to see them included in here. Tournaments are often a procession of top seeds crushing low seeds (although formats have been improved lately), but showmatches can be made between relatively equal levels, which should help keep the rankings interesting.

For example the Escape showmatch last week was #27 LaaaaaN vs. #23 Villese and it was excellent!

 

[mungo__]

Hey guys,

I know there are some features you wished for, but currently I am quite busy on other projets so I can't really implement them.

However, any feedback is still very appreciated and I will process it as soon as I find the time!

One minor thing that would only take like 1min:
Do you think the "average elo" should be higher, so that for example the top5 are ~2400 Elo?

If yes, we could change that so that the numbers resemble more the voobly rating.

Best,

mungo

 

[enmipho]

IMO it's fine as is, people understand it's a different ladder. You wouldn't necessarily expect the top player of RM 1v1 to have the same rating as the top player on one of the CBA ladders, and what matters is how players are rated relative to one another.

 

[chicklet]

One minor thing that would only take like 1min:
Do you think the "average elo" should be higher, so that for example the top5 are ~2400 Elo?

If yes, we could change that so that the numbers resemble more the voobly rating.

This isn't necessary or worth your effort. The absolute number is meaningless (although some people obsess over arbitrary thresholds). What matters is the relative level as @enmipho says.

 

[Michaerbse]

Hey guys,

I know there are some features you wished for, but currently I am quite busy on other projets so I can't really implement them.

However, any feedback is still very appreciated and I will process it as soon as I find the time!

One minor thing that would only take like 1min:
Do you think the "average elo" should be higher, so that for example the top5 are ~2400 Elo?

If yes, we could change that so that the numbers resemble more the voobly rating.

Best,

mungo

I don't think that there should be any starting elo anymore (i assume the starting elo defines the average elo). I prefer the initial rating of a player being calculated based on his first tournament performance as it is done in chess. This is obviously not possible at the start as you need a few players with a rating at first but the available data as of now should definitely be enough to apply this way.

(If this is not possible I agree with the others - it doesn't really matter much and you can discuss forever whether 100 points more or less would be appropriate)

 

[robdyrob]

I'd prefer a higher value if it is really only 1 min effort. It is purely esthetical, but consistent numbers are easier to understand. Especially for non-casual viewers if ELO-numbers are mentioned in streams.

Given the small number of active pro players, I guess there are still to few data to calculate ELO based on starting performance only. It is maybe an option for potential newcomers though.