Using Player-based ELO to determine team elo score

Please discuss issues concerning the Ducati League here. This is the liaison between league players and the league council.
Post Reply
User avatar
[dmp]
Captain
Captain
Posts: 282
Joined: Mon Dec 09, 2002 3:20 pm
Location: CPH, Denmark
Contact:

Using Player-based ELO to determine team elo score

Post by [dmp] »

I'll try to describe another way to calculate team scores.

The idea is to rate every player and let team score depends on this. Let me try to make it not-so-long :). (For you that knows how my old ladder worked, this is basically the same)
  • The team elo-score will be the average of the members elo-score (perhaps one could exclude the lowest and the highest - so avoid that a single player affects score too much - ie. a new player). This score is used on the website
  • A players ELO score changes ONLY when participating in a match.
  • When calculating the amount of points a team loses/wins - only the players elo-points are used. Not the average score described in the first point. So if two weak players participate, the elo-score for their team is low, compared if two strong players from the same team plays
  • If a team wins or loses 32 points, the participating players split this score. If players sub, they are given points for the time they played. Eg. Player A plays a full match, but B and C subs after 15 mins. Then player A will get 50% and B, C will get 25% each. (Instead of splitting points equally between players - it could be weighted)
  • It may create new problems such as trading players. Suddenly getting the "right" players to join your team will boost your score - so there is not much teameffort in gaining that score. However to increase it, they need to work. This may or may not be a problem. If a very strong player joins a team, I'll bet most of you will consider that alot stronger - even if they didnt play a match since the arrival of that player
This got a few sideeffects:
  • Downside: The current way of reporting matches is insufficient. We need some code to handle match-reporting (which also also for some other neat stuff). However, it is still somewhat more complicated
  • Downside: Focus is shifted abit from team-wise to player-wise scoring. However as player scores only changes in matchs, I dont see this as a major issue.
  • If a team uses two weak players to fight a match, they will not lose as many points as if two strong was playing.
  • Teams will actually benefit from using the weakest players, as they gain point more if they win (which might yeil more equal matches - however i doubt that people will consider this, due to the complexity of elo)
  • Single players can see if they improve directly in their own stats. Only they can change it (so if they want to earn elo, they need to work for it)
I don't need huge pictures here.
Admirarch
Private First Class
Private First Class
Posts: 33
Joined: Sat Sep 11, 2004 9:06 am
Location: Seeking lost whimsy

Post by Admirarch »

Additionally:

Downside: Inactive players on a team can overly influence its rating. A team can keep a permanently high rating by having a few very highly rated inactive players whilst they lose every single match.

As I see it we're almost decided on some sort of seasonal league which rewards activity heavily and then the natural move would be to keep standard ELO in the background with slightly reduced priority. As I see it this comes in as a replacement for the standard ELO in this arrangement and so needs comparing to that in the function of a long term indicator of team strength.

It's complicated and fiddly so hopefully I'll have actually worked out what I think of it in a day or so and will post that then.
User avatar
[dmp]
Captain
Captain
Posts: 282
Joined: Mon Dec 09, 2002 3:20 pm
Location: CPH, Denmark
Contact:

Post by [dmp] »

Admirarch wrote:Downside: Inactive players on a team can overly influence its rating. A team can keep a permanently high rating by having a few very highly rated inactive players whilst they lose every single match.
Yep. Its pretty much the same as now. So the solution to this could be activity-factor or something similar.

And if this was implemented (the idea) then it would be possible to exclude inactive players in the team-score calculations. (Perhaps doing activity-factor on players :-) )
I don't need huge pictures here.
User avatar
sussi
Private First Class
Private First Class
Posts: 142
Joined: Sun Mar 23, 2003 9:39 pm
Location: Germany

Post by sussi »

Here it comes, Just litle different idea, if we have a personal ELO score as mentioned above, than the point gain or lose could calculated from the average personal score per team of the player who played the match. And add/substracted to the teamscore.

That means differently from dmp's proposal, the calculation how many point plus/minus starts from the average personal ELO of actual in that match involved players. That is count to the team ELO. The Elo won't move for joining a team.

I really like the idea of a personal ELO, especally if funmatches would count for it.

That would make it more interessting of let playing new players and makes the league more interessting.
birdie
Lieutenant
Lieutenant
Posts: 700
Joined: Fri Jan 21, 2005 4:46 pm
Location: Belgium

Post by birdie »

i dont like the idea for personal ELO points

CTF is a teamgame, not a personal game. You win a match together (with 2 - 3 - 4) and not on your own

ELO should represent the strength of the team, not from the players

birdie
User avatar
Xell
Private First Class
Private First Class
Posts: 45
Joined: Sun May 15, 2005 1:55 pm

Post by Xell »

I agree with birdie, the teamplay is in a CTF game more important than the individual skills.
For an individual score you can sign up here: http://1vs1.bzleague.com/
User avatar
[dmp]
Captain
Captain
Posts: 282
Joined: Mon Dec 09, 2002 3:20 pm
Location: CPH, Denmark
Contact:

Post by [dmp] »

And I agree too. But currently it dosnt represent the skill of the team. It represents a random number between the two weakest and two strongest players of the team - at best. (Assuming that most matches are 2vs2)

My suggestion is how we can get a better score for a team. And no matter how you turn it, the team is dependent on its players.
I don't need huge pictures here.
chtibidasse
Private First Class
Private First Class
Posts: 14
Joined: Sat Jan 07, 2006 6:52 am

Post by chtibidasse »

[dmp] wrote: My suggestion is how we can get a better score for a team. And no matter how you turn it, the team is dependent on its players.
How do you link this with the cap score? Because the player who caps doesn't care about his personal score, his only goal is to bring the flag home.

(I don't speak about myself because, caping or not, I'm always the last one at the scoring list...)

Chtibidasse :turtle:
User avatar
sussi
Private First Class
Private First Class
Posts: 142
Joined: Sun Mar 23, 2003 9:39 pm
Location: Germany

Post by sussi »

Did i understand rigth , we want a better (team) ELO, what represent better the strengh of a team, it schould take in account the strengh of the actual playing teamies, but it should not be a ladder? That is like wash me but don't make me wet.

Anyway possible is not to publish the ladder, only to publish the difference of strengh from the player from his team. It's a little sick but it should work. And to be sure, i don't like it. ;)
User avatar
Chestal
Dev Guru
Dev Guru
Posts: 171
Joined: Fri Dec 06, 2002 11:56 pm
Location: Siegen, Germany
Contact:

Post by Chestal »

chtibidasse wrote: How do you link this with the cap score? Because the player who caps doesn't care about his personal score, his only goal is to bring the flag home.
You don't look at each individual players' performance during the match, you just use the outcome of the match as usual.

The most obvious problem with the current purely team-based ELO rating happens with teams which have players of vastly different skill levels. So, depending on who of the team's player matches, it might be easy or next to impossible to beat them. This could be abused to only play the team when it's to players aren't around and get some cheap ELO points. On the other hand, if you play that team's to players, it will actually be much stronger than the ELO rating indicates and you'll lose more points then would be fair.

It would also help when players change teams. [dmp] has already enumerated the positive and negative effects in his original post. Maybe we need a good example because there seems to be some confusion.
birdie
Lieutenant
Lieutenant
Posts: 700
Joined: Fri Jan 21, 2005 4:46 pm
Location: Belgium

Post by birdie »

sussi wrote:Did i understand rigth , we want a better (team) ELO, what represent better the strengh of a team, it schould take in account the strengh of the actual playing teamies, but it should not be a ladder? That is like wash me but don't make me wet.

Anyway possible is not to publish the ladder, only to publish the difference of strengh from the player from his team. It's a little sick but it should work. And to be sure, i don't like it. ;)
Birdie wrote: ELO should represent the strength of the team, not from the players
User avatar
sussi
Private First Class
Private First Class
Posts: 142
Joined: Sun Mar 23, 2003 9:39 pm
Location: Germany

Post by sussi »

Just to make it clearer. If the teamsite state prehaps teamELO is1200, player A +10, player B -100 (thats the variation from the present teamELO). And A and B won a match, the effektive teamELO for the calculation is 1210+1100/2=1155.
That number goes into the calculation for gain points. The result of the calcultion si (perhaps) 30 points to add. Than 1200 +30 is the new teamscore of 1230.
User avatar
[dmp]
Captain
Captain
Posts: 282
Joined: Mon Dec 09, 2002 3:20 pm
Location: CPH, Denmark
Contact:

The story of Angie, Andy, Ann, Aron, Betty, ... and CNN?!

Post by [dmp] »

Ok, Ill try to make an example. Please hide your calculators, because the numbers are just for showing the effects - not showing my math skills :)Its pretty long, but I also wanted to show who and why it solves some of the problems I belive the current system have.

If you want to "play" with elo, you can use http://zeebrothers.net/elocalc.php

Lets say:

TeamA consists of newbies: newAngie, newAndy and oldtimers: oldAnn. oldAron.
TeamB consists of oldtimers: oldBetty and oldBart
TeamC consists of newbies: newCaden and newCathy
(Note the names. Every name beginning with A belongs to Team A etc. All names prefixed with "new" is weak players and "old" is strong players. All weak players play equally good, and all strong players play equally good. Yeah, its very unlikely, but I just want to show the problems - not simulate the league)

TeamA and TeamB are well-establised teams except that TeamA just got fresh cannon fodder (newAngie, newAndy). TeamC is a poor newbie team. Poor souls.

Current rating
First I want to show the bad effects of the current implementation.

Looking at the "league", we can learn that TeamA and TeamB is equal in score. Both teams got 1400 points. TeamC got 800 point

Now, lets see, if TeamA matches TeamB with the different combination of players.

The wining team gets 25 points in every situation, as both teams got 1400 points. Now is this fair?

Case #1: TeamB (oldBetty and oldBart) wins over TeamA (newAngie, newAndy)
If newAngie, newAndy (TeamA) plays oldBetty, oldBart (TeamB) and TeamB gets an easy win (which probably wont surprise anyone), they will get a full 25 points for the victory, even tho its two newbies they play against. TeamB are strong so the newbies (and the Team A for that matter) are punished pretty hard for losing the match even tho everyone expects TeamB to win. Some would consider it newbie bashing if TeamB does this too much (taking points for rich team by playing their weak players)

Case #2: Team A (newAngie, newAndy) wins or get a draw against TeamB (oldBetty, oldBart)
Against all odds, TeamA win. This is something that would hit CNN.. And they get the same amount of points as if TeamB win! So the weak players earns the same amount of points beating old players, as the old players gets for beating the new ones. CNN picked up the story as "breaking news", but our dear old ELO didnt reward their efforts.

But at least get something for the victory. If they get a draw, they get nothing. zip. Nada. However, CNN isnt that hard. They still get an interview.

And we can continue. If the two teams match eachother with only old players.. They will win/lose 25 points. Or get nothing if its a draw.

I'm not sure about you, but two newbies beating up a strong team, should be rewarded more compared to a more even fight.

Case #3: TeamA (newAngie, newAndy) vs TeamC (newCaden and newCathy)
The problem also appears in the opposite situation. If TeamA (with newAngie, newAndy) players TeamC (newCaden and newCathy) - an all newbie game (and I assume this is an even fight). As TeamA is rated high (with 1400) and TeamC is rated low (with 800) the effects of a draw or loss betweem for TeamA is huge. And if they win, they wont get much point-wise.

TeamA wins over TeamC. ELO rewards TeamA with 2 point (and TeamC loses 2 points)
The match is a draw: TeamA loses 23 points and TeamC wins 23.
TeamA loses: They lose 48 points (which TeamC gets)

Does this seem fair - considering that all the players got similar skill?

The problem
Based on the team score, ELO calculates the probability for one team to win. If the team with high probability wins, the reward is smaller. However if the team with low probability wins, they get a big reward.

SideNote Or YouCanSkipThisParagraph: One can also say: If our prediction (the probability) is correct, then the elo rating must be almost correct (or close to it), so we just need to adjust it in small steps. However, if our prediction is wrong (TeamA won, and we didnt belive they could) then we need to make a larger adjustment. So its vital that the numbers we work with is as good as possible.

The problem is, as you might have figured out. We got one score per team. And since alot of teams consist of players of different skilllevels, playing the team is very much different depending on the participating players. (Eg. Strong TeamA is oldAnn and oldAron.. While weak teamA is newAngie, newAndy). So it seems that one score per team isnt enough.

My proposal
My idea is pretty simple. Begin rate every player. Determine if a player is a newbie or a old hardcore tanker. If we know this, our estimation of how strong a team is, can be based on who is participating - and more importantly, how good they are.

Let's say that all new players got 800 as personal elo score, and old players got 1400 as elo scores.

Case #1: TeamB (oldBetty and oldBart) wins over TeamA (newAngie, newAndy)

First of all. We can do better than belive that TeamA and TeamB are both a 1400 points strong teams. We actually know who is playing.

TeamA = (800 + 800) / 2 = 800
TeamB = 1400 + 1400 / 2 = 1400

Now the ELO score for the teams are more correct. TeamA with its newbies are rated much lower than TeamB with its oldies.

In this situation, TeamB will only get 2 points for beating the newbies.

If oldAnn, oldAron was playing instead of newAngie and newAndy, the team would be considered even. And TeamB would have won 25 points. Much much better, imho.

Case #2: Team A (newAngie, newAndy) wins or get a draw against TeamB (oldBetty, oldBart)

CNN is back! And this time Team A gets more than a interview. They get 48 points for their amazing victory. (as two newbies are not supposed to beat two oldies)

Even if they dont do that good, and gets a draw, they'll still get 23 points. TeamB should be able to win this one, according to the numbers.

Dosnt this approch seem more fair; that a newbie team beating a older team, gets a bigger reward, than a old team beating newbies?

CNN is bored, let's see what happens with the last case.

Case #3: TeamA (newAngie, newAndy) vs TeamC (newCaden and newCathy)

Before: TeamA was rated too high, because of the previous matches by oldAnn, oldAron. If the new players lose a match, they shouldnt be punished for this.

As all players got 800 in score, the "teamscores" will be:

TeamA = 800 + 800 / 2 = 800
TeamB = 800 + 800 / 2 = 800

.. No surprise. Teams got even scores, so depending on who wins, they will earn 25 points. Draw gives nothing.

So, how does personal ELO changes?

If a team wins a match, earning them 25 points, the 25 points are shared between the players. If we look at case #3, and TeamC wins, newCaden and newCathy will get additional 12.5 points each (each getting a total of 812.5). TeamA's newAngie and NewAndy gets a new elo score of 787.

If TeamA and TeamC rematches. Now, TeamA have swapped newAngie with oldAnn. So TeamA is stronger than before, even trough they have just lost a match. Let's see how the the system reacts.

TeamA: 787 + 1400 / 2 = 1093.5
TeamC: 812 + 812 / 2 = 812

Now, ELO will give them TeamA probability that they will win, as they just got a stronger player (oldAnn) participating.

If TeamA wins they will earn 8 points
If TeamB wins they will earn 17 points
If its a draw, TeamA wins 42 points.

Now, because of TeamA's swapping of players, TeamB is considered as the underdogs - so if they win, they will gain many points.

But how about the teamscore shown on the website

Ok, if all players got a personal ELO stats and the team elo score depends on the participating players - what score should we then use as the "overall" teamscore - e.g. for showing/sorting teams on the league webpage?

One can use the average of all players in the team to calculate how good the team is. That would probably be the best. There is alot of others ideas, but its basically boils down to, using the player elo's to find a team elo.

Final words

You wont belive this, but I've left out some details :). And please understand, this is not a perfect idea. And it is not final. I've outlined some good/bad things about it in my first (and alot shorter) post.
I don't need huge pictures here.
chtibidasse
Private First Class
Private First Class
Posts: 14
Joined: Sat Jan 07, 2006 6:52 am

Post by chtibidasse »

sussi wrote: That is like wash me but don't make me wet.
Give me some soap and sand, I'll wash you without getting you wet... :twisted:


Chestal said : Maybe we need a good example because there seems to be some confusion.

For me there was a lot of confusion. So thanks [dmp] for the example. It's more easy for me to get it now. And as far as I do understand it realy, I agree with your proposal.

I hate maths.

Chtibidasse :turtle:
Admirarch
Private First Class
Private First Class
Posts: 33
Joined: Sat Sep 11, 2004 9:06 am
Location: Seeking lost whimsy

Post by Admirarch »

dmp: Agreed with everything there except the calculation of team scores to be shown on the website. There I'd prefer the team scores to be solely based on the matches, as per sussi's idea, rather than changing on the addition and removal of players as they would with averaging.
User avatar
[dmp]
Captain
Captain
Posts: 282
Joined: Mon Dec 09, 2002 3:20 pm
Location: CPH, Denmark
Contact:

Post by [dmp] »

Yeah. That part is something I didnt give much attention to (my focus was suggest that we need more data to make sensible teamscores), so it isnt really something I feel very strongly about. But when thinking about it, sussi's idea makes more sense.
I don't need huge pictures here.
CapN
Private First Class
Private First Class
Posts: 42
Joined: Sat Dec 07, 2002 7:21 pm
Location: Maine, USA

Post by CapN »

DMP: I think I like your idea, it does seem to make sense, and I think your right in that it might create more activity from newer players. They should be more comfortable taking on a match they know they might lose, as the risk of losing a lot of points for their team is less.

It's still a little fuzzy in OldCapN's head though. So, obviously the key to this whole idea is a person's individual ELO, Right?

So, I think a real good discussion of what makes up the individual ELO might be in order. (or maybe just for me, because i dont understand).

So initially where would we start with an individual's ELO? It seems like it would need to be pretty subjective? How many different "levels" would there be? Or would it kind of all work out as soon as more matches were played. Meaning everyone start out relatively the same...and over time people's elo would probably end up where it should...based on wins or losses.

One part that sort of sticks out to me is the whole notion of "individual" elo. Is it really individual?? It seems very grounded and tied to a teammate or teammates. For example, many of the top teams in the league now have 8+ members. In some of those teams there are a few dominant players, players that can pretty much carry a match, especially against a weaker opponent. If OldDominantPlayer, and his AverageTeamie beat most opponents, isn't AverageTeamie's ELO a little skewed toward the high end? I'm just not sure how we would really create a truly individual ELO if it's soley based on TEAM victory points. I think this comes into play in reality quite a bit. Especially given all the talk of creating more active teams. One good player starting a new team, takes on some new players etc. I could see that team beating some of the average teams based a lot on the skills of the good player. What effect would that have his/her teammates ELO's?

Like I said up top, I do like the idea and think it makes sense, just not sure about the whole individual ELO calculations.

-CapN-
User avatar
sussi
Private First Class
Private First Class
Posts: 142
Joined: Sun Mar 23, 2003 9:39 pm
Location: Germany

Post by sussi »

Teamies have to start at the bottom (my proposal is 1000), teams at the present 1200.
smalltux
Private
Private
Posts: 4
Joined: Sun Jun 18, 2006 2:15 am

Post by smalltux »

Just wanted to add some redundant force to what sussi said. 8) The point of the league must surely be to encourage playing matches and not having good players per se, right? In that case, I would have an easy choice. The team score should reflect the matches, not necessarily the players.

So, judging the idea so far, I think it'll be in line with the current spirit of the system. Thus, I'm "carefully optimistic" as we say in Sweden. Most of my concerns have been adressed - I feel the overall effect will be very nice.

/smalltux (David_V)
User avatar
sussi
Private First Class
Private First Class
Posts: 142
Joined: Sun Mar 23, 2003 9:39 pm
Location: Germany

Post by sussi »

THe idea is to alter the team score if newer/weaker player going to match. So the player ELO is only calculated/shown in contrast of his team. Thats also very nice to show were a single players stays in the team.
Post Reply