+- Official Battle Tanks Community (https://btanks.net/forum)
+-- Forum: Battle Tanks - Warcraft III (https://btanks.net/forum/forumdisplay.php?fid=3)
+--- Forum: Suggestions (https://btanks.net/forum/forumdisplay.php?fid=8)
+--- Thread: ELO calculating in team balance (/showthread.php?tid=3217)
ELO calculating in team balance - progg - 2012-03-30
Hi.
I just had an idea.
All have some win straeks and lose streaks. This provides a situation where 1600 player (i mean rly good player) can have like 1550 points after some games lost staraight and 1500 (average decent player) can have 1550 after winning streak.
It's not that bad when there are 1-2 players like this in game, but when there are more teams can have equal alo but be super ultra imba^2 lanced.
Why not use average from last 20 games to calculate elo for balance.
I mean - for those who care - in ranking your current elo.
In balance counting - average elo from last 20 games.
This is an example of game i was talking about.
http://league.btanks.net/lastgames.php
Alton better than bom.
Jojoy better than baaam.
Sohai equal to bombodron
Dami equal to redemption.
Who played some games with those players knows this is total bs.
My suggestion could fix this problem at least partially.
RE: ELO calculating in team balance - Alexandro - 2012-03-30
When i look at the teams they are really equal for me^^
RE: ELO calculating in team balance - Max - 2012-03-30
(2012-03-30, 07:28:27)Alexandro Wrote: When i look at the teams they are really equal for me^^
You are good enough to...
Good idea i think. We need sth to balance teams when some good guys have a ponctual low ELO and unbalance teams.
RE: ELO calculating in team balance - LIoOoOoIL - 2012-03-30
i dont see any sense in it.
if a "average" player wins like ~15 of his last 20 games (and is maybe still not pretty good), he gets high elo balanct. If an "good player" loose around 10 of his last 20 games he gets balanct as average. So?
The problem u mentiond would be still there.
or did i understood something wrong?
RE: ELO calculating in team balance - Max - 2012-03-30
Kinda hard to find "bad player" winning 15 games on 20... It still could balance a little i think. Even if he wins 15 of 20, he still gets his "bad ELO" into the average which low his real one in balance, same on the opposite for good players. I guess...
RE: ELO calculating in team balance - LIoOoOoIL - 2012-03-30
did pick the ~15wins just to clear out the example, and progg just took the average elo from the last 20 games, nothing about to add it with his real elo
RE: ELO calculating in team balance - Bob666 - 2012-03-30
Using an average ELO of past games is actually not as absurd as it may seem. ELO is basically good for 1vs1 situations, but no matter how good or bad a player is, there may still be a win or lose streak because of team mates, just out of random reasons. Using like 5 past games COULD help here, but I doubt it.
Rating a single player in a game using support points and so on would be the only real solution, but this will never be able to map the real strength of the player, since not all tactics result in kills, not all tactics result in assists, or in a high tank value.
So all in all, there's nothing one can do about it.
However, I am not sure if or how this is possible, but the league predicts the change of one team to win. Maybe players with a win streak can be swapped into the team with the lower chance to win and vice versa. This way streaks could be reduced. Winning more games in a row then just gives more points, because the predicted result was a loss. This could also help in finding the real strength of a player faster.
RE: ELO calculating in team balance - progg - 2012-03-30
Alex i rly don't understand why u don't get it.
Let's take next example.
Player named turboboob has elo between 1580-1620 for 80% of time. However sometimes he plays drunk etc and goes down to 1520 points after 2 days of playing. Sometimes also he gets a streak of imabalanced games when there's no way to balance the elo of teams and he gets elo of 1660.
Obviously hangover is finished and he playes a game with 1520 elo he's imba. Same as when lucky balance streak is finished. Both vaules 1520 and 1660 (remember these are just and exmaples) are non representative deviations. Making elo (only for game balancing) as an average elo from past x games would mitigate a risk of having elo different than from range of 1580-1620 (in this particular case) It would obviously lower the risk of having imba games coused by non representative elo.
To be clear couse after reading your post i think U might understood it as an perposition of calculating whole elo only on basis of last 20 games - i will make an example.
Games Elo
1. 1530
2. 1537
3. 1529
4. 1540
5. 1543
6. 1550
7. 1555
8. 1546
9. 1538
10. 1547
11. 1538
12. 1528
13. 1535
14. 1544
15. 1534
16. 1524
here "hangover games start"
17. 1512
18. 1504
19. 1494
20. 1483
What i'm saying is this guy's elo is 80% time in 1520-1550 range. After only few games it goes down to 1480.
Not that big difference? Yes.
But imagine second guy, who's normal elo is 1480 but he had winning streak and he has 1530 now.
Now imagine 4-5 guys like this in one game and in extreme situations (like in game i linked bad in first post, but now proper linke - http://league.btanks.net/game.php?gameid=29132 ) 7-8 guys like this.
Result?
24 min game. 52-12 in kils.
No doubts if we take their average elo from last x games to calculate elo for balance only it would be more reliable.
RE: ELO calculating in team balance - eSVau - 2012-03-30
I do like the idea of this balance change
RE: ELO calculating in team balance - ssl - 2012-03-31
(2012-03-30, 20:13:49)Bob666 Wrote: since not all tactics result in [...] a high tank value.
I'd like to contradict this opinion. Can you name an example where this is true?
RE: ELO calculating in team balance - progg - 2012-03-31
I give u an example just from last game i played.
We had 4cps and both towers down in base. We focued in def and took 2 other cps. Then we've lost mid.
While other guys attacked mid i was buing bombs and facts to pwn the base. I spent like 15k on it in 5 mins. No tank value increase.
RE: ELO calculating in team balance - ssl - 2012-03-31
Valid example.
RE: ELO calculating in team balance - UnifiedDoom - 2012-03-31
While the logic to balance the teams more effectively is a good idea, I think a change like this will actually counter-act what should. By taking the average elo of the last 20 games a player's "balancing elo" will lag far behind his "streak elo." Now I think this is what progg was actually going for because now players will be balanced based on their overall skill (in the past twenty games at least) rather than the skill represented by their current "streak." But by making the balancing elo lag behind the real elo we will give players on "win streaks" consistently better teams than they ought to get. Conversely players are "losing streaks will" be given worse teams than they should because their previous average elo was high. In a way a system like this will encourage having many wins or losses in a row by skewing the team balance to the past, not the present.
I think the main problem with the elo system as people mentioned before is that it isn't perfect and doesn't account for all the variables. It's very unlikely in fact almost impossible that two people with the same elo like 1550 have the exact same skill level, however the system can't tell. Randomness like having "alton better than bom" or "dami equal to red" is just part of league. If we gave those players their "real" elo's it's very likely bom and dami would be at a disadvantage because their current "losing streak" has lowered their morale. I like the current system because it rewards people who are having hard times with better teams but challenges people who are having good times with harder opponents.
My problem with the system is simple, giving all 0 ranked games players 1500 elo. While 1500 elo may seem average because that's basically wins=losses, I think someone who can consistently sustain 1500 is actually above average. Legitimately any "noob" who only has 25 games under their belt is NOT at this level of skill. Yet that same player will be ranked that way. look at this game: http://league.btanks.net/game.php?gameid=29105
both teams had a player with 1500 elo, yet one was very clearly better than the other. I'd say the real value of teal was around "1400" and his presense on the darkforce made it a really tough win for that team.
I've already proposed this, but I think I was ignored so I'll propose it again: Give new players fresh out of qualify the balancing elo "1450" for their first 5 games ONLY. Then take whatever they get in wins/losses in their first 5 games and add/subtract it to 1500 to get their elo for the 6th game. So for instance if the player turns out to be like teal above and most likely the cause of many loses for his team his elo will "normalize" at a low level without having to "ruin" his first 5 games. Conversely if the player is a pro in disguise after 5 games he will be given an elo fitting to someone on a win streak.
I also think that because league ranking is indeed given on an individual basis, more emphasis should be placed on individual "game performance." Right now the range is only 0 - .15 for losers and .85 - 1.00 for winners. Now ideally I would say that the ranges should be 0 - .50 for losers and .50 - 1.00 for winners so if you play well enough as an individual you can sustain a loss without losing positions in the individual ranking. Like in this game where I ended up losing with 49-14 stats: http://league.btanks.net/game.php?gameid=28670
However I think such a radical change would shift the league system too much. I would like to see it made incrementally perhaps starting at .0 - .30 for losers and .70 - 1.00 for winners?
RE: ELO calculating in team balance - Prog - 2012-03-31
(2012-03-31, 04:29:20)UnifiedDoom Wrote: I also think that because league ranking is indeed given on an individual basis, more emphasis should be placed on individual "game performance." Right now the range is only 0 - .15 for losers and .85 - 1.00 for winners. Now ideally I would say that the ranges should be 0 - .50 for losers and .50 - 1.00 for winners so if you play well enough as an individual you can sustain a loss without losing positions in the individual ranking. Like in this game where I ended up losing with 49-14 stats: http://league.btanks.net/game.php?gameid=28670
However I think such a radical change would shift the league system too much. I would like to see it made incrementally perhaps starting at .0 - .30 for losers and .70 - 1.00 for winners?
That would be really bad, because it encourages egoistic play too much.
RE: ELO calculating in team balance - progg - 2012-03-31
Doom i think we're partially both right. Elo and balance system is too compleacted to forsee exact effects of change like this. Probably some talented maths guy could build a model of this and tell us but we all know it's not gonna happen untill we try it.
Some time ago i also proposed to make ranking for both quali and ranked games with tresholds limiting join "advance" and "begginer" games (instead of quali and ranked).
This system i think could solve the problem u mentioned (1500 elo 0 games) and maybe encurage good players (there are quiet lot of them) form quali to join raanked games and force smoe guys we all know to play "novice" server;)
RE: ELO calculating in team balance - UnifiedDoom - 2012-03-31
(2012-03-31, 12:08:25)Prog Wrote: That would be really bad, because it encourages egoistic play too much.
Nobody has ever called me out for ego playing, please explain how this is true.
RE: ELO calculating in team balance - Prog - 2012-03-31
If I can get to 0.5 in the loser team based on my own player stats, I will obviously steal as many kills/creeps/towers as possible and never lose any points (or just very little). That is to a certain extent already the case with the 0.15 maximum, but I think anything higher than that is unreasonable.
RE: ELO calculating in team balance - UnifiedDoom - 2012-03-31
(2012-03-31, 14:09:39)Prog Wrote: If I can get to 0.5 in the loser team based on my own player stats, I will obviously steal as many kills/creeps/towers as possible and never lose any points (or just very little). That is to a certain extent already the case with the 0.15 maximum, but I think anything higher than that is unreasonable.
Perhaps but with other variables like assists, cp's captured, and survival (to prevent/discourage suiciding for extra kills) game performance is not based on selfish play alone. I agree with you that building kills, creep kills, tank worth and individual kills are weighted far too high in this though. You talked about the situation on the losing team, but what about someone on the winning team who clearly doesn't deserve a win like in this game http://league.btanks.net/game.php?gameid=29105 (which i've already mentioned). Teal's endgame stats may seem alright (tied with grizzly) but his earlygame was easily 0 10 in the first 5 minutes and he didn't actually get any kills until after his team had taken all of the cps. Yet he's gaining points in the ranking anyway despite being an obvious detriment to his team. And because of this win he will be treated like a good player with 1508 elo in his next game, despite the fact that he clearly isn't. I have this replay if you want to see it (I admit I pulled a super selfish build in middle, but I did what I felt was nessecary to win)
RE: ELO calculating in team balance - GEN_Schwarzkopf - 2012-04-01
(2012-03-31, 04:29:20)UnifiedDoom Wrote: My problem with the system is simple, giving all 0 ranked games players 1500 elo. While 1500 elo may seem average because that's basically wins=losses, I think someone who can consistently sustain 1500 is actually above average. Legitimately any "noob" who only has 25 games under their belt is NOT at this level of skill. Yet that same player will be ranked that way. look at this game: http://league.btanks.net/game.php?gameid=29105
both teams had a player with 1500 elo, yet one was very clearly better than the other. I'd say the real value of teal was around "1400" and his presense on the darkforce made it a really tough win for that team.
I agree new players should start with 1450 ELO.
You can have more wins than losses and still have <1500 ELO if you are a support player. Like chess this game is about winning, not just farming and player kills.
The ELO system originated in chess. http://en.wikipedia.org/wiki/Elo_rating_system You win you get the same points. The correction factors for skill shifts ELO below 1500 for equal wins:losses for support players and above 1500 for ego players.
- Rob
RE: ELO calculating in team balance - El_Polacco - 2012-04-02
I think the main problem is that some players who have poor elo under 1400 who are destroying quality of ranked (i think once i met 1320 ^^) imagine the guy who have 1750 elo got that player in team then he receives huge advantage since beginning. Another example top ranked players , were in one team and played against players who have similar ELO . Such ELO imparities are doing negative effect so many times i played
I guess that its really unfair for that team cause of 2 reasons .
a)Playing against better skillled people which require perfect teamplay from worse players
b)Possibility to play next boring game due to leavergold ( after votekick worse player according to elo ) in enemy team and losing fun to play (as i was doing
)In my opinion there should be possibility like after disconnect in any team its needed to make players balance in also another team if they agree
Then player who wanted to leave could do it without penalty points. For most of players playing against leavergold team is pretty boring i suppose then how about this but i realise it can be influential factor
It's just a suggestion not sure if this one might make more fairness in games