Alternative balance

Ive noticed the teams auto balancing does some wierd things sometimes, it has a habit of say putting the highest elo players and some of the loweest on the one team and making the other team full of mid range, this may be mathematically balanced but its not what u would expect. Wouldnt it make more sense to simply take it from the top and pair off each player to each side. It would be nice to add an alternative balance command !balance-alt or
!balanceclassic to execute when the balance seems a bit messed.

True but if there was more than one command then at least multiple combinations might be possible.

quote: The original balance would make: 1800 + 1100 + 1400 = 4300 1500 + 1500 + 1200 = 4200 But ur suggestion (If I did understand you right): 1800 + 1500 + 1200 1500 + 1400 + 1100 But this balance is really sux

What does give much better results is pairing the highest with the lowest, again and again, until you have remaining pair - which you split between the two teams, giving the highest to the lowest total elo side and the lowest to the highest total elo side.

eg.

1800 + 1100 (+ 1400)

1500 + 1200 (+ 1500)

It gives better results than the default balance in practically every example, and will certainly result in 1000% more fun for players with ridiculous high elo who get unwinnable teams of spastics every game under the current system.

The reason for this is that there is no linear relationship between the elo number and the player skill, so simply getting the two 'total numbers' of each team to be equal does not create even games. A 2500 elo player will not really sway a teamgame that much more than a 2000 elo player - they get the same amount of resources, which can only achieve so much - but they will certainly get a far, far worse team. Its better to simply rank the players from highest to lowest and assign from them to teams, if your objective is a fun and balanced match. I will repeat this forever.

Apparently, a bunch of balancing algorithms had been tested before and the current one got picked because all others sucked.

As has been pointed out in every other balance thread, we have a huge database of played games; if you think you can devise a better balancing algorithm than the current one, prove it by checking it against those games.

Actually there should be a stronger relationship between elo and skill than [GBC]1v0ry_k1ng said. Theoretically the balance of any player setup is the same if every player has 500 more elo. Only elo differences matter. If an elo 2500 player doesn't do so much more for his team than an elo 2000 player, he should not be able to hold this elo. In reality he may be because there are other deformations in the elo system:
- High elo players are more likely to get 2nd com.
- There are no seperate elos for 1v1, team and ffa yet.
- Low level players are misvalued , mostly overvalued even though 250 offset, experienced TAstyle rts players/smurfs undervalued. (The offset also causes the player average to differ from 1500 (a value that is usually held constant in an elo system). However the game average is higher than 1500 because high skill players are likely to play more, which is no problem.)


If 4hundred 's proposal means alternately adding players to teams from the top, the team with the highest elo player is always better (or same) than the other team for even total numbers of players and it's randomly well or bad balanced for odd numbers. If it only means pairing every two players with neighboring elos and then dividing pairs to teams by minimizing teams' average elo difference, it may be better. But this would give elo variance equality a higher priority than elo average equality..

In [GBC]1v0ry_k1ng 's system there is a high but decreasing oscillation of teams' average elos during the apportionment and the resulting difference between teams' average elos can be low for certain numbers, but often will be very high.

@Hower 's system doesn't eliminate the irregular distribution for high player numbers and increases the difference between average elos compared to the current system (especially for low player numbers).


You won't get really good balance if you only rank players by 1st, 2nd and so on. The current system minimizes the difference between teams' average elos by calculating it for all possibilities with desired team sizes afaik, which is the most important thing. An improvement is only possible if there are a few possibilities with low avg elo difference. You can always find the ideal solution by specifying the value that is to be minimized. If A is team 1's elo distribution, B team 2's elo distribution, I suggest minimizing (avg(A)-avg(B))²+c(sqrt(Var(A))-sqrt(Var(B)))², where c is a constant that should be picked not too high for good balance. I guess I could easily write an algorithm that minimizes this with same scale of complexity as current one.

quote: Its better to simply rank the players from highest to lowest and assign from them to teams, if your objective is a fun and balanced match.

It is trivial to prove this wrong.

Just given the order of player's elo (not the elo themselves, just as you suggested), you cannot conceive a matching algorithm that creates balanced games for all cases.

Just as an example, the player elos 1900, 1701, 1700, 1501, 1500, 1300 need to be balanced differently than the elos 1901, 1900, 1701, 1700, 1501, 1500, but they would inevitable yield identical matchups with your approach (since the elo order is the same).

TheEloIsALie

quote: we have a huge database of played games

Le me

quote: impart database! (preferably in excell format)

where is my database?

Brackman But making total elo (or average per team) equal isnt he way to balance teams.

Look -

2k + 2k + 1k + 1k = 6k
1,5k + 1,5k + 1,5k + 1,5k = 6k as well.

Balanced? Not really.

There are at least 2, sometimes 3 or even more possibilities to balance elo numbers equealy. Problem is, current balance amost always pick the worst one.

With those numbers good balance would be:
2k + 1,5k + 1,5k + 1k = 6k
Both teams would be actually equal, but the Balance system would pick the 1st way propably.

[GBC]1v0ry_k1ng
with those numbers given every single, even most retarded balance algorithm would work (not saying ur is retarded). I was just saying that every random group of players can be divided to two teams in at least two (or more) ways, while keeping total (average) elo in both teams equal. But equal elo division doesnt mean equal skill division.