Rework ELO-formula?

[SirMalum]

I have read some things about the ELO-system and I think, the formula you are using is not the best.

Example:
You have an ELO of 1000, due to balancing both teams have an ELO-sum of 7500. The balancing reassures, that you have a rather balanced game and a 50% chance to win.
But according to the formula at the http://league.btanks.net/ you get an

expected_value = 1 / ( 1 + 10 ^ ( ( elo_sum_of_light_force - ( 2 * elo_player_of_dark_force + 3/5 * elo_sum_of_dark_force ) ) / 2000 ) )

which gives expected_value=1/4.16 = 0.24.
This way you can win 15.2 points, but only lose 4.8 points even though the chances to win are 50% and you have rather the higher ranked players in your team.

I don't say your formula is wrong, cause there is no fixed definition for ELO-ratings afaik. (Maybe you can tell me where you got the formula, cause wikipedia only shows is for 1v1-games). But the formula leads to the fact, that good players can not get a very high rating because they get noob mates AND lose more points than they can win.
It is not generally bad that good players have only 1676 ELO (current leader), but in a system where you can easily lose 15 points in one game, a difference of 176 to the starting rating is not very significant.

How about just using the formula
expected_value = 1 / ( 1 + 10 ^ ( ( elo_sum_of_light_force - elo_sum_of_dark_force ) / 2000 ) )
casue then the expected_value really represents the expected chances to win the game.

DISCUSS!!

[eSVau]

I have read some things about the ELO-system and I think, the formula you are using is not the best.

Example:
You have an ELO of 1000, due to balancing both teams have an ELO-sum of 7500. The balancing reassures, that you have a rather balanced game and a 50% chance to win.
But according to the formula at the http://league.btanks.net/ you get an

expected_value = 1 / ( 1 + 10 ^ ( ( elo_sum_of_light_force - ( 2 * elo_player_of_dark_force + 3/5 * elo_sum_of_dark_force ) ) / 2000 ) )

which gives expected_value=1/4.16 = 0.24.
This way you can win 15.2 points, but only lose 4.8 points even though the chances to win are 50% and you have rather the higher ranked players in your team.

Nothing wrong, so far - besides 1000 elo would be a real worse (most unlikely) value

I don't say your formula is wrong, cause there is no fixed definition for ELO-ratings afaik. (Maybe you can tell me where you got the formula, cause wikipedia only shows is for 1v1-games). But the formula leads to the fact, that good players can not get a very high rating because they get noob mates AND lose more points than they can win.

ELO is only defined for 1vs1 games, that why we modified the formula to account the team aspect - you can't win or lose a game alone, you're always depending on your mates.

It is not generally bad that good players have only 1676 ELO (current leader), but in a system where you can easily lose 15 points in one game, a difference of 176 to the starting rating is not very significant.

It's part of the idea: as good player you're supposed to win. On the other hand as worse player you win 15 points beating a (expected) stronger enemy.
League has been started two weeks ago, we had couple of month of beta testing but top player were still raising up. But not every player do have the same activity or game count, some even got more than a handfull of games within month ob beta testing (there is a mapper who didn't even managed to qualify himself).
It's more common to use ELO with a more tournament wise manner, read: every player has more or less the same amount of games which increase the explanatory power of those elo ratings. But that is not the case, our playerbase fluctuate to much, which causes from a point of view of a good player to play always versus close to start rating players, which will result into a small point gain.
It will take quite some time till ELO rating will stabilize (and auto balance will work fine).

How about just using the formula
expected_value = 1 / ( 1 + 10 ^ ( ( elo_sum_of_light_force - elo_sum_of_dark_force ) / 2000 ) )
casue then the expected_value really represents the expected chances to win the game.

Unmodified version of elo expected values, comparing both teams. Also a valid formula but a boring one in combination with auto balance, everybody would gain or lose the same amount of points every game (ignoring our bias due gameperformance).

Actually our current formula is a trade-off between comparing the sum of both teams and your own elo rating vs the fifth enemy team rating. Trying to account both, team dependency and personal skill.