random setting is not always balanced

[rootbeer23]

the auto-balancing work ok in most cases, but there are still instances in which it chooses close to the worst matchup.
can you fix it? i can write down a lua function that would calculate the best possible balance if that helps.

[Ze_PilOt]

No, I can't do miracles.

If I do it the best way possible (computing all possibilities and take only the best one), I have to compute all the combination, minus the duplicates.

The formula being (x! / (x-y)! y!) / y (x being the numbers of players and y the numbers of teams).


in 2vs2 :
4! / (4-2)! 2!

that's :

(4*3*2*1) / (2*1)*(2!) = (12 / 2) / 2 = 3.

(ie)
A B vs C D
A C vs B D
A D vs B C

For a 4 vs 4 :

((8*7*6*5) / (4*3*2*1) ) / 2 = (1680 / 24) /2= 32.

For a 6 vs 6 :

((12*11*10*9*8*7) / (6*5*4*3*2*1) )/2 = (665280 / 720) / 2 = 462.

In lua, the evaluation of one team combination is about 0.5 sec.

That's mean that for a 4v4, it will takes roughly 16 sec to compute the team combination.
In a 6v6, roughly 4 min.

That's not possible.

And even if it was, it can't do miracles. If they are big disparities in the players, you won't have a game quality > 40% whatever you do.

[Softly]

How does the current system alter balance?

I've noticed that you're more likely to spawn in the spot you were in in lobby than anywhere else.

[rootbeer23]

In lua, the evaluation of one team combination is about 0.5 sec.

assuming 6v6 do this once:
1) add all 12 ratings, divide by 2. A = (R1 + R2 + ... + R12) / 2

for each possible team composition:
add up 6 numbers and compare to A. keep the combination where the difference is minimal.

[Ze_PilOt]

It's not that simple.

Remember that your rating is not a simple number but a gaussian curve.

Here is the code that compute the game quality : https://bitbucket.org/thepilot/forged-a ... eskill.lua

[Gowerly]

12!/6!(6!) = 479001600

That's how many ways there are of arranging 6 people from 12. Will take a while to choose the best one there.

[Ze_PilOt]

12!/6!(6!) = 479001600

That's how many ways there are of arranging 6 people from 12. Will take a while to choose the best one there.

Actually 12! / (12-2) 2, divided / 2 as team A vs team B is the same as team B vs team A.

And i'm wrong in the previous post (written before mu coffee:). But that's not important : Lua can't handle it easily.

[Gowerly]

Does that work when the suggestion was just to find 6 people with scores that added up close to the average?

[Ze_PilOt]

That's one of the worst solution to do that actually. (and it won't work as we have to compute gaussian curves, not numbers).

The current solution is simple, fast, and work most of the time (unless you have new players with high deviations but high mean).

[rootbeer23]

are you trying to arrange the teams in a way that the uncertainty of the rating has approximately the same magnitude
in both teams? If i look at my curve it still gives me a single number that is the best guess for my rating.
can you not compute the curves and then use the best guess as an integer to bruteforce the team composition?