Poll: Do you want a 20:10 ELO w/l cap for all LIHL games?

[ILOCOS_NORTE]

a 2s at 25/5 elo for example is perfectly balanced. its just means at least one player is very good and others from opposite are bad.
however it is not a game worth playing for the 5 elo team ( risk vs reward ), should they care about elo.

@pewpew lasergun
If you think the 25:5 game is perfectly ballanced do you also think that if those 4 players play 6 games, the stronger team would win 5 out of 6 ? If you think that you are right with the perfect ballance, If you think it will result in something like 4:2 the game is not perfectly ballanced.

[ILOCOS_NORTE]

Without taking the poll into consideration from this topic (https://entgaming.net/forum/viewtopic.php?f=83&t=54688), do you think your suggestion would solve the fears voiced in that discussion?

Yes

[ILOCOS_NORTE]

20/10 is indeed arbitrary. his comment to uakf.b not withstanding as no 1500 ELO players are with 950 ELO players against 1000-1100 players in 2v2s - it's not viable for ELO gain as he said, so isn't within the statistics. the 66% figure based off high ELO players dodging 2v2s and only playing with low ELO players anyways to keep gains 15/15 doesn't reflect a true ELO (eg. one with scaling limiting the imbalance)

First I said 1400, not 1500. Secondly, I did play many 2s games last season with an ELO around 1400 and 20:10 up to 25:5.

You are right if you say I dont have valid data of the win rates since most high ELO players dodge the 2s games. That is why I put the cap @ 20:10, wich is still high since it represents a 66,66% win rate of the strong team.

While you are on the hand correct that the 20:10 value is not 100% correct, this value is not arbitrary. Im using a mathematical trick here called "interpolation". Means I approximate the win rate (setting a value from what I think could be true, in this case the 66,66% win rate). I do that because I dont have enough data. After the 20:10 ELO cap is fix, I need the players to reach their "true ELO" and keep playing 2s games. Then I can compare the "real winning rate" with the 20:10 ELO distribution. Maybe I come to the conclusion that a 22:8 cap or a 19:11 cap would be better. But I cant say that for now without valid data. And I dont have valid data until games are actually played for 20:10. It would be best if some people (doesnt matter if high or low ELO) help me and provide me with information about their win/lose rates from 2s (or 3s) games with 1 high + 3 low or 3 high +1 low ELO players.
I hope that was not too confusing, I tried my best to explain it to people without a deeper mathematical understanding for that. If you find something unclear feel free to ask me.

the 20/10 ELO limit could be exploited heavily in 4v4 - if you put a team of 1400+ experienced players on one side, and only players who've never reached 1100 on the other I would expect a win-rate of at least 80%. this won't happen without challenge mode, but wow swap-abuse will be powerful.

I doubt this will happen neither with challenge mode nor with swapping. Who in the low ELO team would agree to swap all good players on one team? Or what captain (low ELO himself) picks all low ELOs while the other captain (high ELO guy) picks all 1400+ ? Without taking into account that there was no game in lihl history with 4x 1400+ in a 4s game (afaik )
This case is a little made up

@eldryanthewise

[ArMeDaNdDeAdLy]

Even the 17-12 games arent fair to me.This is better than the current system but not an ideal solution.You have my support.

[Diablo_]

I also voted yes, but I still think altering the formular would be better, because the Elo gain/loss gets off too quickly, like 18/12 should be 17/13, 20/10 should be 18/12, 25/5 should be 20/10 etc.
But this suggestion would at least kill the most extreme (and most inaccurate) cases.

[eldryan]

20/10 is indeed arbitrary. his comment to uakf.b not withstanding as no 1500 ELO players are with 950 ELO players against 1000-1100 players in 2v2s - it's not viable for ELO gain as he said, so isn't within the statistics. the 66% figure based off high ELO players dodging 2v2s and only playing with low ELO players anyways to keep gains 15/15 doesn't reflect a true ELO (eg. one with scaling limiting the imbalance)

First I said 1400, not 1500. Secondly, I did play many 2s games last season with an ELO around 1400 and 20:10 up to 25:5.

You are right if you say I dont have valid data of the win rates since most high ELO players dodge the 2s games. That is why I put the cap @ 20:10, wich is still high since it represents a 66,66% win rate of the strong team.

While you are on the hand correct that the 20:10 value is not 100% correct, this value is not arbitrary. Im using a mathematical trick here called "interpolation". Means I approximate the win rate (setting a value from what I think could be true, in this case the 66,66% win rate). I do that because I dont have enough data. After the 20:10 ELO cap is fix, I need the players to reach their "true ELO" and keep playing 2s games. Then I can compare the "real winning rate" with the 20:10 ELO distribution. Maybe I come to the conclusion that a 22:8 cap or a 19:11 cap would be better. But I cant say that for now without valid data. And I dont have valid data until games are actually played for 20:10. It would be best if some people (doesnt matter if high or low ELO) help me and provide me with information about their win/lose rates from 2s (or 3s) games with 1 high + 3 low or 3 high +1 low ELO players.
I hope that was not too confusing, I tried my best to explain it to people without a deeper mathematical understanding for that. If you find something unclear feel free to ask me.

the 20/10 ELO limit could be exploited heavily in 4v4 - if you put a team of 1400+ experienced players on one side, and only players who've never reached 1100 on the other I would expect a win-rate of at least 80%. this won't happen without challenge mode, but wow swap-abuse will be powerful.

I doubt this will happen neither with challenge mode nor with swapping. Who in the low ELO team would agree to swap all good players on one team? Or what captain (low ELO himself) picks all low ELOs while the other captain (high ELO guy) picks all 1400+ ? Without taking into account that there was no game in lihl history with 4x 1400+ in a 4s game (afaik )
This case is a little made up

@eldryanthewise

frankly, the math is too basic. your explanation is obvious, yes, but incorrect. approximating is using an arbitrary data set in this instance that isn't correctly modeling this system. whether you change it after it's implemented doesn't change the "20/10 rule" from being misrepresenting. you can use a linear regression model to calculate the likely result of this model before it's even implemented however as I said the data is flawed.

clearly you have never played challenge modes frequently. as this happens all the time - remember joke games against BA that you were in, in fact, with all 1200s on one team va me + 800s on other while I was like 1200. was funny (this is what started trend of low-balling ELO picks) but would be stupid for me if this rule was implemented as it wouldn't be +2/-28.

[ILOCOS_NORTE]

I also voted yes, but I still think altering the formular would be better, because the Elo gain/loss gets off too quickly, like 18/12 should be 17/13, 20/10 should be 18/12, 25/5 should be 20/10 etc.

Already working on that. I texted uak following yesterday:
[...] The question remains in wich way we alter the formula. Keep it the same and just set a cap at 20:10 or create a new function which approachs 20:10 (wich I think is better). [...]
I didnt make that public yet because I wanted uaks confirmation first.


As soon as he answers I'll you provide you witch fresh infos and maybe a new formula already. Dont know yet if uak giving me the current one, if he wants me to alter it himself or me to do it etc.

[MarshMallows]

Yeah games that have 1400 vs 1325 elo average should be almost 15:15 imo, the system seems to be bonkers!

[ILOCOS_NORTE]

frankly, the math is too basic. your explanation is obvious, yes, but incorrect. approximating is using an arbitrary data set in this instance that isn't correctly modeling this system. whether you change it after it's implemented doesn't change the "20/10 rule" from being misrepresenting. you can use a linear regression model to calculate the likely result of this model before it's even implemented however as I said the data is flawed.

Yes the data is flawed because of high ELOs dodging 2s. We simply dont know the average expected winning rate of high ELO players in 2s. And without a solid data base it is impossible to identify the correct ELOs !scores, be it 20:10, 22:8 or 18:12. If you can do that with your linear regression before the system is implemented and find the correct ELO cap I would nominate your for the nobel price.
So, what do you suggest to do before implementing the new system? No good me @ future mathers, pls help him

***Edit:

frankly, the math is too basic. your explanation is obvious, yes, but incorrect. approximating is using an arbitrary data set [...]

You never chose an arbitary data set when doing approximations because the computing would take too long. You think first, identify the possible number range (which is in this case 15:15 to 29,999:0,001) and chose a good value in between this range that might be the correct number. For example, if I just roll 15,5 and 14,5 and implement that into lihl, using the data after the season, find out its to low, knowing the new data range is in between 15,5 and 29,99, roll a 29.5 now, run the next season with 29,5:15,5 cap and find out its to high leading to the new number range of 15,5 - 29,5, roll now a 28,0 .... etc. These are truely arbitary extrema just to demonstrate to you that I chose to start with 20:10 for a reason. If you can identify the expected value approximately you never start with arbitary values.***

clearly you have never played challenge modes frequently. as this happens all the time - remember joke games against BA that you were in, in fact, with all 1200s on one team va me + 800s on other while I was like 1200. was funny (this is what started trend of low-balling ELO picks) but would be stupid for me if this rule was implemented as it wouldn't be +2/-28.

Give me 3 game links of lihl games with that !scores

[ILOCOS_NORTE]

I just noticed no one pressed the "THANK YOU" buttom !!!
What do you people think why I'm doin all this ???

[aRt)Y]

What kind of data sets would you need too make rough approximations?

[ILOCOS_NORTE]

What kind of data sets would you need too make rough approximations?

Best would be the 2s !scores and results from future LIHL games were 20:10 is already implemented ...

Since thats not possible,
lets say there have been 100 2s games last season with 20-28:10-2 scores, in average 24:6. The mathmatical winning chance for the high ELO team is in average 80%. The high ELO team wins 4 out of 5 games => 4*6 ELO = 24 ELO, the low ELO team wins 1 out of 5 games = 24 ELO. Teams and scores are ballanced and after 5 games the ELO of each player will be the same as before, in theory ...
The problem in reality is that the high ELO team didnt win 4 out of 5 games. They won, maybe, only 3 games and lost 2. Now the high ELO team gained 3*6 ELO and lost 2*24 = -30 ELO. This would tell me the scoring was wrong. If the high ELO team wins 3 out of 5 their winning chance was 60% and for the low ELO team 40% which would result in 18:12 scores.

What I would need now is for every game that was not 15:15 (99% of games) the winning chances of the high ELO team and the actual result. Overall, in average, I can tell if the high ELO teams won more or less games then they should and identify the correct !scores.

But that's really a butt load of work and absolutly not neccessary. The 20:10 score cap is good enough. Only if one player would go for 2s only, plays with high ELO 20:10 games only and would reach a winning rate of >66,66% (wins more then 2 out of 3 games) he can gain infinit ELO. If someone can farm low ELOs like this we had to increase the cap.

At this point I would just try out a new scoring formula, awaitening uaks confirmation now. Dont need more discussion since 2 lihl admins and 100% of the voters support.

I would close the thread now.

[ILOCOS_NORTE]

Copy this function and use a plotter to see the graph:

12*lg(1+9*(1-(1/(1+(x/1000)))))+15

set following number range:
x 0-500
y 15-30
x interval => 25
y interval => 1

This is a first first design, the parameter can be adjusted easely. Since it's an lg function there is no real ELO cap at 20 to stop people from reaching infinite ELO. But you need >325 ELO difference for a game >20:<10.


3d function:
f(x,y)= (+)12*lg(1+9*(1-(1/(1+((x-y)/1000)))))+15 ELO reward for the low ELO team
f(x,y)= (-)12*lg(1+9*(1-(1/(1+((x-y)/1000)))))+15 ELO reward for high ELO team
with:
x = average ELO of high ELO team
y = average ELO of low ELO team
x ∈ [0,∞]
y ∈ [0,∞]

2d function (easier for plotting):
f(x) = (+)12*lg(1+9*(1-(1/(1+(x/1000)))))+15 ELO reward for the low ELO team
f(x) = (-)12*lg(1+9*(1-(1/(1+(x/1000)))))+15 ELO reward for high ELO team
with:
x = a - b
a = average ELO of high ELO team
b = average ELO of low ELO team
a ∈ [0,∞]
b ∈ [0,∞]

@uakf.b

[dark_magician]

IMO only - if it's possible, would have a condition to say:
20-10 cap only applies if it's 1, 4 v 2, 3 (highest and lowest ELO vs 2nd and 3rd),
else cap is 'void' and ELO is as it would be.

Can't exactly have 2 people wanting to play together knowing there's a 'safety' cap.
Probably a rare situation any way, only likely to apply to Patar and Martiny if anything

[dark_magician]

PS. Locos - that formula is too complex for me haha.