Town of Salem 2.0 - Ranked ELO Changes

Announcements made here about the game and the company.

Re: Town of Salem 2.0 - Ranked ELO Changes

Postby WittyRecluse » Wed Jun 07, 2017 12:16 pm

rickms wrote:
Daselend wrote:Strange right? It's almost as if there is no bug at all disadvantaging lower ranked players, just that they're bad at this game.


Debating if there's a bug or not is senseless.. It existed. It's the exact same math the privileged higher ELO users.

The ELO calculation comes up with a number between 0 - 1, depending on the difference in ELO between you and your opponent. As an example say the ELO difference was large enough between Player A and Player B. Player A was high enough where, for a win his number would be 0.25, and for a loss his number would be the inverse, 0.75

Say we use a k-factor of 20 in this example (for easier math)..
Player A, the superior player having would win 5 (20 * .25) ELO, or Lose 15. (20 * .75)
Player B, would win 15, or Lose 5.

Our error was that we were using the win amount in both cases, so what was happening was:

Player A, the superior player, would win 5 ELO or lose 5 ELO
Player B, would win 15 ELO or lose 15 ELO.

This is why higher players now see larger losses. Lower ranked players who would lose more often than win against higher ranked players would have their ELO drop even further. Widening the gap between them and worsening the problem.

I'm not claiming that ELO is now perfect. The luck factor, dying night 1, etc are all things we look to improve going forward.

The other problem is that this makes it harder to climb up from around 1000-1100 elo. In these elo brackets you often get into the lowest elo queue, so the system will deflate your elo as if you're high elo with +1/-10 every time.

Town of Salem has too much luck involved to make elo based so much on team's elos; make your faction factor way more into the elo you get and your elo factor in way less.

ALSO make NB (surv/amne) a flat rate of -1 and surv a flat rate of +1. This will encourage NB to take risks and side scum; the new +1/-12 forces survs to always side town since it's safest which is really bad for the balance of the game.
User avatar
WittyRecluse
Vigilante
Vigilante
 
Posts: 618
Joined: Sun Jun 12, 2016 8:02 pm

Re: Town of Salem 2.0 - Ranked ELO Changes

Postby rickms » Wed Jun 07, 2017 12:37 pm

WittyRecluse wrote:The other problem is that this makes it harder to climb up from around 1000-1100 elo. In these elo brackets you often get into the lowest elo queue, so the system will deflate your elo as if you're high elo with +1/-10 every time.


Winning amounts have not changed. The bug was purely in the amount lost. A vast majority of ranked players are in the 1000-1300 range so over time the vast majority of their games should be against relatively equal opponents

Town of Salem has too much luck involved to make elo based so much on team's elos; make your faction factor way more into the elo you get and your elo factor in way less.

Your Role/Faction factors HEAVILY into the final amount you lose. Being a SK for example limits your lost to 10% of what the ELO calculation says you should lose.

ALSO make NB (surv/amne) a flat rate of -1 and surv a flat rate of +1. This will encourage NB to take risks and side scum; the new +1/-12 forces survs to always side town since it's safest which is really bad for the balance of the game.


This is a much better argument to make than "ELO is bad/broken". This comes into the territory of our modifications to the ELO number after it's calculated. Survivor has about ~50%, so the swing for survivor can be . +/- 12 at a wide elo gap. Amnesiac has a +17/-7 on the wide range. Witch/Vamp (now that it's fixed)/SK/WW all have very low ~10% modifiers making their losses less painful. We can def revisit this in our Ranked update.
rickms
Witch
Witch
 
Posts: 47
Joined: Mon Jan 11, 2016 1:08 pm

Re: Town of Salem 2.0 - Ranked ELO Changes

Postby Gracee » Wed Jun 07, 2017 12:38 pm

I'm not claiming that ELO is now perfect. The luck factor, dying night 1, etc are all things we look to improve going forward.


Question: Why do BMG insist on using the ELO-system when it's not made for games like these?

Then I would like to talk about the Wiki-page you guys are throwing around, making it seem like we don't know how ELO work.
From the Wiki: The difference in the ratings between two players serves as a predictor of the outcome of a match.
Problems: There is no way to predict the outcome of a game in Town of Salem. There are too many factors that could easily change the outcome without any action from the player itself.

From the Wiki: Two players with equal ratings who play against each other are expected to score an equal number of wins.
Problems: Equal rating does not exist when the players in one game can vary 1.5k ELO or more, from top to bottom.

From the Wiki: The lower rated player will also gain a few points from the higher rated player in the event of a draw.
Problem: All of your draws are 0, which means you're not enforcing the ELO-system in case of a draw.

From the Wiki: The gradation of the K-factor reduces ratings changes at the top end of the rating spectrum, reducing the possibility for rapid ratings inflation or deflation for those with a low K factor.
Problem: You're not using a gradation of the K-factor, meaning it's even worse for high ELO to randomly be put into games with lower ELO-players. Seeing how you're trying to avoid the extremes in the ELO-spectrum, why are you not using a K-factor gradation?

I would like educated answers to these points, reasoning behind it and how it's being worked-around.
Gracee
Escort
Escort
 
Posts: 75
Joined: Thu Feb 11, 2016 1:54 pm

Re: Town of Salem 2.0 - Ranked ELO Changes

Postby ProblemsIRL » Wed Jun 07, 2017 12:44 pm

Can't help but feel that this would not be anywhere near as bad if it came with a hard reset of everyone's ratings.
( ͡° ͜ʖ ͡°)
User avatar
ProblemsIRL
Lookout
Lookout
 
Posts: 89
Joined: Sat Oct 22, 2016 7:59 am

Re: Town of Salem 2.0 - Ranked ELO Changes

Postby WittyRecluse » Wed Jun 07, 2017 12:54 pm

HISSSSSSS wrote:
rickms wrote:
HISSSSSSS wrote:It is sad to see that the average elo of all players is around 1100 (from shapesifter13), considering it really is not that hard.
Please realize that every single player started from the same spot (1200).


Strange right? It's almost as if there was some bug disadvantaging lower ranked players.

It's almost as if players are so inexperienced and brainwashed from youtubers saying VFR is the devil that they always lose.

And then they leave games all the time which means that they're definitely amazing players with 80% winrates that just happen to be low elo.
User avatar
WittyRecluse
Vigilante
Vigilante
 
Posts: 618
Joined: Sun Jun 12, 2016 8:02 pm

Re: Town of Salem 2.0 - Ranked ELO Changes

Postby Daselend » Wed Jun 07, 2017 1:20 pm

rickms wrote:
Daselend wrote:Strange right? It's almost as if there is no bug at all disadvantaging lower ranked players, just that they're bad at this game.


Debating if there's a bug or not is senseless.. It existed. It's the exact same math the privileged higher ELO users.

The ELO calculation comes up with a number between 0 - 1, depending on the difference in ELO between you and your opponent. As an example say the ELO difference was large enough between Player A and Player B. Player A was high enough where, for a win his number would be 0.25, and for a loss his number would be the inverse, 0.75

Say we use a k-factor of 20 in this example (for easier math)..
Player A, the superior player having would win 5 (20 * .25) ELO, or Lose 15. (20 * .75)
Player B, would win 15, or Lose 5.

Our error was that we were using the win amount in both cases, so what was happening was:

Player A, the superior player, would win 5 ELO or lose 5 ELO
Player B, would win 15 ELO or lose 15 ELO.

This is why higher players now see larger losses. Lower ranked players who would lose more often than win against higher ranked players would have their ELO drop even further. Widening the gap between them and worsening the problem.

I'm not claiming that ELO is now perfect. The luck factor, dying night 1, etc are all things we look to improve going forward.


If you think that either +15 or -15 is a bug.. it isn't a bug +1 or -10? Give the high elos +15 or -15, instead of this shitty +1 or -10, i'll take it any day of the week, even i'll take it over my games before the dumb fix you guys did, when i was getting +1-4 and -1-4 the overwhelmingly majority of my games.

Now, instead of having +15 or -15 in some games, low elos have +10 or -1 every freaking game. Wow, perfect system.

Having either +15 or -15, wasn't a bug, it was a blessing for those low elo players, if they win against high elo, proving themselves, they get a big buff of elo. Yeah, it came with a great price.. but they didn't queued regularly against high elos, how many games did every 1200 elo player played in a high elo queue? 2-3 every month? That's not enough to be unable to raise elo, i've proven that, with the screen of my alt, in less than 400 games i'm 1660, 500+ higher than the average ELO.

Even if you think that that's a bug, and you need to fix it, why the hell are you implementing an EVEN WORSE system? You can't say that the previous system was bugged, and this is intended, because no one can maintain his ELO at high elo.

Oh wait, "that's because it's inflated".

You know that's fucking bullshit, i can understand if you say that Man U's ELO is inflated and he can't maintain ELO, i can buy it, but 2450 is inflated? Bullshit.
Daselend
Donor
Donor
 
Posts: 120
Joined: Wed Oct 08, 2014 6:50 pm

Re: Town of Salem 2.0 - Ranked ELO Changes

Postby rickms » Wed Jun 07, 2017 1:34 pm

Daselend wrote:[
If you think that either +15 or -15 is a bug.. it isn't a bug +1 or -10? Give the high elos +15 or -15, instead of this shitty +1 or -10, i'll take it any day of the week, even i'll take it over my games before the dumb fix you guys did, when i was getting +1-4 and -1-4 the overwhelmingly majority of my games.

Now, instead of having +15 or -15 in some games, low elos have +10 or -1 every freaking game. Wow, perfect system.

Having either +15 or -15, wasn't a bug, it was a blessing for those low elo players, if they win against high elo, proving themselves, they get a big buff of elo. Yeah, it came with a great price.. but they didn't queued regularly against high elos, how many games did every 1200 elo player played in a high elo queue? 2-3 every month? That's not enough to be unable to raise elo, i've proven that, with the screen of my alt, in less than 400 games i'm 1660, 500+ higher than the average ELO.

Even if you think that that's a bug, and you need to fix it, why the hell are you implementing an EVEN WORSE system? You can't say that the previous system was bugged, and this is intended, because no one can maintain his ELO at high elo.

Oh wait, "that's because it's inflated".

You know that's fucking bullshit, i can understand if you say that Man U's ELO is inflated and he can't maintain ELO, i can buy it, but 2450 is inflated? Bullshit.


I'm gonna quote this again, since what you want the rating system to be is not how ELO works...

"A player's Elo rating is represented by a number which increases or decreases depending on the outcome of games between rated players. After every game, the winning player takes points from the losing one. The difference between the ratings of the winner and loser determines the total number of points gained or lost after a game. In a series of games between a high-rated player and a low-rated player, the high-rated player is expected to score more wins. If the high-rated player wins, then only a few rating points will be taken from the low-rated player. However, if the lower rated player scores an upset win, many rating points will be transferred. The lower rated player will also gain a few points from the higher rated player in the event of a draw. This means that this rating system is self-correcting. A player whose rating is too low should, in the long run, do better than the rating system predicts, and thus gain rating points until the rating reflects their true playing strength."

Source: https://en.wikipedia.org/wiki/Elo_rating_system
rickms
Witch
Witch
 
Posts: 47
Joined: Mon Jan 11, 2016 1:08 pm

Re: Town of Salem 2.0 - Ranked ELO Changes

Postby Gracee » Wed Jun 07, 2017 1:59 pm

@rickms so I actually make an effort to read the wiki-page you're throwing around on the forums for everyone who questions anything about the ELO-system, and I can't even get an answer? Yet you're posting replies to other players who's repeating themselves -.- /ignored.
Gracee
Escort
Escort
 
Posts: 75
Joined: Thu Feb 11, 2016 1:54 pm

Re: Town of Salem 2.0 - Ranked ELO Changes

Postby rickms » Wed Jun 07, 2017 2:11 pm

Gracee wrote:@rickms so I actually make an effort to read the wiki-page you're throwing around on the forums for everyone who questions anything about the ELO-system, and I can't even get an answer? Yet you're posting replies to other players who's repeating themselves -.- /ignored.


I thought I covered your comments in my other responses, what exactly would you like me to respond to?
rickms
Witch
Witch
 
Posts: 47
Joined: Mon Jan 11, 2016 1:08 pm

Re: Town of Salem 2.0 - Ranked ELO Changes

Postby Gracee » Wed Jun 07, 2017 2:32 pm

Question: Why do BMG insist on using the ELO-system when it's not made for games like these?
Motivation: The difference in the ratings between two players is supposed to serve as a predictor of the outcome of a match. There is no way to predict the outcome of a game in Town of Salem. There are too many factors that could easily change the outcome without any action from the player itself. Another aspect of the ELO-system is that two players with equal ratings who play against each other are expected to score an equal number of wins. Equal rating does not exist when the players in one game can vary 1.5k ELO or more, from top to bottom.

Another Question: The ELO-system does not always support 0-gains for Draws. Why does yours always end up with +/- 0?

Another Question: Seeing how you're trying to avoid the extremes in the ELO-spectrum, why are you not using a K-factor gradation?

I cannot find answers to these questions straight up in the answers you've provided in this thread. If I missed something, feel free to direct me to that answer and I'll read through it again.
Gracee
Escort
Escort
 
Posts: 75
Joined: Thu Feb 11, 2016 1:54 pm

Re: Town of Salem 2.0 - Ranked ELO Changes

Postby rickms » Wed Jun 07, 2017 3:11 pm

Gracee wrote:Question: Why do BMG insist on using the ELO-system when it's not made for games like these?
Motivation: The difference in the ratings between two players is supposed to serve as a predictor of the outcome of a match. There is no way to predict the outcome of a game in Town of Salem. There are too many factors that could easily change the outcome without any action from the player itself. Another aspect of the ELO-system is that two players with equal ratings who play against each other are expected to score an equal number of wins. Equal rating does not exist when the players in one game can vary 1.5k ELO or more, from top to bottom.


I was not around when they made the initial decision. It's not that we are insisting on using ELO, it's that it is what is in place currently. Sky's the limit with the Ranked update coming. If you have a suggestion for a better system, please share.

Given that everyone in ranked faces the same RNG, over time the randomness generally cancels out. The RNG gods can be cruel at times, however. Why does one person win more than another over a long stretch of time? The best answer is because one person plays the game with better skill. This is why we have a rating. While it may seem unfair at the micro level, for a particular game, over the course of many games it is a fair indicator of skill. The misconception is that very high numbers mean very high skill. The ELO value you have only means something in the context of the other players.

This is the important part for people to know.

Take an inflated elo for example.

If the top ELO is 3000 and you have 2500, then you know where you are relative to the top. You have ~83% of the rating the top player has.

If we reduce this by 10%, If the top elo is 2700 and you have 2250, you still have same indicator of relative skill. You have 83% of the rating the top player has.

Aside from the absolute value, nothing has changed as far as indication of skill. Which is always relative to the others playing.

**HOWEVER**

The ELO equation has a constant in it's equation. The value 400. This is important.

The 3000 vs 2500 would result in ELO factors of 0.05/0.95
The 2700 vs 2250 would result in ELO factors of 0.08/0.92

I know that it seems like a small difference, and it is, but it's important to note that the ELO range MATTERS when it comes to the math. This is why it was important to fix this bug, so the ceiling can come down and the floor can come up and the ELO changes be more inline with what might be expected given the caliber of play.

Another Question: The ELO-system does not always support 0-gains for Draws. Why does yours always end up with +/- 0?


We explicitly do not adjust Elo when there is a draw. Yes, this is contrary to the natural ELO algorithm. Not trying to punt on this question but I do not know the original rational behind this decision. I'll ask around and see if anyone knows. Draws are pretty infrequent so it's not seen as a particular major issue.

Another Question: Seeing how you're trying to avoid the extremes in the ELO-spectrum, why are you not using a K-factor gradation?


This is something we are considering for the Ranked update.
rickms
Witch
Witch
 
Posts: 47
Joined: Mon Jan 11, 2016 1:08 pm

Re: Town of Salem 2.0 - Ranked ELO Changes

Postby Jakerams » Wed Jun 07, 2017 9:22 pm

Lol lost -9 though I did want to lose I just to see what would happen I was veteran and tried to carry town I told town who to vote they instead listened to the 4 mafia the surv sided with the mafia.

Jailor executes the proven trans and claims he was a liability to town (I reported it so you have proof if you check the report)
Me: Don't vote the arso vote the mafioso.
Town listens to the the mafia and vote the arso.
I'm pleading to innocent the arso but the players being as bad as they are refused.


I told town to vote with me but they refused. And guess what veteran can't carry a town. Atleast in other games with elo ratings you can technically carry your teams but in town of salem that is technically not possible to do in many situations unless it's very late game.
Jakerams
Jester
Jester
 
Posts: 11
Joined: Sat Jan 10, 2015 7:29 pm

Re: Town of Salem 2.0 - Ranked ELO Changes

Postby rickms » Thu Jun 08, 2017 11:09 am

Jakerams wrote:I told town to vote with me but they refused. And guess what veteran can't carry a town. Atleast in other games with elo ratings you can technically carry your teams but in town of salem that is technically not possible to do in many situations unless it's very late game.


Valid argument. Noted.
rickms
Witch
Witch
 
Posts: 47
Joined: Mon Jan 11, 2016 1:08 pm

Re: Town of Salem 2.0 - Ranked ELO Changes

Postby WittyRecluse » Fri Jun 09, 2017 8:35 am

It's worth pointing out that a lot of times (at least as town), the absolute elo of your teammates can matter more than the relative elo of your opponents. In towns where everyone else in the queue is 1200 elo, I would get about -14 for a loss because I would inflate my team's elo, but I could only win like 50% of those games because teammates are so bad. In that elo range I've seen transporters transport revealed mayors with confirmed town, vigs shoot d1 med claims, townies that don't keep wills, town that don't cc an obvious scum causing the scum to be "confirmed TP" due to the lack of ccs, transporters who transport a confirmed town with the person that confirmed vig says he's shooting (that was an awful game).... like Jakerams said there is only so much you can carry your teammates and teammates like the ones I mentioned make it impossible to win even if you carry a lot.

Also, as mafia, the new strategy is killing the high elo townie first because even if he's TPed, a loss is only -1 so it doesn't matter. In the old days it was risky to do this because you would get -6 if you failed, so low elos didn't want to use this cheap "strategy". Now there's no reason not to do it.
User avatar
WittyRecluse
Vigilante
Vigilante
 
Posts: 618
Joined: Sun Jun 12, 2016 8:02 pm

Re: Town of Salem 2.0 - Ranked ELO Changes

Postby kosmo16 » Fri Jun 09, 2017 3:05 pm

I am sick of crying and complaining of high elo players. According to the devs, elo system was broken really, really badly. In previous system a new very smart player with let's say 90% winrate who faces much stronger opponents (according to elo) was still losing elo in long-term. High elo players are all veterans. Im sure that there is no player who has more then 1.8k elo and is playing less then 6 months. This fix was really needed. Thank you!

About elo system, Im a bit worried that you still pasting this link:
https://en.wikipedia.org/wiki/Elo_rating_system

The Elo Rating System assumes that if two opponents with equal strength face each other, the probability of winning of player/team A is 50%. It is not true in the Town of Salem. First of all there are more teams then 2. Secondly, you need to realize that Town of Salem is an asymetric game, so if Mafia and Town have both the same Elo, probability for Town to win the game is much more then 50%. The adjustment for this in needed.

Thankfully equation to calculate elo change depends only on difference in elo. I will present my proposition of elo calculation on an example:

Let's say we have 9 Town members: Jailor, Spy, Sheriff, Medium, Escort, Doctor, Vigilante, Sheriff, Veteran with average elo 1600.
we have 3 Mafia members: Mafioso, Godfather, Consort with 1600 elo
and NK Serial Killer also with 1600 elo
Survivor and Jester (are not that important).

Now very important part. I hope that the BMG has some statistic winrate for this specific rolelist (more or less). Let's say that statistically Town wins 80% of games, Mafia - 15% of games and SK - 5%.

Let's split this to 3 sets of 2 teams:
Town (84%) vs Mafia (16%)
Town (94%) vs SK (6%)
Mafia (75%) vs SK (25%)

Equation for new ranking is:
R = CR + K * (E - WR)

where
CR - current ranking
K - constant coefficient
E - result, 1 for win, 0 for loss
WR - win rate (for 84% it i 0.84)

Let's check what will happen in the game if the Town wins when K is equal to 12 (this means that max elo change will be 24, for 4 teams should be 8):
Town ranking change:

Town vs Mafia change:
Town Ranking = 1600 + 12 * (1 - 0.84) = 1602
Mafia Ranking = 1600 + 12 * (0 - 0.16) = 1598

Town vs SK change:
Town Ranking = 1600 + 12 * (1 - 0.94) = 1601
SK Ranking = 1600 + 12 * (0 - 0.06) = 1599

Of course we should sum up it so Town is now 1603, Mafia 1598 and SK 1599.

Let's check what will happen when the Mafia wins:

Town vs Mafia change:
Town Ranking = 1600 + 12 * (0 - 0.84) = 1590
Mafia Ranking = 1600 + 12 * (1 - 0.16) = 1610

Mafia vs SK change:
Mafia Ranking = 1600 + 12 * (1 - 0.75) = 1603
SK Ranking = 1600 + 12 * (0 - 0.25) = 1597

After sum up Town is now 1590, Mafia 1613 and SK 1597.

And finally SK wins:

Town vs SK change:
Town Ranking = 1600 + 12 * (0 - 0.94) = 1589
SK Ranking = 1600 + 12 * (1 - 0.06) = 1611

Mafia vs SK change:
Mafia Ranking = 1600 + 12 * (0 - 0.75) = 1591
SK Ranking = 1600 + 12 * (0 - 0.25) = 1609

After sum up Town is now 1589, Mafia 1591 and SK 1620.

This happens when Town, Mafia and SK are equal strength. What if it's not true?

Let's see how approximately a function probability of win of elo difference looks like:
http://imgur.com/a/TZXqm

And let's check the graph where for difference 0 we have 0.86 probability of win (Town vs Mafia):
http://imgur.com/a/TkoEM

Now consider Town of elo 1200 winning against mafia of elo 1475:
Town vs Mafia change:
Town Ranking = 1200 + 12 * (1 - 0.5) = 1206
Mafia Ranking = 1475 + 12 * (1 - 0.5) = 1469

This is an example how elo could be calculated. I believe we should have at least few win ratios for different Any role faction. Retributionist present/not present. NK is ww/arso/nk.

I am not saying this is ideal, but is at least a good point to start.

Also please devs don't listen to people who are saying that elo should depends on Town role (you play as a Team) or what is worse from executed evils as Jailor or shot evils as Vigilante. Why?

Example:
4 people left:
Medium, Escort, Vigilante, Investigator
Role list:
Random Town, Town Killing, Town Investigative, Random Mafia

Now, because of elo Vigilante refuses to vote up the Medium claim, because he wants to shoot it at night and gain more elo. At night Vigilante shoots the Medium and unexpectedly realizes that an Escort claim was Random Mafia. Town loses. If he voted up the Medium and lynched him, he would shoot an Escort claim at night. Town wins.
Legacy Season Rank: Master
Season One Rank: Master
Season Two Rank: Diamond
Season Three Rank: Master
Season Four Rank: Master
User avatar
kosmo16
Jailor
Jailor
 
Posts: 733
Joined: Sat Nov 21, 2015 7:37 am
Location: Wonderland

Re: Town of Salem 2.0 - Ranked ELO Changes

Postby Jakerams » Fri Jun 09, 2017 6:02 pm

kosmo16 wrote:I am sick of crying and complaining of high elo players. According to the devs, elo system was broken really, really badly. In previous system a new very smart player with let's say 90% winrate who faces much stronger opponents (according to elo) was still losing elo in long-term. High elo players are all veterans. Im sure that there is no player who has more then 1.8k elo and is playing less then 6 months. This fix was really needed. Thank you!

About elo system, Im a bit worried that you still pasting this link:
https://en.wikipedia.org/wiki/Elo_rating_system

The Elo Rating System assumes that if two opponents with equal strength face each other, the probability of winning of player/team A is 50%. It is not true in the Town of Salem. First of all there are more teams then 2. Secondly, you need to realize that Town of Salem is an asymetric game, so if Mafia and Town have both the same Elo, probability for Town to win the game is much more then 50%. The adjustment for this in needed.

Thankfully equation to calculate elo change depends only on difference in elo. I will present my proposition of elo calculation on an example:

Let's say we have 9 Town members: Jailor, Spy, Sheriff, Medium, Escort, Doctor, Vigilante, Sheriff, Veteran with average elo 1600.
we have 3 Mafia members: Mafioso, Godfather, Consort with 1600 elo
and NK Serial Killer also with 1600 elo
Survivor and Jester (are not that important).

Now very important part. I hope that the BMG has some statistic winrate for this specific rolelist (more or less). Let's say that statistically Town wins 80% of games, Mafia - 15% of games and SK - 5%.

Let's split this to 3 sets of 2 teams:
Town (84%) vs Mafia (16%)
Town (94%) vs SK (6%)
Mafia (75%) vs SK (25%)

Equation for new ranking is:
R = CR + K * (E - WR)

where
CR - current ranking
K - constant coefficient
E - result, 1 for win, 0 for loss
WR - win rate (for 84% it i 0.84)

Let's check what will happen in the game if the Town wins when K is equal to 12 (this means that max elo change will be 24, for 4 teams should be 8):
Town ranking change:

Town vs Mafia change:
Town Ranking = 1600 + 12 * (1 - 0.84) = 1602
Mafia Ranking = 1600 + 12 * (0 - 0.16) = 1598

Town vs SK change:
Town Ranking = 1600 + 12 * (1 - 0.94) = 1601
SK Ranking = 1600 + 12 * (0 - 0.06) = 1599

Of course we should sum up it so Town is now 1603, Mafia 1598 and SK 1599.

Let's check what will happen when the Mafia wins:

Town vs Mafia change:
Town Ranking = 1600 + 12 * (0 - 0.84) = 1590
Mafia Ranking = 1600 + 12 * (1 - 0.16) = 1610

Mafia vs SK change:
Mafia Ranking = 1600 + 12 * (1 - 0.75) = 1603
SK Ranking = 1600 + 12 * (0 - 0.25) = 1597

After sum up Town is now 1590, Mafia 1613 and SK 1597.

And finally SK wins:

Town vs SK change:
Town Ranking = 1600 + 12 * (0 - 0.94) = 1589
SK Ranking = 1600 + 12 * (1 - 0.06) = 1611

Mafia vs SK change:
Mafia Ranking = 1600 + 12 * (0 - 0.75) = 1591
SK Ranking = 1600 + 12 * (0 - 0.25) = 1609

After sum up Town is now 1589, Mafia 1591 and SK 1620.

This happens when Town, Mafia and SK are equal strength. What if it's not true?

Let's see how approximately a function probability of win of elo difference looks like:
http://imgur.com/a/TZXqm

And let's check the graph where for difference 0 we have 0.86 probability of win (Town vs Mafia):
http://imgur.com/a/TkoEM

Now consider Town of elo 1200 winning against mafia of elo 1475:
Town vs Mafia change:
Town Ranking = 1200 + 12 * (1 - 0.5) = 1206
Mafia Ranking = 1475 + 12 * (1 - 0.5) = 1469

This is an example how elo could be calculated. I believe we should have at least few win ratios for different Any role faction. Retributionist present/not present. NK is ww/arso/nk.

I am not saying this is ideal, but is at least a good point to start.

Also please devs don't listen to people who are saying that elo should depends on Town role (you play as a Team) or what is worse from executed evils as Jailor or shot evils as Vigilante. Why?

Example:
4 people left:
Medium, Escort, Vigilante, Investigator
Role list:
Random Town, Town Killing, Town Investigative, Random Mafia

Now, because of elo Vigilante refuses to vote up the Medium claim, because he wants to shoot it at night and gain more elo. At night Vigilante shoots the Medium and unexpectedly realizes that an Escort claim was Random Mafia. Town loses. If he voted up the Medium and lynched him, he would shoot an Escort claim at night. Town wins.

As stated before the elo system does more harm than it does good with the changes just because the majority of players are of lower intelligence, making it technically impossible to carry games in many situations as I've stated above. If we actually had a chance to carry the entire game singlehandedly it would be more of a skill based game. But in actuality people in ToS are pretty bad and being matched with low elo players is unfair because they determine your elo for the game.
Jakerams
Jester
Jester
 
Posts: 11
Joined: Sat Jan 10, 2015 7:29 pm

Re: Town of Salem 2.0 - Ranked ELO Changes

Postby Daselend » Sat Jun 10, 2017 10:34 am

In previous system a new very smart player with let's say 90% winrate who faces much stronger opponents (according to elo) was still losing elo in long-term. High elo players are all veterans


Not true. You have the screen of my alt account on the previous page.

A smart player with 90% winrate, could get to high elo easily. But, of course, he has to be smart, which low elo players aren't.

High elo players are smart, that's why they're high, everyone starts at the same place, and everyone had the same ELO system until days ago, people were (and still is) high elo because it was not impossible to reach high elo.
Daselend
Donor
Donor
 
Posts: 120
Joined: Wed Oct 08, 2014 6:50 pm

Re: Town of Salem 2.0 - Ranked ELO Changes

Postby Jakerams » Sat Jun 10, 2017 11:33 am

Daselend wrote:
In previous system a new very smart player with let's say 90% winrate who faces much stronger opponents (according to elo) was still losing elo in long-term. High elo players are all veterans


Not true. You have the screen of my alt account on the previous page.

A smart player with 90% winrate, could get to high elo easily. But, of course, he has to be smart, which low elo players aren't.

High elo players are smart, that's why they're high, everyone starts at the same place, and everyone had the same ELO system until days ago, people were (and still is) high elo because it was not impossible to reach high elo.

To be fair a 90% winrate is nearly impossible because it's a team game where you have to work together and odds are that you will likely get many lower intelligence players on your team. As i've stated above it's not a game where you can solo carry as town, you can technically solo carry as mafia but it's unlikely, and NK can also solo but as a team game it's impossible to solo carry as town.
Jakerams
Jester
Jester
 
Posts: 11
Joined: Sat Jan 10, 2015 7:29 pm

Re: Town of Salem 2.0 - Ranked ELO Changes

Postby rickms » Sat Jun 10, 2017 12:42 pm

kosmo16 wrote:
About elo system, Im a bit worried that you still pasting this link:
https://en.wikipedia.org/wiki/Elo_rating_system

The Elo Rating System assumes that if two opponents with equal strength face each other, the probability of winning of player/team A is 50%. It is not true in the Town of Salem. First of all there are more teams then 2. Secondly, you need to realize that Town of Salem is an asymetric game, so if Mafia and Town have both the same Elo, probability for Town to win the game is much more then 50%. The adjustment for this in needed.


You are correct, after we calculate the expected win rate of differently rated elo opponents we then apply a role/faction based win-rate modifier to that value. Not exactly as your math laid out but very close. It's why you never lose more than 1-2 elo as SK.

Thanks!
rickms
Witch
Witch
 
Posts: 47
Joined: Mon Jan 11, 2016 1:08 pm

Re: Town of Salem 2.0 - Ranked ELO Changes

Postby kosmo16 » Sat Jun 10, 2017 2:29 pm

rickms wrote:
kosmo16 wrote:
About elo system, Im a bit worried that you still pasting this link:
https://en.wikipedia.org/wiki/Elo_rating_system

The Elo Rating System assumes that if two opponents with equal strength face each other, the probability of winning of player/team A is 50%. It is not true in the Town of Salem. First of all there are more teams then 2. Secondly, you need to realize that Town of Salem is an asymetric game, so if Mafia and Town have both the same Elo, probability for Town to win the game is much more then 50%. The adjustment for this in needed.


You are correct, after we calculate the expected win rate of differently rated elo opponents we then apply a role/faction based win-rate modifier to that value. Not exactly as your math laid out but very close. It's why you never lose more than 1-2 elo as SK.

Thanks!


You should also take into account distribution of teams: 9 Town vs 3 Mafia, should have different modifiers then 8 Town vs 4 Mafia.

I believe that elo modifier for role isn't needed, you play as a team, so each team member should gain/loss the same amount of elo.
Sometimes the player with more fluctuate modifier dies n1 and has 0 impact for the game. Shouldn't be rewarded/punished for this.
Legacy Season Rank: Master
Season One Rank: Master
Season Two Rank: Diamond
Season Three Rank: Master
Season Four Rank: Master
User avatar
kosmo16
Jailor
Jailor
 
Posts: 733
Joined: Sat Nov 21, 2015 7:37 am
Location: Wonderland

Re: Town of Salem 2.0 - Ranked ELO Changes

Postby Daselend » Sat Jun 10, 2017 3:27 pm

Jakerams wrote:
Daselend wrote:
In previous system a new very smart player with let's say 90% winrate who faces much stronger opponents (according to elo) was still losing elo in long-term. High elo players are all veterans


Not true. You have the screen of my alt account on the previous page.

A smart player with 90% winrate, could get to high elo easily. But, of course, he has to be smart, which low elo players aren't.

High elo players are smart, that's why they're high, everyone starts at the same place, and everyone had the same ELO system until days ago, people were (and still is) high elo because it was not impossible to reach high elo.

To be fair a 90% winrate is nearly impossible because it's a team game where you have to work together and odds are that you will likely get many lower intelligence players on your team. As i've stated above it's not a game where you can solo carry as town, you can technically solo carry as mafia but it's unlikely, and NK can also solo but as a team game it's impossible to solo carry as town.


Yep, it's very unlikely that someone will walk into the game and come out with a 90% winrate, but i wasn't discussing that, i was discussing on how it is absolutely impossible that someone with a 90% winrate will lose elo on the old system and """""""his bug""""""". With basic math you can easily check how this is just not possible, but people will just made up things just to support his PoV, then cry about how tired he is of high elo people crying
Daselend
Donor
Donor
 
Posts: 120
Joined: Wed Oct 08, 2014 6:50 pm

Re: Town of Salem 2.0 - Ranked ELO Changes

Postby rickms » Sat Jun 10, 2017 6:21 pm

kosmo16 wrote:You should also take into account distribution of teams: 9 Town vs 3 Mafia, should have different modifiers then 8 Town vs 4 Mafia.


Very good point. This will be looked into for sure.

I believe that elo modifier for role isn't needed, you play as a team, so each team member should gain/loss the same amount of elo.


As it stands now ,they should. It's your teams rating modifier vs your opponents. When I say "faction/elo" I say that because the neutrals don't have a faction.

Sometimes the player with more fluctuate modifier dies n1 and has 0 impact for the game. Shouldn't be rewarded/punished for this.


N1 deaths is absolutely something we're actively talking about. We are still discussing specifically what, but most likely there will be some sort of rating loss modifier when you die on N1.
rickms
Witch
Witch
 
Posts: 47
Joined: Mon Jan 11, 2016 1:08 pm

Re: Town of Salem 2.0 - Ranked ELO Changes

Postby LRyuzakiL » Sun Jun 11, 2017 1:59 pm

This should be seriously considered for the ELO change, considering all the back and forth debate about the seasons and ELO in general

viewtopic.php?f=14&t=69012&p=2213556#p2213556
LRyuzakiL
Bodyguard
Bodyguard
 
Posts: 234
Joined: Sat Dec 26, 2015 10:13 pm

Re: Town of Salem 2.0 - Ranked ELO Changes

Postby kosmo16 » Sun Jun 11, 2017 2:40 pm

One more question, skipping all role modifier. Are all people in the same team gaining/losing the same amount of elo, no matter what their current ranking is?
Legacy Season Rank: Master
Season One Rank: Master
Season Two Rank: Diamond
Season Three Rank: Master
Season Four Rank: Master
User avatar
kosmo16
Jailor
Jailor
 
Posts: 733
Joined: Sat Nov 21, 2015 7:37 am
Location: Wonderland

Re: Town of Salem 2.0 - Ranked ELO Changes

Postby ScardyBob » Sun Jun 11, 2017 11:33 pm

kosmo16 wrote:Equation for new ranking is:
R = CR + K * (E - WR)

where
CR - current ranking
K - constant coefficient
E - result, 1 for win, 0 for loss
WR - win rate (for 84% it i 0.84)

Let's check what will happen in the game if the Town wins when K is equal to 12 (this means that max elo change will be 24, for 4 teams should be 8):
Town ranking change:

Town vs Mafia change:
Town Ranking = 1600 + 12 * (1 - 0.84) = 1602
Mafia Ranking = 1600 + 12 * (0 - 0.16) = 1598

Town vs SK change:
Town Ranking = 1600 + 12 * (1 - 0.94) = 1601
SK Ranking = 1600 + 12 * (0 - 0.06) = 1599

Of course we should sum up it so Town is now 1603, Mafia 1598 and SK 1599.

One issue that may mess things up is how you calculate the ranking for the factions (i.e. Town, Mafia, Coven). It would be better to use a robust statistic (like the median) rather than the average to minimize the skew due to having heavy Elo outliers on a team (i.e. think a team of 2000+ players paired with one or two <1200 players or a team of <1200 players paired with one or two 2000+ players).
ScardyBob
Investigator
Investigator
 
Posts: 283
Joined: Tue Aug 09, 2016 11:02 pm

PreviousNext

Return to Announcements

Who is online

Users browsing this forum: No registered users and 10 guests