wozearly wrote:The Elo system is entirely and solely based on the concept of win = +points, lose = -points.
The first of two main problems with trying to attach "credit" for certain actions is that it risks incentivising individuals to act in a way that, counterintuitively, may be detrimental to the team. And the second is that not all roles are equally suited for "crediting" in this way.
As an example of the first point, take a Vigi who gains points for shooting and killing a Mafia member, loses points for shooting and killing a Town member, and gains and loses no points if the person they shot survives. And a Jailor who gets points for executing evils and loses points for executing Townies.
The Vigi and Jailor now have a perverse incentive in certain situations; say that Town has narrowed down its list of potential last Mafioso suspects to one of three - two unproven Vigi claims and a Transporter claim who may be a Hypnotist-turned-Mafioso. Jailor is still in play and has executions. Any other Townie in play is confirmed beyond reasonable doubt, all other evils are dead.
Someone leading the Town would likely propose a low-risk solution for a Town win; Vigi claims shoot each other, Jailor jails and executes the Transporter. Whatever happens the Mafioso dies and Town wins. But Vigis and Jailor may not like it. If the Trans claim is Mafioso, both Vigis will lose ELO for their "bad" decision but the Jailor will gain ELO. Alternatively, if one of the Vigis is Mafioso then the Vigi will gain ELO while the Jailor loses ELO for their "bad decision".
As it's basically a 50/50 chance of ELO loss, you'd expect Jailor and Vigi to object. Jailor would be happier with a more complex approach of jailing the most likely suspect without executing, then executing the following night if there are no Mafia kills (so the Jailor gets max ELO and not the Vigi or all Town via voting, natch). Vigis would prefer to shoot based on their own scumreading and gut instinct than be instructed to lose ELO if they're pretty sure the Trans is a fake. Expect a series of unhelpful arguments as each defends their own ELO risk.
There are many other variants of this (Doc refusing to sit on Jailor because no-one will attack a Jailor known to be guarded, so they have better ELO opportunities by trying to heal others, even if that is detrimental to the team, etc.)
For the other problem, take a role like Transporter. How would you determine what was or wasn't a successful transport for ELO purposes. Or a Medium? It gets even worse on the evils side. How do you credit a Blackmailer? Or whether a Witch controlled the "right"people.
Annoying though it is to be stuck with poorer players from time to time, TOS is a team game and Town benefits from all of its players recognising that, in most cases, they're entirely sacrificable provided it takes down or identifies at least one Mafia member in return.
Benn3 wrote:[b]Seriously, ELO distribution is horrible in Ranked. You either get some from winning, or lose some from losing.
James2 wrote:Benn3 wrote:[b]Seriously, ELO distribution is horrible in Ranked. You either get some from winning, or lose some from losing.
That is precisely how elo is supposed to work. The goal of the game is to win, not to keep townies alive, transport mafia into themselves, etc.
Joacgroso wrote:I feel like I went from Light Yagami to Keiichi Maebara.
Benn3 wrote:You should gain more ELO depending on how well you play the game, such as:
- Shooting evils as a Vigilante
- Executing evils as the Jailor
- Protecting people from attacks as a TP
- Voting and lynching(pressing guilty) members of the opposing faction
- Winning the game with as many faction members alive as possible
- Making a Mafia Killing target themselves as a Transporter
kyuss420 wrote:devs arent gonna code an AI to watch every match and allocate ELO on how much someone contributed to the win.
kyuss420 wrote:Or whether their genius play worked or not -which is another problem, if you make an optimal play, say with 20% chance of failure and 80% chance of success, if you hit the 20% you will lose ELO, even tho it was the most optimal move at the time.
Joacgroso wrote:I do believe elo gain/loss should be reduced for people who died n1/n2.
Joacgroso wrote:James2 wrote:Benn3 wrote:[b]Seriously, ELO distribution is horrible in Ranked. You either get some from winning, or lose some from losing.
That is precisely how elo is supposed to work. The goal of the game is to win, not to keep townies alive, transport mafia into themselves, etc.
One could argue that then elo isn't the best scoring system for this game (after all, it was created for chess) and that ToS should use a better one. Then again, this game doesn't have that many players, so even if we had a fairer algorythm we would still have a huge skill gap between players in every game. There are higher priorities.
The sad thing is that this game is inherently unfair, so it's hard to make a fair scoring system. I do believe elo gain/loss should be reduced for people who died n1/n2. Just because the game can't be perfectly fair it doesn't mean we can't try.
James2 wrote:The goal of the game is to win, not to survive the first two nights (or die within the first two nights). There are cases where a person can benefit their team by risking early death (or benefit a probably losing team by not risking early death), so that should not be a factor.
Joacgroso wrote:I feel like I went from Light Yagami to Keiichi Maebara.
James2 wrote:The goal of the game is to win, not to survive the first two nights (or die within the first two nights).
Superalex11 wrote:
As for the rest of your babbling, kyuss, it seems like you're just gut-reacting against this because... I'm not sure actually. None of the rest of your comment really makes sense or is applicable. To any others reading this, I'd recommend ignoring kyuss, as he has a history of making stupid incoherent posts like this.
Joacgroso wrote:I feel like I went from Light Yagami to Keiichi Maebara.
Joacgroso wrote:James2 wrote:The goal of the game is to win, not to survive the first two nights (or die within the first two nights). There are cases where a person can benefit their team by risking early death (or benefit a probably losing team by not risking early death), so that should not be a factor.
There may be cases where that happens, but I feel like reducing the elo impact if you die early will do more good than harm, even if it's not perfect. We can give exceptions to ambushers or bodyguards if they die guarding, and maybe veterans too. How often do you think plays would be punished by this feature compared to the amount of times the feature would actually help?
Superalex11 wrote:James2 wrote:The goal of the game is to win, not to survive the first two nights (or die within the first two nights).
Do you believe the average player who dies n1/n2 and wins has earned their win through enough of their own skill/effort? Do you believe the average player who dies n1/n2 and loses has deserved their loss by failing to perform as well as their opponents?
If your answers to the above two are yes, then we're at a fundamental point of disagreement. If your answers are no, then I fail to see the point of your statement here.
Is it because you believe what basically amounts to luck should be this impactful towards elo (again, where elo is meant to be representative of skill)?
Or do you not believe elo should be (or, in general, is) representative of skill?
Joacgroso wrote:I feel like I went from Light Yagami to Keiichi Maebara.
Joacgroso wrote:We shouldn't refuse to improve the system because people might gamethrow.
This feature will do a lot more good than harm, I think. Even if unfair elo changes "balance themselves out", reducing player frustration will only make the game better. Also, not everyone plays ranked that often. People can be (un)lucky and not play often enough for the games to balance the elo out.
Joacgroso wrote:I feel like I went from Light Yagami to Keiichi Maebara.
Joacgroso wrote:I mean, we already have achievements, roles and gamemodes that reward gamethrowing. And in much more blatant ways than this. Just because people may exploit a mechanic in very specific scenarios it doesn't mean the feature won't help in more cases. I honestly don't understand why you care so much about those rare scenarios.
If you really think people can predict the outcome of the game by night 2, then we can compromise and only reduce elo changes if you die n1, since there's no way the game is decided by then.
And just because something is inherent to something else it doesn't mean it's automatically fair, especially in a game where you can't pick your teammates or prepare with them before the game. And I do think the game should be improved for everyone, not only for those who plays dozens of games every day. We're not speaking about balance, we're speaking about a fairer scoring system.
Joacgroso wrote:I feel like I went from Light Yagami to Keiichi Maebara.
James2 wrote:Noise is not a problem in ranking systems. There are some cases where a person can help their team while dead, and in those cases good players will tend to gain elo and bad players will tend to lose it. That the majority of n1 deaths don't have such opportunities doesn't matter, as players in such cases will gain elo as often as they lose it. No algorithm can detect all cases where a person who dies early can help their team, and even if it could it'd create perverse incentives (by encouraging players to get themselves killed in "potentially influential" ways, or discouraging them from risking such deaths for probably losing teams).
James2 wrote:Joacgroso wrote:We shouldn't refuse to improve the system because people might gamethrow.
Actively rewarding people for gamethrowing isn't improving the system. In fact "throwing" to improve one's elo isn't really the same type of behavior as throwing out of frustration or to troll. The former is presently impossible, and should stay that way.
James2 wrote:Having one's win chances affected by teammates is inherent in a team game and therefore isn't unfair. If people get frustrated because they don't understand expected value the solution is to educate them, not to change the game.
If people only play a few ranked games then they aren't the people ranked should be balanced around.
James2 wrote:That something is inherent in a type of game means it's fair by definition. I don't think it's possible for a ranking system to accurately measure skill after only a few games, especially in a team game. Besides, rank matters less to a person who seldom plays ranked.
Joacgroso wrote:I don't know, I understand what you say but I fail to see how it is so important. It's not like townies can do much to get themselves killed n1. Mafia is aware of vet baits. And even if everyone does this, only one player can get away with it, without really affecting the outcome of the game.
And again, you can't predict how the game is going to end by night 1. If we assume everyone is so obssesed with elo that they are willing not to play the game just to optimize their net elo gain, then we can also assume we are talking about high elo players, who always follow the meta anyway. So you can't really tell how bad the town is going to be.
Superalex11 wrote:So, you didn't answer my questions, which makes me think you realize your reasoning is broken, so you're intentionally avoiding the mental confrontation. What you did say also points to this conclusion that you're simply not following the points presented.
The goal of the proponents in this thread is to change the ranking system such that players' elo is more reflective of their skill, and less of simply whether they win or not.
It does not matter if, in the long term, all noise balances out and we get an expected 0-elo change from true elo.
What matters is the frustration that comes when one gets undeserved reward or penalty. Would a literal coin-toss post-game be an appropriate ranking system? After all, with 50/50 odds, long-term all noise will balance out, and the game will be equalized in term of elo (Hint: the answer is no).
Also, regarding your comment about perverse incentives (specific to early deaths, at least), there is no case where a player would make an effort to die very early in order to avoid changing their elo by much. It would be similar to expecting a player to attempt to draw. Even if it might be technically in one's best interest in very rare and specific cases, people play this game to play the game, not to minimize elo loss in uncomfortable situations by not playing the game.
One of the largest reasons people push for a medium rework is not to rebalance it into something more stronger, but to give the medium something to do other than drudgingly copy paste dead chat. Agency is the issue, not power.
...what cases do you think would be created by these changes? A mafioso running into a vet n4 because his other teammates are already dead and he'd rather lose 10 elo instead of 12? Anything of this nature would either end up being too rare, or too meaningless to matter. And as Joacgroso is saying, even if we take you at your word and assume you're 100% correct in the pervasiveness of throwing this would create, it's still a net good.
Thinking a bit more, it also becomes evident that if the devs don't just slap this idea together with tape, it wouldn't be too much work to find a balancing point in specific elo change modifiers for these cases. Take the mafioso->vet example I gave: If on n4 the mafioso believes he has some high probability of losing anyway, thus believing he's better off to die earlier and lose less elo, the true, mathematically expected difference in elo loss could be calculated from the expected probability of the mafioso actually losing even if he doesn't suicide into the vet. Again, this isn't trivial, but if the devs wanted to, they 100% could.
James2 wrote:Superalex11 wrote:So, you didn't answer my questions, which makes me think you realize your reasoning is broken, so you're intentionally avoiding the mental confrontation. What you did say also points to this conclusion that you're simply not following the points presented.
To answer your earlier questions, the average player in a given game isn't individually decisive in the game's outcome. Regardless of the night they die on. Being affected by the actions of one's teammates is an inherent part of team games. Thus it's entirely proper that the skill of one's teammates influences whether one gains or loses elo in a given game.
James2 wrote:Superalex11 wrote:The goal of the proponents in this thread is to change the ranking system such that players' elo is more reflective of their skill, and less of simply whether they win or not.
Skill is an ethereal quality that can't be objectively measured. The closest one can get to directly measuring it is by wins and losses.
James2 wrote:Superalex11 wrote:It does not matter if, in the long term, all noise balances out and we get an expected 0-elo change from true elo.
Accurately representing a player's ability to win games ("skill") is literally the most important feature of a rating system.
James2 wrote:Superalex11 wrote:What matters is the frustration that comes when one gets undeserved reward or penalty. Would a literal coin-toss post-game be an appropriate ranking system? After all, with 50/50 odds, long-term all noise will balance out, and the game will be equalized in term of elo (Hint: the answer is no).
A coin-toss would be pointless and therefore bad, but it'd be a lower order injury to the game than basing elo on in-game actions other than wins and losses. The former would make elo less efficient (as a function of time), while the latter would make it cease to measure skill at all.
James2 wrote:Allowing players to "draw" the first night of any game by dying would affect d1/n1 play in every game.
James2 wrote:any role can have an outsized influence during the day, if the player has the requisite will power.
James2 wrote:Superalex11 wrote:...what cases do you think would be created by these changes? A mafioso running into a vet n4 because his other teammates are already dead and he'd rather lose 10 elo instead of 12? Anything of this nature would either end up being too rare, or too meaningless to matter. And as Joacgroso is saying, even if we take you at your word and assume you're 100% correct in the pervasiveness of throwing this would create, it's still a net good.
Any game where a townie expects (based on d1 chat and/or recognizing other players) that town is more likely (than average) to win, the townie would have more of an incentive to keep themselves alive than if they were only motivated by winning. This could lead to townies being excessively averse to visiting potential veterans, BGs being averse to protecting people likely to be attacked, townies being more likely to false vet bait (or be silent if at an elo where mafia is likely to attack vet baits), etc. And the converse would apply when a townie thought town was less likely than average to win. Mafia would have similar perverse incentives, and to a greater extent since they know which players are on their team and can freely talk with their teammates n1.
Even if the time-of-death based difference in elo gains/losses was mild (such that players would be unlikely to overtly throw), the changed incentives would still subtly affect behavior in the ways described above.
Superalex11 wrote:1) To the extent that the concept of skill is too abstracted to be meaningfully represented quantitatively, just about anything to do with game design is as well. This is a meaningless remark. Representations such as elo scores, while not perfectly representative of every possible aspect of skill, are absolutely extant objective measurements of skill.
2) The closest one can get to measuring skill is entirely dependent on the type of skill. Going by wins and losses is not the most accurate - it is the easiest, and often most efficient (in terms of dev work, not mathematical convergence).
Again, you're just missing the point, and I don't know why. I agree that accurately representing a player's ability to win games (*though this is not necessarily equivalent to skill) should be the primary goal of a rating system. However, what you seem to keep missing is that simply going by wins and losses, and pointing to the infinite long-term result of an expected 0-elo change, you're ignoring more useful methods of identifying skill.
A coin-toss would be pointless and therefore bad, but it'd be a lower order injury to the game than basing elo on in-game actions other than wins and losses. The former would make elo less efficient (as a function of time), while the latter would make it cease to measure skill at all.
And you manage to miss the point... again...
The reason using a coin-toss to determine elo change is bad is because it has zero correlation to skill. The reason the propositions in this thread are broadly good is because they have high correlation to skill. The current system, just going off of wins (parameterized by faction), is weakly correlated to skill.
Now, weak correlations do give identical long-term results as high-correlations (this is how casinos make money), but if we're looking to improve the system, obviously higher is better. The higher the correlation of elo change factors to skill, the quicker elo converges on a player's actual skill level.
As a final remark on this comment, I have no idea what you're talking about at the end there, almost to a point where it looks like a typo. What makes you think in-game actions have no capacity to measure skill?
- Townies being averse to visiting vets should be promoted...
- Bgs dying defending would obviously not be penalized as much as otherwise; on the surface I'd imagine they'd actually be rewarded, thus having even more incentive than now to die defending
- Townies attempting to confuse mafia into not attacking them should be promoted...
- In the opposite townie case, townies would still be averse to visiting vets...
With most of these and others similar, tailored adjustments would be trivial to implement. Even in the worst case where some particular example is shown to be a legitimate problem, such example could be tailored to by removing the factor involved.
You even continue on to mention the option of adjusting these modifiers, yet you dismiss it under the notion that the incentives would still be too large. To this, I again point back to the point of these changes being a net good. Sure, maybe we'd start seeing 14 vet baits d1 (though I'd bet we still wouldn't, people aren't so mechanical). Even if we do, I think it would be worth it if it means halving the convergence time of one's elo to its true value, in addition to significantly less frustration for players across the board.
Superalex11 wrote:- Townies being averse to visiting vets should be promoted...
- Bgs dying defending would obviously not be penalized as much as otherwise; on the surface I'd imagine they'd actually be rewarded, thus having even more incentive than now to die defending
- Townies attempting to confuse mafia into not attacking them should be promoted...
- In the opposite townie case, townies would still be averse to visiting vets...
- In the opposite townie case, given we expect bgs to be rewarded for dying defending, there is again no reason a bg would choose to do nothing at night rather than defend a valuable target
- Mafia cases have already been covered above in this post
Users browsing this forum: No registered users and 7 guests