How the MMR works, and why what Patrick said makes absolute sense (from a dev POV)
Comments
-
'The ends justify the means, and there is no better way.'
There, I condensed the entire post into 10 words.
This just goes to show you that Behavior fanboys will say anything at all to defend something stupid and out of touch so long as it came from their favorite dev company.
Everything said in the comment about reducing people to numbers and factoring in their performance sounds nice on paper. Gotta give it to the poster for blatantly admitting the horrendous flaws of the system.
The strange thing is it seems to be functioning off the notion that there is no better system. That there's no way to correctly calculate skill in an asymmetrical setting. That such a thing can't exist or doesn't already.
The thing is, SPM.
Score Per Minute.
It is already a system that's been in use in popular games like Call Of Duty for years. Turns out, if you calculate the score of a player and average it out across their playtime, you can get a very clear picture of their skill. It doesn't have to be lopsided, it doesn't have to be confusing, and it doesn't have to be unfair.
In fact, one of the major complaints is that MMR encourages scummy playstyles (Camping, hiding in corners, bringing only the most meta perks) because these things make you survive matches or get 4K trials. Not because you are being skilled or playing well. In MMR 'The ends justify the means', and that's as good as it gets.
However, if you change the scoring events to reward more bloodpoints for positive gameplay behavior (+Fresh Meat scoring event for chasing down a new survivor that was not recently unhooked | +Braveheart scoring event for when a survivor has been in chase for over 45 seconds. | +Tireless event for both sides for chaining several scoring events together for a long period of time, etc.) this will cause players to rank up when they preform good actions and play well, as their Score Per Minute will rise when they do well and get out fast, or overwhelm survivors and crush them quickly. You will not be punished for short games like the old system, you wont reel in agony as all 4 survivors escape after a hard fought match with 8 hooks - or run the killer all game just to die to NOED. If you do well with the time in your match, your average goes up. If you do not do well with your time, your average goes down. Getting points for a hatch escape isn't going to pip you up the ranks if you did nothing for your team all game.
That, and the fact that Killers and survivors are all soft-locked to 32k BP per match means that there is a finite speed to the game, and killers and Survivors can be measured by the same metrics. One simple number can determine how well you can preform in the match.
The developers saying that they MUST have this clunky system means they lack professionality. The developers defending the negative aspects of their own system makes them pompous and short sighted. The developers refusing to implement a better system in spite of community wide backlash shows a criminal lack of competence that should not be ignored.
14 -
"Massively prioritize speed" is probably "pull in the closest number ignoring ranges". That's effectively what you said, but also accounts for the potential that it pulls in completely wrong people.
I think in general the matchmaking usually rejects people outside of the range then gradually increases the range the longer you wait. They've said before it works like that IIRC. But, usually this still results in something close. So if you've got an MMR of 1200 and it's looking for people between 1190-1210 and the only other person who can fill your lobby is 1700, it will ignore that person and wait for someone who might be closer. After some amount of time it increases the range to 1180-1220 and etc.
When backfill happens, it completely ignores the range and could pull in that person of 1700 even though it makes no sense and they would never get included with the original range. But, if it found a person who was 1200 at the time of backfilling, it would just fill the lobby with that person. And it would pull in someone at 1500 before the one at 1700.
This is probably more an issue at times when the queues are very unbalanced and it can't find a person at the appropriate MMR.
At least that's my guess. I think it's pretty close to yours but I thought I would just add on.
0 -
Read the whole thing. Not only does this excuse not justify anything, but if it truly was this ‘difficult’ to implement then they shouldn’t have implemented it in the first place and left the rank where it was
2 -
First off, amazing post and follow ups. Thank you for taking the time out of your day to help break down these very complex design challenges into a more readable and understandable form!
PS – Love Trueskill, I remember reading papers about it many years ago. It solved some problems with matchmaking that are unique to videogames (ie building a fresh team for every game) in interesting ways.
It’s absolutely a challenge to communicated about this topic, as it is both extremely technical and complex. When doing a Q&A stream, I am trying to boil down years upon years of research and experience into a few sentences that are understandable, correct, and quick to deliver so that we can answer as many questions as possible. As much as I would love to talk to everyone for hours about Matchmaking, the mathematical nuance of Glicko-2, and how all of this needs to be modified to apply to an asymmetrical game… At some point I need to get back to working on the game. 😊
I’ll offer up a bit of clarification about this point: There is a “soft” cap for MMR as it is considered by the Matchmaking system. A player’s numerical MMR value is allowed to go as high (or as low) as the math says it should. However, when the Matchmaking system is looking at building matches from players in the queue it will effectively round down (or up) the extreme outliers to help them find a match. This was done in direct response to one of the MMR tests we ran last year: Some players were completely unable to find a match because they were such extreme outliers, no matches could actually be formed with them involved. The soft cap is a dynamic value that we can change at any time (and we have adjusted a few times already). Right now I think it’s around 2,000, so any players who have an MMR above that value are treated by the Matchmaker as though their MMR is 2,000 for the purposes of building a match. Post-match MMR adjustment is still based on their ACTUAL MMR, though.
It's worth noting that this isn’t done because there are tons of players above the cap, it’s actually the opposite! The vast majority of our players fall within our lower and upper soft caps. The caps exist to handle the rare, but still real, edge case of extreme outliers. They deserve to play the game, too!
Sadly, in practice, it’s much more similar to the worst-case scenario. It still weighs MMR a bit when backfilling, but not in a meaningful manner. This is partially a hold-over from the Rank-Based Matchmaking days. But, as mentioned, it has been identified and is something we are actively working on addressing!
Improving the algorithm to take into account team results is also underway. Originally, we didn’t pursue this approach because Survivors can leave a game and start searching for their next one before the previous game had completed, meaning there were no results available for updating their MMR. This is one area where the solo-result system has an advantage, as soon as you are done in a game your MMR can be updated.
In reality, the MMR adjustment from a single game is small enough that you likely won’t notice a difference in opponent from match to match as MMR adjustments are meant to show impact over many games, not just 1 or 2. Our hope is that we can hybridize the system and use both team results and personal results to generate an even more accurate MMR that contains the benefits of both approaches!
As another note about the various posts talking about the 1v4 aspect of DbD as it pertains to the maths:
Our batch update logic (that’s what updating with multiple results simultaneously) is based on the Glicko algorithm. It (Glicko) was designed to handle situations where you’ve got multiple results in a short period of time where there’s unlikely to be much difference in the skill of the player between those games. To put it another way, think about someone going and playing in a one-day tournament: They win some and they lose some, but their skill level at the game was probably the same the entire time. In DbD, we use this for Killers to process the results of a single game: 4 matchups that all happened simultaneously are “batched” together to make use of this approach.
If you’re more curious about the maths, I encourage you to check out Mark Glickman’s website about it! He provides papers explaining the formula in detail. A quick search for “Glicko” on your search engine of choice should lead you there.
We have made many customizations to the formula to reflect the asymmetric nature of DbD, but the core mathematical concepts remain the same. 😊
24 -
The game already tracks skill based player objectives in the emblem system, so saying that it can't be done makes zero sense. The only problem was that the numbers needed to be changed to properly reflect the value of each of the actions. Also, there were some actions in that system that don't require skill and they should be removed.
Plus, survivors play as a team but are ranked individually, so if one of the players plays extremely well and carries the team, but dies on their first hook, they get put onto a worse team next match, while players that do little in the match but hide, get higher mmr scores and get placed with better teams that will carry them even more. How is that remotely fair and balanced? Who could possibly call that a good system for ranking players?
Now, no one really cares that bad players get ranked up. They will eventually get to a place where their playstyle just doesn't work with far more skilled killers. The problems is that good players can end up in a snowball of losses because they keep getting teamed with less skilled players.
And, to act like Patrick's comments were valid, when he used a hockey analogy that doesn't reflect the situation in DbD, is ridiculous. This is not a team game. You don't win or lose as a team. You win or lose as individuals, so individual skill should absolutely be taken into consideration. And since they already track that stuff, and were using it in the previous MMR system, you can't say it can't be done. It just needed to be tweaked.
2 -
"This just goes to show you that Behavior fanboys will say anything at all to defend something stupid and out of touch so long as it came from their favorite dev company."
Or, the counter position is that this just goes to show how people who've no clue what they're talking about will make any/every accusation and resort to logical fallacies in order to try and make a point.
I'm in no way a BHVR fanboy. Quite the opposite, because I see all the stuff they could be doing but aren't, and often get very wrong.
But honesty and integrity also means standing up for the things that are right. And in this, they are right.
Unless we adopt some weird paradigm where literally everything they do is wrong... is that what you're trying to suggest?
3 -
Have you never listened to them talk about MMR and their balancing in-depth beyond this one QnA? Wanting 2 kills and 2 escapes every match is their paradigm. It's what the community has been getting at them for for a long time now, because their logic and philosophy are flawed.
3 -
I have, but you're still missing the point! It's more than likely that you've misunderstood what they're saying. Particularly because I've heard them make statements which contradict what you're saying when taken holistically.
I'm telling you the horse pulls the cart, and you're arguing the cart pushes the horse because you believe the devs said the horse should be in front of the cart (how else is it supposed to push the horse otherwise).
1 -
Cheers :)
0 -
I'm not sure where you've seen 2 kills 2 escapes every match, but if you have it on hand, please let me know and I'll see if we can correct it. The actual target is 2 kills 2 escapes average. i.e. All matches together average out to a 50% kill/escape rate.
20 -
I'm just not seeing any indication that MMR is better than the old matchmaking system.
My games have stayed roughly the same in terms of fairness (if not less so than they were under the Emblem system) and the rest of the game appears to have suffered for it.
I don't think the system was worth it at all, especially if we are going to continue to make glacially slow balance changes.
27 -
Mmr is literally worthless if it's so forgiving because it prioritizes q times
The only time mmr was worth anything imo was the last sbmm test. Sure, the first few matches were a bit of a clown fiesta but I found myself enjoying solo q for once.
2 -
Firstly, I abhor the way some members of the community react and argue woth a mentality not far off the people from the witch-trials. Patrick didn't deserve the abuse. People are able to question something with a certain amount of decorum, but not start demanding for his job.
The system, as I've always maintained, is a vast improvement on the complete chaos the original one was. It's still flawed, and I was disappointed with it being based on wins and losses. As pointed out, the system won't take into account every tiny flick of a spin. But to call something skill-based is confusing. This system could simply be explained by the devs as a league table, and you go up or down, depending on your result. It's match-based, and they need to stop refering to it as skill.
An SBMM system would work differently. You asked for an example, and mine would be using the Emblems as 4 separate scoring systems. People are paired with others who score similarily in every emblemic score. So, people who score with 6,000bp in boldness and objectives would be paired with others who score the same on average. The killer can have their emblems paired with those survivors. The BP rewards for work and skill in each category.
The above system may be complicated or may get messy, and I'm certainly not a tech guy. But this would be a skill-based system.
I guess I'm backing you for the most part, but the devs have got to stop refering to this as skill-based. It's not - it's result-based, and they ought to refer to it as such to avoid confusion. But to me, it works better and can be worked upon. Mind you, even if they were to enjoy winding up the communitu, I cannot blame them because of the abuse they get. Pretty much those members of the community are pathetic, but it's not the whole community. There are good ones out there.
0 -
Completing generators is not 100% of the survivor objective, escaping is part of it. Besides, players can very well die before all gens are completed, even in games where eventually all gens do get completed and 2 survivors escape, and so those players that died earlier did not complete the gen objective before the game was over.
Regardless, they aren't aiming for either 2 escapes every match or an alternating cycle of 4-escape and 4-kill matches, they are aiming just for a 50% kill/survival average, "per-player" but over all matches, which in the results of the individual matches can mean much of anything, inlcuding games where 1 or 3 people escape.
What a perfectly-balanced round of DbD (or any game) ultimately truly is is another question entirely, and one that basically cannot be answered since different people see different things as differently important. But contrary to what you said, defining surviving and killing as win conditions is... the opposite of ridiculous in Dead by Daylight. Those are the clearest, most obvious criteria by which to measure success, and they are also what most people mostly care about. Maybe DbD could be better if it instead revolved around hooks, and killers could win with less-than-3 kills if they get enough hooks, and lose with more-than-2 kills if they didn't get enough hooks... but that is not the reality of the game, and so regardless of whether people would like such a game, the MMR cannot possibly be based on hooks as is, that would actually be ridiculous and lead to absurd consequences, the game would have to be redesigned in some major ways to make this possible.
As for @TheSubstitute's question whether MMR - the success-based, "skill-correlated" matchmaking system - is preferable to other matchmaking systems like the one we've had before... I for one think it is, both based on my personal and anecdotal gameplay experiences, and also on my conviction that people ultimately care about killing and surviving above all, and will have more "fun" with the game if they do so often enough, which MMR is meant to ensure, as opposed to the old system where again, average kill rates were very high, due to having very lopsided matches. I think it is no coincidence that most of the complaints about MMR seem to come from killer players, who have on average suffered a decrease of 10-20% in kill rates down from what they were before. They are actually averaging around 2 kills now as the system is intending (and good players still average much above that, since the system not seldomly does not and often simply cannot supply them with similarly-rated players, due to lack of such players in the pool), but they are having a harder time than in the past, getting less kills, and are often seemingly not happy with averaging around 2 kills.
Of course, if enough players (on either side, for whatever reason, justifiably or understandably or not) are dissatisfied with the game due to the system, having less "fun", playing less or not playing at all anymore, and if this actually leads to the game being less successful, and does so to such an extent that BHVR considers it to be too detrimental to (new) player retention, then it probably should and in all likeliness will be changed. Perhaps the game needs to be designed around average kill rates of around 70% since humans just want to be favoured in their average playing experience of being a killer in such a game, more than they feel they should be as a survivor. Those could be reasonable assumptions. But so far I have no reason to assume this is the case, the decline in player numbers over the last few months is not major and a more steep downtrend to be expected given that the game had experienced its by-far biggest surges in player activity in the months before that, with the RE chapter and Covid affecting things - and beyond that, the fact that BHVR went and reworked such an integral system to their game's experience to begin with that they knew would notably decrease global kill rates if it worked as intended, and are still not showing signs to be reconsidering on (to scrap the system), suggests to me that they have concluded (based on many more insights than we have of course) that it is actually beneficial for the game.
You are basically taking every point the OP attempted to explain, and showcasing again the flaws in understanding those explanations are meant to alleviate.
We've already had a matchmaking system based on Emblems, and as you yourself point out, it had problems. In fact, BHVR can see with hard numbers that it was vastly more imbalanced than the current system. As you allude to, there are problems with the things Emblems count and how those things are weighed. And there are things it does not count at all. The main problem with the ranking system based on Emblems and the matchmaking in turn based on ranks was that pretty much everyone ended up at rank 1 sooner or later, and that's how ridiculous kill rates approaching 80% even on average could happen in red ranks. They just did not track player skill and performances well at all, otherwise you would not have had 100 hour Megs at rank 1 every season. The problem with trying to track all the things that go into playing DbD (or indeed any game) is that they are basically uncountable and unquantifiable, and even if you could define, determine and detect all of them, you would then run into the next essentially-impossible task, of deciding concrete tiers of how important they all are in relation to each other.
The solution is to simply look at the results that all of these things lead to: wins. Looking at that will account for all of those things well-enough, and it will automatically weigh the many different skills because those skills that are the most conduvice to winning, naturally lead to winning more than others that aren't as conducive to winning.
So much for the explanation why skills are best accounted-for by looking at results in terms of wins/losses (every MMR/Elo system in existence does this, it's what MMR is designed to be). Keep in mind however that accounting for skills is not actually the reason why these systems look at win results - winning is what people play for, they compete in these player-versus-player games in order to succeed, the vast majority of players does not play in order to not succeed. As such, the system that pairs players should give them a chance to be able to succeed. Ultimately games are about fun, and people usually have fun in games if they succeed, especially if the games are designed to yield good (engaging) gameplay in competitive contexts. The MMR system gives players this chance, it uses algorithms to predict outcomes and pair players such that roughly equal chances of success are at play more often than not. You don't want a chess grandmaster regularly playing against beginners - but not because the chess master is more skilled and therefore a superior person that a beginner doesn't deserve to play against, but simply because neither player involved would enjoy that experience much, let alone if they had to play such lopsided matches all the time (and more problems with this exist if they're playing for prizes of course). Skill merely correlates with winning, it is not the reason why we use MMR in games in and of itself.
For the team aspect thing... While it is true that a team-based rating could (and apparently will in the future) be used, you are failing to see how the per-survivor survival rate ultimately will also account for their teamplay performance. Teamplay also leads to increased chances of succeeding (surviving), and so while it is certainly possible to die despite having done a lot for the team, a player that is good at teamplay will still over many matches survive more often than a player that is not good at teamplay and does not contribute a lot. Your assumption that such players that do not contribute get even better teammates as they rise in rating and players that do contribute worse teammates as they fall is glaringly flawed in various respects. First of all, this is one match, the rating adjustment won't mean much. Secondly, if the survivor that did not contribute much would consistently survive, so would others that do not contribute, and so at higher ratings they would be less and less likely to encounter teammates that can carry them, instead they would all hide. You see quickly how this leads to an absurd scenario. And this is because people that do not contribute on average do not actually survive, they will never surpass players that do contribute in average success rates. Over enough matches, the players that do stuff, are even good enough to "carry" a team among which are people that barely contribute at all, will always outperform those people that do not contribute, in terms of success (survival) rates. You are also failing to account for the fact that players that do nothing but rise in rating would face increasingly tougher killers against which surviving is less likely, especially playing in that way, and that players that do a lot but fall in rating would face increasingly less tough killers, against which surviving is more likely, especially playing in that way.
As for your last point, it's funny how starkly you contradict your earlier point about how this game is played as a team, but suddenly hockey is not fit for an analogy because it is a team game. Regardless, as pointed out, the analogy didn't pertain all of those aspects of hockey, just like it didn't pertain the aspect of hockey that people play on ice and wear helmets. The analogy specifically pertained the fact that as in hockey, the team that wins (scores more goals) is also automatically seen as a better team and matched with better teams in turn. They do not have to use very complicated means to evaluate a team's performance/skill levels to decide who they should play against, and they definitely don't let the team that lost (scored less goals) proceed in the brackets and leaderboards to face the better teams just because someone analyzed the game and said that the team that lost in terms of goals is however still better... No, they just look at goals for that stuff, which does also account for performance and skill levels, but ultimately is just what the game is about, winning, by achieving clearly-defined, diametrically opposed win conditions, advancing if you do so, not advancing if you do not. That's the point of the analogy, and Patrick could have used essentially any competitive PVP game in existence to make this point, he just chose hockey because he's Canadian.
1 -
Global average kill/escape rates are now much nearer 50% than they had been with old matchmaking, there couldn't really be much better of an indication that this new system has overall led to a more balanced state of matches in this game, which was its purpose.
1 -
As somebody who stopped playing the game entirely a few days after MMR dropped because I just stopped having any fun (and has been thinking about giving the game another try after it decays some due to inactivity), I think using pre-MMR data under the old system, not factoring in items/addons, and not being able to see our MMR as an option in the settings were big mistakes that make the system hated and feel awful.
3 -
This is beautiful and I kinda want to lick your brain now. But not in a weird way.
0 -
Can you imagine the controversy if the actual MMR number was exposed to an entire community that have no understand of what it means??
"Waaah... why is my mmr X?? My friend's is Y, but I'm a better player! Check out my cosmetics!! Also I did a sexy juke and t-bagged the killer. My MMR should double because of that alone but it didn't. The system is broken. Waaaahhhh!!"
0 -
err.... ok, I'm game :D
1 -
I think it is interesting that you mention that survivors not being efficient (hiding well, not doing gens) is a "rare" incident when it is more common than you think. It is a huge reason why solo queue survivor is in bad shape.
Like, you can think whatever you want about the system, but the fact still remains. Games are unfun with people who are so inefficient they end up forcing others to do everything. If the game is constantly giving them "wins" then (hypothetically) they are bound to be at the matchmaking level of those who escape who also actually perform objective actions. I don't see how pointing that out is in any way shape or form a bad thing.
Even while on a hook you can see the auras of survivors who are actively doing SOMETHING, and I don't believe anyone is expecting recognition for being efficient and contributing to the match, but to me its the same thing as when you ask a teammate why they didn't take a hit or why they didn't do a gen...in other words "contribute" and their immediate response is "I am alive you are the dead one", because they don't want to admit what that answer is.
So I am sticking with what we know right now, and that is that just because you escape doesn't mean you are good at the game. Which is what has always been there anyway.
0 -
"I'm not who you think I am, making me correct by default. I'm right and you're wrong"
Intellectualism at its finest and most condensed.
You wrote your entire comment trying to bring logic to a flawed perspective. You are now attributing thoroughness and truth as delusions of detractors. You are labeling those who want an accurate representation of skill and a less toxic model of scoring as **"People who've no clue what they're talking about will make any/every accusation and resort to logical fallacies in order to try and make a point."**
Calculating viability, skill, and value off of an end result of a complex mixture is the very definition of judging a book by its cover. You explain away many parts of development and what makes such a system easy for developers to use, or a streamlined effect that generates somewhat accurate results - but that is looking at the picture and not the landscape beyond.
You will never, ever, ever glean an accurate representation of player skill, value, or contribution by measuring W's and L's. You will always create a more boring and predictable game where the only objective is winning at all costs. You will never entice new players to stay once they start winning and face the mediocracy of all those W's.
Trying to wrap your head around a game once you strip away perks, tiles, killers, powers, items, maps, technical aspects, and interactions is a lot easier. Its very simple to strip away everything and look at a binary end result. But just because something is easy doesn't make it true. Just because something is simple does not make it healthy, and just because you think you can wrap your mind around something's development does not make you the arbiter of what is right and what is wrong.
3 -
sbmm doesn't make any sense in a casual game like DBD, players got used to the previous system and this new one is just killing fun out of everyone's games ending with people quitting the game.
Since sbmm came out i only have frustrating games with 4men and double chance perks nostop, i stopped play killer. Old MMR was way better at least i was able to chill and have a fair match
4 -
That's all very cute and all but as an end user of the game all I care about is if I get fun, evenly-matched games that I feel I could have won if I played well. That happens in about 10% of my play time, the rest are stomps or being stomped.
2 -
Dude, you are trying so hard to defend the system that you are not comprehending the things you are disagreeing with. I clearly stated that the emblem system was flawed, but it contained all the skill based information needed to create a skill based system. I also said that it ranked people up too easily. There is no flaw in my understanding. I pointed out that all that was really needed was to tweak the numbers so that wasn't the case and to remove anything that boosted rank but had nothing to do with skill.
I also did not contradict myself. I pointed out that it is a team-based game, but it ranks players as individuals, not as a team, therefore the analogy to hockey is completely flawed. How did you not understand that? If the survivors all won or lost as a team, then you could compare it to a team sport....maybe, but this system does not rank players as a team. Some members of the team go up, some go down, so nothing like hockey. If you are going to try and correct me, get it right. Yes, hockey looks at goals when it comes to who wins the game, but that does not reflect the skill of each individual player. Every player has their own stats, which is how they are ranked in the league regardless of how the rest of the team plays, and which is why good players on bad teams still get invited to the all-star games.
Also, that whole "it will balance out" thing does not work. When you lose, you are matched with even worse players, so you have an even greater chance of losing the next match and matches after that. That logic expects that you will somehow solo carry a 4 man team to victory, which is ridiculous.
6 -
I have the same experience of the new players of vhs then, MMR or not.
Let's say i wanna play a game of hillbilly, 4 survivors with AT LEAST 2000h when i have barely 100h(at killer), the game thinks im at their level? I'm not because by the time I end one single chase 3 gens pop and i can't do anything about it, they do all the gens and I'm force to facecamp or bring noed, sadly this is not my playstile so they all wait at the exit with their clicky clicky and tbags. I should get players of my level according to MMR but it's a nostop sweatfest of iridescent 4stacks swf with dh bt ub Ds/coh.
As i said this is not fun and I'm done with it, just feelsbadman for new players
2 -
You know what's funny about this whole dribble that you wrote... it's this sentence: "You will never, ever, ever glean an accurate representation of player skill, value, or contribution by measuring W's and L's."
Err... that's exactly what I've been saying, if only you'd care to read and understand instead of approaching it with a preconceived viewpoint that you needed validating.
In fact, I'd go further and say it's impossible to glean an accurate representation of player skill... full stop! Because how on earth are you going to measure it, without coming up with some consistent objective mechanism to measure it that produces consistent results in all match conditions, despite the fact that matches can vary wildly based on the actions of others and RNG, both of which are completely outside your control.
You're telling me to wrap my head around stuff... I'd humbly advise you to follow your own advice. It seems, from the comments in this thread, that the people who know how to built software and systems like this agree with how I've explained it. If you know better than us, then prove it. Explain in concrete terms where the flaws are and how to fix it. Let's see you implement what you think you know.
0 -
So, let me get this straight... 10% of your games were winnable had you played right (so we can call them even)... and the remaining 90% you're either winning or losing such that (I'm guessing here) overall you win/lose around 50% of the time?
Seems like the system is doing what it was designed to (which is not necessarily the same as what you're expecting or wanting it to)
0 -
You've probably heard these critiques before but since I don't want to read every single post on this thread I'll voice them anyway.
First and most important issue is I disagree strongly with viewing DBD as a 1v1 game in any capacity. This would be a critique of the base assumption that the goal/ideal for a SBMM system in DBD is to achieve a state where each individual survivor has a 50% chance to survive and the killer has a 50% chance of killing each survivor. This is so wrong because that just isn't what DBD is. DBD is a 1v4 game and any SBMM system is at the very least required to reflect this nature to be even remotely effective. While it seems there is some plan to create a hybrid system, I will repeat, considering DBD a 1v1 game ever is deeply flawed and can't work. Either all survivors win or none do.
Secondly on the subject of a facecamping Bubba you're coming at this just from the wrong perspective. Better players do not have solutions to a facecamping Bubba. While you're trying to distinguish this event from player skill you are already assuming player skill to be at issue here. It is functionally impossible even for the literal best teams in comp to do anything about a Bubba that intends to facecamp. They will always get at least 1 kill, potentially even 2 regardless of their skill and regardless of their opponent's skill. This isn't even the only version of this strategy, several killers can guarantee at least 1 kill per match regardless of their opponent's skill. And even those that can't guarantee it can make it vastly more likely by giving up on 'winning' to instead earn as many kills as possible. A SBMM system that uses individual kills/escapes fails to capture these basic gameplay facts.
Very similar to the previous critique, in fact this even comes naturally as a result of the previous critique, but a survivor that does a 5 gen run and dies losing MMR is a very significant problem. You speculate that pulling off a 5 gen run and dying must be a fluke, something not necessary to be accounted for by the system. That is wrong. If you 5 gen run the killer they have quite literally no better choice than to try and confirm their kill on you. If you had allowed yourself to go down earlier the killer would have had alternative options than to camp you. The system is actually consistent in this. Being better at running the killer will increase your odds of dying in a match. Now this doesn't apply when all 4 survivors are very good but more likely than not an above average player will be matched with players a bit lower than them. The players worse than them get hooked after short chases and the killer moves on. But when they get to a higher skilled player, the chase takes much longer. So long all the gens get finished and the killer is left with no real viable option other than to try and confirm their kill. This is not a fluke it is the standard way the game works. The system not only fails to capture skill but it can actually rate a player negatively because they were just too good. There is no greater failing for a SBMM system than to consistently lower the MMR of the best player in the match but that is what this system does.
You seem to have a very deep understanding of the statistics and algorithms used for this but if you don't understand the emergent metagame in DBD you will apply your statistics incorrectly. And that's what I believe I've seen from your post. It's also quite likely why we've ended up with this system in the first place. The MMR works perfectly for some game that doesn't actually exist but it absolutely fails DBD.
1 -
You misunderstood. I don't think inefficient or otherwise bad survivors are rare (I know all too well that they aren't, believe me), it's the scenario that they survive that is rare, and certainly, they do not survive as often as more efficient, better survivors. The system is about consistently surviving.
Those "non-players" (as I usually refer to them) do in fact not constantly get wins. Unless such a person plays in an SWF with players that are worlds better than them and does so exclusively, the scenarios where "deadweight" survivors are hard-carried by their teammates' 5-gen chases and whatnot are more than rare enough for them to not be able to maintain even just 50% escape rates at those levels, let alone more than that such that they would climb in rating. And again, even if this would somehow be the case for some absurd reason, you enter the next stage of absurdity where then a lot of such non-players would populate higher rating brackets, get paired with each other, but somehow still survive?
Your issue is that you think the MMR system is some form of "recognition", and I can understand why it would then be extra-annoying to sometimes see a player that didn't do much of anything useful get to escape while you die. But realize that the MMR system is not actually a recognition of anything, the only thing that could happen if that player's rating actually notable increases from that escape or other such escapes, is that they get matches where they are less likely to escape, and more likely to be brutalized by the killer, a thought you can maybe find some solace or a chuckle in.
The "recognition" if anything comes from the Bloodpoint/Grade stuff, which will definitely reward you more if you do all that stuff in the round, than it will the non-player.
Either way, the matchmaking definitely still has issues, it's not like I don't see survivor players in my games (on either side) that clearly do not belong in rounds with me. Maybe part of that is that the system actually concludes that it has to sometimes mix in much more lower-rated survivors to achieve balanced outcomes, and while that can be totally frustrating, maybe you can take it as a sort of compliment too: the system predicts that you are good enough to be able to carry these Megheads over the finish line against actually-competent killers that they would otherwise never stand a chance in hell against. And maybe if you took stats on your games, you would find that you actually do still average 50+% escapes even despite these matches, or hell, even when looking only at those matches.
I know that latter part doesn't do too much to make up for some of the frustations of playing with (and also against) such players, and I do hope that BHVR will tighten the matchmaking algorithms to actually pair more closely-rated players even if that affects queue times and turns out to skew kill/survival rates away from 50% (then actual game balance changes could be invoked of course), and reconsider on mixing much-lower-rated survivors into higher-rated lobbies (if this is actually an intended function of the matchmaker logic currently). But yeah, beyond that the only advice I can give is that you could try and change your perspective, to care less about actually surviving and more about just whatever gameplay you do get in those rounds, and see such games just as more of a challenge, give it your best, and don't mind so much if you don't make it. Easier said than done, but at least you should have an easier time with it now when you consider that MMR does not actually reward them or punish you.
Also, BHVR will apparently make the rating adjustments be dependent on the entire team's survival results after all, so in the future you won't even have to feel bad about the "rating" in that scenario, since that Meghead escaping increases yours too!
I don't think the emblem system contained all the skill information needed for such a system, and I don't believe it would be possible to tweak it such that its metrics and how it measures and weighs them could yield an as well-balanced outcome in terms of kill and survival chances/rates as the kill and survival-based system does. I also think BHVR would have realized and made that happen if it were readily possible, rather than create an entirely new system.
The difference between hockey and DbD in that regard is that in hockey teams are actually teams that play together all the time, "teams" in DbD are usually only playing together for that one match, so of course it would be absurd to rate them "as a team", those ratings would have to be discarded instantly. They can certainly be rated on basis of their performance as a team, but the ratings of course have to be applied individually. In hockey, teams are rated by wins as a whole, because they are the player entities that get matched against other teams, and while individual players are rated separately from their team with more metrics, those separate ratings are not used to make matches happen, matches are based on the team rating (goals - wins/losses), which are the equivalent to individual (survival/kill-based) player ratings in DbD. You can perhaps read more about that when you follow the trail left by OP/DesignDad, who mentioned that "PS – Love Trueskill, I remember reading papers about it many years ago. It solved some problems with matchmaking that are unique to videogames (ie building a fresh team for every game) in interesting ways."
Either way, my pointing out that the critique of the hockey analogy is misguided didn't hinge on correcting you on the teamplay thing, even if hockey were a 1V1 game the point would stand that the parts of it that Patrick drew analogy to, are analogous between DbD and hockey. The "matchmaking" systems only care about wins, only those hockey teams that score more goals than their opponents win, only they advance to face better teams that themselves also won.
I still don't get how you don't understand that it will balance out because you are matched against worse opponents (less killy killers in this case).
Look, if you consistently get matched with teammates that are so much more awful than you, I feel your pain, and have had more than my share of those experiences. I find things have improved a lot in that regard with MMR, and I get really rather alright teammates more often now when I queue solo (noticably more tournament players and other players that I recognize than I ever did prior to MMR, and general veterans of the game, many high-hour gamers), but if you are actually stuck in "Elo hell", I won't deny that that can happen, might depend on region too. What region are you in, maybe we could duo?
As a by the way: Much as it may seem here, I'm not usually eager to "defend" the devs or be optimistic about plenty of the things they do. I just genuinely think the MMR system - while certainly not perfect, for the reasons I was happy to point out in multiple posts - was an actual, tangible improvement over the old matchmaking system, and I am glad the devs "dared" to touch such an integral game function in this way. But ultimately none of what you or I or anyone here says will greatly sway them, they'll decide what to do anyway, and if you have any hopes that they will work on the Emblem system and bring it back to use for matchmaking, I doubt you'll see that happen.
Post edited by zarr on0 -
if it was designed to deliver 90% of games not being fun, yep it's a wild success.
0 -
Global averages are meaningless without the data behind them.
There's far too many variables. For all we know, it could be CoH and CoH alone that is pushing the global Kill Rates down.
They could've stopped counting DC's as Kills. We literally don't know.
3 -
I actually like the kill vs escapes thing, I just wish the team was considered as a whole rather than 4 1v1's.
0 -
CoH has definetly influenced a lot regarding general kills. Especially i feel disgusted if rng gives u linked pallet within boon zone. With medkit + predrops you can be healthy in <20s mid chase.
If and which is a honestly a must with DH pandemic, if it gets fairly nerfed, kill rates will go sky rocketing. Which will lead to new additions for survivors or waves of nerf on killer side.
It is not in balancing matches via SBMM.
The problem lies in mechanics, meta and maps.
As a killer I know that gens will flyby, hoping boon is not in a basement or a tower, and that every survivor is most likely running DH BT DS IW COH.
DH will always squeeze additonal lap or two in loops which after all take fair amount of time.
Going for 4K as objective will most likely require tunneling.
As a survivor I will most likely spawn on gen. Sit on gens for majority of time. Maybe have a chase or two. If I get hooked I need to be much more careful, as I might be obsession (being tunelled).
TLDR; Gen rush or tunneling is what is primary objective. Either way, it is not fun
1 -
Guys. These discussions are meaningless. Arguing how to do a proper MMR system in an asymmetrical game with player imbalances is completely pointless. The system could assign numbers perfectly and still not work because once you are in queue for a certain amount of time, the system doesn't work anyway. If you make it more strict in terms of skill, people quit because they don't want 10 minute minimum queues. If you make it less strict, like we have now, you have literally first game players going against people with thousands of hours. In both of these scenarios, the MMR might be calcuated perfectly, but none of this matters whatsoever in an asymmetrical game like Dead by Daylight that has an uneven amount of players on each side required to play the game.
Imagine if you queued into a League or Dota match but you had to queue as solo mid, support, jungler, etc. Your queue times would either be astronomically higher or the system would widen the search range, leading to extreme variance in skill.
League, Dota, Counterstrike - any game with functional MMR is not asymmetrical. You just need ten players to start the match. That's it. The roles do not matter. With Dead by Daylight, you need the exact correct ratio to make queue times not very long, and once they are long, as they are, the system does not work.
These discussions are meaningless.
7 -
DCs have after some date always been excluded from these stats. So if they have for some reason not been excluded for the MMR-era stats, that would mean those stats are actually inflated by around 5% (Peanits at some point mentioned that this is roughly what the increased kill rate when including DCs amounts to, probably a bit less now due to DC penalties), putting them even closer to 50% when subtracting games with DCs.
Either way, the "too many variables" argument is often misplaced. There are certainly too many variables at play to be able to use those stats to say everything about the game, there are arguments you cannot make in inference on global average stats alone... but you can absolutely say something and some things about the game based on them. One such thing is "the global, average state of balance when including all factors and variables is such and such", because that's what the stats literally say, they are the "live balance" that accounts for all possibilities that do affect matches, and they tell the story of the average, global win/loss experience more than anything else ever could, by literally showing the results of millions upon millions of matches, every single (pub) match every single player played during that time, "factoring in" every single thing that affected them.
To illustrate this: Even if DCs had not been excluded from the stats, there can be the argument that DCs are important to include; DCs unfortunately do in fact affect the gameplay experience of DbD, and insofar they affect global, average win/loss rates, they are for better and worse things that have to be accounted for when trying to shape the experience your average player will have in the game. That doesn't mean you balance the actual game for the idea that players can DC, but trying to keep global average rates within a reasonable margin even when including DCs is still an obvious goal, ideally of course by trying to reduce the incidence of DCs to begin with. Same for the countless other factors (e. g. you know Nurse has a higher skill floor than any other killer in the game and players new to her struggle playing her and that's the main factor why her rates are low, but yet you still want her to have global average kill rates closer to 50% so that people overall can be assumed to struggle less with playing her, so you buff her, give her add-ons that make her more of a "regular" killer by increasing her base movement speed, fix her bugs, implement MMR that gives beginner Nurses easier matches, ... and that all despite her also being the undisputedly strongest killer in the game in the hands of a skilled player, which has always been true and yet she received various buffs throughout the game's existence and will soon receive a big fix pass, again without any nerfs).
...Anyway, comparing global average rates from then to now is definitely rather telling in terms of what the global average win/loss experience is like with MMR, and sorry, I have an impossible time seeing that COH is what would affect rates this much. But even if it did, that would ultimately just add to the argument, as the stats would say MMR can keep kill rates around 50% despite such an unbelievably (literally) impactful perk having been introduced.
0 -
You got called out hard by Peanits, where is your response. If you don't put one out I will call you out whenever you post anything in this forum.
0 -
Using false equivalences like that hockey analogy fallacy didn't really help their arguments.
1 -
I'm not convinced queue times would get prohibitively long if we made the matchmaker a little more strict. Crossplay off queues (at least prior to MMR when lots of people still played on it) used to be pretty reasonable in my region, yet also much more consistently paired notably higher-level players of both roles. The players are out there, queueing at the same time, the matchmaker just seems to usually match them away before they can find each other.
Obviously we don't want Dowsey-tier hour-long queues, but if the system would at least look for much stricter rating matches during the first 5 minutes, I think things could already be much more consistent. And this could even be implemented with a cut-off point, where only players above a certain MMR threshold (starting at high mid to high range) are more strictly queued among their brackets in those first minutes, which might help with queue times overall since less of them will be paired away into lower brackets, more remaining available for "bracket-exclusive" pairing, and since it also won't affect lower bracket queue times much as they can still be matched more liberally even in the first 5 minutes, including with players from the upper brackets that have at that point already been waiting for more than 5.
Obviously I don't know any of this, we would have to see actual numbers on this stuff, and it might be fair to say that since BHVR is not doing it they may have already decided faster queue times are more important... but I for one would be okay with ~5 minute average high MMR queues if it more consistently led to closer matches. Although you as a streamer are probably not quite okay with that.
1 -
That's the saddest thing I've heard all week. "You got called out by a dev whose trustworthiness is paper-thin, so I'm going to hound you wherever you go." News flash, trying to get attention from me won't make up for the attention you didn't receive from your parents growing up.
Basically, Peanits said that their desired average is a 50% kill rate, meaning that their desired average (meaning the majority of and ideally all) matches are a 2k2e. There's no other way you can spin that in a way that isn't "We want every other match to alternate between a 4k and 4e" or "We want matches to still have wild amounts of variance between the results to the point where they are individually meaningless but hey, they collectively average out to this random number we abstractly decided was the pinnacle of balance."
Peanits more or less said "Find me where we said wanted 2k2e matches on average. What we actually want is 2k2e matches on average."
1 -
Interesting. People who have turned crossplay actually notice their games are even more mismatched than ever. I think CoconutRTS did an experiment with that as well (as anecdotal as it is). It would make sense, as you would have longer queues with it turned off, thus leading to wider variance.
Its also a slippery slope. You might be okay with 5 minute queues every time, but a lot of people wouldn't be. This would in turn, lead to less players, leading to longer queues, leading to wider variance, in an endless cycle.
1 -
Wow, insult a developer and put words into his mouth. And honestly you're the one here who's words are paper thin. You've yet to put anything up, where's the clip to back-up your claim. Now I won't hound you, but I might remind you that you have yet to put out anything substantive.
0 -
To be fair, I haven't been queueing much on cross off post-implementation of MMR, I'm one of the people that transitioned back into the ""casual"" pool after it became more bearable thanks to MMR (I know in your experience things haven't changed much, but I for one really do see top-level players relatively frequently in my lobbies now that I only rarely did prior to MMR). So I can't really say what cross off is usually like now, but it used to be good enough.
I get your second argument, and as easy as a suggestion like "make matchmaker a little more strict for high MMR" sounds, in reality tweaking these numbers likely indeed leads to more and unforeseen knock-on effects. It's definitely rather complicated, I mean, there's a reason BHVR has for years been and still is tinkering with this stuff. So yeah, like I suggested, maybe the fact that BHVR seems to favour queue times to be as short as they are now already suggests making them any longer in favour of ratings would lead to bad things.
Maybe they could test a split "casual"/"competitive" queue for a week or so after all...
0 -
Pretty much my same standpoint, except I don't think its the best system the game can have, there's too much that can go on in a match of dbd that an mmr system really just can't keep up with it, no matter how perfected it is, but overall good post and pretty good take
0 -
Yes, I didn't say completing generators was the entire Survivor objective. I said completing generators and opening the gates is. You could pedantically add "and then walking out of the gates", but the gist is that for a 2E you still need all mechanical impediments to escaping completed--i.e., Survivor goals must be completed.
And I said the ridiculous part was that the win condition is defined per-Survivor. i.e., you can have two Survivors win, two Survivors lose, and the Killer technically has two wins and two losses. We get 8 individual MMR updates (neat that the Killer's batched into a glicko-like update rule, but that just converts it over to four terms in a sum--they're still liable to have a net shift). But this is a draw. And if one person dies but 3 people escape, that's a clear win for Survivors. But here we have it as three wins and one loss.
Let's say we have a SWF group who make new accounts to always play together. You'd think they should have the same MMR, but instead if one person is specifically the one who draws attention and dies the most, then you'll get this divergence where their MMR is drastically lower (if not decreasing), even if they maintained a 3-1 escape average. But by approaching the team-focused side as a team, the current matchmaking gets thrown off.
And this actually matters if the team MMR is averaged! Let's say they're particularly extreme about it and always die. Maybe this then gets to the other players and one of them dies or they all die--a pretty bad performance for this person individually, sure, but maybe they're just an average-ish group otherwise and the others die even faster. Nevertheless, they always "lose" by the individual standard, so their MMR drops to 500. But on average, the team escapes enough for the others to rise to 1500--theoretically starting at 1k. Average MMR is (1500*3+500)/4: 1250. If you consider that the fourth person's skill (regardless of deaths) is integral to the results, they're being consistently undermatched until the average MMR is this 1500-point. Which may take a while, as the remaining player's MMR hasn't quite bottomed out, and the other individual players are consistently out-MMRing the Killer. This might even stabilise them at a lower MMR than they belong for their match results, due to the gap in individual MMRs getting too big to really adjust against the points they're matching.
If the team MMR isn't averaged then it's simultaneously matching against noobs as well as good Killers, and we know that's a failure.
i.e. defining the win condition based on individual results can be demonstrably aberrant if the team side functions as a team.
1 -
Hey, I actually have experience with Crossplay off! I can help!
Crossplay off basically doesn't exist on the Series X and PS5. The pools are way way way too small and it took me far too long to find a lobby. I gave up after 30 minutes. I suspect this is because the consoles are difficult to get and the queue is limited to only next-gen systems.
On the previous gen, my Killer queues were anywhere between 5 minutes to over 45. I would get matched with the same 4-ish SWF's pretty consistently, with a game of randoms thrown in between matches. Survivor queues were also pretty long. 10-20 minutes on average.
I now have Crossplay on. My queue's for Killer are instant but extremely inconsistent. My queue's for Survivor are roughly 7-10 minutes long and are also inconsistent, as the data I've been collecting shows. I can, of course, give you a PDF of the spreadsheet my friends and I are keeping, we've got 270 games so far, iirc. It's pretty interesting, if I do say so myself, so if you're interested in seeing it let me know.
In short, the system often goes beyond the 5 minute range in search of matches (at least on the Survivor end) and I suspect if we narrowed the search parameters, we'd be looking at much longer queues. I do believe people would rather play an unbalanced game than no game at all and it's no fun being in queue longer than you are in a match. Of course, this is all speculation, as Scott so succinctly put it, it's meaningless in the end. BHVR will do what BHVR does.
1 -
The problem with MMR is, even if it’s a PERFECT system, it won’t work if people aren’t playing one side *cough cough* killer so the MMR system has to expand the range at which people are included in the search for a lobby, rendering it ineffective
1 -
Thank you for this post. I've realised recently that I hate this community anytime the devs do anything. It's not the devs fault this game is dying, it's our entitled community.
3 -
How exactly can be community fault?
If the development brings bad decisions into the game, its the community fault cos they dont enjoy the game anymore?
5 -
Whilst I wouldn't lump every member of the community together, you have a good point.
Imagine creating a game which you feel is good and understand needs some tweeks, and yet every day you read messages, comments and stuff saying: "You have no ######### clue about your game?", "This game is terrible because the devs suck", "Call yourself devs? You're pathetic!". Whoever you are, you will build a resentfulness. You've built up a game for years and made a success, and yet still berated. I'm certainly not saying that means BHVR ought to respond with jibes - that's unprofessional. But it's clear why some are as they are.
The better aspects of the community needs to make sure that the childish comments the minority make are dealt with.
3 -
Ok, here we go...
"First and most important issue is I disagree strongly with viewing DBD as a 1v1 game in any capacity. This would be a critique of the base assumption that the goal/ideal for a SBMM system in DBD is to achieve a state where each individual survivor has a 50% chance to survive and the killer has a 50% chance of killing each survivor. This is so wrong because that just isn't what DBD is. DBD is a 1v4 game and any SBMM system is at the very least required to reflect this nature to be even remotely effective. While it seems there is some plan to create a hybrid system, I will repeat, considering DBD a 1v1 game ever is deeply flawed and can't work. Either all survivors win or none do."
I get what you're saying. But the problem is everyone has to be ranked individually because they come into the game as individuals. I play solo most of the time (woe is me, I have no friends), so most of my matches are with 3 randoms (and a random killer)... how on earth is any MMR system supposed to apply a ranking to reflect this particular unique grouping that may never happen again??
It can't! It's pretty much impossible (to the point where, if there were a solution, it'd require so much effort to discover, implement, and validate that it almost becomes a self-defeating exercise).
The system HAS to look at each player as individuals to give each player their own ranking, so it can match them with other players. If you know of a better way, then I'll present you the same option as everyone else... tell us what it is!
"Secondly on the subject of a facecamping Bubba you're coming at this just from the wrong perspective. Better players do not have solutions to a facecamping Bubba."
Yes they do. Firstly, they're better in chase, so less likely to be caught. They're also more map aware so if they do go down, they'll go down where a rescue is most likely to succeed. Furthermore, better teams can co-ordinate rescues in difficult circumstances. As killer, I'll camp the hook in the EGC, and I've seen many a time survivors make a successful rescue. It absolutely can be done. Your problem now is that you think you possess the sum knowledge of all possibilities, and that's clouding your judgement severely.
"They will always get at least 1 kill, potentially even 2 regardless of their skill and regardless of their opponent's skill."
So, you're saying it's completely impossible for a Bubba player to exit a trial with 0-kills? Really? That's your argument?? Oh dear.
"You speculate that pulling off a 5 gen run and dying must be a fluke, something not necessary to be accounted for by the system."
I said no such thing, and this just goes to show you didn't read what I said with intent to understand, but rather you just tried to quote mine and cherry pick points you think support your argument.
I said in that scenario, it was either a fluke chase, or something the survivor is skilled enough to repeat. If the survivor can repeat it, they will repeat it, and when you look at their games overall you'll see they escape more often and their level will rise.
Aside from not reading and not understanding, your other problem is this supposition that going up/down in MMR level is a good/bad thing. It's not. There's no kudos or reward from having a higher level. It's literally just a mechanism to match you with similarly skilled players. It's not a comment on your level of skill. The system isn't designed to stroke your ego. It's designed to match make.
1
