Website Feedback
Closed Threads
IRC Chat
irc.quakenet.org #teamliquid
IRC Web ClientTeamSpeak 3 (47 users) | |
|
| SKC Brazil. June 16 2012 09:57. Posts 3257 | Profile # |
On June 16 2012 09:44 DodgySmalls wrote:
@SKC and @BurningSera
Obviously you aren't going to do as well at some heroes and others but having a total range of 40% skill disparity is pretty high. If there is a hero you actually can't play that well it's just probable that the hero is worse than other heroes and you shouldn't pick it.
IE I do much better at antimage than I do at alchemist, that's definitely not because I play better or worse every game. In fact, I probably play to within a 1 or 2% skill margin.
So saying "mmr is stupid because there are different heroes and different heroes are stronger/weaker" is nonsensical because it's your own fault for picking a shitty hero or one that you're not good at.
You should be able to play at least a few heroes from any role NEAR the skill level you normally remain at. That narrows it down from 100ish to 10-20 heroes that you need to be good at.
I think the final piece of the puzzle is that there's no way to ladder CM which would really be necessary to have valid mmr as even random draft can create strong imbalances based on certain picks.
Noone said MMR is stupid because people's plays isn't consistent? It was simply stated that the quality of people's play varies depending on several factors, as does their MMR, so you cannot simply take MMR as a definitive way to say how well they will play that single game, for better or for worse, within a reasonable skill gap obviously. Specially when everyone's MMR is fairly close together, as a matchmaking game should be.
Obviously noone always picks the best hero to win the game in AP. Well, not noone, there are a few people that pick Lycan every single game, but it's a rarity. People pick to have fun, which ussually mean not picking an awful team because getting stomped isn't fun, but it's far from insta picking Furion/Lycan/CK/Invoker/etc every single game. A lot of people play far more than 20 heroes, quite often close to the full roster avaible. Some times you simply want to play a hero you are not that good at. What about people that random? Some times you want to random, and there isn't anything wrong with that. AP pubs isn't a competitive system, you can't treat it as such. The game would be extremelly boring if people actually did their best to win every AP game. It would be a rush to pick the best heroes you would always see the same heroes every game.Last edit: 2012-06-16 10:10:24 |
|
|
| paralleluniverse Australia. June 16 2012 13:25. Posts 2946 | Profile # |
On June 16 2012 06:59 SKC wrote:
Show nested quote +On June 16 2012 05:34 Itsmedudeman wrote:
On June 16 2012 05:23 SKC wrote:
On June 16 2012 05:16 Itsmedudeman wrote:
On June 16 2012 05:10 SKC wrote:
On June 16 2012 05:06 Itsmedudeman wrote:
On June 16 2012 04:57 SKC wrote:
On June 16 2012 04:45 paralleluniverse wrote:
On June 16 2012 04:33 SKC wrote:
On June 16 2012 04:10 Itsmedudeman wrote:
[quote]
Is this a joke dude? Look at HoN and LoL. Both their rating systems use MMR and most people stay incredibly close to 50%. Even the highest rated players float around 50-55% which is incredible. It's actually much more volatile in 1v1 situations you don't even know what you're talking about. Look at the top of GM with people sitting at 80-90% winrates. The MMR system in Dota/HoN/LoL keeps everyone at a roughly 50% winrate which should be the entire goal of a matchmaking system.
I kinda messed up the wording, the issue are the balanced conditions, but that was poor wording, the game often starts inbalanced simply because of picks and random teammates, for example. It will obviously bring you back if you go too far from your true MMR and get close to 50% winrate with enough time, but that's no surprise, any MMR system will do that. The issue is how MMR can be skewed by other factor and is not a very precise in any way when dealing with people close to each other. You can easily have a win streak and go from 900 to 1000 or a loss streak and go from 900 to 800. That doesn't mean you got better or worse. That doesn't mean a 1000 person is necessairly better than a 900 person. Adding the fact wins or losses may not have been your fault, sometimes you get carried, sometimes your team makes you lose, sometimes the other team picks Lycan + Furion while your team randoms adds a lot of randomness to Dota, making MMR unrealiable for a quick analysis. 1v1 SC is a lot simpler, it's a lot more balanced and the MMR is still volatile.
All of those factors, like being carried is implicitly accounted for in MMR, because they affect your chances of winning and MMR considers whether you win or not. Obviously MMR goes up if you have a win streak. So a good measure of skill should stay the same, in the face of evidence that you are better because you're winning more? The question is a simple one, is MMR a creditable measure of skill? The answer if you look at the evidence is overwhelmingly yes given that all players have a close to 50% win ratio, even in team games. This already accounts for the fact that MMR changes during win streaks and lose streaks, and yet there are STILL 50% win ratios. If it didn't measure skill correctly, this would not be the case.
Those factor are not accounted in the MMR because the game doesn't know about them. I'm not talking about when people always get carried or always carry someone. The issues happen in situations like: You have a shitty friend. He always plays with you. You sometimes play alone. Both of you start at 500 and 1000 MMR playing together. Your points will start merging, since you will lose more points in losses and win less points on wins. after sometime you both are 600 and 800. Now you decide do play alone. Are you a 1000 or 800 player? Is he better or worse than a 900 player? An MMR system obviouly needs to work like it does, but that doesn't mean you can't accept there is a margin of doubt, a region of incertainty where it is simply not acurate. This is obviously a quick and slightly exagerated example, but things like that add randomness to the system that doesn't exist in SC. Speaking of which, here is a method that people believe can show you your MMR: http://www.teamliquid.net/forum/viewmessage.php?topic_id=332391And it's mentioned how wildly it varies quite frequently. It was an issue when developing the method.
So? The MMR system is doing its job and when you play with someone constantly over and over your average true MMR is shown which is the only that matters when you queue with the person going into the next game. If you want to exaggerate circumstances then the only thing that should matter are those in the exaggerated circumstances. In this case, what your teams MMR total is at and your chances if you've played with your friend all the way through for a long period of time. If we up the ante and change this to a 5 person group, then the MMR your team has is the ONLY thing that matters for your next matched game, not what your individual skill rating is.
Yes, but that only works if you keep the 5 person group. The next time either you or your friends solo queues, your MMR will be off. Then it will fluctuate again to your "true" MMR. Then if you play again, if may change again. But those games also have an influence on other games and other people's MMR's. It's just another factor that adds to the randomness of it.
On June 16 2012 05:10 Itsmedudeman wrote:
On June 16 2012 05:03 SKC wrote:
On June 16 2012 04:57 SKC wrote:
On June 16 2012 04:45 paralleluniverse wrote:
On June 16 2012 04:33 SKC wrote:
[quote]
I kinda messed up the wording, the issue are the balanced conditions, but that was poor wording, the game often starts inbalanced simply because of picks and random teammates, for example. It will obviously bring you back if you go too far from your true MMR and get close to 50% winrate with enough time, but that's no surprise, any MMR system will do that. The issue is how MMR can be skewed by other factor and is not a very precise in any way when dealing with people close to each other. You can easily have a win streak and go from 900 to 1000 or a loss streak and go from 900 to 800. That doesn't mean you got better or worse. That doesn't mean a 1000 person is necessairly better than a 900 person. Adding the fact wins or losses may not have been your fault, sometimes you get carried, sometimes your team makes you lose, sometimes the other team picks Lycan + Furion while your team randoms adds a lot of randomness to Dota, making MMR unrealiable for a quick analysis. 1v1 SC is a lot simpler, it's a lot more balanced and the MMR is still volatile.
All of those factors, like being carried is implicitly accounted for in MMR, because they affect your chances of winning and MMR considers whether you win or not. Obviously MMR goes up if you have a win streak. So a good measure of skill should stay the same, in the face of evidence that you are better because you're winning more? The question is a simple one, is MMR a creditable measure of skill? The answer if you look at the evidence is overwhelmingly yes given that all players have a close to 50% win ratio, even in team games. This already accounts for the fact that MMR changes during win streaks and lose streaks, and yet there are STILL 50% win ratios. If it didn't measure skill correctly, this would not be the case.
Those factor are not accounted in the MMR because the game doesn't know about them. I'm not talking about when people always get carried or always carry someone. The issues happen in situations like: You have a shitty friend. He always plays with you. You sometimes play alone. Both of you start at 500 and 1000 MMR playing together. Your points will start merging, since you will lose more points in losses and win less points on wins. after sometime you both are 600 and 800. Now you decide do play alone. Are you a 1000 or 800 player? Is he better or worse than a 900 player? An MMR system obviouly needs to work like it does, but that doesn't mean you can't accept there is a margin of doubt, a region of incertainty where it is simply not acurate. This is obviously a quick and slightly exagerated example, but things like that add randomness to the system that doesn't exist in SC. Speaking of which, here is a method that people believe can show you your MMR: http://www.teamliquid.net/forum/viewmessage.php?topic_id=332391And it's mentioned how wildly it varies quite frequently. It was an issue when developing the method.
On June 16 2012 04:55 Itsmedudeman wrote:
On June 16 2012 04:33 SKC wrote:
On June 16 2012 04:10 Itsmedudeman wrote:
[quote]
Is this a joke dude? Look at HoN and LoL. Both their rating systems use MMR and most people stay incredibly close to 50%. Even the highest rated players float around 50-55% which is incredible. It's actually much more volatile in 1v1 situations you don't even know what you're talking about. Look at the top of GM with people sitting at 80-90% winrates. The MMR system in Dota/HoN/LoL keeps everyone at a roughly 50% winrate which should be the entire goal of a matchmaking system.
I kinda messed up the wording, the issue are the balanced conditions, but that was poor wording, the game often starts inbalanced simply because of picks and random teammates, for example. It will obviously bring you back if you go too far from your true MMR and get close to 50% winrate with enough time, but that's no surprise, any MMR system will do that. The issue is how MMR can be skewed by other factor and is not a very precise in any way when dealing with people close to each other. You can easily have a win streak and go from 900 to 1000 or a loss streak and go from 900 to 800. That doesn't mean you got better or worse. That doesn't mean a 1000 person is necessairly better than a 900 person. Adding the fact wins or losses may not have been your fault, sometimes you get carried, sometimes your team makes you lose, sometimes the other team picks Lycan + Furion while your team randoms adds a lot of randomness to Dota, making MMR unrealiable for a quick analysis. 1v1 SC is a lot simpler, it's a lot more balanced and the MMR is still volatile.
That has absolutely nothing to do with "undeserved MMR" or anything of the sort. The fact is every person is volatile and has a range of performance. If someone performs badly that may cause you to lose, then that's just how it is, but it doesn't mean they're bad, it just means they had a bad game and suffered for it. But a range of 100 or so rating is not a big deal. That's as best as you can hope for. To say you need to win or lose based on your individual performance each game is dumb, and there's simply no way to measure that. K/D is the worst possible thing you could ever use for any rating. Anyone with experience in dota or any other game knows this. Simply put, if you can propel your team to win past their weakest link and carry your weakest link, then you deserve the rating you've gained and vise versa. "Wahhh, we had a feeder on our team so I lost like 4 games in a row". Chances are they had a bad player as well, but everyone else around them played better.
I never said you should use K/D or something like that. I said a perfect system would analyze if you actually played well or not. That's impossible and winning or losing is a better method to do that than taking K/D into account would be. I also didn't say you ever "don't deserve" your MMR because your teammates always suck. In time, things like that are averaged. But these things generate a good amount of uncertainty if you look at a single point in time and fairly small diferences in MMR. There are a lot of factors that can generate a pretty good level of uncertainty about your "true" MMR and it can easily vary widely in time without any diference on your actual skill. But people only see that single number you carry right now and take too many conclusions that they shouldn't about what it means. Damn quote instead of edit.
I'll say this again. Under no system do you have a non volatile individual MMR. It's not just your teammates or picks that are volatile, it's you yourself and every person in the world who plays dota. There is no "actual" skill. Your actual skill is just as volatile as how your team performs every game.
Which is exacly what I was trying to say to the guy that said your MMR stays mostly on the same place and is not volatile. It is. It changes wildly sometimes.
So what's even your point? You've been trying to argue that there is somehow this holy grail of individual performance measurement, but there isn't. I'll say it right now, any suggestion you put forth for a better game to game matchmaking is just plain worse and dumb and you shouldn't suggest it cause I'll just flame you. And no, it doesn't change a "lot".
It started with the argument that people exagerate on the value of MMR, they will take a look at the number and believe it means everything. That a 50 points diference means they are a better player and could tell everyone what they did. Then it went to an argument on how acurate MMR is, and it is not that acurate if you look at a single amount of time. That's why you can use it for matchmaking, in fact you should use it, there is no better way, but the fact it's volatile is an issue. Even for SC2, which has a much simpler system. People don't understand that there's a pretty big amount of uncertainty in these numbers. And yes, it does change wildly sometimes. Get a new account for Dendi and see his MMR skyrocket for an easy example, but there are obviously other situations where it can happen. Like the arranged teams issue. The fact the game will be free to play may generate a region where there's a lot of uncertainty on your MMR if smurfs become common as well. I didn't say it changes a lot often.
Why are you accounting for obvious abuses to any rating system? Yeah, get a smurf to tank someone's rating, cool. But this in no way is a flaw to the MMR system, only a ranked system of any sort where you're allowed to reset your rank. And it's not volatile. You will never be matched against significantly worse players than you if you are that good. Only slightly worse or slightly better, which is the best one can hope for a 5 man team game.
I'm not saying the MMR system is a problem. I'm saying there are reasons not to show it, not that there are reasons not to use it. You have to take obvious abuse into account because obvious abuses exist. They are one of many little things that add to the uncertainty of your MMR. Let's try again. You can't say someone skill level is equal to 1243.56 MMR. You should say it's around 1200-1400, because there is a degree of uncertainty and variability. Now let's get to the issue. When you create a game, you ussually find people around your skill level, a game should have 1200-1400 players, for example. Inside that game, it's safe to say everyone is around the same level. But that's not how people act. They ignore the uncertainty of the system (except for themselfs, they are always their peak MMR, if not more) and act like just because they are 1350 they are better than a 1250 and should be able to tell them what to do and how to play. In truth, the worst ranked player in a team could be the best player if he just had a unlucky streak and lost a lot of games in a row. That doesn't mean he will play worse, his skill level didn't change. People will find any reason they can to act like a moron to someone else, and MMR is quite often something that triggers it. I don't think hidding it does more harm than good. That also doesn't mean people will stop being asses if MMR isn't shown, it only mean there is no reason to give them one more way to do it. That doesn't mean MMR can't tell if you are bronze or platinum. That irrelevant, you won't find plat players as bronze unless there's something very wrong. The issue is that the range of uncertainty of the MMR is very close to the range of people you play with. I also don't see the point of simulating anything unless you add all variables that make your MMR fluctuate. Did you consider wether you had a good night of sleep? You should't have to take shit from a 1000 if you dropped from 1100 to 950 last night for whatever reason. And again, I'm not saying you shouldn't have lost those points. I'm saying people react the wrong way about it and don't understand how the system works. If they did, there would be less bitching about "ELO hell". As a side benefit, you don't get people posting their ranks as a confirmation that they know what they are doing while ignoring anything a lower ranked player says. There is a reason the idea of showing your rank on the SC2 strategy forums is always denied.
http://research.microsoft.com/en-us/projects/trueskill/details.aspx
You continue to argue from ignorance about how these skills systems work, and disregard how they explicitly take into account uncertainty about skill. In fact, that's one of the hallmarks of these types of skill measuring systems when compared to ELO which does no such thing. As I've already said, and as you've ignored, the uncertainty is measured by sigma, and can be taken into account by calculating mu-3*sigma, which is the lower bound of a 99.7% confidence interval, as TrueSkill does. Or it can even be given as a 95% confidence interval mu+/-1.96*sigma.
Also sigma always naturally decreases over time, i.e. uncertainty decreases as players play more games. Notice the factor in front of the sigma^2 equation is always between 0 and 1, showing that uncertainty decreases. In fact, they have to add a small uncertainty term (this is the gamma^2 in the TrueSkill paper previously link) to prevent it from going to 0 and never changing. Skill is determined by a normal random variable MMR~N(mu,sigma^2) with decreasing variability, it's performance that is determined by the random variable N(MMR,beta^2) which has constant uncertainty beta^2, as determined by the skill level of the game. Therefore, the variability of your performance is not important, it's stripped out of MMR, because MMR measures skill which is defined by consistency. MMR uncertainty decreases and can be taken into account as TrueSkill shows.
Things like whether or not you have a good night of sleep is taken into account by the variability of the PERFORMANCE, which is incorporated into the beta^2, so I did take it into account. You need to understand that there's TWO different things that TrueSkill and other MMR systems model, i.e. MMR which is skill, and performance which is modeled by as a normal random variable, centered at MMR with a variability of beta^2, and thus the variability of performance can be separated from the uncertainty about skill.
The only valid point that you have given is education, i.e. people, like you, don't understand the system. But that can be fixed by more education or better communication, e.g. giving MMR as a range, not a number, or giving MMR as a conservative estimate as TrueSkill does, or showing unmodified SC2-style points (as long as they can be compared, which they can't be in SC2), or even showing a percentile, or a percentile range like top 25%-30%.Last edit: 2012-06-16 13:59:28 |
| | Check out "Shadow of the Eternals" on Kickstarter |
|
|
 Excalibur_Z United States. June 16 2012 13:32. Posts 10287 | Profile # |
| It's worth noting that TrueSkill reduces uncertainty much much more rapidly than Blizzard's system does, and this has the side effect of "locking" a player into a particular TrueSkill value. It takes quite a few unexpected results for TrueSkill to expand your sigma again so that you can start shifting over toward a new value. Blizzard's uncertainty never gets this small, it's designed to remain flexible enough so that situation doesn't happen. |
| |
|
| paralleluniverse Australia. June 16 2012 13:40. Posts 2946 | Profile # |
On June 16 2012 13:32 Excalibur_Z wrote:
It's worth noting that TrueSkill reduces uncertainty much much more rapidly than Blizzard's system does, and this has the side effect of "locking" a player into a particular TrueSkill value. It takes quite a few unexpected results for TrueSkill to expand your sigma again so that you can start shifting over toward a new value. Blizzard's uncertainty never gets this small, it's designed to remain flexible enough so that situation doesn't happen.
Actually, while sigma^2 would naturally decrease to 0 in TrueSkill as the above equations show, they add a term to sigma^2 that is updated after the game results to prevent this. That's the gamma^2 term, which is a parameter that can be chosen, tuned or optimized, at the top of the last page of the TrueSkill paper: https://research.microsoft.com/pubs/67956/NIPS2006_0688.pdf
Obviously, Blizzard would be doing something similar, I suspect that Blizzard doesn't use a constant, but a function of a few other variables, like the unexpectedness of the results of the last few games.Last edit: 2012-06-16 14:29:34 |
| | Check out "Shadow of the Eternals" on Kickstarter |
|
|
| paralleluniverse Australia. June 16 2012 14:02. Posts 2946 | Profile # |
On June 16 2012 07:27 Excalibur_Z wrote:
Show nested quote +On June 16 2012 05:37 paralleluniverse wrote:
On June 16 2012 05:23 Excalibur_Z wrote:
On June 16 2012 04:13 paralleluniverse wrote:
On June 16 2012 03:48 SKC wrote:
On June 16 2012 03:38 paralleluniverse wrote:
"Overview At the heart of the system is a hidden value known as the matchmaking rating, or MMR for short. Matchmaking rating helps to ensure you play against players around your skill level and influences how many points you stand to gain or lose per match. Your points will drift toward your hidden MMR over a period of time, but because MMR is more volatile than points, your MMR is never cemented at a fixed value. For this reason, it is extraordinarily difficult to reverse engineer MMR from points." Taken from http://www.teamliquid.net/forum/viewmessage.php?topic_id=195273. It's a well respected guide in TL. Show me Blizzard saying the opposite first, because I'm also pretty sure they said MMR is volatile.
He's saying that MMR isn't fixed, but fluctuates a bit around your true skill, which is pretty much what I said. And you would already know this if you play WoW arena's where you're essentially mostly fluctuating +/- 100 points (on a scale from 0-3000, so 100 is very small). How does he know this given that he admits that "it is extraordinarily difficult [impossible] to reverse engineer MMR", where is his source? The source is information about WoW's system.
He says it's never cemented at a fixed value. He says MMR is more volatile than points. You said MMR is not volatile, it hardly ever changes. Basically that it's fixed and points change more easily than MMR. Basically opposite views. Are you trying to work his explanations into yours or simply saying he is flat out wrong? I can't really see it from your post. If the first, I don't see how it's possible. If the second, I believe he spent a little more time than you studying the system, I'll take his word over yours. Plus my memory, since I think I remember Blizzard saying the MMR system is volatile. I think that's more credible than "the information about WoW's system". Maybe if you could find the source of Blizzard's quote that it isn't volatile, I'm pretty sure ExcaliburZ pays attention to things like Blizzcon and wouldn't contradict Blizzard so heavily.
There is not a shred of evidence that MMR is more volatile than points. It's not possible to tell what MMR is, so how can he conclude it's more volatile than points? I never said that MMR is cemented at a fix value, I said it converges to your true skill and fluctuates around that point. But a +/- 100 fluctuation on a 0-3000 scale which is what you get in WoW is so small that it's unimportant. He can pay all the attention he wants, the fact of the matter is that Blizzard has never said anything substantive on the SC2 ladder system to warrant anyone's attention, so it's all (mostly correct) extrapolation from WoW's system, which Blizzard has said a lot about. They did give him a nod that the broad general idea of his post is correct at Blizzcon, but not all the details. Again, I challenge you to simulate MMR on a computer. It's not volatile. It fluctuates a little around a point. If MMR works so well to measure skill in 1v1, then how can it be volatile? Player skill is mostly consistent. Diamond players don't suddenly wake up with the skill of a silver player.
Actually if you follow Not_That's F function values, it seems that MMR is indeed more volatile than points, but only slightly. For example, if evenly matched, both players would earn +/-12 points but their MMRs would change +/-14 (might be 16, I forget). So, even though MMR is more volatile than points, it's not like it's spiking up and down with extreme peaks and valleys after every single game. Generally MMR is going to hover around a particular range like you said. You're not going to go from Bronze MMR to Diamond MMR after like 20 straight wins, it takes way way way longer than that. The same should hold true for Dota or any other Elo-based game. On a similar note, players typically brag about their Elo peak and use that to gauge their skill. Peaks are irrelevant. The average is what actually speaks in terms of skill. Someone with a peak of 2500 but who has 90% of his games at 2000 isn't going to be as skilled as someone who has a peak of 2300 with 90% of his games played at 2150. Anyone can luck their way into a temporary high rating, but it's consistency that defines skill. Valve has been pretty adamant about not showing MMR (to players or to anyone) so that's probably not going to change, but it's probably for the best. I haven't played HoN or LoL but I've heard the stories over "Elo elitism" and the completely fabricated "Elo hell" which allegedly exists below a certain threshold.
I haven't looked at this work in it's full details since it's first posting, but: 1) Are the points and the MMR on the same scale? 2) Does it account for the fact that the MMR update depends on the MMR and uncertainty of MMR of both players (does this vary linearly with points? In which case it wouldn't matter.) 3) What significance is there in that MMR is allegedly a tiny bit more volatile than points, when points are pretty much nonvolatile anyway, accounting for bonus pool. Points usually fluctuate at around like 800 +/- maybe 24, after 2 games for a player whose points has converged to his MMR, about halfway through the season. Also, you never responded to my post about ranking system in the Arcade beta post.
1. Not entirely sure about "same scale" because there's a point floor which may or may not coincide with the MMR floor, and there's an MMR cap which is of course smaller than points which are (theoretically) unlimited. Aside from those two extremes, though, points are pretty close to MMR after a relatively short number of games as long as you're factoring out all the obfuscators. 2. According to Not_That's models, change in MMR is not influenced by either player's uncertainty. Uncertainty appears to just define the range of opponents. 3. That's right, points generally stabilize over time as well. You'll have peaks and valleys but you're not going to be fluctuating by like thousands of points. I did read your other post but you seemed pretty adamant about the virtues of decay systems which kind of brought the discussion off the tracks, so I didn't reply to it.
Good answers. Not to bring this discussion off the tracks, but I hope I've articulated why decay systems (or kicked off the ladder systems) are better in that post.Last edit: 2012-06-16 14:03:08 |
| | Check out "Shadow of the Eternals" on Kickstarter |
|
|
| Excalibur_Z United States. June 16 2012 14:09. Posts 10287 | Profile # |
| Yeah you did explain it, but I wasn't going to start an argument over which is better because it's immaterial. I was just explaining why Blizzard did it the way they did, and why it wasn't an "incorrect" decision because there are sensible reasons behind it. |
| |
|
| DodgySmalls Canada. June 16 2012 14:35. Posts 101 | Profile # |
On June 16 2012 09:57 SKC wrote:
Show nested quote +On June 16 2012 09:44 DodgySmalls wrote:
@SKC and @BurningSera
Obviously you aren't going to do as well at some heroes and others but having a total range of 40% skill disparity is pretty high. If there is a hero you actually can't play that well it's just probable that the hero is worse than other heroes and you shouldn't pick it.
IE I do much better at antimage than I do at alchemist, that's definitely not because I play better or worse every game. In fact, I probably play to within a 1 or 2% skill margin.
So saying "mmr is stupid because there are different heroes and different heroes are stronger/weaker" is nonsensical because it's your own fault for picking a shitty hero or one that you're not good at.
You should be able to play at least a few heroes from any role NEAR the skill level you normally remain at. That narrows it down from 100ish to 10-20 heroes that you need to be good at.
I think the final piece of the puzzle is that there's no way to ladder CM which would really be necessary to have valid mmr as even random draft can create strong imbalances based on certain picks.
Noone said MMR is stupid because people's plays isn't consistent? It was simply stated that the quality of people's play varies depending on several factors, as does their MMR, so you cannot simply take MMR as a definitive way to say how well they will play that single game, for better or for worse, within a reasonable skill gap obviously. Specially when everyone's MMR is fairly close together, as a matchmaking game should be. Obviously noone always picks the best hero to win the game in AP. Well, not noone, there are a few people that pick Lycan every single game, but it's a rarity. People pick to have fun, which ussually mean not picking an awful team because getting stomped isn't fun, but it's far from insta picking Furion/Lycan/CK/Invoker/etc every single game. A lot of people play far more than 20 heroes, quite often close to the full roster avaible. Some times you simply want to play a hero you are not that good at. What about people that random? Some times you want to random, and there isn't anything wrong with that. AP pubs isn't a competitive system, you can't treat it as such. The game would be extremelly boring if people actually did their best to win every AP game. It would be a rush to pick the best heroes you would always see the same heroes every game.
1) Yes people did say that people would vary in skill based on hero selection, which is untrue because you will still be moderately close to the skill you were at the other heroes (the only exception being pro players who severely focus on certain roles).
2) Nobody should pick to win in AP because it's obnoxious and pointless, ap games literally mean nothing.
But without visible mmr and a healthy group of CM matchmakees there is no "competitive ladder" at all. Period. It's analogous to a SC2 ladder where you have to play but every game starts out with 2 base play and there are no ladder divisions, and a large range of players can be grouped together. In that system there would be no competition, because it differs so greatly from the professional play, coupled with the fact that matches are not as apparently balanced and it's difficult to gauge whether you should have won or lost a match based on games skill disparity.
I find it enjoyable to lose to a masters level player (as a diamond sc2 player) because I know that they probably SHOULD beat me, because they are almost certainly better than me. The same cannot be accomplished in dota2.
Personally I think it's nonsense and we need a competitive mode in MM alongside some sort of rating system (even something like sc2 divisions would make me much happier).
I don't know why you feel the need to disagree with everything I say SKC. You seem to think the current system is perfect, which it obviously isn't. Even if my proposed changes are worse than current by your standards you should recognize there are flaws in the current system so acting as if it's perfect is a little silly.
|
| | Please remove nyx assassin | |
|
|
| BurningSera United Kingdom. June 16 2012 18:08. Posts 5196 | Profile Blog # |
^the discussion of this thread is about 'MMR should be made visible or not'.
Nobody is saying current MMR is perfect, SKC certainly didn't. In fact, using a static MMR (range or whatever) to accurately indicate one's skill on using 100+heroes in a 5v5 game is IMPOSSIBLE.
Lets make it very clear, a good carry may have ZERO IDEA how to play a supporter/ganker. Vice versa. Don't tell me you haven't seen some shitty players (in the sense of bad attitude) only know how to farm all day and have zero map awareness or help in team fight when needed. And many of those players get to high mmr while they can get away with that playstyle because they are mainly whoring wins with their premade team. And then, when they go solo queue or didnt have the full team, the average pubs team mates didn't allow them to do so, he raged/grieved so hard and ruined people's game.
There are just too many factors and variables in a 5v5 game but we still need a MM system in the end day, what to do? That, should not be discussed in this thread. because we only care about 'MMR should be made visible or not' but not 'current MM calculation sucks or not'.
My point of view is that the cons of showing MMR/ranking outweighs its benefit because no MMR can accurately indicating one's skill anyway, why bother. If you have played HoN you should know how a lot of the players behave so hostile and endless flaming/grieving in game due to some shitty players thought they are good because they have high mmr.
What about those actual good players with high mmr you say? Well then they will be automatically placed against good players and have some good games. If you placed against dendi = you are super good. Showing MMR or not has nothing to do with that.
And no, for the record, many players play many heroes and enjoy playing them. Feel free to check my profile screenshots if you want an example, i can carry, i can support, i can roam or jungle, 800games in, i have no hero played more than 60 games in total (most of my games are in filter 'high skill').
edit:just checked with dota2 stat.de, ~850games in, my top five heroes are slardar (51games total), mirana (51), weaver (44), wr (35), sandk (23), the rest of the heroes are mostly between 10-20 (including huskar, am, void, qop, od, dk, sk, doom ).Last edit: 2012-06-16 18:17:53 |
| | Spawn more Overlords - Zerg player at heart. 2009, 820, Yaphets, YamateH <3 | |
|
|
| SKC Brazil. June 17 2012 02:23. Posts 3257 | Profile # |
On June 16 2012 14:35 DodgySmalls wrote:
Show nested quote +On June 16 2012 09:57 SKC wrote:
On June 16 2012 09:44 DodgySmalls wrote:
@SKC and @BurningSera
Obviously you aren't going to do as well at some heroes and others but having a total range of 40% skill disparity is pretty high. If there is a hero you actually can't play that well it's just probable that the hero is worse than other heroes and you shouldn't pick it.
IE I do much better at antimage than I do at alchemist, that's definitely not because I play better or worse every game. In fact, I probably play to within a 1 or 2% skill margin.
So saying "mmr is stupid because there are different heroes and different heroes are stronger/weaker" is nonsensical because it's your own fault for picking a shitty hero or one that you're not good at.
You should be able to play at least a few heroes from any role NEAR the skill level you normally remain at. That narrows it down from 100ish to 10-20 heroes that you need to be good at.
I think the final piece of the puzzle is that there's no way to ladder CM which would really be necessary to have valid mmr as even random draft can create strong imbalances based on certain picks.
Noone said MMR is stupid because people's plays isn't consistent? It was simply stated that the quality of people's play varies depending on several factors, as does their MMR, so you cannot simply take MMR as a definitive way to say how well they will play that single game, for better or for worse, within a reasonable skill gap obviously. Specially when everyone's MMR is fairly close together, as a matchmaking game should be. Obviously noone always picks the best hero to win the game in AP. Well, not noone, there are a few people that pick Lycan every single game, but it's a rarity. People pick to have fun, which ussually mean not picking an awful team because getting stomped isn't fun, but it's far from insta picking Furion/Lycan/CK/Invoker/etc every single game. A lot of people play far more than 20 heroes, quite often close to the full roster avaible. Some times you simply want to play a hero you are not that good at. What about people that random? Some times you want to random, and there isn't anything wrong with that. AP pubs isn't a competitive system, you can't treat it as such. The game would be extremelly boring if people actually did their best to win every AP game. It would be a rush to pick the best heroes you would always see the same heroes every game.
1) Yes people did say that people would vary in skill based on hero selection, which is untrue because you will still be moderately close to the skill you were at the other heroes (the only exception being pro players who severely focus on certain roles). 2) Nobody should pick to win in AP because it's obnoxious and pointless, ap games literally mean nothing. But without visible mmr and a healthy group of CM matchmakees there is no "competitive ladder" at all. Period. It's analogous to a SC2 ladder where you have to play but every game starts out with 2 base play and there are no ladder divisions, and a large range of players can be grouped together. In that system there would be no competition, because it differs so greatly from the professional play, coupled with the fact that matches are not as apparently balanced and it's difficult to gauge whether you should have won or lost a match based on games skill disparity. I find it enjoyable to lose to a masters level player (as a diamond sc2 player) because I know that they probably SHOULD beat me, because they are almost certainly better than me. The same cannot be accomplished in dota2. Personally I think it's nonsense and we need a competitive mode in MM alongside some sort of rating system (even something like sc2 divisions would make me much happier). I don't know why you feel the need to disagree with everything I say SKC. You seem to think the current system is perfect, which it obviously isn't. Even if my proposed changes are worse than current by your standards you should recognize there are flaws in the current system so acting as if it's perfect is a little silly.
I don't get it. I've said times and times again the issue isn't with the system, it's on how people don't understand it an use with wrongly.
You are saying you need a competitive mode. Well, guess what, this thread isn't about a new mode with MMR and -CM only. This thread is about adding visible MMR to the CURRENT system, which is mostly played in not competitive modes, mainly AP and SD. You don't need to add competitiveness to a system that doesn't work as a competitive system. The purpose of the current system is to create the most fun games for most people. Not the most fair, not the most balanced and definatelly not the most competitive. Of course a good amount of fairness and balance is needed for the game to be fun, but they are not the main goal, else they would remove AP, -random, SD and random teams vs arranged teams. Adding MMR is a way to add an extra competitive spirit to a mode where too much too much competitiveness can actually be harmful.
Do you like playing competitive Dota? Join an inhouse. Find CFs. Petition Valve to create another ladder with rankings and CM only. But don't preted the current matchmaking that is mostly played as AP is competitive in any way.
About the heroes, I can find you several players that have win rates far over 50% with certain heroes, sometimes easily into 75%, while having far under 50% win rates with other heroes. I can't see how it's hard to see a bad player would have a much, much easier time playing a hero like Huskar or Lich than Meepo or Chen.Last edit: 2012-06-17 02:28:20 |
|
|
| ineffablepwnage June 18 2012 10:07. Posts 21 | Profile # |
| I would prefer not to see MMR. What pisses me off the most in pub games is when people start talking trash about their own teammates. I think the negatives outweigh the positives when people are bitching about how horrible their team is before the game even starts. Also, MMR means way less in a game like dota where the roles are so different and some heroes require far more skill to play. I've played with people in HoN with 1900+ MMR who played absolutely horribly, and people with 1400 MMR win the game for my team (usually as a support). That said, I wouldn't be heartbroken if they put it in. I would quit dota if they put in public K:D ratios. I've seen way too many people quit playing a game that wasn't anywhere near lost because they wanted to save their glorious K:D ratio. |
|
|
| Daozzt United States. June 19 2012 01:28. Posts 587 | Profile # |
| Elitism issues aside, visible mmr does work even if it's only decided by w/l ratio. In my experiences playing HoN, I've never seen a 1600 player that was better than an 1800 player, vice versa. |
|
|
| rob.au June 19 2012 02:08. Posts 369 | Profile # |
| I think this kind of discussion validates the LoL system of having seperate queues for people who want to play with visable MMR. |
|
|
| DodgySmalls Canada. June 19 2012 03:03. Posts 101 | Profile # |
On June 19 2012 02:08 rob.au wrote:
I think this kind of discussion validates the LoL system of having seperate queues for people who want to play with visable MMR.
I agree, as I said before I think the best method is to have separate queues for the serious and the casual. It just makes more sense for everyone. |
| | Please remove nyx assassin | |
|
|
| semantics June 20 2012 01:47. Posts 8556 | Profile Blog # |
Like in hon BADS will be BAD. People in hon use excuses and blame stats for their bad play, i carried myself mostly up and out to 1800 mmr, i have friends that carried themselves mostly though solo play up to 1900mmr. MMR is a measure of wins vs losses in relation to other people's wins vs losses. In essence it counts the only stat that matter if you're a consistent winner or a loser.
To that extent it should be used to gauge other players general skill, any excuse such as i don't play support often etc is irrelevant as you remove all other factors and only count one thing if you can win.
That being said i've always taken multiplayer stats as such. It should be an option for the user to share the stats, any stats to the public, if that player doesn't want to share his stats that is fine. But elitism and being a baby about sucking at a game you just picked up is not an excuse to disallow viewing your own personal stats. That should always be available to yourself, the option is if you want to share that.
Hon's community stat obsession is just a funnel for whiners, you can't blame the stats for causing whiners, there will always be whiners in any community all stats did was provide one of many outlets. Eliminating the viewing of stats wont stop people from being dicks/cry babies and rage quiters all it does is eliminates them from calling you bad and referencing your shitty stats, instead they will find other ways to pick at you. IE what hero you choose, your items, your kdr in game.
You can't stop people from being dicks online the best thing you can do is mute/ignore themLast edit: 2012-06-20 01:48:15 |
|
|
| Absurd Bunny June 20 2012 02:23. Posts 140 | Profile # |
If someone with a decent mmr, at 1600, can beat other people at 1600 and lower, but gets ranked with players 1900+, gets beat, and he gets a lower mmr, it makes him look like a lower player, but he can still beat the other people at his level. If mmr gets shown, all the players that have a higher mmr than him will think that they can't trust him because he's not good.
There are cases where mmr should be shown and used, and there are other places where it can't. Probably the good people wouldn't mind it being shown, and the new/bad players don't want it to be shown. |
|
|
| MrTortoise June 20 2012 02:59. Posts 965 | Profile # |
All i know about MMR is that it turns the borderline rational people into crybaby dickheads who seem to think that losing a game means they are playing with noobs + gives them the right to rage at the support who just let them farm non stop for 25 mins.
What I really love is playing with people at my own lofty skill level and them thinking they are better but have no idea how retarded they look to me. Or alternativley have the awareness to see others suck but not the self awareness to deduce that they therefor must look as hilarious to everyone else.
The big problem MMR had in hon was that noobs start at 1500 and then drop rapidly. So if you are climbing up from noobdom the level of the game actually drops when you get close to 1500 again. At 900 psr as it was called back then every game someone bought courier and people would try to ward as you have got into the guys that actually want to play the game but as they read guides and fight for support chars they cant carry hard enough with str heroes to punish the 1200-1400 games when you get some total noobs in the game.
Its one less thing to qq about / be a dick with and that is great as it lets me play and enjoy the game.
In hon you used to get situations where people would kick all people who were +-100 psr from themselves ... yet its easy to win or lose that much in 1 night ... and they dont then realise that the variance is gigantic (or deduce that a 900 psr with 400 games is probably better than the 1400 with 20 games). The problem is that they dont understand how snowballing effects the game and assume someone who rapes in 1 game must do so in all games ....
Last edit: 2012-06-20 03:08:59 |
|
|
| Bigjuicy Brazil. June 20 2012 17:47. Posts 3 | Profile # |
| I think the mmr system made hon players the epitome of ragers. People care way tooo much about a single number that they just bitch and moan the entire game if shit doesn't go their way. |
|
|
| MrTortoise June 21 2012 02:44. Posts 965 | Profile # |
| [deleted - basically a double post ] Last edit: 2012-06-21 02:46:33 |
|
|
| RicksonXCIV June 21 2012 03:29. Posts 10 | Profile # |
Dont really know if this has already been posted/suggested cuz i didnt read the whole thread.
Why not make it like SC2 like the ladder system put people in league's? |
|
|
| DodgySmalls Canada. June 21 2012 03:36. Posts 101 | Profile # |
On June 21 2012 03:29 Rickson )) wrote:Dont really know if this has already been posted/suggested cuz i didnt read the whole thread. Why not make it like SC2 like the ladder system put people in league's?
I think this is an excellent idea, and yes a few people did sort of suggest it.  |
| | Please remove nyx assassin | |
|
|
| Prev 1 2 3 4 5 6 7 8 9 10 Next All | |
|
|
|
| |
|
Sidebar Settings...
|
Store
