It doesn't mean much if you look at the big picture. When someone inflates their elo, they'll be put in the team with less skilled players and eventually stomped back to the level they are actually at. You can't ride other people's elo infinitely because shuffle will put less and less such people to your team.
Except that in the big picture, people change team...
They usually follow the big heads (or at least the ones they know).
You will always find someone like that in any game.
If people change teams and the average skill becomes uneven, the elo gain/loss becomes much lower, down to a single digit. You only gain any significant elo when the round was fair, i.e. the averages were close. Since these numbers may change during the round, it's also accounted for. The difference is more important at the beginning of the round than it is at the end. For example, if you started the game with 2450/2430 averages and closer to the end it became 2700/2300, it wouldn't matter much for the amount of elo the teams gain and lose. The opposite is also true.
You only gain any significant elo when the round was fair, i.e. the averages were close.
To my knowledge, this is incorrect. Unfair teams with significant differences such as 100+ in average team skill have greater changes in skill value. The more unexpected the win, the bigger the adjustment is to each individual players skill value. Hiveskill / Hivescore is just a number that shows how likely it is whether You can make your team win in your average environment, and a very unexpected win can dramatically change that probability. I managed to win a game not that long ago where we were down 1000 hive skill, and my skill value went up something like 100 points. The adagrad stochastic gradient for new players can make the amount gained or lost for each player impossible to predict for a human. I don't see anything else wrong with your comment though.
In addition to your comment, here is the expected win or loss probability of any given team by how big the difference in average team skill is. As you can see, even a 100 difference in average team skill gives one team a significant advantage.
Yes, the elo changes even more if the team with less average skill wins. It's just not that common so I forgot about that possibility.
I believe this formula is used in NS2+ or Shine, I saw the winning probabilities once on TTO. Can't tell if they're accurate enough or maybe that was another formula. Anyway, these probabilities were turned off after several rounds to not discourage players. And yes, 100 difference usually guarantees the outcome. Too bad it's a common situation when the round starts with even teams, then someone leaves and some other player with elo greatly above or below average joins instead. The balance immediately breaks and we need to shuffle again which leads to more people leaving because the game doesn't start and they become frustrated. I wish there was an enforced rule to prevent that.
Did you ever see the graphs I made showing various rookies or veterans hive skill values overtime. I mention this because it shows how much each players hive skill oscillates.
Here is a collection of 20 of the graphs. These are in no way a random selection of players. https://imgur.com/a/R2iHz
Ironhorse also gave a few details that explain why it is really hard to inflate your elo by switching teams, or joining the more skilled players.
@Ourukai
* If one team smashes another, but the teams were stacked to make that outcome almost certain, nobody’s skill level changes at the end of the round. Only unexpected victories change skill levels.
* The effect on your score is determined by the time you spent playing the round rather than your points, with the beginning of the game weighted much higher than the end. This means that it doesn’t harm you to join a game that is already lost, because you’ll have played for a very small fraction of the weighted time spent in the game.
* If the two teams have a different number of players, this is compensated for automatically by normalizing by the total time across both teams rather than just within a team, and the model predicts that the team with more players will win.
Yes, many times in fact. Those graphs never tell if the team chosen was the one who actually won. It shows only the climbing after a certain number of round. Are they still active ? 2016...
As for the citation. The skill factor changes even if you're on the winning team and you're supposed to win. 100% sure. Even if it is a small number it is a number. And as it would me less risky... people just do it.
Plus an unexpected victory can take many form in NS2 with a non zero load of luck. That is why some people complained the most about Gorge tunnel. Unfortunately they were listen to.
AND if it is unexpected and most of the time accompanied by luck; how can you associate this with a skill factor ?
If people change teams and the average skill becomes uneven, the elo gain/loss becomes much lower, down to a single digit. You only gain any significant elo when the round was fair, i.e. the averages were close. Since these numbers may change during the round, it's also accounted for. The difference is more important at the beginning of the round than it is at the end. For example, if you started the game with 2450/2430 averages and closer to the end it became 2700/2300, it wouldn't matter much for the amount of elo the teams gain and lose. The opposite is also true.
Let me explain it better. the game doesn't start if people see there is a big ELO gap.
Follow for a moment. The "stickers" are the one who attach themselves on the tail of some big head players. Others are just regular players or rookies that just don't care about it (for the sake of simplicity).
After a shuffle (not an official thing as it was adequately put) some stickers change team. The gap is bigger (not every time). The game will not start as some player noticed it. But the sticker is glued.
At this point 1 on 2 shuffle will be voted again. Or some other players will move naturally. The usual scenario is the shuffle vote. The stickers will change team as many time as they need. It going like that up to a point the teams get balanced according to ELO. With one specific trait. The sticker stacked with their respective "heads". In itself it is stacking and they get what they want. And as i stated just before. You still earn point doing that. You can argue it's gonna be slower. It doesn't matter for them.
Bad behavior 1 ; the rest 0.
Oh. And what about this: 1 player plays 1/4 of the game in one team, then change team. He's got a lot of intel' to make sure it will 'pinch' somewhere it hurts. Then he comes back (for the last quarter) in the team he left... what happens in terms of ELO ? Hint: the score of this particular individual is well above 3K.
This ELO system has always encouraged the bad behaviors .
Stacking, need i say more...
Smurfing for individual war (try to make somebody you don't like loose) or simply messing up on a server
Fake games. Training rookie in exchange for points as they loose the game.
The ELO can go up to infinite which is totally in opposition with human skills or capabilities. It's like a wallet.
A good load of things i anticipated in other threads (Hive v1&2) years ago did happened. To my total despair, unfortunately.
After a shuffle (not an official thing as it was adequately put) some stickers change team. The gap is bigger (not every time). The game will not start as some player noticed it. But the sticker is glued.
Not true and not always. Shuffle doesn't care about stickers and can move them and the top players in the opposite teams just as they should be. If the teams are full, you simply can't switch to the opposite team if you want as it's full. I usually play on TTO and I haven't seen such behavior there. I mean, I also like to ride some top players elo if possible because they usually always win so what's the point of losing because of just one player? It doesn't reflect my skill anyway, I haven't become worse because someone joined the other team. But I don't stick all the time because it's not possible as already said and just plain tiresome.
Smurfs, yes, but they're easily detected and banned because no one likes to play against smurfs.
About the infinite elo, well, try it? I wonder how far you could go like that. Theory and gut feeling is one thing but in practice it's much harder if not impossible to do. The system is made specifically to predict the round outcomes and correct itself if it's wrong. If you can exploit some property to gain as much elo as you want, why not to try and become the top Sanji out there? Well, you probably can do that with 11 fake accounts where you constantly beat them (I believe only 6x6 rounds and up are ranked). Those accounts should also have elo close to your own or else it would take ages.
This ELO system has always encouraged the bad behaviors .
Stacking, need i say more...
Smurfing for individual war (try to make somebody you don't like loose) or simply messing up on a server
Fake games. Training rookie in exchange for points as they loose the game.
The ELO can go up to infinite which is totally in opposition with human skills or capabilities. It's like a wallet.
Stacking - not encouraged, since relative skill diff is taken into account
Smurfing for individual war - yeah, that's a possibility, but how seriously? I've seen like 5 smurfs in NS2 in ~3 years. Statistically this can be ignored.
Fake games - yet again, if the relative skill diff is large enough (should be in your scenario), resulting skill change is dampened drastically
The ELO can go up to infinite which is totally in opposition with human skills or capabilities. - How come? How do you know what is the absolute pinnacle of competence? Even in real life in their specific trade, there are god-like people out there (compared to others). Oh ok, sure, "Infinite" is impossible, but from a modeling perspective, it's good enough.
I usually play on TTO and I haven't seen such behavior there. I mean, I also like to ride some top players elo if possible because they usually always win so what's the point of losing because of just one player? It doesn't reflect my skill anyway, I haven't become worse because someone joined the other team. But I don't stick all the time because it's not possible as already said and just plain tiresome.
Smurfs, yes, but they're easily detected and banned because no one likes to play against smurfs.
Admins aren't what they're used to be. It depends also on the server rules. Some are really forgiving. And banning one smurf sometimes just create another one. You gotta get in the head of those people. They're really willing to go to some extreme extents.
About the infinite elo, well, try it? I wonder how far you could go like that. Theory and gut feeling is one thing but in practice it's much harder if not impossible to do. The system is made specifically to predict the round outcomes and correct itself if it's wrong. If you can exploit some property to gain as much elo as you want, why not to try and become the top Sanji out there? Well, you probably can do that with 11 fake accounts where you constantly beat them (I believe only 6x6 rounds and up are ranked). Those accounts should also have elo close to your own or else it would take ages.
What i describe is: At some point you don't learn anymore (map, positionning..), You can 't be faster anymore, you can't kill more anymore (not enough time, targets...). Humans have limits, and this game has limits too. The logical question is : What limit for the skill factor ? There is none that i know of.
It gives the present situation in which you have people with 4K and rookies with almost none. Not only it'll be more difficult for the shuffle to sort things out, but those ELO can't be accurate. Not one bit. I mean is a player at 5k better than one at 3K after 500 games ?? Especially if they never played on the same server. I'm not sure.
In fact i've seen better 1,5k players than those 3k, just because they were unfortunate to loose the wrong games. Or simply because they use to play in other "circle of friends". A pity.
They go back to RR and then wait for an opportunity. Too easy.
And a spectator takes their slot immediately. Not that easy. I don't know what servers you play at, I heard some of them has a really toxic community (mostly NA?). Maybe that's the main source of all the negativity.
Admins aren't what they're used to be. It depends also on the server rules. Some are really forgiving. And banning one smurf sometimes just create another one. You gotta get in the head of those people. They're really willing to go to some extreme extents.
Then play on the servers with better rules and admins — easy!
What i describe is: At some point you don't learn anymore (map, positionning..), You can 't be faster anymore, you can't kill more anymore (not enough time, targets...). Humans have limits, and this game has limits too. The logical question is : What limit for the skill factor ? There is none that i know of.
Why should there be a limit? If you beat some players, you have a better skill. If you beat the best, your elo is higher than theirs. When you start to lose against people with lower elo, your elo goes down and theirs goes up until it reaches the balance point.
It gives the present situation in which you have people with 4K and rookies with almost none. Not only it'll be more difficult for the shuffle to sort things out, but those ELO can't be accurate. Not one bit. I mean is a player at 5k better than one at 3K after 500 games ?? Especially if they never played on the same server. I'm not sure.
I'm not sure either, they can play against each other and settle that, no? Is your point that there's not many average players of rank 5 on the servers you play? If that's it then yes, you can't balance such extreme teams with sanjis and rookies only. But then again, in what game is that possible at all?
In fact i've seen better 1,5k players than those 3k, just because they were unfortunate to loose the wrong games. Or simply because they use to play in other "circle of friends". A pity.
Elo is not a number to brag about. It's a number to make you win and lose with approximately 50% chance. If you win too much that means someone else loses too much and elo compensates for that putting better players in the opposite team and the weaker ones in your own. It shouldn't concern you if someone has 3k or 5k if they don't lead to purely one-sided rounds. Unfortunately, sometimes they do if there's no one of comparable skill on the server. But there's not much anyone can do about it (of course, if that player is not a cheater with aimbot or wallhack). If someone's good, he can destroy the opposite team single-handed and elo reflects that.
Sure, if you only play with several other people your elo can be wrong in the global scale. It's like that everywhere in the world, not just in NS2 or games in general. You can be top in your class in school but then you go to college and suddenly you're not that great anymore or may even fail exams and drop-out. Elo shows how you perform compared to others, so if you want a more objective elo, play with as many players as you can or else you'll be confined to your bubble and be the best among your friends but worst among the pub players.
The problem could be solved by a better shuffle algorithm. It might account standard deviation or other statistical metrics so that we don't end up with teams like 5k+n*1k vs n*2k or so, i.e. one has a sanji + bunch of rookies and the other one has average players. You can expect much better teamplay from a team of averages than from the former. I played such rounds recently, my team had 1 sanji (7th rank) and others were rank 5 at best, me including; the other team had 3 or 5 players of rank 6th and others were 5-4, but mostly lower than 5. Of course we lost, because rank 6 can teamplay much better. And the average skill score was pretty close that time.
Yet again, it doesn't have anything to do with elo itself but with the shuffle algorithm.
And a spectator takes their slot immediately. Not that easy. I don't know what servers you play at, I heard some of them has a really toxic community (mostly NA?). Maybe that's the main source of all the negativity.
Basically spectators go spectate with no intent on playing in the first place. If that would be the case for some it would far worse than what i described before.
Then play on the servers with better rules and admins — easy!
You're right there so many servers online. If only they were crowded... The last ~200 players left won't save the day on that matter is it ? Now if you're asking them to seed a server. You've gone too far mister!
Why should there be a limit? If you beat some players, you have a better skill. If you beat the best, your elo is higher than theirs. When you start to lose against people with lower elo, your elo goes down and theirs goes up until it reaches the balance point.
As Mofo1 mentioned before. If you're not good at playing marine but excellent at alien. you will loose or win 'skill' (as you name it). You see were it goes right? Thre is no balance point in that regard. Plus in that case you have a mean to manipulate your own skill factor (and others as collateral damage).
Worse. Imagine i'm a super Fade and decide to go gorge then the team loose. If you say i made a bad decision this would mean i have only one way to play NS so my 'skill' can be accurate. Super fade vs bad commander. Super gorge vs Lerk. Super marine / bad JP.
It's kind of anti-NS-versatility don't you think ?
That's were it's all wrong. If it was a ranking system it could be left as it is. Even if the team victory is the determinant (see the many bad scenario described a while ago in which you suffer from idiots doing stupid things). But it's not. It's used for shuffling games.
Skill and skills are quite different especially at NS. Even if it would be more difficult (i consider it to be mainly an bad excuse for laziness) the way to evaluate a NS player cannot be a single number.
I'm not sure either, they can play against each other and settle that, no? Is your point that there's not many average players of rank 5 on the servers you play? If that's it then yes, you can't balance such extreme teams with sanjis and rookies only. But then again, in what game is that possible at all?
This shed a light on the team A,B & C,D issue. You can't have a decent idea of the strengths until the system corrects the value. That means that everyone have to play over and over with everyone. It's kind of exponential as son as you add more people... therefore impossible. Unfortunately we're human. Need to sleep, eat and such.
It's a number to make you win and lose with approximately 50% chance. If you win too much that means someone else loses too much and elo compensates for that putting better players in the opposite team and the weaker ones in your own. It shouldn't concern you if someone has 3k or 5k if they don't lead to purely one-sided rounds. Unfortunately, sometimes they do if there's no one of comparable skill on the server. But there's not much anyone can do about it (of course, if that player is not a cheater with aimbot or wallhack). If someone's good, he can destroy the opposite team single-handed and elo reflects that.
Sure, if you only play with several other people your elo can be wrong in the global scale. It's like that everywhere in the world, not just in NS2 or games in general. You can be top in your class in school but then you go to college and suddenly you're not that great anymore or may even fail exams and drop-out. Elo shows how you perform compared to others, so if you want a more objective elo, play with as many players as you can or else you'll be confined to your bubble and be the best among your friends but worst among the pub players.
The problem could be solved by a better shuffle algorithm. It might account standard deviation or other statistical metrics so that we don't end up with teams like 5k+n*1k vs n*2k or so, i.e. one has a sanji + bunch of rookies and the other one has average players. You can expect much better teamplay from a team of averages than from the former. I played such rounds recently, my team had 1 sanji (7th rank) and others were rank 5 at best, me including; the other team had 3 or 5 players of rank 6th and others were 5-4, but mostly lower than 5. Of course we lost, because rank 6 can teamplay much better. And the average skill score was pretty close that time.
Yet again, it doesn't have anything to do with elo itself but with the shuffle algorithm.
The school example isn't a fit for the matter at hand. In this example everything changes. In NS2 we have the same game all over the world. A decent system could make USA A USA B play AUS A or AUS B with a prediction that would not be perfect but better than what i could witness.
Waiting for a system to correct itself means playing bad games after bad games waiting until that holy day... that never comes actually.
How can a shuffle algorithm can see if you're better at marine than alien with one single number ?
If a proper "set" of values is available there won't be long until it is used in a shuffle algorithm the proper way. That would be a better shuffle algorithm only because a better skill measuring system is behind.
Hey-hey-hey, you're now mixing in a completely different problem, namely the per-side elo separation. That's a known issue and it's meant to be solved in the upcoming 2.0 release (at least according to the Trello). I wholeheartedly agree that a single number for two different skill sets is not a right thing. As for the life forms, I have no solution for that, do you? If you shine as a fade but intentionally go gorge and therefore make your team weaker, that just means you're playing as a lower-skill player so if you lose it's partially your fault and your elo is corrected then. But it doesn't matter in the long run if you mostly play as a fade, you'll win next time. Basically, if you play as a life form you're not a master at you drag the team down so your high elo doesn't match your actual skill and it goes down until you master that life form. The game assists you in that putting you against weaker players as you lose elo until you start to win. I don't see a problem.
And don't forget that you're not playing solo, you have a team. Even if someone for some reason doesn't play good enough just today others can compensate for that (a little). I don't believe that at one side everyone chose their best weapons/lifeforms and on the other side each player somehow plays as a rookie because of reasons. I guess being a versatile player also counts towards your skill. If you lost your super-fade, went gorge and made a rush tunnel that brought you victory, it's as good as swiping half the team with a fade. If you're a one-trick dog, well, you better do your trick as best as you can all the time, otherwise your defeat is well deserved.
!!!! sic... really ??? We have some people who do the opposite.
I know, and it's not an NS2-only issue. For smurfs, it happens in all ranked games because computers can't tell if a new player really a rookie or a smurf. Even people can't tell sometimes. Otherwise, even if people brag about their skill I don't see a problem. If they don't cheat and play according to their elo, it's fine. My point was that you personally shouldn't be so conscious about that number. It doesn't mean you're worse as a person than somebody else, just that you're not as good in this particular game. And I'm afraid most of us are already stuck at the levels we are now because of physical limitations and age. That's fine, the goal of the game is to have fun. Not like you can sell those elo points for dollars or anything.
Waiting for a system to correct itself means playing bad games after bad games waiting until that holy day... that never comes actually.
That's quite pessimistic and I don't see where this comes from. Are you constantly losing, like 20 times in a row? Or you're angry that you're losing at all? Do you expect to win all the time? That's not how the competitive games work.
If it was a ranking system it could be left as it is. Even if the team victory is the determinant (see the many bad scenario described a while ago in which you suffer from idiots doing stupid things). But it's not. It's used for shuffling games.
What is the point of the ranking system if not putting the players of similar skill against each other? It's exactly like that in all sports and e-sports. The rank itself is useless if it's not used for balance. If you don't do that you might as well drop 12 sanjis vs 12 rookies and watch the massacre. Is it fun? Doubt so.
Let's start anew. What's your main problem with elo, ranking, shuffle etc.? Does the balance seem wrong to you? Do you lose or win too much? Give your observatory profile URL so that we could talk more productively. Here's mine. As you might see, my skill has barely changed since the last year.
Hey-hey-hey, you're now mixing in a completely different problem, namely the per-side elo separation. That's a known issue and it's meant to be solved in the upcoming 2.0 release (at least according to the Trello). I wholeheartedly agree that a single number for two different skill sets is not a right thing. As for the life forms, I have no solution for that, do you?
The approach cannot be a single number or a side number. It has to be smarter than that.
Basically you have several choices. Some think about K/D mixed with X and Y and a weight on top, etc.
I do prefer profiling. We know the game can recognize one thing or two. Like the NS2 success in Steam. There is quite many different things in there. Why not using it ? With less time than Hive we will be able to see if a player prefer fade / gorge, shotgun/LMG, to build and protect etc. The good thing is it can answer the need of a team. As a commander from time to time i really don't care having 10 Rambos in my team if no one will build that damn RT.
In the end a shuffle system getting those values can see if a person has chances to go gorge instead of fade (as an example). So the system can put 1 fade with 0% chance to go gorge, and one other fade with 50% chance to go gorge instead of that last fade alone in the team. Not easy to explain but i'm sure you get the point.
Or exploiting other things. We know a rookie is more likely to go gorge often. But with time he will try something new. Say about 50 hours. Chances are he will probably weaken the team.
Of course the shuffle system will need to know how good you are in each category but in the end it will have a truly better set of values to "guess" what would be a good mix. Or at least avoid the worst. Things that occur more than once per night when i play.
So people can have games with what's needed for a decent start (builder, hunters) and they can freely try something new. Up to them to organize it properly.
This wont avoid those who try to trick the system, or to be a jerk. But it'll be far better than the ugliness i could witness.
If you shine as a fade but intentionally go gorge and therefore make your team weaker, that just means you're playing as a lower-skill player so if you lose it's partially your fault and your elo is corrected then. But it doesn't matter in the long run if you mostly play as a fade, you'll win next time. Basically, if you play as a life form you're not a master at you drag the team down so your high elo doesn't match your actual skill and it goes down until you master that life form. The game assists you in that putting you against weaker players as you lose elo until you start to win. I don't see a problem.
Not encouraging people to try something new is it ? No need to go further. It's simply plain wrong.
And don't forget that you're not playing solo, you have a team. Even if someone for some reason doesn't play good enough just today others can compensate for that (a little). I don't believe that at one side everyone chose their best weapons/lifeforms and on the other side each player somehow plays as a rookie because of reasons. I guess being a versatile player also counts towards your skill. If you lost your super-fade, went gorge and made a rush tunnel that brought you victory, it's as good as swiping half the team with a fade. If you're a one-trick dog, well, you better do your trick as best as you can all the time, otherwise your defeat is well deserved.
So maybe it's not only about the life form? Maybe the skill lies somewhere else? If that so, you're getting close.
I know, and it's not an NS2-only issue. For smurfs, it happens in all ranked games because computers can't tell if a new player really a rookie or a smurf. Even people can't tell sometimes. Otherwise, even if people brag about their skill I don't see a problem. If they don't cheat and play according to their elo, it's fine. My point was that you personally shouldn't be so conscious about that number. It doesn't mean you're worse as a person than somebody else, just that you're not as good in this particular game. And I'm afraid most of us are already stuck at the levels we are now because of physical limitations and age. That's fine, the goal of the game is to have fun. Not like you can sell those elo points for dollars or anything.
I know 1 or 2 who would if they could.
I'm not cautious about that number. Only against the logic behind that brought every single bad behavior anybody can witness. Sooo annoying to be right.
That's quite pessimistic and I don't see where this comes from. Are you constantly losing, like 20 times in a row? Or you're angry that you're losing at all? Do you expect to win all the time? That's not how the competitive games work.
To me, a bad game in the shuffle context is a game that shouldn't happen if the proper data was there to decide how to balance the game. I've seen that many time. ELO is supposed to be fine in both teams but in less than a couple of minutes, the game is already over and many player on both teams know that.
What is the point of the ranking system if not putting the players of similar skill against each other? It's exactly like that in all sports and e-sports. The rank itself is useless if it's not used for balance. If you don't do that you might as well drop 12 sanjis vs 12 rookies and watch the massacre. Is it fun? Doubt so.
Let's start anew. What's your main problem with elo, ranking, shuffle etc.? Does the balance seem wrong to you? Do you lose or win too much? Give your observatory profile URL so that we could talk more productively. Here's mine. As you might see, my skill has barely changed since the last year.
What's wrong ? Look
What's the reason of the changes (with a big magnitude somehow) ? I tell you; i didn't changed a bit. That's the funny thing about it. I'm quite mr constant. But the skill factor moved up/down. With not much activity indeed. It's more like i ended up in the wrong team sometimes.
To sum it up; 50% growth in 1 year 9 months and with not much activity. My aim didn't changed, i still rush like "for the horde" (because i play in pub) for the fun of it. It's not like a third hand grew on me and i can now type faster (or hit the map key ten times a sec). I'm no better, other regulars didn't get better (they do the same things).
That's a big issue. The ELO is used in place of skill (in shuffle). But the ELO is determined using the team win/loose. I'm not blaming shuffle. The input data is the input data.... But do admit that a team result based ELO cannot be seriously taken as a personal skill factor/score/whatever. Because simply it depends on the team result first. I mean i hope you see the dichotomy. That not a crack in the logic of trying to evaluate players individually. More like a canyon. And i do know it's not the only one.
Not encouraging people to try something new is it ? No need to go further. It's simply plain wrong.
The game doesn't encourage or punishes. It measures your performance, that's it. Let's pretend the game would encourage you to try the life forms you're inexperienced at. That would mean it should put weaker players against you so you're not at disadvantage. What if you "cheat" the game and instead go the life form you've aced? That would be totally unfair for marines. To make it fair you should setup some sort of a contract with the game that prevents you from choosing another life form so you could only go the one you've chosen before the round. This system is easily exploitable and not reliable. It's up to the player whether they want to become a pro in a single life form or master each one a bit. For rookies it's a non-issue anyway, they don't have much elo to gain or lose and the pro players should already feel the game pace. For example, when you see your team pushes marines with ease, you could go a life form you haven't tried yet, like fade or lerk and do some practice. If the game is hot and hard it's better to stick to your best life form.
Your "profiling" system is also easily tricked. It basically makes predictions that are based on nothing (maybe some prior statistics that might not be very reliable for a particular round) and that makes it vulnerable. Because those predictions are made before the round starts. The game thinks that PROBABLY this player will do that and that player will do this so let's pretend that player will be stronger and this one weaker and let's try to build a team out of what we have according to some guy's concept of a "dream team". But if these players don't follow that prediction the model goes haywire, I even suppose it will be much worse than the current system. I personally don't believe this detailed statistics would do more good than harm. It's not totally bad, it could work for dota-like match making, i.e. when the master server takes players from a huge pool of waiting players and tries to make two good teams out of that. But we simply don't have that many players for this method to be reliable. We usually only have 20-24 players and should work with that and this detailed stats simply wouldn't work.
To me, a bad game in the shuffle context is a game that shouldn't happen if the proper data was there to decide how to balance the game. I've seen that many time. ELO is supposed to be fine in both teams but in less than a couple of minutes, the game is already over and many player on both teams know that.
This happens because of lack of communication or overall low skill sometimes. Like, 4 players going gorge, not making tunnels, tanking marines and quickly dying etc. Sometimes it's a purely "marines day" or "aliens day" until the server population changes drastically because a significant share of the players are only good on one side. I hope this will change with separate marines/aliens elo though.
What's wrong ? Look
What's the reason of the changes (with a big magnitude somehow) ? I tell you; i didn't changed a bit. That's the funny thing about it. I'm quite mr constant. But the skill factor moved up/down. With not much activity indeed. It's more like i ended up in the wrong team sometimes.
Now we're talking. You actually don't play much and that's the root of the problem. Six days in a whole month? If I'm counting right, you only have like 120 data points since Jan 2017 (not rounds as this is updated once a day, not after each round), that's a very low number and your skill might not be calculated reliably yet. More importantly, put aside the actual elo numbers how does that affect your game experience? Why are you concerned at all? Let it fluctuate up and down, who cares?
I'm not saying the current elo and shuffling system is perfect but it's the best we have. I once thought that accounting for K/D ratio should be a must but then gave it a secod thought and realized it's not that important. You can kill a couple of marines and bite 1.2 RT/min as a skulk and thus bring your team victory. Or you can kill a few skulks but 2-3 fades and several lerks destroying the entire aliens life form pool. The game tries to highlight that with score points but I'm not convinced this should affect the elo. Why? Because that would shift priorities. Instead of teamwork to actually win people would prioritize frags and points (because that would save them most of elo in case of losing or bring them more if they win). It has been discussed already and the consensus was that only the fact of winning/losing should affect elo just like in chess, dota and other games.
TL;DR: "Don't worry about hiveskill too much. It's a sophisticated system and probably already takes everything into account that you are worried about."
Sometimes it's a purely "marines day" or "aliens day" until the server population changes drastically because a significant share of the players are only good on one side.
Finally someone else that agrees with me that some people are better at one side over the other!
I hope this will change with separate marines/aliens elo though.
That would be nice, if this new elo can accurately predict whether a person is better at one team or another... Which seems unlikely since it will still use a single number.
I'm still waiting on a "layman's" explanation (without the mathematical mumbo jumbo) of how it will know when a player with say 1700 hive score is really a 2000+ on one side and a -500 on the other...
Finally someone else that agrees with me that some people are better at one side over the other!
That's pretty obvious, no? It's even implied in the client config menu in Shine where you can set the shuffle preference (if you have a preference, you're probably better on that side). I really don't see why this can't be implemented, there's no fundamental difference between the current system and that with two separate elos. Shuffling OTOH might prove to be more difficult. I don't know how it works now but this problem I believe is NP-complete so it can be quickly verified (compute and compare the average skills) but could take a while to solve, up to bruteforcing all the permutations. And even that shouldn't be a lot for any computer newer than 80486. Just for the sake of it, let's calculate how many permutations there are for 24 players.
Combinatorics to the rescue! We want to know how many unordered sets of 12 players can be formed out of 24 players (because the rest goes to the opposite team). The formula is pretty simple, n! / (m!(n - m)!) where ! is factorial, i.e. n! = 1 * 2 *3 ... * n. For our example n is 24 and m is 12 so: factorial(24) / (factorial(12) * factorial(24 - 12)) = 2704156. Just 2.7 million iterations. Now, if we don't really care about the team preferences we can safely divide this by two because half of the sets are the opposite (for example, if players from 1 to 12 go to marines and 13 to 24 to aliens, the opposite set would be 1 to 12 — aliens, 13 to 24 — marines which is the same in terms of balance using the current one-number elo). So we have 1.35 million variants which is quite a small number and can be calculated in under a second I guess. This straight forward algorithm doesn't scale well though, for 13 out of 26 players (just 2 players more) the permutations number becomes 10 millions and for Wooza's 25 out of 50 it's an enormous 126410606437752. But for 24 players it's indeed possible to find a perfect balance in reasonable time.
The mirror hack (dividing by two) wouldn't work if we have two separate elo numbers, of course.
FrozenNew York, NYJoin Date: 2010-07-02Member: 72228Members, Constellation
Just wondering if my college has worn off, but wouldn't a balance be done by randomizing, checking the difference, then swapping semi-random players with smaller and smaller differences in their skill until it's close enough?
And that would just take like, idk, probably less than 100 iterations to converge? And if there was 2 hive skills you would just be checking the appropriate skills when trading players?
The algorithm is described here in the comment. Not a full search but must be a good enough method. It also considers the team preference as "happiness" so that if you're constantly put into the team opposite to your preference your shuffle weight for your desired team will grow. If there are more unhappy players than happy the teams are simply swapped. So your team preference works but in a strange and interesting way. This will also break if/when separate elo is implemented.
The game doesn't encourage or punishes. It measures your performance, that's it. Let's pretend the game would encourage you to try the life forms you're inexperienced at. That would mean it should put weaker players against you so you're not at disadvantage. What if you "cheat" the game and instead go the life form you've aced? That would be totally unfair for marines. To make it fair you should setup some sort of a contract with the game that prevents you from choosing another life form so you could only go the one you've chosen before the round. This system is easily exploitable and not reliable. It's up to the player whether they want to become a pro in a single life form or master each one a bit. For rookies it's a non-issue anyway, they don't have much elo to gain or lose and the pro players should already feel the game pace. For example, when you see your team pushes marines with ease, you could go a life form you haven't tried yet, like fade or lerk and do some practice. If the game is hot and hard it's better to stick to your best life form.
If as said the system measures the performance and correct values it means it will never be close to accurate.
In order for new players and veterans to fully enjoy the game diversity they must be given a chance in those areas without harm. On a 24 slot public server it's quite manageable. Far more than a 5K who decide to be the jerk of the day and ruin every game. Not to mention that as a Strategy game every unit has a power that can be devastating if used properly.
Otherwise it looks like a CS in space for them. Just like another toilet paper, they will throw it without discovering anything about this game (for the new player), and without enjoying some other stuff for regulars. Hence player retention.
I'm not for imposing a contract on players. Just observing and accumulating stats of their actions. On the technical layer: Storage, calculation and all, aren't a problem at all.
Your "profiling" system is also easily tricked. It basically makes predictions that are based on nothing (maybe some prior statistics that might not be very reliable for a particular round) and that makes it vulnerable. Because those predictions are made before the round starts. The game thinks that PROBABLY this player will do that and that player will do this so let's pretend that player will be stronger and this one weaker and let's try to build a team out of what we have according to some guy's concept of a "dream team". But if these players don't follow that prediction the model goes haywire,
Exactly the same goes for Hive system. Easily tricked, makes prediction based on less data (so, less than nothing is it?). The hive thinks this player is better than this one and shuffle tries to build a team out of a "i don't know what concept" in fact. But if these players don't follow that prediction the model goes haywire
Problem is; it's an proven absolute now. To my despair again.
I even suppose it will be much worse than the current system. I personally don't believe this detailed statistics would do more good than harm. It's not totally bad, it could work for dota-like match making, i.e. when the master server takes players from a huge pool of waiting players and tries to make two good teams out of that. But we simply don't have that many players for this method to be reliable. We usually only have 20-24 players and should work with that and this detailed stats simply wouldn't work.
The hive system won't work with more and more players. It's simply impossible. The more the player the less accurate it will be. To me Hive system isn't reliable already...
This happens because of lack of communication or overall low skill sometimes. Like, 4 players going gorge, not making tunnels, tanking marines and quickly dying etc. Sometimes it's a purely "marines day" or "aliens day" until the server population changes drastically because a significant share of the players are only good on one side. I hope this will change with separate marines/aliens elo though.
Nope i was referring to a bad distribution of player because the ELOs weren't accurate at all (if it ever was). They were supposedly good players in that team. Ultimately the were NOT. Just a bunch of "stickers" who couldn't surf the wave of stacking with their idols this time. Everybody loose time because of this. Stacking 1; hive 0.
Now we're talking. You actually don't play much and that's the root of the problem. Six days in a whole month? If I'm counting right, you only have like 120 data points since Jan 2017 (not rounds as this is updated once a day, not after each round), that's a very low number and your skill might not be calculated reliably yet. More importantly, put aside the actual elo numbers how does that affect your game experience? Why are you concerned at all? Let it fluctuate up and down, who cares?
On what basis it's not playing enough ? I play when i want and for the number of games i want.
Who can prove that playing more games will make my Hive:ELO accurate. How many games do i have to wait for this to happen ?
Don't bother it's simply impossible to tell. I already ask this one to the master himself. No sound was heard that day... years ago.
A soon as one new player enter the NS2 realm it's over... Yay Free Week end happened. And probably one will happen before Christmas. This system will never be close to accurate in the shape it is today.
After 3.4 thousands hours i think it should be done now. Any decent system would be able to correct itself in less than 24~30 games (considering a game is 20 min and the target is 8 hours. Enough to fill a week for a beginner). It will NEVER happen for the Hive system.
A new system will have to manage incoming new data the same way as any observing system. The difference is : It will be to create/refine a better picture and not staying with a blurry one like Hive does.
If i play less than before... maybe there's a reason. Bad behavior (enclose the long list), dumbing down the game at each patch, sales... but ultimately bad game because of a poor mix shuffle did, because of a not so accurate skill system (and that's really gentle of me to put it like that).
I don't care about the number. I don't suffer from the wallet syndrome like (many?) others. On the other hand i do hate what this system brought with it and what this system is responsible for : bad games (by the definition mentioned above). So yes it affect my game experience as this game has a huge potential that is constantly ruined by those kind of things.
I'm not saying the current elo and shuffling system is perfect but it's the best we have. I once thought that accounting for K/D ratio should be a must but then gave it a secod thought and realized it's not that important. You can kill a couple of marines and bite 1.2 RT/min as a skulk and thus bring your team victory. Or you can kill a few skulks but 2-3 fades and several lerks destroying the entire aliens life form pool. The game tries to highlight that with score points but I'm not convinced this should affect the elo. Why? Because that would shift priorities. Instead of teamwork to actually win people would prioritize frags and points (because that would save them most of elo in case of losing or bring them more if they win). It has been discussed already and the consensus was that only the fact of winning/losing should affect elo just like in chess, dota and other games.
I was waiting for that one. If it was to save someone who desperately needs a heart transplant i would consider going on the highway with 3 flat tires. But NS2 isn't in that situation. It doesn't need to be saved right now with one injection like in the movies; and 5 sec later the patient is dancing around. It needs the proper assets to grow steadily, like building a wall. Problem is the hive system is more likely to destroy it (see bad behavior).
If you want to promote teamwork make it a score worth more than K/D and that has an influence on the ELO. A lot of player will change fast; trust me (wallet syndrome).The game know how to measure that kind of things. It's part of the Steam success already. Re-using it as a stat (sampled at each game) won't be huge effort. Then there will be a debate on whose to say this action is better than another one.
Frankly whose to say Hive is right in its approach in the first place ?
I seriously dunno, it's like you're playing a different game than me. I have my ups and downs but I can't tell I play bad rounds all the time. It happens sometimes, some days are bad, others are ez win, sometimes it's a real struggle just as in real life. I doubt you or anyone can make a system that guarantees good games consistently each time. Not with the current players number, match making method and game mechanics with not so many ways to come back. Elo calculation and shuffle algorithm are not even the main reasons of that. I play almost every evening and if the game were that bad as you describe I'd drop it years ago. Yet here I am, launching it much more often than any other game and playing for couple of hours staying on the positive side most of the time.
Sometimes it's a purely "marines day" or "aliens day" until the server population changes drastically because a significant share of the players are only good on one side.
Finally someone else that agrees with me that some people are better at one side over the other!
Don't even try mofo. I have told you so many times that people are better at one team or the other. I am better at aliens myself, not unlike yourself. I have also told you that I want separate marine and alien skill values yet you have persisted to act like you are alone in that. I have spent SO MUCH TIME telling you these things and trying to explain as simple as I know how. I knew it was probably a waste of my time going in, but then you go and say things like this where you act like everyone has always said you were wrong.
That would be nice, if this new elo can accurately predict whether a person is better at one team or another... Which seems unlikely since it will still use a single number.
I'm still waiting on a "layman's" explanation (without the mathematical mumbo jumbo) of how it will know when a player with say 1700 hive score is really a 2000+ on one side and a -500 on the other...
TL;DR: "Don't worry about hiveskill too much. It's a sophisticated system and probably already takes everything into account that you are worried about."
True, but UncleCrunch doesn't like how it doesn't actually measure skill but Hiveskill / Hivescore is just a number that shows how likely it is whether You can make your team win in your average environment...
... with truly and utterly poor results so far. I had to fix it.
You missed it. I don't care about what measurement is used, or what equation was ripped from a statistical cookbook. I don't have feelings about that. I'm only looking at the results in the end. So far...
FrozenNew York, NYJoin Date: 2010-07-02Member: 72228Members, Constellation
edited September 2018
The "in your average environment" thing is what needs to be taken for granted, to acknowledge that nothing can be perfect about it.
When i started playing again in like 2016 and the 8v8 captains server was around I had never got a hive skill before. I only played that server and it was "high" skill. I absolutely crushed until around 3000-3200 then started fluctuating. So comparing my 3200 to someone playing 12v12 anywhere was useless as far as I was concerned, but it was damn right in that environment.
Same would go for when competitive scrims and better gathers actually existed, you're mixing games where people lose a ton of hive skill when realistically it should probably just gift them hive skill for playing that kind of round just so it would match others better on a public server.
So, could you give an option for a server to track a hive skill based only on ELO won and lost ON that server? Then you could choose to use that number in shuffle?
It's all dumb anyway no matter what when you consider drunk games count too.
HandschuhJoin Date: 2005-03-08Member: 44338Members, NS2 Playtester, NS2 Community Developer
@Frozen the Serverside approach also has issues, since even on the same server there are different players depending on daytime... EU morning TTO for example has 1k avg skill while EU nighttime it goes to 3k
Generally the more ppl "mix themselves up" from different servers/countries from time to time the more accurate can the hiveskill be, to determine how useful you are in your team... but this is still an average number of your gameplay. Since you gain or lose hiveskill rather quickly there is not to much an issue if you play drunk.
Hiveskill is far from perfect but in my opinion the best approach to somehow quantify your skilllevel to make fair teams since it is far to complex to use another method. The only thing that needs to be adressed are in my opinion changes for Hiveskill regarding Alien/Marine/Comm plus that Old- or Smurfaccounts need a more instant hiveskill-increase.
moultanoCreator of ns_shiva.Join Date: 2002-12-14Member: 10806Members, NS1 Playtester, Contributor, Constellation, NS2 Playtester, Squad Five Blue, Reinforced - Shadow, WC 2013 - Gold, NS2 Community Developer, Pistachionauts
I don't think "your average environment" makes a huge difference because players mix between servers frequently, and that's enough to keep different servers skills in equilibrium. This could make a big difference internationally, as EU and US players rarely play each other on even footing, but within one region where pings don't vary enormously, it should even out.
I don't think "your average environment" makes a huge difference because players mix between servers frequently, and that's enough to keep different servers skills in equilibrium. This could make a big difference internationally, as EU and US players rarely play each other on even footing, but within one region where pings don't vary enormously, it should even out.
I don't think there is enough random mixing. Ns2 servers are community based, and it is not uncommon for a majority of players to find one community/server they like and then play on that one server the majority of the time. There is also not enough mixing between the skill extremes, which I think is a good thing, but it adds variance.
For a variety of reasons, there is a lot of variance in a single players skill value. One reason is a single skill value for marine, aliens, and commanders. Another reason is not enough mixing. I am sure there are many reasons, but variance is undoubtedly high in my view.
For example, I can go play mid day with a bunch of low skill players and have my skill value go up 5 games in a row. I can then go play on TA in the evening where the average skill value is 2500 and lose 5 games in a row. I could play commander for a few games in a row and drop 200 skill value. I could then lerk a few games in a row and gain 200 skill value. Yes, my skill value is a oscillating around the correct equilibrium but the variance is quite high. That 200 skill difference on one player being under or over valued can be the difference between a win and a loss for his team.
As a player with a hive skill of 2700 right now, it isn't hard for me to drop 200-500 hive skill just by playing differently such as gorging or commanding. I also peaked at 3000 hive skill when I was trying hard. The system is adjusting my skill value the correct way based on how I have been playing recently. My skill value right now at 2700 is an equilibrium between me trying hard and and playing gorge, but the range is from 2200 to 3000. Yes it has a mostly right equilibrium based on my average environment, but it also has a lot of variance.
I know the system works. In some basic testing years ago, I found that hive correctly predicted which team would win about 90% of the time. This seems unnaturally high to me, and has needed verification. Tik from the south America server has similarly found a about a, 70% prediction accuracy for the hive system on his server alone. The system does work, but the individual player variance is probably higher than it should be even before accounting for international differences.
I don't think "your average environment" makes a huge difference because players mix between servers frequently, and that's enough to keep different servers skills in equilibrium. This could make a big difference internationally, as EU and US players rarely play each other on even footing, but within one region where pings don't vary enormously, it should even out.
I beg to differ. I only play on 8bit!
Not so much because of server quality, rather than who plays there reqularly.
Essentially this says that if you have a small number of random edges (about as many as you have nodes) then the entire graph will be connected with very high probability.
For our purposes, this means that if there are as many players who choose servers randomly as there are servers, then every server will be able to transfer skill to every other server, and the differences should even out over time.
Wow! Never heard about this theory. I was concerned about global P2P networks like bitcoin or I2P having clusters of nodes not connected to the rest of the world. I thought it could happen because they connect randomly but couldn't find any information on this particular subject, it seemed like no one cared about this aspect. Turns out, it has already been proven wrong, thanks for mentioning that!
Comments
In addition to your comment, here is the expected win or loss probability of any given team by how big the difference in average team skill is. As you can see, even a 100 difference in average team skill gives one team a significant advantage.
I believe this formula is used in NS2+ or Shine, I saw the winning probabilities once on TTO. Can't tell if they're accurate enough or maybe that was another formula. Anyway, these probabilities were turned off after several rounds to not discourage players. And yes, 100 difference usually guarantees the outcome. Too bad it's a common situation when the round starts with even teams, then someone leaves and some other player with elo greatly above or below average joins instead. The balance immediately breaks and we need to shuffle again which leads to more people leaving because the game doesn't start and they become frustrated. I wish there was an enforced rule to prevent that.
As for the citation. The skill factor changes even if you're on the winning team and you're supposed to win. 100% sure. Even if it is a small number it is a number. And as it would me less risky... people just do it.
Plus an unexpected victory can take many form in NS2 with a non zero load of luck. That is why some people complained the most about Gorge tunnel. Unfortunately they were listen to.
AND if it is unexpected and most of the time accompanied by luck; how can you associate this with a skill factor ?
Let me explain it better. the game doesn't start if people see there is a big ELO gap.
Follow for a moment. The "stickers" are the one who attach themselves on the tail of some big head players. Others are just regular players or rookies that just don't care about it (for the sake of simplicity).
After a shuffle (not an official thing as it was adequately put) some stickers change team. The gap is bigger (not every time). The game will not start as some player noticed it. But the sticker is glued.
At this point 1 on 2 shuffle will be voted again. Or some other players will move naturally. The usual scenario is the shuffle vote. The stickers will change team as many time as they need. It going like that up to a point the teams get balanced according to ELO. With one specific trait. The sticker stacked with their respective "heads". In itself it is stacking and they get what they want. And as i stated just before. You still earn point doing that. You can argue it's gonna be slower. It doesn't matter for them.
Bad behavior 1 ; the rest 0.
Oh. And what about this: 1 player plays 1/4 of the game in one team, then change team. He's got a lot of intel' to make sure it will 'pinch' somewhere it hurts. Then he comes back (for the last quarter) in the team he left... what happens in terms of ELO ? Hint: the score of this particular individual is well above 3K.
This ELO system has always encouraged the bad behaviors .
- Stacking, need i say more...
- Smurfing for individual war (try to make somebody you don't like loose) or simply messing up on a server
- Fake games. Training rookie in exchange for points as they loose the game.
- The ELO can go up to infinite which is totally in opposition with human skills or capabilities. It's like a wallet.
A good load of things i anticipated in other threads (Hive v1&2) years ago did happened. To my total despair, unfortunately.Smurfs, yes, but they're easily detected and banned because no one likes to play against smurfs.
About the infinite elo, well, try it? I wonder how far you could go like that. Theory and gut feeling is one thing but in practice it's much harder if not impossible to do. The system is made specifically to predict the round outcomes and correct itself if it's wrong. If you can exploit some property to gain as much elo as you want, why not to try and become the top Sanji out there? Well, you probably can do that with 11 fake accounts where you constantly beat them (I believe only 6x6 rounds and up are ranked). Those accounts should also have elo close to your own or else it would take ages.
They go back to RR and then wait for an opportunity. Too easy.
Admins aren't what they're used to be. It depends also on the server rules. Some are really forgiving. And banning one smurf sometimes just create another one. You gotta get in the head of those people. They're really willing to go to some extreme extents.
What i describe is: At some point you don't learn anymore (map, positionning..), You can 't be faster anymore, you can't kill more anymore (not enough time, targets...). Humans have limits, and this game has limits too. The logical question is : What limit for the skill factor ? There is none that i know of.
It gives the present situation in which you have people with 4K and rookies with almost none. Not only it'll be more difficult for the shuffle to sort things out, but those ELO can't be accurate. Not one bit. I mean is a player at 5k better than one at 3K after 500 games ?? Especially if they never played on the same server. I'm not sure.
In fact i've seen better 1,5k players than those 3k, just because they were unfortunate to loose the wrong games. Or simply because they use to play in other "circle of friends". A pity.
Then play on the servers with better rules and admins — easy!
Why should there be a limit? If you beat some players, you have a better skill. If you beat the best, your elo is higher than theirs. When you start to lose against people with lower elo, your elo goes down and theirs goes up until it reaches the balance point.
I'm not sure either, they can play against each other and settle that, no? Is your point that there's not many average players of rank 5 on the servers you play? If that's it then yes, you can't balance such extreme teams with sanjis and rookies only. But then again, in what game is that possible at all?
Elo is not a number to brag about. It's a number to make you win and lose with approximately 50% chance. If you win too much that means someone else loses too much and elo compensates for that putting better players in the opposite team and the weaker ones in your own. It shouldn't concern you if someone has 3k or 5k if they don't lead to purely one-sided rounds. Unfortunately, sometimes they do if there's no one of comparable skill on the server. But there's not much anyone can do about it (of course, if that player is not a cheater with aimbot or wallhack). If someone's good, he can destroy the opposite team single-handed and elo reflects that.
Sure, if you only play with several other people your elo can be wrong in the global scale. It's like that everywhere in the world, not just in NS2 or games in general. You can be top in your class in school but then you go to college and suddenly you're not that great anymore or may even fail exams and drop-out. Elo shows how you perform compared to others, so if you want a more objective elo, play with as many players as you can or else you'll be confined to your bubble and be the best among your friends but worst among the pub players.
The problem could be solved by a better shuffle algorithm. It might account standard deviation or other statistical metrics so that we don't end up with teams like 5k+n*1k vs n*2k or so, i.e. one has a sanji + bunch of rookies and the other one has average players. You can expect much better teamplay from a team of averages than from the former. I played such rounds recently, my team had 1 sanji (7th rank) and others were rank 5 at best, me including; the other team had 3 or 5 players of rank 6th and others were 5-4, but mostly lower than 5. Of course we lost, because rank 6 can teamplay much better. And the average skill score was pretty close that time.
Yet again, it doesn't have anything to do with elo itself but with the shuffle algorithm.
You're right there so many servers online. If only they were crowded... The last ~200 players left won't save the day on that matter is it ? Now if you're asking them to seed a server. You've gone too far mister!
As Mofo1 mentioned before. If you're not good at playing marine but excellent at alien. you will loose or win 'skill' (as you name it). You see were it goes right? Thre is no balance point in that regard. Plus in that case you have a mean to manipulate your own skill factor (and others as collateral damage).
Worse. Imagine i'm a super Fade and decide to go gorge then the team loose. If you say i made a bad decision this would mean i have only one way to play NS so my 'skill' can be accurate. Super fade vs bad commander. Super gorge vs Lerk. Super marine / bad JP.
It's kind of anti-NS-versatility don't you think ?
That's were it's all wrong. If it was a ranking system it could be left as it is. Even if the team victory is the determinant (see the many bad scenario described a while ago in which you suffer from idiots doing stupid things). But it's not. It's used for shuffling games.
Skill and skills are quite different especially at NS. Even if it would be more difficult (i consider it to be mainly an bad excuse for laziness) the way to evaluate a NS player cannot be a single number.
This shed a light on the team A,B & C,D issue. You can't have a decent idea of the strengths until the system corrects the value. That means that everyone have to play over and over with everyone. It's kind of exponential as son as you add more people... therefore impossible. Unfortunately we're human. Need to sleep, eat and such.
!!!! sic... really ??? We have some people who do the opposite.
The school example isn't a fit for the matter at hand. In this example everything changes. In NS2 we have the same game all over the world. A decent system could make USA A USA B play AUS A or AUS B with a prediction that would not be perfect but better than what i could witness.
Waiting for a system to correct itself means playing bad games after bad games waiting until that holy day... that never comes actually.
How can a shuffle algorithm can see if you're better at marine than alien with one single number ?
If a proper "set" of values is available there won't be long until it is used in a shuffle algorithm the proper way. That would be a better shuffle algorithm only because a better skill measuring system is behind.
And don't forget that you're not playing solo, you have a team. Even if someone for some reason doesn't play good enough just today others can compensate for that (a little). I don't believe that at one side everyone chose their best weapons/lifeforms and on the other side each player somehow plays as a rookie because of reasons. I guess being a versatile player also counts towards your skill. If you lost your super-fade, went gorge and made a rush tunnel that brought you victory, it's as good as swiping half the team with a fade. If you're a one-trick dog, well, you better do your trick as best as you can all the time, otherwise your defeat is well deserved.
I know, and it's not an NS2-only issue. For smurfs, it happens in all ranked games because computers can't tell if a new player really a rookie or a smurf. Even people can't tell sometimes. Otherwise, even if people brag about their skill I don't see a problem. If they don't cheat and play according to their elo, it's fine. My point was that you personally shouldn't be so conscious about that number. It doesn't mean you're worse as a person than somebody else, just that you're not as good in this particular game. And I'm afraid most of us are already stuck at the levels we are now because of physical limitations and age. That's fine, the goal of the game is to have fun. Not like you can sell those elo points for dollars or anything.
That's quite pessimistic and I don't see where this comes from. Are you constantly losing, like 20 times in a row? Or you're angry that you're losing at all? Do you expect to win all the time? That's not how the competitive games work.
What is the point of the ranking system if not putting the players of similar skill against each other? It's exactly like that in all sports and e-sports. The rank itself is useless if it's not used for balance. If you don't do that you might as well drop 12 sanjis vs 12 rookies and watch the massacre. Is it fun? Doubt so.
Let's start anew. What's your main problem with elo, ranking, shuffle etc.? Does the balance seem wrong to you? Do you lose or win too much? Give your observatory profile URL so that we could talk more productively. Here's mine. As you might see, my skill has barely changed since the last year.
Basically you have several choices. Some think about K/D mixed with X and Y and a weight on top, etc.
I do prefer profiling. We know the game can recognize one thing or two. Like the NS2 success in Steam. There is quite many different things in there. Why not using it ? With less time than Hive we will be able to see if a player prefer fade / gorge, shotgun/LMG, to build and protect etc. The good thing is it can answer the need of a team. As a commander from time to time i really don't care having 10 Rambos in my team if no one will build that damn RT.
In the end a shuffle system getting those values can see if a person has chances to go gorge instead of fade (as an example). So the system can put 1 fade with 0% chance to go gorge, and one other fade with 50% chance to go gorge instead of that last fade alone in the team. Not easy to explain but i'm sure you get the point.
Or exploiting other things. We know a rookie is more likely to go gorge often. But with time he will try something new. Say about 50 hours. Chances are he will probably weaken the team.
Of course the shuffle system will need to know how good you are in each category but in the end it will have a truly better set of values to "guess" what would be a good mix. Or at least avoid the worst. Things that occur more than once per night when i play.
So people can have games with what's needed for a decent start (builder, hunters) and they can freely try something new. Up to them to organize it properly.
This wont avoid those who try to trick the system, or to be a jerk. But it'll be far better than the ugliness i could witness.
Not encouraging people to try something new is it ? No need to go further. It's simply plain wrong.
So maybe it's not only about the life form? Maybe the skill lies somewhere else? If that so, you're getting close.
I know 1 or 2 who would if they could.
I'm not cautious about that number. Only against the logic behind that brought every single bad behavior anybody can witness. Sooo annoying to be right.
To me, a bad game in the shuffle context is a game that shouldn't happen if the proper data was there to decide how to balance the game. I've seen that many time. ELO is supposed to be fine in both teams but in less than a couple of minutes, the game is already over and many player on both teams know that.
What's wrong ? Look
What's the reason of the changes (with a big magnitude somehow) ? I tell you; i didn't changed a bit. That's the funny thing about it. I'm quite mr constant. But the skill factor moved up/down. With not much activity indeed. It's more like i ended up in the wrong team sometimes.
To sum it up; 50% growth in 1 year 9 months and with not much activity. My aim didn't changed, i still rush like "for the horde" (because i play in pub) for the fun of it. It's not like a third hand grew on me and i can now type faster (or hit the map key ten times a sec). I'm no better, other regulars didn't get better (they do the same things).
That's a big issue. The ELO is used in place of skill (in shuffle). But the ELO is determined using the team win/loose. I'm not blaming shuffle. The input data is the input data.... But do admit that a team result based ELO cannot be seriously taken as a personal skill factor/score/whatever. Because simply it depends on the team result first. I mean i hope you see the dichotomy. That not a crack in the logic of trying to evaluate players individually. More like a canyon. And i do know it's not the only one.
Your "profiling" system is also easily tricked. It basically makes predictions that are based on nothing (maybe some prior statistics that might not be very reliable for a particular round) and that makes it vulnerable. Because those predictions are made before the round starts. The game thinks that PROBABLY this player will do that and that player will do this so let's pretend that player will be stronger and this one weaker and let's try to build a team out of what we have according to some guy's concept of a "dream team". But if these players don't follow that prediction the model goes haywire, I even suppose it will be much worse than the current system. I personally don't believe this detailed statistics would do more good than harm. It's not totally bad, it could work for dota-like match making, i.e. when the master server takes players from a huge pool of waiting players and tries to make two good teams out of that. But we simply don't have that many players for this method to be reliable. We usually only have 20-24 players and should work with that and this detailed stats simply wouldn't work.
This happens because of lack of communication or overall low skill sometimes. Like, 4 players going gorge, not making tunnels, tanking marines and quickly dying etc. Sometimes it's a purely "marines day" or "aliens day" until the server population changes drastically because a significant share of the players are only good on one side. I hope this will change with separate marines/aliens elo though.
Now we're talking. You actually don't play much and that's the root of the problem. Six days in a whole month? If I'm counting right, you only have like 120 data points since Jan 2017 (not rounds as this is updated once a day, not after each round), that's a very low number and your skill might not be calculated reliably yet. More importantly, put aside the actual elo numbers how does that affect your game experience? Why are you concerned at all? Let it fluctuate up and down, who cares?
I'm not saying the current elo and shuffling system is perfect but it's the best we have. I once thought that accounting for K/D ratio should be a must but then gave it a secod thought and realized it's not that important. You can kill a couple of marines and bite 1.2 RT/min as a skulk and thus bring your team victory. Or you can kill a few skulks but 2-3 fades and several lerks destroying the entire aliens life form pool. The game tries to highlight that with score points but I'm not convinced this should affect the elo. Why? Because that would shift priorities. Instead of teamwork to actually win people would prioritize frags and points (because that would save them most of elo in case of losing or bring them more if they win). It has been discussed already and the consensus was that only the fact of winning/losing should affect elo just like in chess, dota and other games.
Finally someone else that agrees with me that some people are better at one side over the other!
That would be nice, if this new elo can accurately predict whether a person is better at one team or another... Which seems unlikely since it will still use a single number.
I'm still waiting on a "layman's" explanation (without the mathematical mumbo jumbo) of how it will know when a player with say 1700 hive score is really a 2000+ on one side and a -500 on the other...
Combinatorics to the rescue! We want to know how many unordered sets of 12 players can be formed out of 24 players (because the rest goes to the opposite team). The formula is pretty simple, n! / (m!(n - m)!) where ! is factorial, i.e. n! = 1 * 2 *3 ... * n. For our example n is 24 and m is 12 so: factorial(24) / (factorial(12) * factorial(24 - 12)) = 2704156. Just 2.7 million iterations. Now, if we don't really care about the team preferences we can safely divide this by two because half of the sets are the opposite (for example, if players from 1 to 12 go to marines and 13 to 24 to aliens, the opposite set would be 1 to 12 — aliens, 13 to 24 — marines which is the same in terms of balance using the current one-number elo). So we have 1.35 million variants which is quite a small number and can be calculated in under a second I guess. This straight forward algorithm doesn't scale well though, for 13 out of 26 players (just 2 players more) the permutations number becomes 10 millions and for Wooza's 25 out of 50 it's an enormous 126410606437752. But for 24 players it's indeed possible to find a perfect balance in reasonable time.
The mirror hack (dividing by two) wouldn't work if we have two separate elo numbers, of course.
No, but we have the advanced AI commenting on this forum
And that would just take like, idk, probably less than 100 iterations to converge? And if there was 2 hive skills you would just be checking the appropriate skills when trading players?
In order for new players and veterans to fully enjoy the game diversity they must be given a chance in those areas without harm. On a 24 slot public server it's quite manageable. Far more than a 5K who decide to be the jerk of the day and ruin every game. Not to mention that as a Strategy game every unit has a power that can be devastating if used properly.
Otherwise it looks like a CS in space for them. Just like another toilet paper, they will throw it without discovering anything about this game (for the new player), and without enjoying some other stuff for regulars. Hence player retention.
I'm not for imposing a contract on players. Just observing and accumulating stats of their actions. On the technical layer: Storage, calculation and all, aren't a problem at all.
Exactly the same goes for Hive system. Easily tricked, makes prediction based on less data (so, less than nothing is it?). The hive thinks this player is better than this one and shuffle tries to build a team out of a "i don't know what concept" in fact. But if these players don't follow that prediction the model goes haywire
Problem is; it's an proven absolute now. To my despair again.
The hive system won't work with more and more players. It's simply impossible. The more the player the less accurate it will be. To me Hive system isn't reliable already...
Nope i was referring to a bad distribution of player because the ELOs weren't accurate at all (if it ever was). They were supposedly good players in that team. Ultimately the were NOT. Just a bunch of "stickers" who couldn't surf the wave of stacking with their idols this time. Everybody loose time because of this. Stacking 1; hive 0.
On what basis it's not playing enough ? I play when i want and for the number of games i want.
Who can prove that playing more games will make my Hive:ELO accurate. How many games do i have to wait for this to happen ?
Don't bother it's simply impossible to tell. I already ask this one to the master himself. No sound was heard that day... years ago.
A soon as one new player enter the NS2 realm it's over... Yay Free Week end happened. And probably one will happen before Christmas. This system will never be close to accurate in the shape it is today.
After 3.4 thousands hours i think it should be done now. Any decent system would be able to correct itself in less than 24~30 games (considering a game is 20 min and the target is 8 hours. Enough to fill a week for a beginner). It will NEVER happen for the Hive system.
A new system will have to manage incoming new data the same way as any observing system. The difference is : It will be to create/refine a better picture and not staying with a blurry one like Hive does.
If i play less than before... maybe there's a reason. Bad behavior (enclose the long list), dumbing down the game at each patch, sales... but ultimately bad game because of a poor mix shuffle did, because of a not so accurate skill system (and that's really gentle of me to put it like that).
I don't care about the number. I don't suffer from the wallet syndrome like (many?) others. On the other hand i do hate what this system brought with it and what this system is responsible for : bad games (by the definition mentioned above). So yes it affect my game experience as this game has a huge potential that is constantly ruined by those kind of things.
I was waiting for that one. If it was to save someone who desperately needs a heart transplant i would consider going on the highway with 3 flat tires. But NS2 isn't in that situation. It doesn't need to be saved right now with one injection like in the movies; and 5 sec later the patient is dancing around. It needs the proper assets to grow steadily, like building a wall. Problem is the hive system is more likely to destroy it (see bad behavior).
If you want to promote teamwork make it a score worth more than K/D and that has an influence on the ELO. A lot of player will change fast; trust me (wallet syndrome).The game know how to measure that kind of things. It's part of the Steam success already. Re-using it as a stat (sampled at each game) won't be huge effort. Then there will be a debate on whose to say this action is better than another one.
Frankly whose to say Hive is right in its approach in the first place ?
Intelligence doesn't mean it's smart...
Here is a link about it..
True, but UncleCrunch doesn't like how it doesn't actually measure skill but Hiveskill / Hivescore is just a number that shows how likely it is whether You can make your team win in your average environment. If you want to see what he would rather have, take a look at this thread where he describes it.
You missed it. I don't care about what measurement is used, or what equation was ripped from a statistical cookbook. I don't have feelings about that. I'm only looking at the results in the end. So far...
When i started playing again in like 2016 and the 8v8 captains server was around I had never got a hive skill before. I only played that server and it was "high" skill. I absolutely crushed until around 3000-3200 then started fluctuating. So comparing my 3200 to someone playing 12v12 anywhere was useless as far as I was concerned, but it was damn right in that environment.
Same would go for when competitive scrims and better gathers actually existed, you're mixing games where people lose a ton of hive skill when realistically it should probably just gift them hive skill for playing that kind of round just so it would match others better on a public server.
So, could you give an option for a server to track a hive skill based only on ELO won and lost ON that server? Then you could choose to use that number in shuffle?
It's all dumb anyway no matter what when you consider drunk games count too.
Generally the more ppl "mix themselves up" from different servers/countries from time to time the more accurate can the hiveskill be, to determine how useful you are in your team... but this is still an average number of your gameplay. Since you gain or lose hiveskill rather quickly there is not to much an issue if you play drunk.
Hiveskill is far from perfect but in my opinion the best approach to somehow quantify your skilllevel to make fair teams since it is far to complex to use another method. The only thing that needs to be adressed are in my opinion changes for Hiveskill regarding Alien/Marine/Comm plus that Old- or Smurfaccounts need a more instant hiveskill-increase.
I don't think there is enough random mixing. Ns2 servers are community based, and it is not uncommon for a majority of players to find one community/server they like and then play on that one server the majority of the time. There is also not enough mixing between the skill extremes, which I think is a good thing, but it adds variance.
For a variety of reasons, there is a lot of variance in a single players skill value. One reason is a single skill value for marine, aliens, and commanders. Another reason is not enough mixing. I am sure there are many reasons, but variance is undoubtedly high in my view.
For example, I can go play mid day with a bunch of low skill players and have my skill value go up 5 games in a row. I can then go play on TA in the evening where the average skill value is 2500 and lose 5 games in a row. I could play commander for a few games in a row and drop 200 skill value. I could then lerk a few games in a row and gain 200 skill value. Yes, my skill value is a oscillating around the correct equilibrium but the variance is quite high. That 200 skill difference on one player being under or over valued can be the difference between a win and a loss for his team.
As a player with a hive skill of 2700 right now, it isn't hard for me to drop 200-500 hive skill just by playing differently such as gorging or commanding. I also peaked at 3000 hive skill when I was trying hard. The system is adjusting my skill value the correct way based on how I have been playing recently. My skill value right now at 2700 is an equilibrium between me trying hard and and playing gorge, but the range is from 2200 to 3000. Yes it has a mostly right equilibrium based on my average environment, but it also has a lot of variance.
I know the system works. In some basic testing years ago, I found that hive correctly predicted which team would win about 90% of the time. This seems unnaturally high to me, and has needed verification. Tik from the south America server has similarly found a about a, 70% prediction accuracy for the hive system on his server alone. The system does work, but the individual player variance is probably higher than it should be even before accounting for international differences.
I beg to differ. I only play on 8bit!
Not so much because of server quality, rather than who plays there reqularly.
https://en.m.wikipedia.org/wiki/Giant_component
Essentially this says that if you have a small number of random edges (about as many as you have nodes) then the entire graph will be connected with very high probability.
For our purposes, this means that if there are as many players who choose servers randomly as there are servers, then every server will be able to transfer skill to every other server, and the differences should even out over time.