Interesting ideas but the system we have in place right now is a lot better. DBR/MMR is way too important to neglect.
@Ivanero
Maybe this would happen but it's very simple to change it, just by changing constants.
Besides, the less the games you have played, the less points you have
@åäö
How can something that does not exist anymore be important
Top 50 needs to have a new algorithm without DBR
Its unquestionable that the top 50 heroes is the best DotaBuffPLus feature so far. I'm almost sure the dotabuff plus users are decreasing gradually because there's no way of being in this ranking. Why? DBR.
Ok ok.... most of people check the ranking to get some references about how to play with X hero, but imo makes more sense putting ppl who play more often with X hero than a pro player who plays one time per month
+1
@sano, holy shit this is dumb. As åäö said, a rating system is incredibly important, weighting wins and losses according to the skill level of opponents, factoring in stacking and stuff - all the kind of shit that is sorted out with a rating system is VITAL for this to work.
I mean, I could make a new account, and complete these steps;
1. Lose 9 matches (pick hero)
2. Pick a specific hero, say Nature's Prophet or some mid hero, any really easy pub hero
3. Win easily as I crush the complete and utter scrubs that play in this bracket
repeat
I stay in a super low bracket, and I'll have a 100% win rate and a ridiculous KDA too. Weighting those wins is an absolute must when it comes to these rankings. The current top 50 are somewhat silly because they are using old DBRs (not sure if this is correct), but if they were using updated ones, it would be a lot better than any formula that ignores weighting these wins.
You can cry all you want about it, the people at the top level MMR are there for a reason, and in most cases they will play that hero better than the person who has played the hero ten, twenty or thirty times more than them.
Terrible, try to be more polite when discussing :)
The example you pointed out is just unrealistic. Let's say I want to be the top Lone Druid. Then I feed 9 times and pick Lone Druid and stomp, repeat. Ok. Remember that you need 30 games with the hero to be in the top, so you would have to play 300 games, while feeding 270, just to try to be in the top. I don't think many people would have the patience to do this. And if they did, we would easily detect them and ban them, just like admins already do regarding Limited Pool exploiters or dire Roshan abandon exploiters. And if we change the number from 30 to 50 games, one would need 500 games instead of 300. It's just too much.
However, I have to agree that measuring a player's MMR is important. But keep in mind that DBR doesn't exist anymore, so people have the same DBR they had 1 year ago, or they just don't have DBR. This is ridiculous because it's possible that I started playing 1 year ago and my DBR was terrible at that time, but today I'm a very high bracket player. And it's even more ridiculous because, as someone pointed out, if AdmiralBulldog started playing after the end of DBR, guess what? He wouldn't be in the top Lone Druid! This is absurd.
So, in order to have a good, consistent top50, we have 2 options:
1) Don't use any kind of DBR, like the one I did in the first post.
2) Make a new DBR algorithm.
Personally, I think a new DBR would be great, but we have to know if the admins are willing to do so: maybe it's not their interest. I actually think they are not very interested in earning any money with Dotabuff since there are so many features which could be awesome and even I would be willing to pay for it, but they are just terrible right now (top50 is a great example).
Now, if they're willing to create a new DBR, I'll be happy to help. It's not impossible at all.
"You can cry all you want about it, the people at the top level MMR are there for a reason, and in most cases they will play that hero better than the person who has played the hero ten, twenty or thirty times more than them."
Not always. Let's say, for example, Beesa is bad as Invoker (I have no idea if he's good or bad, just an example). He's at the top MMR. If he plays it 30 times and gets 55% winrate, he will probably be in the top50 today, and Grimorum is not, even though Grimorum has 80% winrate and 10 KDA, because Grimorum doesn't play in the top bracket and plays badly with other heroes. But we can say that the understanding Grimorum has about Invoker, and his gameplay with the hero, is higher than Beesa's.
how about this, rating doesn't mean shit since this is a team game after all. all you should care about doing is improve your own gameplay and
not chase after some stupid top 50 heroes bullshit
@nublets come hitler
Don't you agree that one nice way of improving your gameplay with a specific hero is to watch someone who's very good with it playing? I think that's the main point of the top50.
Its a tough call to make as I am sure that people would like to see this implemented it can give illusions of grandeur which don't, or shouldn't exist.
Some of my top heroes (not the ones on front page) have very good winrates and KDA's and a decent number of games. (e.g. Luna, Ursa, Veno). Lets use Axe, I have only once or twice come across an axe player who i considered as good as myself but even then a lot of my games were in High as well as Very High.
My games played in Very High is probably only a 55% win rate with half the games on a decent MMR. So.....
High MMR - 55% - 3KDA - 50 games : this wouldn't be better than someone who had
Very High MMR - 70% - 5KDA - 30 games : see my point?
Someone who is good with Luna in a lower bracket can never be the same player as someone who can Luna in a really top end game. I like the idea of giving others a chance but it would be a tough call to make everyone happy and have a system that is fair.
TL;DR Why am I not in top Axe?
@Havoc Badger
I understand and I completely agree, the problem is that we don't have any ways to measure MMR nowadays, and even bracket checking was recently removed by Volvo. So this means all the solid information we have are games played, win rate and KDA.
Remember that when we talk about the top players with a hero we're talking about people who have very high winrate and KDA with a high number of games. I agree that it's possible to have players in normal bracket who are like this, but even if it's possible, it's not likely to have many cases.
And I'm not in my PC right now, but I'm pretty sure my algorithm would put a higher score to the Very High MMR - 70% - 5KDA - 30 games player compared to the High MMR - 55% - 3KDA - 50 games.
I want, too, to have a way to measure ratings in order to use it in the top50, but I don't think there's a good way of creating a new DBR. We don't have information from Valve regarding bracket, and some enemies/allies have hidden profiles.
Well i think its correct to use matches played and KDA rate instead of brackets
Then i would maybe be in top drow and potm
Well actually I already did make a new rating system that does not reward any of the various methods of gaming the system to achieve inflated stats. It cannot be blocked by Valve like DBR because it does not require everyone to have stats on. It only requires win/loss records of those with stats on and uses strength of schedule to determine the value of the games instead of in-game stats. You receive negligible points for playing with or against players far weaker than you. This prevents any kind of smurf stacking to achieve rating points and allows you to be undamaged by games played with nub friends or games played against smurf stackers or games where Valves matching just screwed you over.
I am looking for a platform to do the final testing and host it. I think dotabuff would be ideal but if they are not interested I will eventually find someone else who can do it. Or in the long term learn enough to just make the website myself.
@Relentless that's pretty cool man, it would be awesome to have a decent rating system again. Feel free to add me if you want to discuss anything regarding this kind of stuff, I may be of some help (or not).
@Relentless, can you tell us the specifics in regards to your strength of schedules method?
@Sano, your algorithm isn't perfect, but it is much better than our current top 50 with DBR. Just add a few tweaks and it is good.
@Jussi, Dotabuff's only way to improve is to look forward and implement a new top 50. We will never improve if we live in the past (DBR).
It's really something that needs to be perfected and set up before announcing specifics. If dotabuff wants to use it they will want to make sure everything is working and announce what they want to announce when they are ready. The most recent web developer I talked to about would not set up a secure site to test it. Whenever you are releasing a new product it should be ready to be used. Otherwise people make negative judgments based on partial functionality.
I would ultimately like this rating system to be able to be used to find teams and set up stacks for pub games. Here are some examples of attempts to do this that have failed.
[1] Team Matching....Even though this is the primary official way to make teams in the game very few use it. The number of teams initially grew to over 160k teams, but since it did not work well people stopped using it quickly About 120k of the teams that were created quit in 1 or 2 games. Many more quit after trying it for a week or a month. Now there are about 20k teams that have a tMMR and most of those are inactive.
[2] Guilds...Valve announced guilds hoping that people would use them to create something like that small communities that created themselves in dota 1. But very few have joined and fewer still actually use them to play games because they are to hard to find. http://dev.dota2.com/showthread.php?t=103079 , there was no build-in way to advertise them and still is not http://dev.dota2.com/showthread.php?t=105170 and they don't work http://dev.dota2.com/showthread.php?t=97516 , http://dev.dota2.com/showthread.php?t=103079 , Even if a Guild is done well and lots of people join it can't actually handle the capacity required to have enough people to fulfill the purpose of guilds so everyone can find each other for games and teams. http://dev.dota2.com/showthread.php?t=98074 The maximum for a working guild is 1000 members...far short of what is needed. Because of these problems the initial excitement for guilds quickly died. As you see here.... http://www.reddit.com/r/DotA2/comments/1mg4uy/reddit_guild_it_feels_dead/ "guilds are dead"
[3] Inhouse leagues... In dota 1 there were a lot of leagues that grew organically over the years because battlenet was completely inadequate for finding game and making teams. But in dota 2 this system has struggled. The same sorts of things cannot be created because it is too difficult for players to find each other. Like guild initially many will join a new inhouse league, but when they find that things don't work and there are not that many players involved interest quickly dies. http://www.reddit.com/r/DotA2/comments/18qysa/inhouse_leagues/
(A) Keita Gaming http://www.gosugamers.net/dota2/news/19606-keita-gaming-hosts-inhouse-league Big announcements... lots of hype and promise, even a cinematic trailer but it all rapidly comes to nothing. http://keita-league.com/ All that is left now is the trailer See why it dies here http://www.keita-gaming.com/index.php?p=forum&t=383 , partial release... did not work, everyone quits.
(B) Ngage dota 2 league http://dota2.n-gage.tv/index.php ...big hype, lots of ideas, complex rules...but in just a few months has already crashed to less than 10k users. See why it has failed here... http://en-us.reddit.com/r/DotA2/comments/1l2w3s/ngagetv_join_now/ It was too hard to join. It was too hard to find matches. Things did not work. People quickly gave up on it.
(C) Dotabro http://en.reddit.com/r/DotA2/comments/14mewd/dotabro_crew_proudly_presents_dotabro_league_our/ website doesn't even exist anymore... http://www.youtube.com/watch?v=tibJaSlzjG0 see what it was here
(D) http://strategyinrush.com/forum announced in May, zero posts
(E) ixdl...died, but now is back because they made it for pros. North American Elite League has casters, a real prize pool, in-game ticket and has expanded to EU. But this is only for pros, and extremely good pub players. Still millions of ordinary players have no way to make teams other than accidentally encountering each other in games. Friends lists are far to small. Even Guilds are too small.
Everyone knows they want something but they can never get it too work... http://gather.dota2.no/info.php here is the newest attempt, only for Scandinavians. Sadly it is suffering the same fate as all the others. The last two "seasons" there were not even enough players to fill up the top 100 list.
_____________________________________________
So, this is why you just can't talk about things like this in detail until they are ready.
I'm re-reading the posts here and I think something isn't clear for most people - the goal of this topic isn't to propose my algorithm as a new one. It's way too simple and it's just an example. I just want to make things clear - that we can and we should have a new algorithm and that it's not hard to make a new one. We just need someone to do it and the Dotabuff staff to aprove it.
the problem is unless everyone gives their data to dota buff they can not make accurate data
If a group of players choose not to show their matches the 'unknown' variable will always be a factor
As Sampson points out the missing data is the essential problem, the main reason DBR has not updated in nearly a year.
That, and a few other problems is what my new system is supposed to solve. While it may seem that creating such an algorithm is not difficult because the basic components may be obvious, really it requires a novel solution. Such things may seem simple and obvious after they have been explained but before someone tells you how to solve a complex equation it sure doesn't seem simple and obvious to you. That is why even though everyone wants it such an algorithm it has not been presented before.
Anyway, to make a highscore for each hero there needs to be a base factor that is the difficulty level of the games played. DBR filled that roll very nicely when it was started. But as time goes by it becomes less and less useful, and now its been frozen for 11 months so a new one is really required for the highscores to be considered accurate.
@Relentless
When they got rid of skill brackets they also increased the MMR range of players in one team, so I think you can no longer estimate anything.
Ex: http://dotabuff.com/matches/393693158
If you look at ShadowFiend's games from 2 months ago, you'll see he was always on high / very high. On the other hand look at gyro in my team and at weaver on the other team. And yes, it was a "Solo queue" game, so there were no stacks in this game.
Yes, the first idea I had for this won't work anymore. But the new one is actually a lot better. The old one was simple...computationally efficient, but the new one is robust. It will work even if some of your games have 9 Anonymous players and people turn stats on and off. It does not estimate stats based on a bracket, its just calculates them factoring in the quality of the information for each game.
Actually we are not. In fact my system doesn't use any in game stats. If you want to make a hero highscore like the top 50, then in-game stats become important. But to give an overall rating you just want to know how difficult the game was (who was on which team, what are their ratings) and (did you win?). Once you know how difficult the game is, then your in-game score becomes relevant. A win against a stack with Dendi and Xboct is worth a lot. A win against a bunch of nubs is worth very little. Maybe against the nubs you got 200 KDA... that means nothing. But if you averaged 3.5 KDA on a support hero and won 55% of games against stacks with 2 or 3 pro players... that will put you on the highscore.
Defeating nubs very consistently and dramatically means you are clearly better than nubs. And if that is all you have done it should give you a score solidly at above nub status... but not a top rating. If you want a top rating, you have to beat people who have one.
I am not into any statistics or anything however I would have some ideas I just want to spit out. Maybe something useful will come out of it.
First off all if no games have been played a player starts with a certain score such as 5000. The performance within will be rated by his stats however with influence of the enemies stats(average of all players). If your score is 5000 and other once is 2000 you will get up to nothing. And vice versa.
However your own teams score should also be taken under consideration and should even be averagized. If you would play wth players with a low score you would get almost nothing however if you play with players with a simialar scores to yours you would get more, leading to a position where friends would have compareable stats sind they play a lot together (and somewhat teamperformance wiill be measured)
So we sup up for now, we got a score system which rates every game depending on your oponents and your teammates
What would be the benefits ? Lets say you would go soloQ and suddenly have dendi in your team, which already means your MMR is high it would give your credits for that and puts you in the same corner for him, however if you have baddies in your team it will do vice versa.
Solely stats should be considered but also within the team (Teamparticipation, roledistribution etc.) and should somehow be brought to relation with a certain hero.
This system considers already that a man with 2000 games wiill not skyrocket cause it takes other players under consideration however it will give him a small bonus.
Regarding the heroscore, an amount of points won within a game can be add to a players heroscore however this might be the wrong approach
Im just spilling out some ideas
imagine the flames after the game when their rating drops by -50:
"you fucking 2000 scrub stop playing doto"
"your mom is so fat"
players with such amount of games would have a positive effect on you when they would play agaisnt you since their score would be higher then the average of people in their/your skillbrakct and rendering more points and losing against them or with them would grant you a lesser penality than average.
Jst gving some thoughts :P
There are several problems a rating system for dota 2 needs to confront. These things need to be systematically evaluated for each game. All games are not of equal value.
[1] There is information about the game that may be missing. Some players have stats off. The quality of the information about each player must be evaluated.
[2] Despite Valves intention to make games 50/50 win/loss each individual game is not. Sometimes its close, sometimes the match is very bad. The balance of the game must be evaluated. Winning easy games should be worth few points. Losing easy games will cause you to lose more points. Winning hard games will be worth a lot of points. Losing hard games will lose only a few points. Just like when betting rares risk and reward should be commensurate.
[3] Whether by intentional attempted abuse, merely playing with nub friends, or poor matching from Valve some games are played with players that have an enormous range of skills. You should not be punished for playing with nub friends so games with a large range of rating scores are worth few points. You should not be punished for being stuck in a bad match where Valve put some nub on your team... OR REWARDED just because Valve screwed the other team with a nub feeder for them, so such games are worth few points. You cannot gain points defeating players who are rated far below you. This removes all forms of abuse from the calculation. You can only get to the top by defeating top rated players. Therefore the range of rated scores in the game must be evaluated.
All of these factors synergize. To get a lot of points from a game it must have high quality information, be balanced, and have a small range of scores. This guarantees that the highscore will be legitimate. The trade-off is that large changes in skill level will take more games to be revealed. But I think everyone will be much happier with a system that is designed to rate Main accounts and gives you a stable rating rather than one like MMR where a smurf plays page 1 games immediately because he pwned nubs 50-0. I can't fix matching, but I can at least provide proper scoring so that when some smurf is jumped up to your level by Valve, or some nub feeder is placed on your team the game doesn't count much against you.
Yeah, but most noobs/feeders have stats off. How are you going to know your game got difficult because you have to carry the noob with stats off against a noob with stats on?
If you only play games against bad players then, yes many of them will have stats off. This means that you will never gain many points and will stay near your initial rating no matter how many games you play. Accounts that are 100-0 with all games against anonymous and low ranked players will be rated somewhat above the nub starting rating. That is all the higher they will get unless and until they play against good players.
The fact is when you play in top level games, nearly all players have stats on and have high ratings. To the extent that your games have missing data, your rating will change more slowly. If you are the only player with stats on, the game will result in a very small change in your rating.
댓글을 작성하시려면 로그인이 필요합니다.
This topic http://dotabuff.com/topics/2013-10-29-top-50-luna brought up a discussion about the current top 50.
I personally think it's pretty bad right now, as it doesn't say for real the people who play a lot with X hero and who have great winrate/KDA with it. It just shows high MMR players who have around 30 games with the hero, that's the truth. I've seen people with 30-35 games and 55% or less winrate in the top 50, which is just ridiculous.
I was thinking about it and started to play with some algorythm possibilites in Excel and I think it's pretty easy to make a better one for this, and I'm talking about one that doesn't depend on DBR, because it's gone after all.
Check out these simulations I made with my algorythm (a REALLY basic one to be honest, but still I think it's way better than the actual one):
5 player models and their scores:
http://i.imgur.com/8lmSVtN.png
How the number of matches impacts the score:
http://i.imgur.com/oNgszOe.png
Here you can see how the number of matches matters a lot in the beginning but with higher numbers it tends to have zero impact. This is because it's kinda obvious a player who has 2000 SF games doesn't deserve a score way higher than a player who has 1000 games.
How the win rate impacts the score:
http://i.imgur.com/qZD5SJA.png
Just a linear relation.
How the KDA impacts the score:
http://i.imgur.com/bslt4UW.png
And now two simulations with numbers I got here in Dotabuff from some players to use as example.
Zeus:
http://i.imgur.com/1IZh6Is.png
As you can see, even if I'm giving score based on number of matches to prevent a lot of 30 matches players, the top Zeus is still the same guy due to extremely high win rate and KDA. Merlini would come second but he has only 60 games while other guy has similar stats with 140 games, so he comes first. Wagamama (he's in the top50) doesn't even get close, since he has only 32 games and a win rate that's not so high.
Tinker:
http://i.imgur.com/3b6iAQu.png
Of course Chains is still first, this guy has 554 fucking games and insane winrate, the top 2 would still be the top 2, but N0tail and SingSing would be really worse than they are right now. Even I would be over SingSing, since he has only 40 games with kinda normal win rate.
Any thoughts? I'm not saying this algorythm is awesome: I actually think it's way too simple for such a complex matter. However, this simple math is, at least to me, better than the top50 system we have now.
PS: Some Relentless insight would be cool I guess.