There's a side point I'd like to address with your example.
I think there should be a ranked or laddered search and a seperate unranked search. In the ranked search you are split into 10 or an X number of tiers that is appropriate for the player base. The party can only have players within 3 tiers.
The example you gave might be a "fair" game given by MM, and both teams have a 50% chance to win. It is however based too much on the fact that 50% of the time Mr 2099 will feed uncontrollably and more than Mr 2112 or vice versa.
If Dendi queued with his grandma who has no idea how to play dota + 3 average skilled players, how would the MM ever get a "balanced" game. It might be able to make a "fair" game where both teams have a 50% chance to win in the long run. However it almost entirely depends on if his grandma feeds too much or Dendi gets fed by the other team too much. The "in betweeners" gets annoyed with his grandma's feeding 50% of the time when her feeding is too much, and gets bored being completely carried by dendi 50% of the time when he owns too much.
If anyone can stack and queue with anyone else this will happen, and it happens a lot.
Implying that:
1) Retards here can read large masses of text
2) They know basic logic/math
3) Even if your right, they wont troll disregarding everything.
Maybe there are 3-4 people who can handle all 3, but mostly this is kindergarten battle arena for all I care
lets get real this is team game so invidual ELO is nothing
no matter how good u are u will never win with 4 bad players in ur team (unless 5 others are even worse )
when queing there should be option to choose ur role like in world of wacrraft when u raid u choose ur role EG Healer /dps / tank /blablabla
so in dota it should be Mid/Offlane/Carry/support/junggler support
@Safe Base
1: There has been a long standing demand for a ladder system in dota2(which i highly doubt any time soon).
2: As mentioned there are too many variables for Mm to take into account while fixing matches. So it tends to generalize according to data available to it and even out the strength of the teams accordingly.
@Vaikiss`742.
Consider we are given the option to choose our role before game starts. Player X chooses support as his role. After game starts what should be the hero pool available to him. As proven by Navi a few days earlier a support Alchemist is pretty much viable albeit highly unconventional. One of the biggest strengths of Dota is versatility. The hero you control can play a vast number of roles.. some more efficient than others.
well if a guy chooses his role as a support then he should be familiar of what he is supposed to do and so that would prvent multiple players picking mid heroes and carries resulting in 0 supports in ur team (yes this happesn even in games with 2k+ wins)
Problem with picking role would be that if I pick support naga, I WILL get flamed that I didn't picked support. That happened with me just few days ago when I was playing abbadon support. So good idea, but it needs some sort of improvements, because multiple heroes can have multiple roles, and person who picked "support" might not be aware that being support isnt just getting courier at the start.
The data is contaminated, as there are high elo players that are horrible. The effect spread to the whole system very quickly.
What you need is a ladder system. Allow a 7-9 min mark timing window to forfeit unbalanced games. And check WITHIN game for the progress of each game see how they played their role.
There's still a long way to go ahead.
the problem is that these players that are below the actual mmr system's capibilities they should simply as a rule not be matched with people over 100 wins, If anything they should only play with other people like them. What's the point of sending a guy with a questionable mmr into a fight with ppl with 1000 wins.
Honestly, If I lived in Washington I would just go to Valve and tell them to hire me to fix this...they clearly seem to know why its failing, but not how to fix it.
I can only guess they are waiting for huge enough numbers of people to play so it works...or maybe they just don't care about the balance of very high solo que. It is a tiny fraction of the games. Maybe they don't judge it work the time to fix it.
For all the attention this gets the fact is its only bad for the best players...and even then only if they don't group up into a stack, which they should be doing.
-----------------------------------------------------
Or maybe I'm totally wrong. Is the matching a problem for Gold level type players? -- people who have near 50% wins. Do you get many games with some 10 wins player who feeds and ruins it for you? Is it really rare or common?
---------------------------------------------------------
There is always a weak player and a strong player...someone on their best hero or their worst and such. But the only time I really get games that are terrible is Sunday. When the low priority que people get released, then I get intentional losers.
I think that if Dendi ques with grandma...they get what they deserve. But to protect yourself from Dendi+grandma teams you have to que with 5 stacks. I think that it is good for dota for the system to push people into stacking more. The game is a team game, not a solo game and everyone should learn to think about it that way.
However, for the solo que option MM should just take the long (maybe 1 hr) time required to find a match of 4k ELO players for me when I use it instead of going 2 min and grabbing whoever happens to be in the que, and putting 3 feeders on my team to balanced it.
also i have found that when i solo queue (rarely but it happens) im either in a game with bronze scrubs or i'm with a four-man against a strong stack
Murs MMR is probably more like 6K, idk how you would ever find a solo que match in the simple version of my system. For me over the course of maybe 20-30 minutes 10 people of my skill level, in my region will go into solo que so eventually a good game is possible.
But for Murs I think you might just sit in the solo que forever UNLESS!!!!....
-------------------------------------------------------
This is how I would fix it. Every player chooses their own range of MMR they want for the match. If they set it too small (+/- 50 points) maybe they never get a match. If they want to play now, they set to +/- 2k MMR and get a crappy match in 10 seconds.
Market forces will determine what people decide for themselves is a good enough match and how long they want to wait for it. Everyone can get what they want, unless it does not exist. 1 min wait times for matches to a full 10 players all above 4k MMR do not exist because the players are not available.
In the aerospace lab at Purdue there is a sign on the wall. It says "We do things 3 ways: good, fast, and cheap. If its good and cheap, it ain't fast. If its cheap and fast, it ain't good. If its good and fast, it ain't cheap."
Btw, here is another bit I think would complement my "your choice" matching system.
It would really be great if it would tell you how many qualified players had been found matching you so far so you could decide for yourself if your range was too small. If you set it to +/- 200 and 10 min later had only 2 players...you could then decide to expand to +/- 500 so it would go faster...or keep waiting. Your choice.
@Relentless your range thing does not work very well for people with very low or very high rating.
@Karn evil 9: The people with "low wins" game you encounter are what Mm assumes to be either Smurf accounts or booster accounts. If Mm does not accelerate these accounts then it will lead to total annihilation of new players who have no idea of the game for a very long time(100 games? ) Most will probably quit by then.
@Relentless: Any wait time above the 7min mark will probably cause a riot.
Personally I feel a ladder system will solve a lot of the issues.. If valve decides to grind it that is.
Thanks for the Anti-Smurf article; it's good to know that Valve actually does something against those sissy players ;-)
The side effect of people complaining about being matched with 10-win-players (who mostly got enough experience from others accounts or DotA1/HoN/LoL to be in this MMR range) is negligible.
karkkilainen, yes people with extremely low or extremely high MMR would have to wait much longer for a good match...but its impossible to get a good one quickly. My system at least allows such people the option to wait longer and get a good match. In the current system extremely high or low MMR means you are forced to play with people far above or far below you.
The Team Matching also needs a challenge system. Instead of just blindly playing whoever is ranked closet to you...the top MMR teams should be visible. You send them a challenge and they accept or reject it. Then you won't waste your time as a 4k tMMR team stomping a 3.6k tMMR team and get zero points for the win. Also you won't have to play a 4.8k tMMR team and get crushed if you don't want to try it.
If you think you are up the challenge and the other team doesn't mind taking a game that will give them no points...fine play a really lopsided match. But if teams would rather wait for a better option to show-up (maybe 5 min after the bad match would have started)...then wait 5 or 10 more min and see who else enters the que.
Love the content guys. One thing I do hate with this current MM is. I'm at 1000+ wins, level 100+, 3000+ hours, 50.5% winrate. But i'm being forced to carry every single game, since I"m playing vs. people under 400+ wins or less than 500 hours. Also have to deal with feeders, emo & smurfs. Every 4-5 games there's a DC (4v5). I was wondering what's elo if you are a good player, but the current MM is forcing you into lower and lower tier.
I for one don't feel like I should play with anybody below 49% win-rate, under 300 wins, under 500 hours etc... Just really ruffles my jimmies when I check 9 of the other accounts and 7 of them are private with under 400 hours. I feel like new accounts are starting up higher in the tier than me and I don't think that's FAIR.
One thing I did notice in HON was the TSR drop/pluggers. People who had this mod, that showed TSR (Trueskill Rating) instead of the HON rating actually knew or it's about 60+% accruate which time would win. 8/10 the team with the higher TSR will win even if the teams had the same MMR roughly.
"Given this reality, if players used Elo to measure "progress", we would constantly be reminding them that they are NOT making any. That would be really bad."
Holy fucking shit
This man is an actual valve developer
Mother of god
Please sign in to post comments.
Ok for all the unhappy souls out there trashing the new Dota 2 Match Making system, here is something that I thought was really informative... so I am sharing with you all.
"It's pretty obvious that many player feel like the MM is broken and the latest fixes from VALVe's side made it worse. But is it true? Well, we know MM is not perfect, and it'll never achieve perfection. But it's not true that MM is not working as intended.
To explain it, MM uses some kind of ELO, if it's truely ELO, or some modified ELO. Knowing this we know already that we have some kind of rating going on next to the percentile.
But sometimes it happens that people with 300-1000+ wins play with people having not even 10 wins. What causes this? Quite simple MM can't accurately tell the ELO by 10 wins (approximately 20 games), but can start to accurately tell the ELO by 75 wins (approximately 150 games). So it can happen that new players played well in their "placement" matches and achieved already a high ELO that can't be true, but because the amount of games were played is not enough there's no accurate telling.
Further smurfs tend to appear as well, and most likely smurfs're those 10 wins players in your game.
But how is it possible that smurfs with 10 wins are being matched with people that have 300-1000+ wins? Well, VALVe is using some kind of anti-smurf system. What does it mean? Well it could be understood like this: Both systems are cooperating and when they notice a smurfer MM'll accelerates the "placement" matches / ELO, thus making it possible for accounts with 10 wins to be matched with people having 300-1000+ wins. Again, 20 games are not enough to accurately tell the ELO but if it's a smurfer MM is accelerating the whole process.
Now I want you to take a look at the example I'll show you now. A game with the ELOs of the players.
R D
3154 3094
2876 2913
2701 2700
2432 2399
2099 2112
We see the ELO of the players, but the first question that comes to everybody minds: Why is there such a huge difference in the ELO?
To hit the point: Radiant is a 4-man stack, where the player with the highest ELO and the lowest ELO is present, and the other two being the one with 2701 and 2432 ELO. The 5th man (a random player) is between the one with the highest ELO player, and the one with the lowest ELO. Never it happens that the random'll be outside of both.
Now MM has to find opponents who've ELOs matching the ones from the Radiant team. It can happen that it's done like this: 2x 2-man stacks (one being the highest ELO / lowest ELO player, and the other with two in the middle) and one random. MM tries to find opponents who's ELOs are equivalent to the ones from the Radiant, so based from the ELO the teams are even (R average ELO: 2652,4 - D average ELO: 2643,6).
So it happens in this game that the random can happen to flame the lowest ELO player for underperforming completely, but is it the fault of the MM? No, it isn't. It tried to make the game even, both teams having a chance of 50% to win.
But then it can happen that players (all 10 players similar skilled / same ELO) play bad. Nobody has a stable performance, further there're many variables that MM can't foresee that can affect the outcome of the match:
- Bad day of one of the players
- A player playing a role he can't play well
- Bad picks
- Bad communication (includes flaming)
- Individual play / teamplay
- A player was "boosted" by a friend to such a high ELO
By boosting I mean a friend playing with a lesser skilled friend and carrying him the games, so the lesser skilled player wins more games and it happens that the ELO is raising.
Sources that summarizes the infos on MM / Anti-Smurf system:
- MM: http://dotametrics.wordpress.com/2013/07/19/valve-employee-comments-on-matchmaking/
- Anti-Smurf: http://dotametrics.wordpress.com/2013/01/09/looking-at-dotas-anti-smurf-detection/
I hope that this'd somehow helped some people to understand how the MM is working, and to solve some misunderstandings from people on the MM.
If there're any mistakes in the text tell me, so I can fix them."
*These numbers haven't been compiled by me(kindly don't flame me).
**All credits to MeCJo (http://steamcommunity.com/id/meCJo/) for this wonderful article.