General Discussion

General DiscussionOn how complicated matchmaking is and how hard is to assign rating

On how complicated matchmaking is and how hard is to assign rating in General Discussion
st@lker

    does anybody understand how complicated matchmaking would be.. people flame a lot about matchmaking.. i am a programmer and will tell u some of the challenges involved.

    1)Considering 50,000 people playing and an average game lasting around 45 mins, for every 5 mins, you have around 2500 players fitting in the pool ready to join a game.

    2)Now of these 2500, you probably would find around 500 or even less fitting the same server profile..

    3)Now matchmaking has to find 10 people with similar rating.. this is further complicated by the fact that dota has a very big learning curve and you put people distributed all over a 10 point scale... 0 for noobs and 10 for dendi

    4)Wait we have one more problem.. this list of 500 players may not be evenly scattered and all of these 500 might be above 5 with the exception of 10 oddballs who have 2 rating

    5)Lets look at another scenario: what if the players are scattered all over the place and it is hard to find players with a similar rating for a match

    6)Wait.. there is something called a low priority which should also be considered for this algo..

    I hope people understand how complex and difficult is matchmaking

    And as a afterthought on the rating..

    Levels: No. they dont matter at all
    Win Rate: No.. it doesnt decide where u go in matchmaking either
    {Edit: And the rating is probably assigned in relevance to rating of ur opposition team.If u pawn a team whose average rating is lower than you, your rating might not go up considerably.. but if do pawn team whose average rating is much higher than you, your rating should go up considerably}

    It is more complex and there are a lot of parameters. below is my brief theory on how it might work

    1) your K/D/A ratio (though this is overrated.. hard supports wont get great kda ratio)
    2) the total damage your hero has dished out to heroes and towers
    3) the total time you stunned your opponents
    4) the total time you disabled(slows and stuff) ur opponents
    5) the total heal you have given
    6) the amount of damage given out by your amplifies
    7) the amount of gold you spent for your team ( courier, wards, things like jango which give ms)
    8) other complex parameters

    yes.. these are recorded in matches.. look at the console for these...
    EDIT:and again these are taken into account only in relevance to your team rating and to your opponents team rating. dendi and me with a same stats on a puck mean nothing, until you take all the players rating in the match. These are not absolute, but relevant

    for example if you play puck.. this is probably how your are judged
    damage: spells and attack damage for towers and heroes
    stun: the stun after a coil break in addition to hex
    disable: the silence and the duration the opponent was in the dream coil

    lycan would be judged on his on the damage he deals to heroes and towers(pushing) k/d/a ratio
    lion would be judged on his stun and damage
    veno on his slow and damage (his performance would also be considered on how much damage his wards did too)
    enigma on he pushes (tower damage) and how good his blackholes ( a full 3 man blackhole would give him 15 secs of stun)
    venge damage and stun and probably his swap is considered a 2 min stun and his contribution to wards and courier and his damage contribution with his aura and negative armour

    and ur k/d/a ratio matters too

    these are a lot of factors..and it might not be exactly how it works.. but i hope it gives u the idea and the effort for a proper matchmaking to work

    TriHard

      If the MMR system worked as you described, we would've had a fine MMR system. Sadly I think that the MMR system is completely different and broken =(

      Slammer

        MMR sucks, I'm always with noobs or Russians and as such I am furious.

        Obadoba

          no fuck this mmr system look at my match history and the people i play with you will see the bull shit

          SeiRa

            I have a theory here? It may be a long shot though.
            But I think, that all people who are whining about MMR is broken, and that they ALLWAYS end up on the worse end of the stick, either don't play that well them selves, or are just ignorent enough to think, that the other team is full of dendis.

            And if you behave, you will never notice that they are russians, and if you bring in some good leadership to the team, you wouldn't notice the noobs, since they are following you orders.

            Take command over your games, and blame only your self for your stats.

            And if i may judge, if you are a playing below 50%, YOU properly the noob :)

            kCrawdad

              @SeiRa: I'll probably get flamed for this since my win rate is below 50%, but I don't think win rate is an accurate indicator of skill after all. That doesn't actually make ANY sense, assuming any MMR system exists.

              Let's look at two scenarios: One player is ranked very high, and often is matched with professional players. That player's win rate is 47.5%.
              Another player is ranked very low, and is frequently matched with really, really terrible players. This player's win rate is 68%.

              Who is the better player?

              (BTW, since I'm sure people will probably check my profile and try to prove how I am a bad player, I am NOT trying to imply I'm great. I know I'm not. But the point still stands that win rate has very little to do with skill level.)

              jams

                You cant have a win rate of 68% and be ranked low, not after any reasonable amount of games. In a perfect rating system, almost everyone would have close to a 50% win rate after enough games, regardless of rating.

                People with no clue about how elo rating systems work should stop coming up with stupid ideas before educating themselves on it.

                kCrawdad

                  Jams, you're proving my point that win rate is not an accurate indicator of skill. I was merely giving an example to show why that is the case.

                  edit: And it technically would be possible, as someone's current rating would (one assumes) be based on more recent performance (over the last X matches) far more than their overall performance, since skill levels change. 68% win rate =/= currently winning 68% of their games.

                  Ce commentaire a été édité
                  jams

                    Your conclusion was somewhat correct. Your example was absolutely wrong.

                    In a good rating system, while 50% win rate should be common, both higher and lower win rate are very good measure of skill. Consider it this way (in a theoretical perfect ranking system):

                    - at 50% win rate players should be close to their real rank
                    - at below 50% win rate, players are actually lower skill than their rank and they will keep on losing rating until they come to their rank.
                    - at above 50% win rate, players are actually higher skill than their rank and they will keep on gaining rating until they come to their rank.

                    Woof Woof

                      lets be honest theres no way to find a game with worse mm than dota s 2 right now

                      Ce commentaire a été édité
                      jams

                        'edit: And it technically would be possible, as someone's current rating would (one assumes) be based on more recent performance (over the last X matches) far more than their overall performance, since skill levels change. 68% win rate =/= currently winning 68% of their games.'

                        No, that's not the way elo rating systems work. You start with a certain rating, lets say 1500. Every game you gain or lose certain amount of elo, lets say +/- 10 per match (average 5). So your current rating is a sum of all your matches, even though it kinda tends to show your current skill. Winning 68% of 200 matches puts you at least at 1800 rating, which would be quite skilled. The only sure thing with elo type rating is that 50% means nothing after lots of matches. 68% means the player is skilled, period!

                        kCrawdad

                          Jams, I'll admit that I intentionally took my example to an extreme to demonstrate a point, but that makes it no less valid. If I simply change "68%" to "55%", the same point is made.

                          But regardless of whether a win rate means you are gaining rating or losing rating, your current rating could be anywhere in the range. So a player with lower win rate could still have an astronomically higher MMR than someone with a higher win rate. So still, absolutely no indicator of skill level.

                          And again, overall win rate =/= recent win rate, which would be more appropriate for determining MMR/skill level. If your recent win rate > 50 %, you're moving up in the world; if recent win rate < 50 &, you're moving down.

                          An example to show the disparity between overall win rate and recent win rate: I play starcraft 2 off and on, with months and months of inactivity punctuated by periods of playing it very, very seriously. My overall win rate is well over 50%. However, when I return from a period of long inactivity, my recent win rate becomes terrible (due to rustiness and being unfamiliar with the new metagame), and my MMR drops accordingly. So while my overall win rate was > 50%, i was actually losing rating, since my recent performance was bad.

                          So really, I would argue that a good rating system would make a 50% RECENT win rate common. Artificially seeking an overall 50% win rate doesn't make sense, since a rating system seeks to rate current skill level. And it seems to me that what we're discussing here is overall win rate, since that is what is displayed.

                          TL;DR: Overall win rate means nothing.

                          edit: Even if we accept that 68% rating (which again, was exaggerated to show a point) means you're moving up in rating, it does not mean you are rated high.

                          Ce commentaire a été édité
                          TriHard

                            @Seira since you're pretty much next level ignorant, I'll try to explain you how the game works since you probably don't play Dota2. Le random untainted player goes to 'Find match' and gets in games with russians who don't speak english (but queue in EU West or US East\West), don't want to communicate at all or listen to you. Or just are plain bad.

                            So what about the +50% Winratio? This is the dumbest argument ever. The only way to achieve this is to play 4\5man stacks every game. I certainly have played some 4\5man stacks but 90% of my games are solo queues that end up with 4russians in my team or other beings that reject to communicate.

                            jams

                              overall win rate for 95% of the players will be 50%, +/- a couple of %. Yes it means (almost) nothing. By seeking to rate the current skill level, the system will slowly put you closer and closer to 50%, overall. Unfortunately, that the way statistics work...

                              To take your Starcraft example - lets say (in our hypothetical) system you had 1900 rating with 400 games and 55% win rate (220-180). Then you come back all rusty and shit and your current skill level is closer to 1700 rating. You play games 20-40 score. You are now at 1700 rating, 240-220, win rate 52%. As you play you get back to your old shape, so you start moving up again, 40-20 score, back to 1900 rating, 280-240 - win rate 53,8%.

                              So you see the system moves you closer and closer to 50% win rate, even though your current skill fluctuated wildly. It even moved you from 55% win rate at 1900 rating to 54% win rate at same 1900 rating. The 50% thing isn't artificial, it's a byproduct of how the system works.

                              st@lker

                                @jams: what example was wrong??
                                your point on theoretical perfect ranking system is correct and it most probable aim of the implementation

                                Ce commentaire a été édité
                                rambosalad

                                  Your 6th bullet is pointless. LPQ would use the same algorithm as regular queue except they use two different pools of players. Other than that, I think it's pretty obvious how difficult it is to rate a player's skill level. There's too many variables involved.

                                  Ce commentaire a été édité
                                  daggius
                                    Ce commentaire a été supprimé
                                    daggius

                                      mm rating is very easy to do and has been done countless times before

                                      you win/lose, your rating increases/decreases an amount dependent upon the rating of your opponent

                                      do this 50 times and you have an extremely good distribution of the players.

                                      winning/losing is the ONLY thing that should be represented in the rating. The rating is an indicator of how good you are at WINNING GAMES. if you get 100 kdr and still lose you are worse than the guy who is 0/30 and wins.

                                      dumbass op

                                      partycat

                                        ^ yeah so if i put a monkey on dota, and had him queue with 4 professionals, that would make him good, right

                                        IlVers

                                          sometimes it doesn't matter at all how well you play because the game is very team oriented.

                                          but a decent player would definitely have a win rate of above 50% and a positive kda. thats not hard.

                                          your past games or your ego doesn't matter. if you didnt play as good as you could have, then dont chat in game like you're elite.

                                          simply apologize for not carrying ur own weight and if someone flames you on it, then ignore it because you know that that wasn't your best game and move on.

                                          its frustrating, and hard but if everyone can behave like adults, we can make dota 2 community a better one than LoL kiddies or dotacash. lmao

                                          SeiRa

                                            @Wicked Eye'd
                                            By a single click into my profile, you will se that i play :), but else good observations.

                                            If you use your clicking device (mouse), u can then click on the "players" up top. Try to click again on the win% thingy, and you will se, that the people with win% ranking 4 and 5 NEVER stack, and murs at nr 3 only stacks sometimes. And if you do have time, please continue clicking down, and you will see, that MANY people can win without stacking.

                                            Sure, these guys are VERY good players but with win% over 80% and 200+ games - It is possible.

                                            Don't know how long you have played dota. But if you're from Europe, you might have heard of Dota-league, an online dota mmr ranking. They only had an MM system, without the possibility of stacking.

                                            I, who truely is a noname, had 800+ games win% of 69%+. Dota is a team game, but a team is full of individuals! And map control and awareness is the key. And stop blaming others Wicked Eye'd. Especially russians, that's really becoming a BAD BAD trend, which i am tired of!

                                            Biggest Nipples

                                              @jams Finally someone who thinks the same as me!

                                              Swamp-Sh33p.hellcase.com
                                                Ce commentaire a été supprimé