Page 6 of 6 FirstFirst ... 4 5 6
Results 51 to 59 of 59

Thread: MMR very low after calibration, even after winning 9 out of 10 games

  1. #51
    Basic Member
    Join Date
    Mar 2014
    Posts
    16
    Quote Originally Posted by Redthirst View Post
    There are a huge difference between low level and high level pubs. You can't create common standarts for all skill levels without system being terminally flawed.
    Can you explain this in a bit more detail? I think that the difference comes mostly from the skill difference and nothing else. Especially when players play solo. And because we are talking about how to rate the skill of players, I think this helps with it.

    Are you saying that if you put a very good player together with 9 noobs, and he stomps the enemy team, his stats for the game would be very different from an easy stomp on his own level? And for the same question for other roles holds too.

  2. #52
    Basic Member
    Join Date
    Apr 2013
    Posts
    2,309
    Quote Originally Posted by totemizer View Post
    Can you explain this in a bit more detail? I think that the difference comes mostly from the skill difference and nothing else. Especially when players play solo. And because we are talking about how to rate the skill of players, I think this helps with it.

    Are you saying that if you put a very good player together with 9 noobs, and he stomps the enemy team, his stats for the game would be very different from an easy stomp on his own level? And for the same question for other roles holds too.
    The thing is, it won't be accurate to judge skill level that way. For example, we give the highest amount of MMR to a player who goes by the exact same build as pro player. But what's next? How are you going to judge how many MMR points other players will recieve?

    Also, there are other problems. There is still problem of heroes who can fulfil several roles. Your system will check that Venge and won't give any MMR points because she's going for a semi-carry build and therefore doesn't have many support items.

    There are also heroes who can change their role throughout one match. Like Alchemist can be a roaming support during the laning phase, but then catch on thanks to Greevil's greed and go for a carry.

    And there are many builds available. Most heroes have more than one build available. And what if there are several pro players playing a certain hero, and each of them has different build?
    Possible solution to foreigners issue
    In the grim darkness of the far future there is only war.

  3. #53
    Basic Member Shmendrich's Avatar
    Join Date
    Oct 2012
    Posts
    2,579
    Quote Originally Posted by totemizer View Post
    It's easier than one would think. You don't need to go and do switch/cases like Raisti suggests. All you have to do is to constantly monitor strategies good players do (that's statistics and it's quite easy to implement it), then you find the patterns, which also can be automated with a bit of human aid for the edge cases, surely not a full time job after the initial system is in place. Then after a noob's game is over, you check against the patterns and find correlations. If there is little correlation you don't give or take much mmr points.
    The problem with that Totem, is that the BEST Players are the ones who get out of "Common strategies" and Improvise whats NECESSARY according to the circumstances. Its like the guy who said "Oh shadow shamans who level up shackless are noobs" Because he saw in DotA2 guide that they recommend you level it up last and he could not think anything beyond that. Thats the recommended/common strategy.

    But its not that simple, sometimes you need shadowshaman to focus on disabling (hex/shackles/ maybe even guinsoo) and sometimes you just need him to hold a lane (in which case yeah shackles is useless).

    It would be relatively simple to implement a basic system that took some of those things in to consideration, but there's a BIG difference in taking them in to consideration, and ACTUALLY making it balanced.....For example if you ask me I would rate a GOOD support with double the rating as a GOOD carry...because I PERSONALLY think support is way more difficult to play (properly). But thats just an opinion/assumption with no mathematical data or any way to actually prove it or any way to measure exactly how much more score should a good support be.

    Its totally worth to do something like this totem, but its not DotA2's current business model. Games like Starcraft 2 DEPEND on a GOOD MM system because their business model is to sell that, competitivity.

    DotA is a very competitive game, however DotA2 is not focused on that, they use the international (which is the competitive side of it) to fulfil their actual business model, which is selling hats. (So many people buy compendiums because of the shiny hats they can get)

    And if you analyse nearly every change Valve has done to DotA, all has been to make it more casual and noob friendly, even in ranked we have no ladder because thats something that scares noobs off, they dont want to feel the pressure of having to climb up a ladder to get in better ranks and play with better people (like in starcraft 2 where the top players are chinese guys who are nearly robots) so this continuous scale system make it seem more friendly and less like a mount everest of gaming.

    Tho...I do HAVE hope after seeing how much money they are making with this international (remember only 1/4 of the money of the compendium went to prize pool, meaning all the other 3/4 is for them X)) so I'm hoping 2 things.
    1 That they will throw the DotA2 development team a bunch of money to spend on developing (which, I think WILL happen, its simply investment)
    2 that they will see the light and use it in actually useful stuff and not just more way to sell us shiny hats (Which...sadly I think they will invest it in more shiny hats and systems to sell us shiny hats...but hopefully not X))
    Make a NON automated report system.
    I would rather wait 10 minutes for a balanced, fun and exciting match that will last 1+ hours. Than wait 10 seconds for a poorly balanced crapmatch that will end in 30- minutes
    Valve, How I'm I supposed to respect certain boundaries to prevent a mute if I dont know them because they are different for each and every person on each and every game I play?
    The only type of player that dislikes having his stats public are the players that suck and cheaters

  4. #54
    Basic Member
    Join Date
    Mar 2014
    Posts
    16
    Quote Originally Posted by Redthirst View Post
    The thing is, it won't be accurate to judge skill level that way. For example, we give the highest amount of MMR to a player who goes by the exact same build as pro player. But what's next? How are you going to judge how many MMR points other players will recieve?
    I don't think "builds" should count much, be it item or skill build. That can vary too much.
    What I am talking about is stats. The same stats they use now, but I would use them in a different manner.

    Quote Originally Posted by Redthirst View Post
    Also, there are other problems. There is still problem of heroes who can fulfil several roles. Your system will check that Venge and won't give any MMR points because she's going for a semi-carry build and therefore doesn't have many support items.
    Again, I would analyze (automatically) what the player tries to do, and not about what's the hero supposed to play. Of course there are patterns for each hero, but there are also patterns which are independent from any hero.
    Quote Originally Posted by Redthirst View Post
    There are also heroes who can change their role throughout one match. Like Alchemist can be a roaming support during the laning phase, but then catch on thanks to Greevil's greed and go for a carry.
    And of course the software should and would recognize that. It's just another common pattern
    Quote Originally Posted by Redthirst View Post
    And there are many builds available. Most heroes have more than one build available. And what if there are several pro players playing a certain hero, and each of them has different build?
    Again, I don't think the "build" matters too much. Builds vary too much. What I talk about is patterns of statistics of good players.

  5. #55
    Basic Member
    Join Date
    Mar 2014
    Posts
    16
    Quote Originally Posted by Shmendrich View Post
    The problem with that Totem, is that the BEST Players are the ones who get out of "Common strategies" and Improvise whats NECESSARY according to the circumstances. Its like the guy who said "Oh shadow shamans who level up shackless are noobs" Because he saw in DotA2 guide that they recommend you level it up last and he could not think anything beyond that. Thats the recommended/common strategy.

    But its not that simple, sometimes you need shadowshaman to focus on disabling (hex/shackles/ maybe even guinsoo) and sometimes you just need him to hold a lane (in which case yeah shackles is useless).
    I mentioned this before. What I would do is 3 steps. 1. find the patterns in the stats of the good players 2. find correlation of the noob players stats with those patterns 3. THE STRONGER THE CORRELATION is, the more you can value the differences.
    If there is little correlation, then the player is either trying out something completely new, which I think is ok, and if they win, they should be rewarded and if they lose, they shouldn't be punished too much, OR maybe they are complete noobs which can be differentiated just by looking at their level/xp and in this case you can punish if they lose and not reward too much if they win, because it was probably luck.

    So in the case you brought up, there would be not much correllation, but the system would know this is a good player already and it could just assume they are experimenting and if they lose don't punish them (you want good players to experiment) and if they win, reward them a lot.

    What you almost nailed is that this system would reward players who play the same thing over and over again, but again, the rules for experimentation would differ a lot for different skills. When you are a noob, or even middle tier player, you still have to learn, only the top 10 or 20 percent knows the game that much so they can come up with really good strats.

    And don't forget, my system would take into consideration if players win, and even if someone from a lower tier comes up with a winning strat, they would not lose points for a win. Especially not because even if the stats would not correlate much with anything, there should be some display of large game impact, you can't win a game with a strategy of not doing much


    Quote Originally Posted by Shmendrich View Post

    It would be relatively simple to implement a basic system that took some of those things in to consideration, but there's a BIG difference in taking them in to consideration, and ACTUALLY making it balanced.....For example if you ask me I would rate a GOOD support with double the rating as a GOOD carry...because I PERSONALLY think support is way more difficult to play (properly). But thats just an opinion/assumption with no mathematical data or any way to actually prove it or any way to measure exactly how much more score should a good support be.
    I somewhat agree with you, but I don't think any kind of mmr system should have hard coded numbers in them like "twice as much". It all should be dynamically calculated based on the circumstances. It's much harder to play support when your carries are feeders than in a game when they at least know to check your mana before they dive


    Quote Originally Posted by Shmendrich View Post
    And if you analyse nearly every change Valve has done to DotA, all has been to make it more casual and noob friendly, even in ranked we have no ladder because thats something that scares noobs off, they dont want to feel the pressure of having to climb up a ladder to get in better ranks and play with better people (like in starcraft 2 where the top players are chinese guys who are nearly robots) so this continuous scale system make it seem more friendly and less like a mount everest of gaming.

    Tho...I do HAVE hope after seeing how much money they are making with this international (remember only 1/4 of the money of the compendium went to prize pool, meaning all the other 3/4 is for them X)) so I'm hoping 2 things.
    1 That they will throw the DotA2 development team a bunch of money to spend on developing (which, I think WILL happen, its simply investment)
    2 that they will see the light and use it in actually useful stuff and not just more way to sell us shiny hats (Which...sadly I think they will invest it in more shiny hats and systems to sell us shiny hats...but hopefully not X))
    There is this thing that I think people assume everyone should play ranked mmr. I think that Valve should actively encourage players to NOT play ranked. I mean playing competitive games and getting good results is it's own reward. Those who opt for more casual games should get either more xp or more random hats, or just some other nice bonuses which you can't get from ranked.

    And I am very skeptical about how much they would care about spending developer time on anything than implementing the remaining heroes and adding more hats. Although, the new spectator features are really great and I love them. So it's not all evil . Really, the only thing I dislike about Valve is how much they screw up the mmr system. Everything else seems fine to me, but this is very annoying.

  6. #56
    Basic Member
    Join Date
    Jul 2013
    Posts
    138
    I think the mmr is working, because on my smurf i won 4/10 ranked games and have a 47% winrate yet i got calibrated right to 4.1k

  7. #57
    Basic Member
    Join Date
    Jul 2015
    Posts
    1
    Quote Originally Posted by GeforceTiny View Post
    On my smurf account I've gotten a 100% win rate so far and I'm lvl 8. All I've done is play co-op bots and non-ranked. My KDA is always high. It will be interesting to see where this experiment takes my smurf MMR. My main account is just over 3K.
    How did that go?

  8. #58
    Basic Member Kryil-'s Avatar
    Join Date
    May 2012
    Posts
    8,065
    Quote Originally Posted by @Vic View Post
    How did that go?
    Not well. The issue with leveling by only doing co-op bot matches has been fixed. If you actually level to 13 without playing a normal game you'll have a requirement of 40-50 games before you actually can play ranked.
    If you think I've closed or deleted a post unjustified, feel free to PM me and I'll explain it to you/discuss it with you further.

  9. #59
    I saw some people say KDA doesnt matter much, but I think this is (under certain conditions) wrong. I have an account that I play with since I first played Dota2. In the beginning, I played terribly bad with terrible things and only much later got better at the game. My MMR ended up being around 3.2K.

    I wondered what would happen if I hadn't had such a bad start but played well right away? I made a new account to test it. I didn't have the time to get it to lvl 13 yet, but I realized that after playing just the first few matches well (and by that I mean with 15 kills or more and a minimal amount of deaths), I had moved from the normal bracket to high skill / very high skill on Dotabuff. So it's not just about how much you win, but also how well you play within those won games.
    I never tried how it works when you play support, but when you play carry, your performance outside win/loss definitely matters, and I found the KDA to be a big factor.

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •