General Discussion

General DiscussionDotabuff Rating

Dotabuff Rating in General Discussion
Inuyaki

    @Infinite Regen
    https://dotabuff.com/matches/94209804
    ...
    No, they don't!

    PS I had very bad games myself, which my team won for me... But what if I never had bad games myself?
    The point is, w/l only can be pretty inaccurate. There are always games where the outcome doesn't reflect your play in that game.
    Btw do support players even have more problems with that. They just rely on their carries, you can't judge them by w/l alone!

    PPS another pretty good example:
    https://dotabuff.com/matches/2914532
    :p

    Este comentario fue editado
    GrappLr

      Infinite, I don't think that's right.

      Case1:

      Man starts in normal, keeps a 50% win ratio the entire time, stays in normal with score of 800 wins 800 losses.

      Case2: Man wins first 3 games, gets sent to high. He proceeds to win 25/27 games, gets taken to very high, he's at 28/2. At very high, his win rate drops down slowly, and he reaches a point where he's equal to his opponents, lets say his win/loss is now 78/40. For the next 700+ games, at around 50% winrate, it would not be unusual for him to reach 800/800, with a bit of bad luck, and still stay in the very high bracket.

      Both are 800/800, yet one of them is a total pub, the other is very competent. So it's not enough to know just the w/l numbers.

      ∞ regen

        Hmm, if that is the case I have no idea. I wonder, is this hypothetical or do you know of cases specifically like that? "with a bit of bad luck" a losing streak I presume which I would also assume would lead to drop in skill level they are playing at, so both who are 800-800 would be playing at the same skill level.

        I think its safe to say that anything being said about the skill brackets is speculation, unless valve outright explains how skill level is done I'm going to assume I'm right. Lots of assumptions here haha. You could be right though.

        Inuyaki, as I already said need a larger sample (like say your entire match history) for w/l to matter. 1 game doesn't mean much.

        GrappLr

          no, i know specific cases. I had one account that was 46% win ratio, in high bracket. I had another account 52% win ratio in normal bracket.

          GrappLr

            And they both had similar amount of total games.

            ∞ regen

              How many games? I know that the first 10 games or so count for more than other games do. Also if you choose advanced it automatically puts you in high skill, where beginner it doesn't. These might have effected it.

              Autism is great

                it doesnt matter very h, high, normal its 99% same shit in ap instead of 7/8/9 ignorant retards u get 4/5/6/7 of them 90% of time

                ∞ regen

                  Like razors that go 3-12?

                  Everyone has bad games, that applies to all levels, but my (admittedly brief) time with very high skill I've noticed people are significantly better.

                  AoeAoe

                    ^this: recently I play veryhigh and people indeed are much better (worst thing you find are 'tryhards').

                    ^^: w/l doesnt matter imho, some complex stuff is prolly used to measure stuff. My winrate is about 52% and for about month I was in normal, then more than month in high and just recently I moved to high/veryhigh. My winratio was always 50-55%.

                    MattieShoes

                      If you think wins/losses is all you need to evaluate skill, you need to rethink.

                      First, a hyperbolic example. If you play Garry Kasparov in chess, and I play your Down's syndrome brother... Is your 0-100 record or my 100-0 record remotely meaningful? Of course not.

                      Now matchmaking doesn't allow something that extreme to happen, but it still happens to some degree.

                      The next problem is your W/L record is cumulative. If I've played 1000 games over the course of years, do the first 900 games that I played two years ago really matter in terms of my current skill? No, it's how I'm playing NOW that matters. Yet 90% of my win/loss record is determined by those 900 games.

                      Sincerely,
                      Somebody on the low end of normal with over 55% wins

                      ∞ regen

                        Mattieshoes, your example isn't relevant at all. No mention of matchmaking at all, it isn't relevant. Because if you start at your down's sydrome brother and win, you won't be playing him again now will you? I got to say though, your second point is quite excellent, and I would love to see a recent win rate stat.

                        As said before your 55% wins isn't relevant, you don't have anywhere near enough games played to have accurate stats. W/l works just the same as any other stats, you need a large sample for it to be accurate.

                        W/L is better than some crumby stat that dotabuff makes up. Kills, assists, and deaths don't matter at all if you don't win the game. An example that is actually relevant: If I play Garry Kasparov in chess and take half his pieces and he only takes one of mine I still lose. Why? Because he checkmated me. Does this mean I am the better player because my "k/d" was better? Hell no.

                        Xeeoph

                          Make it public with an option to hide the rating. If someone has a low rating and wants to carry or play solomid, they should be able to prevent others from viewing their rating and telling them "no you can't play that role."

                          If this ends up causing the community to become DBR-obsessed in a way unfavorable to Valve or the game as a whole, it could simply be shut down?

                          Darth Vader

                            I think that those who favor win rate over a complex computational system do so over the fear that the system will rank them based on the skill level of each individual game played. One good example is this guy named "-emo.equinox" who regularly gets featured on the first page of live games. He plays with and against SEA verified players quite often, but his win rate is only around 48%...

                            يlѦيӉ

                              Another penis war incoming...

                              Layne

                                Public! If it turns into massive flames in games make it private as option.

                                < qt trap >

                                  Will we have it out in a few hours? I mean.. will it be out in the 23rd?

                                  AoeAoe

                                    In WoW, everyone has public rating on his/her 2v2, 3v3, 5v5 and it's not a problem,
                                    In LoL there are 'ranked games', which are also public to my best knowledge (as I understand, people can opt to play 'normal' or 'ranked' though).

                                    DotA is highly competetive and I'm sure there will be 12 years old hurt though.

                                    Another thing that I would love to see is number of 'DC/ragequits' though this can't be obtained parsing matches like dotabuff does.

                                    Naijan

                                      Infinite Regen, you surprised me with an adequate reaction. Thus I will dive into details:

                                      I know that the sample range of 1 can't prove a theorem)
                                      But: it can prove an axiom invalid - if it is an exception to it.
                                      You stated the following "the only statistic that actually matters is win/lose" in imperative form.
                                      I replied that this is wrong, and provided the exception.

                                      Yet, your next post can be resumed as "the only statistic that actually matters is win/lose AND number of games played". Spot the difference. It is MUCH more accurate now.
                                      It would be even true for 100% of cases, if the impact efficiency of player's skill on team performance would be the same for all players.
                                      What do I mean (example):
                                      - a player1, for some strange reason, can and plays only with 1-2 high-impact mid-game heroes. And he has a WR of 60% after 100 games.
                                      - a player2 is a great supporter. But as supporter he cannot really rely on lower-skilled teammates. His impact on own WR is lower than of player1.
                                      This player can stuck for awhile at 50%, because he will keep getting matched with people of that same skill level.
                                      Player2 even if of equal skill to player1, will likely have lower WR after playing those said 100 games.

                                      So it would be even more accurate to say: "the only statistics that actually matters is win/lose AND number of games played AND diversity of heroes used AND relative (per-hero) KDA to other players with similar WR/GamesPlayed". That's a quite long phrase, which we can put into one word: "Rating" :)

                                      Have a nice day)

                                      Im the Bully of my School

                                        i think its a good idea and yes make stats public plz thank you :)

                                        MRJ

                                          To look at evidence for win rate not defining skill, simply go and look at a few of the pro players stats (click players, then the confirmed list at the top).

                                          There are players there who have win rates below 50%. And they're professionals. Therefore it's not an accurate way of measuring skill, at all.

                                          If MM was completely broken and literally through a random group of players together, independent of skill level, THEN win rate would measure skill accurately, but because the MM system sorts players to some degree, win rate can't measure skill accurately.

                                          Litbull

                                            when will this be implemented

                                            Smbodγ

                                              Can't help but provide another explanation why W/L says very little about a player's skill, regardless of the amount of games played.
                                              Simply, since there is some hidden rating created by Valve, the MM system attempts to find the most suitable match, where players' skills are closest to each other. If an onboarding person wins a lot, he/she will quickly get high rating and will begin to lose more than win, which will stabilize the W/L. Actually the amount or wins across all matches says only how effective matchmaking system is. If it is close to 50% in some amount of latest games, it is perfect.
                                              Although there might be some exceptions.
                                              For example, if a player is super-pro, he will have much higher W/L because there is hard to find suitable opponents.
                                              If a player is constantly learning, his W/L will be higher than 50%, since the system is "underestimating" him all the time.
                                              Also there is a point which depends on the rating formulae. If it counts not all games but say last year of games, it changes W/L in random direction. If it is designed to drop rating more quilcky than award it, then higher W/L will correspond to higher rating, but we don't know if it is so.
                                              Also, the rating will be more sensitive of the player who prefers to play carries, since if he learns a lot, it increases the probability of a win greater than of a support player(Naijan's point).

                                              The simplest counter-example is just comparing players in low-skill and medium skill, each of which plays countless games with people of the same skill(rating) and doesn't progress much, staying withing the same bracket(as with most of players). They would have similar W/L but different skills.

                                              W/L would matter only if every game was gathered from random players from the online players pool regardless of their internal ratings.

                                              As to the topic, public ratings are required for good competition. Ability to hide personal rating though is required too, since there will be sensitive people getting trolled and so on, and noone wants angry or dissapointed people in the community.

                                              aethelwulf

                                                @Inifite Regen
                                                It's good that you bring up chess, because I would like to clarify two points about ELO systems and win/loss. Even using a strict ELO system, win/loss will not tell you everything about the player. Firstly, unlike traditional ELO, it appears that dota will start you at different beginning ELOs depending on your answer to your skill. Also, depending on the relative rankings to the opposing team as well as your distance from the standard ELO, each match will be worth different matchmaking values. So, the order of your wins/losses can have a significant impact on your final ranking. So what does win/loss tell us? Generally speaking, it will tell us slope, or whether your matchmaking rating is generally trending upwards or downwards.

                                                Finally, to all the people that want public ratings: We all acknowledge that there is possible downside to releasing public stats. My question is: What's the advantage to releasing public stats? As long as you know the population distribution and your stats, do you really feel the need to know everyone else's?

                                                ∞ regen

                                                  I see I won't be winning this dispute, people don't seem to look at w/l quite the same as I do. As for the comment about supports being less able to get a victory than carries: I disagree. They might not stomp a team like a fed carry, but good supporting can and will force your team to do well and likely win the game. My most successful hero is bane, and I have observed that good use of wards and roaming forces your carries to get fed, and thus winning you the game. Role played doesn't seem to change w/l, is my point (though hero played does!).

                                                  Naijan, "Yet, your next post can be resumed as "the only statistic that actually matters is win/lose AND number of games played"" I don't think you understand. w/l is a stat. A single match is not w/l. It is a stat that tells you wins and loses of a person. This stat INCLUDES number of games played. Your w/l is 31/24, I don't need a separate stat to know you have played 55 games now do I? (Note that the percentage is not w/l but a product of it)

                                                  Now regarding diversity of heroes, why would that matter at all? If I played the worlds best pudge, but didn't play any other heroes I am still the worlds best pudge, and I believe if that were the case my w/l would reflect this. Same for k/d/a, if it actually matters my win lose will reflect this. Now, does that mean that you can look at my w/l and tell my k/d/a? No, but will some score dotabuff is giving us give you that either? No.

                                                  Keep in mind, I'm not trying to be condescending; this discussion is great and I am seeing some flaws with w/l, but none that are addressed by some score (in my opinion).

                                                  Captain Tight Pants

                                                    @dotabuff staff
                                                    How many dbr points can you get/lose in one match? (avg/min/max)

                                                    Bulwark ♪

                                                      I'm excited to see Dotabuff rating but after the poll's results something stood out to me. The second highest rating, Platinum, makes up over 49% of rated players. If you added up all the Silver, Bronze and Gold players it still wouldn't add up to the amount of platinum players. Now, this is only counting the people who voted. So it may just be in this case.

                                                      Midway

                                                        @DotANote

                                                        Skill Bracket Rating Range Current Population
                                                        Diamond 1950+ ~ 1%
                                                        Platinum 1450-1949 ~ 23%
                                                        Gold 1200-1449 ~ 31%
                                                        Silver 900-1199 ~ 33%
                                                        Bronze 0-900 ~ 12%

                                                        https://dotabuff.com/pages/dbr

                                                        As you can see Platinum is 23% of the community. Now as for the voting number, that's to be expected as people who have played more games and more knowledge about the community are more likely to be aware of DotaBuff.

                                                        QWERTZcoRe

                                                          Make an option to hide your DBR from public and everyone will be fine.

                                                          BIGSLAPPY

                                                            No. . . .

                                                            Vimsey

                                                              There should at least be an opt out function. You put up a poll for people who already were using a site that essentially shows peoples skill level already yet you dont include the rest of the community by asking on reddit/joindota or the dev forums. Those you left out will be the recipients of the Brown/Orange treatment of HON. This is irresponsible and reckless.

                                                              Sacred Relics

                                                                +1 Option to hide