Rankings got no sense. 3500 mmr players are with some heroes in top 100 but 6500 mmr with same heroes better winrates are not in top 100 just ridicilous.
I don`t blame our moderators I m sure they do alot of work so that top 100 can be accurate, but they need to know they must work 3x harder so this shit can work a bit. That`s how it is I know it`s hard but.. its a shame 3500-4000 mmr players with also shit kda and win rate to let into top 100 but 6k players can`t get it.
We have a number of significant tweaks we're testing right now which aim to fix this specific problem (as well as other issues with rankings). Thanks for your patience!
There are many problems DB faces from a statistical stand point, for example being unable to differentiate between solo and party games due to the lack of such data in the API. Furthermore the brackets have to be aggregated from opponents who may not have stat-sharing, which means it might be based on their hero performance weighted against the average hero performance on the losing team.
The only two improvements DB can easily implement which I can think of is to remove brackets above Diamond 10, because the stats are taken from pub matches, and in that regard it makes no sense to consider a "professional bracket", especially when Dotabuff's definition of "professional" seems pretty random.
The other is to tweak the algorithms slightly less in favour of KDA (tactical suicides and other stuff are part of the game, perhaps make it more based on K+A weighted for duration of game) and to rebalance the formula for games played so that edge cases don't benefit so much.
my point is that the toplists are in no way accurate, I have found 5 axes who are actually worth anything and I have been watching them. I want to be able to see good players not 3k mmr players stacking with low KDA's winrates but with a high number of games on the hero.
I actually use the toplist to get ideas for changes to heroes or maybe where I am doing something wrong, don't want to watch 3k's nor smurf stackers who play every hero with dagon, that's not skill at all.
I think DB should actually consider removing all players from the rankings below say, Diamond 5, because even if the player in question is actually 5k but only stacks with 2k/3k teammates and ends up in High those games are not necessarily indicative of how the hero should be played in a proper setting.
is there any weighting on the age of games? if i was preparing the stat i'd give each patch a half-life in order that recent patch results would influence the rankings more than old games. it would improve things such as
a) avoid the problem of people who played the hero successfully in the past and don't any more.
b) would also help avoid the distortion of people who have spammed a hero by bringing down the influence of old games
c) would solve the problem of reworked heroes - it isn't really relevent to who is the best PL how they performed pre-6.82 for example.
to be fair to the dotabuff team, it is a very tricky thing they are trying to do but if the purpose of the rankings is to answer 'who is the best player of hero x at the moment?' then it is clear that the 'at the moment' part implies you need to have some time dependency in your algorithm.
I agree with Zenoth.
1) Weaken the influence of the number of games played
2) Establish a minimum division to be qualified to the top 100
3) Buff ranked games over normal games
Будь ласка, увійдіть до системи для написання коментарів.
http://www.dotabuff.com/players/128080734 3.3-3.6K MMR with 2KDA and 52% winrate.