updates will be implemented after ti4. meanwhile, enjoy this year's the international's! biggest esports event of 2014!
mmr should weight much, much more. 7ks can be mechanically weaker than 5k hero spammers, but they know how to play this game and win. it makes a huge differenc, under no circumstances i would choose a 5k hero mainer vs a 7k who never played this hero in his life (with few exceptions like meepo and brood).
decreasing the weight of matches played will open the path to abusers and people who randomly got good stats on a certain hero, so its a bad idea as well.
It would be helpful if you had specific examples (excluding pro players) of players with MMR/games played/winrates that felt way off and should be ranked higher.
For example:
5k MMR 250 games, 65% winrate
vs
7k MMR 70 games, 55% winrate
Who should be higher?
im not sure about the rankings, but from my pov, 65% longrun winrate at 5k is more or less equal to smthng like 40% in 7ks, maybe less.
on the other side, if the new ranking system will include the real best players on a hero, it will become boring for most ppl who do care about these rankings now. a 6k player with 2k games and 65% winrate is still nowhere near any tier1 pro player who barely touched this hero. a 5k dedicated hero spammer is no match for any high 7k-8k player, and there are a lot of them nowadays.
IMO mmr should be weighted as a curve with a steep climb to some arbitrary level point, I put 5k because you can't even call yourself good until this point, though 6k is noticeably better. And climbs less as you go.
for example
1k mmr = 0.1
2k mmr = 0.2
3k mmr = 0.4
4k mmr = 0.70
5k mmr = 1.0
6k mmr = 1.1
7k mmr = 1.15
8k mmr = 1.175
A similiar take can be done for matches played too.
<25 games = 0
25 games = 0.5
50 games = 0.65
75 games = 0.775
125 games = 0.875
175 games = 0.95
225 games = 1.0
275 games = 1.1
325 games = 1.15
375 games = 1.175
And so on. Having 2000 games would only mean like x1.2 Still a higher score, but only marginal, because what you learned probably will only be marginal.
Dota is hard, but not 2000 games to master a hero hard.
Something like that, obviously I just made up these numbers off the top of my head of what feels right, so adjust them or w/e.
BTW while we are on this topic, assuming you rank heroes by the recent year, you should probably weight win rate as well.
The difference between 45% winrate and 55% winrate on a hero is huge. Literally the difference between garbage and good. 55% and 65% is good and great. 65% and 75% is great and amazing. 75% and 85% is party abuse. 85% and 95% is also party abuse.
So weight that shit too.
They should do a similar thing as OpenDota where it resets every month I think, but you should have to play a certain amount each month because people are top 100 for a hero that they played years ago when the hero might of been really good
nah, resetting every month is too short, even their seasonal i think 3 months is a bit too short.
but all-time is too long.
i figure 1 year would be good.
Your 200APM may be useful in-game, but it's not the best for the site. Please wait a minute and try again.
LLMAO?
^ Yes but you need at least some history, 3 months is a bit too small I think. 6 months might be a better indication. Perhaps combine the minimum 25 games played all-time with a minimum of 10 (or something, maybe more?) during the half.
Every patch your hero history is LESS relevant not irrelevant - Just because the hero changes doesn't mean all of your experience and practice with the hero vanished.
This is why they have the score for your matches played. Because experience on a hero is relevant to how good you are with it.
However, if you were amazing at a hero years ago (devilish on meepo) but haven't played in forever, why do you deserve to be in top rankings just because you played a qualification match? (he is shit now btw, he played a few games and it was sad to watch lmao)
On the same coin, what if you were shit at a hero years ago, grinded him and now are good? You deserve to be anchored forever by garbage stats? My example would be myself on antimage, I used to maintain a 30% winrate on him, now I have 70% but my "all time average" is anchored to 59% because of my garbage start.
I mean it might be a big development piece but I think sorting ratings by patch adds a hell of a let more meaning.
1. Have an overall rating top 100 etc. wtv.
2. Have a filters for patch, and ranking (if you can)
Someone with 3k games on a hero who builds the same every game but has 20% winrate in a patch with his build/play vs someone who has found how to make it effective in a patch and has a higher wr% is so much more meaningful than clicking on some 3k who party stacks and hasn't changed his build in years.
Example.
Games played in this patch (1 month) = Relevant.
Top pro guy, played 3 axe games in 4 months = Not relevant.
I am using Axe as an example cause it's what I know but for example....
Play 1 game in 4 months again but still high up the top list.
The whole purpose of stats is to guide decision making, and these should be reliable, accurate and relevant. Most of the people lingering on the toplist for a lot of heroes do not fit this bill.
Someone who plays a game once a month but gets up there cause he's a "pro" or because he has previously playe "3000" games isn't useful to anyone. In a game of ever changing metas having access to the latest trends of successes in heroes is of far more value.
Myspace was one of the biggest, richest social media sites there was but is it relevant to follow their business model in todays times? Its a good reference point in the past but nothing more.
I agree completely. That's why I seriously think the rankings shouldn't be using all-time stats. And I believe it should be rated more on a half bell curve/half reverse J curve or whatever you would call it.
Something like this
There should be a "requalification" of like 15 matches for the season/half.
And every match beyond a certain time is cut-off/not calculated for the rankings.
I think you need to compromise between doing the patch and having enough matches played. Sure, you're right, the patch is important, but so is having enough data.
That's why I believe somewhere around 6 months would be good. Personally I'd choose 1 year, but 6 months would be fine too.
3 months/per patch is too short IMO.
Again, yes, you're right, the patch is important, but you will be lacking data and that will skew the results IMO.
It's a compromise.
The patch is a filter not the sample size.
Sample size should be 6 months with a minimum of 30 games played in 6 months. There should be filters for patches which (I know it might be a big ask) would reshuffle the list and include others who may not have made the overall list but were in the particular list for said patch.
Oh, yeah, just as a filter would be great.
Whenever I random a hero I don't play I just search top ranking players, find someone who played him recently and use his build Xd.
But yeah, as an optional feature that's great. Just shouldn't be mandatory because it would cause inaccuracy due to lack of data, but if it's a filter you can just skim through it yourself and decide who you actually think is relevant. Perhaps it will still show their rank on the global leader board too.
For example:
A = #1
B = #2
C = #3
and then you sort by patch you instead get
C = #1 (#3 global)
D = #2 (#15 global)
E = #3 (#284 global)
or something like that
Shouldn't be too hard to implement because they already sort by patch with other things.
And yeah, 6 months with 30 matches seems fine.
Makes me sad I can't break leaderboard with any of these heroes because the way it works right now :'(
Solo games should have more impact than party games.
Eg this guy - http://www.dotabuff.com/players/61134408/matches?hero=zeus
Is ranked #2 Zeus and undoubtedly his stats are impressive. But he's only* played the hero 162 times compared to most of the other top 10 Zeus players who have played 3 to 5 times more games of Zeus.
On top of that this guy almost exclusively plays unranked party games with a bunch of other 6k mmr equivalent players on the Australian server so usually get matched up against 3-4k stacks depending on the time they play, so of course they just grind easy wins and he gets above average stats on Zeus compared to what he would get if he was playing him solo against other 6k players.
Oh and Lawliepop I know you asked for non-pro player examples, but wings.iceice being #11 despite having a relatively worser winrate, 9th lowest KDA, and less games played (34) than everyone else on the list is a joke.
Just get rid of the bias to 'pro' players, stupid that you can play 50 games of some hero with 55% winrate and get on leaderboards because you play pro games which aren't relevant to pub hero rankings in the slightest.
seems the main issue of contention is ppl are given higher ranking because they are simply "pro". Same reasoning: just because u are higher mmr shouldn't automatically put u on leaderboards if the player doesn't really play the hero.
Well then it is the debate of whether a 7k player with 10 games of x hero is better than a 5k with 1000 games of x hero.
Then there's also overperformance and underperformance of hero, given the bracket you play in. Maybe the 5k player consistently overperforms on x hero in his bracket vs. 7k player underperforms on his hero. Who is better?
Patch influence should decay exponentially. Each patch contributing half as many points as the next. That solves the recent player/legacy player issue nicely.
Havoc is right, I want to look at player rankings to see how the best build/play a hero. the current system doesn't even reset after a hero get reworked.
^This is a good idea.
Just set latest patch to 1.0
and every one behind to be worth less and less.
Probably something along the lines of 1/.9/.7/.4/0
And then you don't even need to have a set cut-off for matches, but will end up being ~1 year anyway, with less points being assigned depending how old it is.
All you mods gotta do is adjust the formulae, EZPZ come on fix this shit
Actually the idea that it should reset makes sense. Well it should at least be something like mmr leaderboards. If u don't play a requisite number of games within a certain timeframe u shouldn't qualify for leaderboards.
It would never reset, but old data becomes un-scored. So the system will always be fresh/up to date. If you don't play in awhile, once the new patch rolls in your games will be worth less. (But so will everyone elses, so it won't cause any instability, but once everyone else keeps playing and you don't, you will fade from the leaderboards for being inactive)
You want to avoid a hard reset because that means shortly after the reset all of your data is unstable and inaccurate.
An easier and more useful thing would be to not have leaderboards and instead have a list of 6k+ players that play a certain hero a lot.
Doesn't even need to factor in kda or winrate just skill level + how many games in the last couple patches
nah it's just a fun thing to keep track of.
do you think athletes run at the Olympics just to "be fast" or because they want to be the best?
You can't be the best without comparing to others.
edit: in fact you can't even be fast if you don't know what slow is.
The hero rankings are handy even now in finding pub players to watch replays of. If they were made better they would be a more effective learning tool. Boosting your e-penis is just a handy side-effect.
im just salty as fuck cuz comparing to people on the leaderboards i should be ~50-60th place as antimage but I'm anchored by my bad stats from 2 years ago when I sucked ass.
Not even including ANY of the gradients/curves I suggested, and SIMPLY just cutting off anything older than a year would already be a huge improvement in the rankings. (Also include a minimum 25-50 games in the year or so)
^ not leaderboards SeemsGood SeemsGood
btw mods another fun improvement would be compressing your dotabuff links.
date=6month
&
hero=anti-mage
etc.
etc.
nobody needs to see that shit.
just use a number code/encryption
should look like http://www.dotabuff.com/players/13450139/matches?6AMAAM
or some shit come on guys
oh and another thing
what the fuck is even diamond 1-10
what's the point of even calling it diamond if the only place we see this tier is in the leaderboards?
nobody who isn't at least diamond is on the leaderboards.
and yea get rid of whatever ridiculous modifier you give to professional division.
what even is professional 1-10 anyway.
how is someone more professional than someone else lmao.
the factor depending on number of game should follow a logarithmic curve (i think it's already the case). and counting max ~500 games for that factor. (personaly, i would put a factor like "L * ( 1/3 ) * log ( 3*K ) " where K is number of game played, and L is a rectification factor so the max of this factor isn't too big, (and/or the min not to little).
Having 2k games will not really improve your play with the hero, compare to 500, while even 300 to 500 will. (ofc those number may vary depending on the hero. Mb higher for invoker, but probably much lower for wraith king. Those number also vary if the hero was flavor of the month or not.
But for the sake of a general formula, we don't need to dwell on some detail like that. First of all we need to improve the quality globaly.
Yeah, a logarithmic curve that's the curve I was meaning, I drew my graph wrong though.
You don't really need a cut-off, the graph will level off eventually.
If say 250 matches is a 1.0 modifier... 500 might be 1.05, 1000 might be 1.07, 2000 might be 1.08 etc.
the MM in normal games at high levels is weird and games are pretty much capped at high 4k, so u can get infinite good stats if u r any good
but if u exclude ranked matches my winrate goes from 70% to 65% :'(
rip stats
I think what sHOULD happen though is there should be a filter like havoc suggested, with one for ranked only as well.
yasp's rankings reset every 3 months, it's based only on mmr and not the divisions bullshit
and it's just better
yasp calculates the rankings basing on good facors, but the formula they use is just bad.
why mods no reply.
how hard can it possibly be to adjust the formulae for rankings.
this is clearly a major improvement.
or has dotabuff made so much money already they dont give a fuck anymore
MARLAN raises a some very good points and tbh I don't even know hero rankings but what you said would require extensive audit before it can be implemented. And even that audit would be fing complex.
If there code is designed with any organization or logic at all it should be pretty easy to change the formula for rankings.
Per favore Accedi per inviare commenti.
thanks.
things to consider
- don't use all-time stats, use the last year maybe.
-- i used to grind meepo, i have barely played him in the past couple years, but apparently i have more experience than someone who has played 50 games of him this month, kappa?>?
-- if you were shit a hero and improved you're forever anchored by garbage stats
- should be less weight on divisons/mmr tbh. i've seen 7k players play a hero much much worse than a 5k who grinds that hero.
- also less weight on matches played. a smart learner can learn in 100 matches what a 4head retard learns in 2000. the difference between someone who has 200 matches on a hero and 2000 is very small. because of this stat you're getting more players who aren't actually good at the hero but just grind the fuck out of it.
by the same coin, i guess a 2k mmr player with 4000 matches is better at dota than a 500 match 5k player ? kappa.
on a serious note: more matches on average means a better player. if you had to chose between someone with 500 matches and someone with 4000 with NO OTHER INFORMATION of course you choose the 4000 match guy. but that's about as far as that stat goes. if you had any other information matches played becomes less and less important because it comes down to how fast you can learn the game. and matches played doesn't represent how fast you can learn, and how much time you spend watching replays/watching videos/practicing in lobby... the stuff that ACTUALLY makes you better.
> suddenly hero rankings are better. you're welcome.
pls fix thanks