why no olee is russian stacker so just due to those 2 facts you have to lower his inflated winrate to see his real skill
russian - 30%
premade with low win players to get games against bad players - 30%
so his true skill is 35.20%
actually it is
dendi
vorksnak
tc
dexter
relentless (forgot you...)
bad to the bone
olee
1. V-god
2.Dendi
3.TC
4.relentless
5.dexter
6.bad to the bone
This guy sucks 5 man stacks LHP.every game same strategy with 4 other guys. my shit shits on him on a real game.---> oleee444
dont even bother to rank him
1. dendi
2. v-god
3. TC
4. Dexter
5. Bad to the Bone
6. Relentless
145,389,281. ole- smurf stacker, exploiter, limited hero pool, dodger
I can see an algorithm that conforms to human judgment being exciting (if it even does), but what would make you think that justifies its efficacy in actually differentiating skill? Let me put this in a different light: It may be easier to predict what the average American thinks AMD stock will do than it could be to actually predict what it will do. Why is the weight of our average words so heavy for you?
In my humble opinion:
1) Dendi - Massive amount of games. Good win rate and kda against top tier enemies. 9 pages of games this month and for that many games he doesn't stack that much.
2) Liquid TC - Great win rate and KDA. Not that many games compared to some others, lots of stacking, some games against lower level enemies (stacked with low MMR friends).
3) D3xtr - Good win rate with a high amount of games. Recent win rate is much lower though.
4) Vroksnak - Tons of games and a good win rate, plays against some top tier enemies.
5) Bad to the bone
6) Relentless
7) Olee444 - Not that picking an old Luna/BM/venge type aura push strat or constantly stacking is a problem but more his games tend to have very low rated players.
Most recent game included someone with 1.4k games and 48.04% win rate, 2 games before that a mid SF with 1.3k games and 49.29% win rate. Even if these 2 players are in a 5 stack the other players in their team don't seem to have that high an MMR either. Another random game 4 days ago have a 400 games 47% win rate player.
My list came out as: (I've added everyone from this topic to it)
1. Liquid TC (quite a bit in front)
2. Dendi
3. Vroksnak
4. d3xtr / Bad to the bone pretty much same score
5. Olee
6. Benao
7. SafeBase
8. Darkness
9. Volvo
10. Relentless miles behind, 47% participation in team kills when everyone averages at about 58+%
I will continue working on the system and create a thread when it is complete. Thanks guys
@Vandal
I see what you are saying, the results i got were close to how I would rank these players but everyone would have a different opinion and I wanted to hear other people out
Good player = mannered player. So:
TC
Dendi
Relentless
D3XTR
Bad to the bone
Vrosknak
Olee (get him out of the list pls)
Its a rough one to gauge to be honest, big kudos to you for trying but I would try someone other than relentless because we are talking about a very methodical support player here.
http://dotabuff.com/players/103176226 - A lot of tree, lich and to add to that probably some of those as offlane.
How much "participation" do you think a lot of these heroes would have. And I think you may want to get a good and accurate representation of how you define skill because it's like the guy from pubstats who basically said to never pick supports if you want your pubstats to look impressive as there is no system that will allow you to shine with playing supports and participating heroes.
On a side note I notice a lot of support players (myself included) enjoy playing Luna, wonder why that is.
@DR INTERNET
Have you quit sharing? I was watching a game of yours yesterday with s7 or at least I thought it was you.
http://dotabuff.com/matches/332011910 The abbadon I believe.
Yes mr. Stalker.
bad to the bone: I'm quite interested in how you can be tied with D3XTR. 64% wr with 50 games is quite low.
Based on his other account http://dotabuff.com/players/104307918
Also, I watch most of s7's games and you were in one of them.
I hope there is something more to it than simply teamfight participation because that is not a reliable metric.
Other things come into it, i'm not gonna disclose the formula but from what i see it is fair. People who play support heroes all the time would suffer because there is no reliable way to rank them (steam API doesn't provide with data such as wards etc)
In game stats should not contribute in any way to a proper ranking system. It should be based entirely on strength of schedule. It is very difficult to appropriately calculate strength of schedule without knowing who a player played with and against...who those people played with and against, who those people played with and against etc.
If you want to rank for a particular hero it makes sense to include in game stats. then part of the score should be strength of schedule and part of it in-game stats. But to compare in-game stats crudely across all heroes is really meaningless.
Diamond level Omniknight averages 1.64 KDA on 52.42% winrate. Diamond level Animiage 2.54 KDA 43.84% winrate.
You could easily have an excellent omniknight player with lower KDA than a feeder Antimage player. This is why you cannot compare in game stats across heroes. People tend to focus on KDAs...but it would be just as silly to use healing as the metric for who is good. Players who use Dazzle, Omniknight, and Tree would look amazing and carry players would apparently suck. You just can't compare in game stats for all heroes.
Even when you are comparing the same heroes, in-game stats alone only have a partial contribution. Lets consider Zues stats.
Average Zues stats for Diamond is 56.93% win and 3.45 KDA (dmg/min would be very nice to use also for Zues, but its not on dotabuff right now for individuals)
Now if we go by win rate its...
▲Vroksnak▲ 11-3 (78% win) 4.26 KDA
Relentless 18-6 (75% win) 4.97 KDA
Bad to the Bone 7-3 (70% win) 3.32 KDA
TC 6-3 (66% win) 4.34 KDA
Dendi 5-8 (38% win) 3.23 KDA
D3XTR 2-4 (33% win) 2.78 KDA
Olee444 (Zues to hard? did not attempt)
If we go by KDA its...
Relentless 18-6 (75% win) 4.97 KDA
▲Vroksnak▲ 11-3 (78% win) 4.26 KDA
TC 6-3 (66% win) 4.34 KDA
Bad to the Bone 7-3 (70% win) 3.32 KD
Dendi 5-8 (38% win) 3.23 KDA
D3XTR 2-4 (33% win) 2.78 KDA
Olee444 (Zues to hard? did not attempt)
But these scores do not show how hard the games were. Am I really better than Dendi at Zues? Maybe...I mean it does seem to be one of his weakest heroes, but he could have played it only in his toughest games where he tried to carry his nub friends against stacks of pros. We don't know without the strength of schedule being measured.
you can't really just go by the strength of schedule when you allow people to stack
all the players in the stack would automatically come closer in their rating and if they kept playing together they'd eventually all end up at the same rating, which would be okay if those players would play 100% of their games together but you cannot make that assumption
another aspect is that games are balanced to 50% except for the very top of the spectrum
so whether you stack with bad players or good players you don't necessarily have an easier game, so losing more points when losing to worse opponents isn't necessarily justified as well
if you actually play in a really good stack you get to play against higher MMR opponents while your game remains very easy because your stack is simply better
so unless you only use a ranking for solo queue, you don't get better results by just looking at the schedule
obviously kda and other things are heavily influenced by playing in stacks as well but you get an overall better picture if you look at all the data you actually have
No, when you play with a good stack you play against other good stacks on page 1 of live games. Strength of schedule is the correct way to determine rating. If you have games with a stack against a weak stack or random players a proper ranking system gives you no points or nearly zero points for the win.
In my new ranking system, such stacks do not have super high scores...except the very best of them who won nearly all those games against teams with 2 or 3 pro players stacked against them also. If you do it right, excellent non-stackers get higher scores than smurf stackers with fake stats.
Now strength of schedule is very hard to measure without access to the full database of games...and dotabuff itself doesn't even have the full record because Valve blocked it. DBR froze because their calculation technique was impossible without complete information. I have actually invented 2 new ones the don't need complete information, one I explained here...
http://dotabuff.com/topics/2013-07-22-how-to-bring-back-dbr-a-practical-approach
But Valve blocked that now too by hiding skill brackets so I created another totally new system that does not need complete information. But its not fully tested yet...even so it needs a lot of information, you have to measure strength of schedule.
But in the absence of a real rating system you can still do some "versatility" type comparisons. Lets see how many heroes these players have excellent scores for...lets say they need 15+ games, better than 60% win rate, better than 4 KDA. You can see that this mark is actually hard to get. Heroes with only a few games sometimes have "lucky" stats.
(15+ games, 60%+ winrate, 4+KDA)
D3XTR 10 heroes
TC 9 heroes
Dendi 6 heroes
Relentless 5 heroes
olee444 4 heroes
Vroksnak 3 heroes
Bad to the Bone 1 hero
Now exactly what metric you pick for versatility has a big impact...i mean TC has 80+ heroes with over 4 KDA, and 80+ heroes over 60% win...but hes had not played 15 games on many heroes. Only 1 of the heroes TC had played 15 games did not have 60% win and 4 KDA. If we lowered games required slightly he would have a lot more. On the other hand Olee444 has basically only played the 4 heroes he has that qualify. Vroksnak plays very difficult games, if we dropped winrates a bit he would have a lot more heroes on this list. Strength of schedule would show that his games have high value.
People don't show a lot of respect of D3XTR but he is clearly an excellent pub player. Still, if his stack is making games easy for him, proper strength of schedule would showi it.
If we had a proper strength of schedule measurement I'm sure my games are worth far more than Olee444 and Dendi's games worth far more than mine. But without actually doing that comprehensive calculation we can only speculate.
Still this versatility test is nice...you can see the power of the best pub players to dominate with a huge range of heroes.
Murs 30 heroes (15+games, 60+ win, 4+ KDA) http://dotabuff.com/players/90586993
Beesa 35 heroes (15+games, 60+ win, 4+ KDA) http://dotabuff.com/players/43908335
Relentless is there a way to see the amount of games, win rate and KDA for all heroes simultaneously?
Yeah same here I have a macro that pulls the text from a page then i just mess around with formulas and stuff to get stats. Its fucking arduous but at least you can get an accurate count of things.
you just have to look at dbr or the top 50 for each hero
they might use kda to a lesser degree but in the end the main factor is win rate and the MMR/DBR of the other team (and your own team, too)
win rate isn't actually used as a factor but obviously winning and losing games being the main factor here when it comes to gaining/losing rating (effects vary on the level of the game obviously)
there's obviously also a strong correlation between kda and win rate (if you actually try to win the game and not play to safe your kda)
and yeah the games definitely get easier if you stack really hard and try really hard, too
there's a reason people can maintain 80% win rates in stacked games and there's no way anyone could maintain that solo queuing
the top players in these rankings usually stack really heavily favored games while these wins still actually net them more points than hard as fuck solo queue games with way worse players on average
on average the system isn't that terrible because you can assume that good players usually stack with other good players and that its weaknesses are as exploitable by everyone trying but it doesn't reliably say you who's the better player in the top segment
i don't know what the appeal is of trying to compare players who play entirely different games with entirely different mindsets, attitudes and structures
like the people looking at someones kda/win rate on a certain hero and then comparing them to their own and thinking that they can judge who's better at the hero
you won't be able to create a system that analyzes the in game stats and tells you whether tc or dendi is actually better
you can prioritize things and manipulate them until you get the player you think is better to a higher rating but that doesn't change the fact that you simply have way too little information to say anything of any value
Zaloguj się aby dodawać komentarze.
I'm developing a dota 2 ranking system which will rank players. I've got a fairly good working formula for working out the points, i'm curious to see if it holds up to the public opinion!
View their latest matches (obviously you can't view every single one but try and have a look at 2-3 pages) and put them in the order of:
1. Highest Skill
.
.
7. Lowest Skill
please use names and not steam ids!
http://dotabuff.com/players/67601693
http://dotabuff.com/players/103176226
http://dotabuff.com/players/70388657
http://dotabuff.com/players/33706185
http://dotabuff.com/players/86738494
http://dotabuff.com/players/145389281
http://dotabuff.com/players/104307918
Thanks