Well say player has a score of 25-0-1. The whole team has 34 kills. So if we play off assists then he didn't contribute much to the team's success did he? So we can't just take assists into consideration.
Yes GPM and XPM would be linked with lasthits but it is also linked with participation. For example if you farm all game and you have a high GPM/XPM but have a score of 3-0-1 when the team has 25 kills then your participation score will suffer.
"Yes GPM and XPM would be linked with lasthits but it is also linked with participation. For example if you farm all game and you have a high GPM/XPM but have a score of 3-0-1 when the team has 25 kills then your participation score will suffer."
This can be good and bad. For example, if you're a hard carry and farmed the entire game only to participate in a major teamfight which wins your teams the game, then your value is being undermined. Certain heroes like Phantom Lancer will also suffer from this type of system since they win games not by participating in fights but by pushing.
I should probably refraze this. the participation score wouldn't suffer - it would balance out your total score since you would have higher gpm/xpm due to farming all game.
@kiNe I've fixed a bug and had to reboot the whole queue. 190 players at the moment * at least 2 minutes each - that is over 6 hours max for your score to appear.
If its like that, then I think this rating system could be decent for positions 1, 2 and 3.
Ranking supports might be hard because some supports could have really shitty scores, really shitty gpm/xpm and sometimes lack in supporting items (wards) but played a big part in winning the game. I don't think its possible to accurately rank supports due to the large variation of scores they might give. E.g. a defensive support dedicated to stacking camps and protecting the carry.
Yeah but the way I plan to develop support rankings is: you are ranked against other people playing the same heroes. So say you play tree - you get a score for it and you are ranked against other players' tree scores.
eh i have 3 and in my last 20 games i've played with/against puppey, viggoss, black, garter, mitch, gordon, poseidon.sakura, T5.Darkfalco
all solo queuing (not solo MM preference)
while i've seen people who play normal bracket games with 7-8
Only problem with comparing tree to other trees is that a killstealing support, or someone playing venge as semi-carry, etc will undermine this, will still have better score than dedicated supports. You just cant rank supports using statistics.
Or you could add some extra points when detect a stat with high assist low kill and a player in team with a lots of gpm/kills. This one would prefer people who are carried by their friends... Still not good.
I guess it just can't be done, but hope you find some way.
Anyway, good idea, good for anything but supports. :)
thanks guys. To be honest i don't expect to develop a perfect ranking system as i said - just have a look at the current dota 2 matchmaking rating system that everyone is hating on lol. There is no way to build a perfect system. But this is just something fun i guess.
hopefully i will figure out something for support heroes.
I realize you wanted to help provide something lots of people were asking for...but you need to know how to do it first.
You are not really favoring carry players. You are favoring people who do the most pvp => gankers. Also because you make no distinction between getting high KDA and high gpm/xpm for easy games and hard games...you are favoring smurfs and you are favoring stackers.
The way your rating system is designed the highest rated players will be those that
1) are actually excellent at dota
2) play all ganking heroes
3) have a stong stack every game
4) have few games
Now support players happen to not do well...but that is not the problem. The problem is you give an equal rating for defeating nubs and pros because you do not measure the difficulty of the game.
Teamfight participation as a measure could be interesting comparing on a per hero basis...but just like KDA, GPM, XPM and so many other in-game stats it is meaningless to compare it across all heroes and meaningless to compare it without evaluating the difficultly of the game.
You say "its impossible to be perfect"...what is your definition of perfect? A proper overall rating system should never consider in-game stats at all. Those can be used in part to compare per hero performance, along with an evaluation of the difficulty of the game...but to compare them for all heroes for all games is just nonsense.
the difficulty of a game means nothing because if you stomp everyone game after game - you will be bumped up to a higher skill bracket.
you must have at least 200 games to be listed and actually everyone who i've checked in top 15 has well over 500 games so i'm not sure why you are even bringing up smurfs
These ratings don't really mean much since there's not enough data available for you to be precise. I was checking the stats of two of my friends, one who is horrible and one who is actually decent. The horrible one has a way higher score than the decent one so yeah... not much else to say.
Its not the same winning a game against mediocre players than against really good players .... and even in the same skill bracket there is diferences of skills.
"didn't even read the rest."
If you actually want to improve this system you cant ignore anyone opinion .... you dont need to be that defensive man no one wants to flame you.
Here is some question/suggestion
1. Is the stat automatically updated everyday?
2. Is it too complicated to share the formula of stat calculation here?
3. Share this to reddit, this will be famous
4. Why the calculation take so long? And what's the best way to solve this?
I'd really love if this thing had a much larger sample size already. Added the majority of people I play with but...would love if you could just find a way to implement everyone already on dotabuff into it.
Also y u stop rating below 3? xP Y u not letting me berate my really really bad friends with stats!? D:
Finally, any eta on when you think you'll get support ranking system in place(even if incredibly faulty)?
@Why do i play this stupid...
4. Why the calculation take so long? And what's the best way to solve this?
probably because the guy cannot afford an expensive server or system i believe, this thins aren't cheap, see dotabuff
Would be nice to know the formula though and also to support the site as a community with some investor maybe
but who knows...
might aswell be going to ruin fast if steam changes again something or implements a rating visible from it own servers
to conclude, i like anything new that is being injected int he community so congrats on the site and keep up the good work, i am well away its not easy
I only get 6,68? I thought i am better than that. Probably i play in SEA server where i presume that SEA server's skill is lower than US east or Europe, hence it's easier to get 58% win rate in SEA server.
Keep in mind...only 1500 players...and most of them appear to be in two groups: support or the pro scene. Everyone in between is just curious people. Give it time to grow.
@IP | Yoshi
nothing to do with the server. everything to do with steamapi restrictions.
@Dipshit
originally i were gonna stop the rating at 5 :)
@Why do i play this stupid...
1. Yes
2. I don't see the point to share the formula - I've already shared how it's calculated though and what things affect the score.
3. People on reddit didn't like it.
4. because of steam :)
@Dorkly
The rating is not about winning - it is about your performance in the games compared to the rest of the players.
One thing that could improve your rating a lot is giving seperate ratings in every role. So, "only" thing you should do is detect roles... You could figure out rules yourself that guess that someone with low kill high assist low last hit more deny is a support, and someone with high kills and not the most cs in team is a ganker, etc, but it would be better if you would use some kind of learning algorythm. So my idea would require people to select their role for some of their games, not all of them, but as many as they have time to do so. After enough data is in the system, your algorythm can start guessing roles for new games and people who wants to help can select if the program was right or not, so some kind of reinforcement learning. If you have enough data it will be closer to perfect. Once you have the games seperated by roles, you can start develop different algorythms.
If you want to try this I can help you out in some months maybe, when you will have enough data and less people queueing per day so more capacity for other computations.
So where do you draw the line if the player was a support in a particular game or a carry? By items? Well what if he supported at first but then got big and fat, farmed a tad more and bought 1-2 carry items by the end of the game
Thats why guessing the rules by ourself wont work. If people themselves say about a lot of games that hey I played sven here but everyone else picked a carry after me, I had to support... So I was playing support. But on his next game, he selects carry couse he feels he played a carry role in that game... It wont define Sven as anything, but the games with him, seperatly every single one of them. If you have enough data, it will work. If you have just that much as now, it definetly wont. Thats why I said some months and hope for the best. :)
You can use other data from sven games as strong rules and make it even better by using data that your algorythm learned from every game as weak rules, just to get a sharper line.
And why couldnt someone select 2 roles for a game, if he feels its better? I guess more than 2 is stupid, but 2 is okey, like as start I was a ganker but then at late game I evolved into a tank, or as in your example, started as support but somehow farmed himself, or the opposite, playing as a ganker and late-game support, couse hero doesnt scale well.
@bad to the bone
Indeed, if we try to approach support rating from that perspective, it's futile.
Because of the lack data, i think it will be too difficult to find good support rating formula. Even if someone find it, its accuracy is doubtful. But if you still want to try to find the formula, let's discuss what indicate a good support? Any suggestion?
I played this one with my friends and Balanar was joking with me all the time so I decided to KS him, like I used my drum to get there and auto-attack his kill, so after game I could mock him that I was even better carry, I got MOAR kills. But I would get way better score for this one with most furmulas. :D
http://dotabuff.com/matches/335585768 (Lich with 18 11 17)
But if you punish high kills, I would get bad score for this
http://dotabuff.com/matches/339835233 (Lion with 13 12 19)
where I wasnt KSing (I guess its that game) but killing a lot of fleeing enemies noone else would be able to, or probably wouldnt be.
Well say player has a score of 25-0-1. The whole team has 34 kills. So if we play off assists then he didn't contribute much to the team's success did he? So we can't just take assists into consideration.
Yes GPM and XPM would be linked with lasthits but it is also linked with participation. For example if you farm all game and you have a high GPM/XPM but have a score of 3-0-1 when the team has 25 kills then your participation score will suffer.
@Sam Fuckin' Peckinpah
well i had to manually add you to the database because the query wouldn't run with your data. Still trying to figure out why..
@Sam Fuckin' Peckinpah turns out you were special :) Because of you i've fixed a very serious bug.
Team player: 47.8%
seems legit
"Yes GPM and XPM would be linked with lasthits but it is also linked with participation. For example if you farm all game and you have a high GPM/XPM but have a score of 3-0-1 when the team has 25 kills then your participation score will suffer."
This can be good and bad. For example, if you're a hard carry and farmed the entire game only to participate in a major teamfight which wins your teams the game, then your value is being undermined. Certain heroes like Phantom Lancer will also suffer from this type of system since they win games not by participating in fights but by pushing.
I registered today morning but I'm still in the waiting queue
I should probably refraze this. the participation score wouldn't suffer - it would balance out your total score since you would have higher gpm/xpm due to farming all game.
@kiNe I've fixed a bug and had to reboot the whole queue. 190 players at the moment * at least 2 minutes each - that is over 6 hours max for your score to appear.
If its like that, then I think this rating system could be decent for positions 1, 2 and 3.
Ranking supports might be hard because some supports could have really shitty scores, really shitty gpm/xpm and sometimes lack in supporting items (wards) but played a big part in winning the game. I don't think its possible to accurately rank supports due to the large variation of scores they might give. E.g. a defensive support dedicated to stacking camps and protecting the carry.
Yeah but the way I plan to develop support rankings is: you are ranked against other people playing the same heroes. So say you play tree - you get a score for it and you are ranked against other players' tree scores.
My rating is 3, why, because i'm playing like a baws with disruptor. Supports are not rated high here.
He is still developing the support ranking.
What would a decent or above average rating be?
anything above 5.0 is good.
So anything above 5 implies they're a decent player? I assume most pro players are above 8?
most pro players are above 9.
Rating: 4.49
Team player: 55.9%
:C
eh i have 3 and in my last 20 games i've played with/against puppey, viggoss, black, garter, mitch, gordon, poseidon.sakura, T5.Darkfalco
all solo queuing (not solo MM preference)
while i've seen people who play normal bracket games with 7-8
i always enjoy more stats, keep up the good work, currently stuck in que but all good
Only problem with comparing tree to other trees is that a killstealing support, or someone playing venge as semi-carry, etc will undermine this, will still have better score than dedicated supports. You just cant rank supports using statistics.
Or you could add some extra points when detect a stat with high assist low kill and a player in team with a lots of gpm/kills. This one would prefer people who are carried by their friends... Still not good.
I guess it just can't be done, but hope you find some way.
Anyway, good idea, good for anything but supports. :)
thanks guys. To be honest i don't expect to develop a perfect ranking system as i said - just have a look at the current dota 2 matchmaking rating system that everyone is hating on lol. There is no way to build a perfect system. But this is just something fun i guess.
hopefully i will figure out something for support heroes.
I realize you wanted to help provide something lots of people were asking for...but you need to know how to do it first.
You are not really favoring carry players. You are favoring people who do the most pvp => gankers. Also because you make no distinction between getting high KDA and high gpm/xpm for easy games and hard games...you are favoring smurfs and you are favoring stackers.
The way your rating system is designed the highest rated players will be those that
1) are actually excellent at dota
2) play all ganking heroes
3) have a stong stack every game
4) have few games
Now support players happen to not do well...but that is not the problem. The problem is you give an equal rating for defeating nubs and pros because you do not measure the difficulty of the game.
Teamfight participation as a measure could be interesting comparing on a per hero basis...but just like KDA, GPM, XPM and so many other in-game stats it is meaningless to compare it across all heroes and meaningless to compare it without evaluating the difficultly of the game.
You say "its impossible to be perfect"...what is your definition of perfect? A proper overall rating system should never consider in-game stats at all. Those can be used in part to compare per hero performance, along with an evaluation of the difficulty of the game...but to compare them for all heroes for all games is just nonsense.
Do you want to know how to do it right?
the difficulty of a game means nothing because if you stomp everyone game after game - you will be bumped up to a higher skill bracket.
you must have at least 200 games to be listed and actually everyone who i've checked in top 15 has well over 500 games so i'm not sure why you are even bringing up smurfs
didn't even read the rest.
I didn't think you wanted to know, but its good to ask.
I'm pretty confident that this is the best that can be done with the limited data that steam provides and all the restrictions they impose.
Why I cannot see mine :((
Rating: 6
Teamplayer: 51,6%
I am pretty new at Dota#2 and never played Dota1. (Not that many games, or good win%) so thats kinda surprising :D
These ratings don't really mean much since there's not enough data available for you to be precise. I was checking the stats of two of my friends, one who is horrible and one who is actually decent. The horrible one has a way higher score than the decent one so yeah... not much else to say.
Its not the same winning a game against mediocre players than against really good players .... and even in the same skill bracket there is diferences of skills.
"didn't even read the rest."
If you actually want to improve this system you cant ignore anyone opinion .... you dont need to be that defensive man no one wants to flame you.
http://www.dota2rating.info/?player=r_augusto
So I'm better than NS? I feel great right now. #roadtoti4
Interesting, waiting for mine
Interesting.
Here is some question/suggestion
1. Is the stat automatically updated everyday?
2. Is it too complicated to share the formula of stat calculation here?
3. Share this to reddit, this will be famous
4. Why the calculation take so long? And what's the best way to solve this?
I'd really love if this thing had a much larger sample size already. Added the majority of people I play with but...would love if you could just find a way to implement everyone already on dotabuff into it.
Also y u stop rating below 3? xP Y u not letting me berate my really really bad friends with stats!? D:
Finally, any eta on when you think you'll get support ranking system in place(even if incredibly faulty)?
@Why do i play this stupid...
4. Why the calculation take so long? And what's the best way to solve this?
probably because the guy cannot afford an expensive server or system i believe, this thins aren't cheap, see dotabuff
Would be nice to know the formula though and also to support the site as a community with some investor maybe
but who knows...
might aswell be going to ruin fast if steam changes again something or implements a rating visible from it own servers
to conclude, i like anything new that is being injected int he community so congrats on the site and keep up the good work, i am well away its not easy
Damn I also got a 3. Yeah this really doesn't rate supports very well. I guess you really will have to have a separate rating for them or something
I only get 6,68? I thought i am better than that. Probably i play in SEA server where i presume that SEA server's skill is lower than US east or Europe, hence it's easier to get 58% win rate in SEA server.
Anyway, i like this rating.
Keep in mind...only 1500 players...and most of them appear to be in two groups: support or the pro scene. Everyone in between is just curious people. Give it time to grow.
relentles offer help and bone acting plain stupid, that ratings cannt be good...
@IP | Yoshi
nothing to do with the server. everything to do with steamapi restrictions.
@Dipshit
originally i were gonna stop the rating at 5 :)
@Why do i play this stupid...
1. Yes
2. I don't see the point to share the formula - I've already shared how it's calculated though and what things affect the score.
3. People on reddit didn't like it.
4. because of steam :)
@Dorkly
The rating is not about winning - it is about your performance in the games compared to the rest of the players.
@way2high
post links
@a5491877
the only people that could help are valve. as i've mentioned they don't provide enough data.
@Bad to the bone
I am still not added to the ranking system. Tried it today morning again and now im still not added. Can you have a look at it pls?
From what i can see, you are in the database.. http://www.dota2rating.info/sl1NE/
One thing that could improve your rating a lot is giving seperate ratings in every role. So, "only" thing you should do is detect roles... You could figure out rules yourself that guess that someone with low kill high assist low last hit more deny is a support, and someone with high kills and not the most cs in team is a ganker, etc, but it would be better if you would use some kind of learning algorythm. So my idea would require people to select their role for some of their games, not all of them, but as many as they have time to do so. After enough data is in the system, your algorythm can start guessing roles for new games and people who wants to help can select if the program was right or not, so some kind of reinforcement learning. If you have enough data it will be closer to perfect. Once you have the games seperated by roles, you can start develop different algorythms.
If you want to try this I can help you out in some months maybe, when you will have enough data and less people queueing per day so more capacity for other computations.
The problem with roles is that people play some heroes differently.
Let's consider sven. Some people consider him a support. I however think that's a waste of this hero's potential.
http://dotabuff.com/players/104307918/matches?hero=sven&game_mode=&match_type=real
Same with warlock
http://dotabuff.com/players/104307918/matches?hero=warlock&game_mode=&match_type=real
So where do you draw the line if the player was a support in a particular game or a carry? By items? Well what if he supported at first but then got big and fat, farmed a tad more and bought 1-2 carry items by the end of the game
Thats why guessing the rules by ourself wont work. If people themselves say about a lot of games that hey I played sven here but everyone else picked a carry after me, I had to support... So I was playing support. But on his next game, he selects carry couse he feels he played a carry role in that game... It wont define Sven as anything, but the games with him, seperatly every single one of them. If you have enough data, it will work. If you have just that much as now, it definetly wont. Thats why I said some months and hope for the best. :)
You can use other data from sven games as strong rules and make it even better by using data that your algorythm learned from every game as weak rules, just to get a sharper line.
And why couldnt someone select 2 roles for a game, if he feels its better? I guess more than 2 is stupid, but 2 is okey, like as start I was a ganker but then at late game I evolved into a tank, or as in your example, started as support but somehow farmed himself, or the opposite, playing as a ganker and late-game support, couse hero doesnt scale well.
I've just realized. The score is carry rating.
@bad to the bone
Indeed, if we try to approach support rating from that perspective, it's futile.
Because of the lack data, i think it will be too difficult to find good support rating formula. Even if someone find it, its accuracy is doubtful. But if you still want to try to find the formula, let's discuss what indicate a good support? Any suggestion?
It's hard to calculate support rating from end-game stats...
I got commended on this game by our carry an I was way more useful than usually:
http://dotabuff.com/matches/339711241 (Bane with 2 5 8)
I played this one with my friends and Balanar was joking with me all the time so I decided to KS him, like I used my drum to get there and auto-attack his kill, so after game I could mock him that I was even better carry, I got MOAR kills. But I would get way better score for this one with most furmulas. :D
http://dotabuff.com/matches/335585768 (Lich with 18 11 17)
But if you punish high kills, I would get bad score for this
http://dotabuff.com/matches/339835233 (Lion with 13 12 19)
where I wasnt KSing (I guess its that game) but killing a lot of fleeing enemies noone else would be able to, or probably wouldnt be.
(Just check the stats)
And higher carry rating doesn't correlate with win rate?
That's weirddddddd
it does actually but as we all know in dota 2 winrate is not everything