|
Latest changes to SEA/KR look better - it suggests that the masters opponents I've been facing are actually in Masters now 
So thanks for that!
Out of interest - what is the scenario for a game that helps you tie down the league offset? Obviously the data needs to be good, but then what's the additional scenario? games with players that are right around the league boundary, and a mix of games vs opponents above and below the boundary? consecutive games like this?
Also, on the Log > Analyse screen, what does the last number represent? I can figure out
56:134 diamond 1150 32
56 = game number,
134 = MMR difference to opponent,
diamond = opponent league,
1150 (SEA) league boundary -
32, what's this? some kind of confidence range?
|
On September 15 2012 12:03 jaycee wrote:Latest changes to SEA/KR look better - it suggests that the masters opponents I've been facing are actually in Masters now So thanks for that! Out of interest - what is the scenario for a game that helps you tie down the league offset? Obviously the data needs to be good, but then what's the additional scenario? games with players that are right around the league boundary, and a mix of games vs opponents above and below the boundary? consecutive games like this? Also, on the Log > Analyse screen, what does the last number represent? I can figure out 56:134 diamond 1150 32 56 = game number, 134 = MMR difference to opponent, diamond = opponent league, 1150 (SEA) league boundary - 32, what's this? some kind of confidence range?
A good offset game is like this:
2 games - direct in a row - both time the player data is collected good and the player is not caped.
The opponents of this to games are from different league.
To give you an idea : out of 10.000 games i get around 30 such couples.
With the assumption that the player dmmr did not change much ( i put in values how much i would expect)
i can take the Dmmr diffrence the player have in this 2 games as offset diffrence of the 2 leagues of the opponent.
Problem is the Dmmr is not calculated exact. So i take the MIn range and the max range of this diffrence.
Than i take the maximum of all minimums and the minimum of all maximums and i get a close range for the offset of the league.
The only error is my assumption about how much his mmr changed because of game 1.
Notthat and me, come up with this idea long time ago.
In the past this was even harder because we had to make it for tiers not leagues. So it had to be known, what tier someone is. That was not directly analysed. Notthat plotted the density functions of this data and you could see the increased density of the tiers. From there on you calculaet with min max like above.
Values: Close!
game number, Dmmr ( in his league) value of the player , opponent league, offset for the player , errormargin of the dmmr function.
Basicly value for me to debug because this screen is not needed anymore.
|
Since there is no division tier now, after you've played enough games to usually get +/-12 points for a win/loss, then your MMR can simply be calculated by points - bonus pool. Why is this tool still necessary?
|
I just got promoted to diamond with this graphs:
![[image loading]](http://i.imgur.com/5j2S8.jpg)
BTW: First time in diamond league for me after over 1 year of platinum! YEAH!!!!
|
On September 15 2012 17:01 paralleluniverse wrote:
Since there is no division tier now, after you've played enough games to usually get +/-12 points for a win/loss, then your MMR can simply be calculated by points - bonus pool. Why is this tool still necessary?
Points minus bonus pool is your adjusted points and not your MMR. Adjusted points can be used to determine MMR in some situations, but very few people know enough how to do this. Point loss and gain will give you more information at a glance now, but not enough information to determine MMR at a glance.
|
On September 15 2012 17:01 paralleluniverse wrote:
Since there is no division tier now, after you've played enough games to usually get +/-12 points for a win/loss, then your MMR can simply be calculated by points - bonus pool. Why is this tool still necessary?
1)
you have to calculate how many change-points you really get. Thats not easy and you have to click a lot of data for it.
You have to do it while you play the game!
2)
To get a game thats real 12/-12 , the opponent must be in the same league you are.
3)
the chance to get such a game is perhaps 1:200 or more.
3)
if all this is ok you know a number that is linear to your dmmr.
4)
You still dont know anything about how people from different leagues.
=
So, if you alt tab out of the game, start calculating while playing,
than you can in one out of 200 games,
compare compare yourself to some random guy in your league.
I calculate mmr for every game you play and put it in comparison to the whole sc2 population.
So you can compare yourself to everyone on your server over a number.
And this for every game, so you see your MMR movement.
(when everything works correct)
On September 15 2012 17:19 enemy2010 wrote:I just got promoted to diamond with this graphs: BTW: First time in diamond league for me after over 1 year of platinum! YEAH!!!!
Im sure about the diamond offsets. Not 100% about the platinum once.
You are way under the line. I will check your games later and see if this are games vs diamond or platinum people and how many of this games are actuell good games.
Can be that -
A) only one of them is a good one and the graph was not stable yet and the errormargin and unluck push it to far down.
B) you have only 1 or 2 good games and a new account? in this case the prediction maker is not accurate and you shoot up like hell - faster than i predict
C) the games are all good and only against platinum opponents = i calculate platinum range to high
PS: rightclick on the graph and click save as to get a easy screenshot.
This one is not even necessary because when you give me your username i can see your whole graph on my back-end version.
|
On September 15 2012 18:29 skeldark wrote:Show nested quote +On September 15 2012 17:01 paralleluniverse wrote:
Since there is no division tier now, after you've played enough games to usually get +/-12 points for a win/loss, then your MMR can simply be calculated by points - bonus pool. Why is this tool still necessary? when you get a game with 12/12 against someone that is the same league you are, and calculate adjusted points you know ... a number that is linear to your dmmr, still not your MMR. You will have such a game perhaps 1 out of 300. Also you still dont know anything about how people from different leagues. So you can only compare yourself to people from your same league when you find for them a 12 / 12 game in this league. So with observing 600 games you will be able to tell how 2 people stand to each other. I calculate mmr for every game you play and put it in comparison to the whole sc2 population. So you can compare yourself to everyone on your server over a number. And this for every game and see your MMR movement. (when everything works correct) Show nested quote +On September 15 2012 17:19 enemy2010 wrote:I just got promoted to diamond with this graphs: BTW: First time in diamond league for me after over 1 year of platinum! YEAH!!!! Im sure about the diamond offsets. Not 100% about the platinum once. You are way under the line. I will check your games later and see if this are games vs diamond or platinum people and how many of this games are actuell good games. Can be that - A) only one of them is a good one and the graph was not stable yet and the errormargin and unluck push it to far down. B) you have only 1 or 2 good games and a new account? in this case the prediction maker is not accurate and you shoot up like hell - faster than i predict C) the games are all good and only against platinum opponents = i calculate platinum range to high PS: rightclick on the graph and click save as to get a easy screenshot. This one is not even necessary because when you give me your username i can see your whole graph on my back-end version.
The fact that skill rating must converge to points in the long run, excluding bonus pool, means that after around 20 or 30 games have been played, your points must equal MMR up to a linear transformation. But that is still enough to compare with other people in the same league. For example, if 2 people in the same league have played over 30 games, and player A has higher (points - bonus pool) than player B, then it follows that player A has higher MMR than player B. If you dispute this, then under what situation would player B have higher MMR than player A? Since this allows for ranking, it is good enough as MMR is essentially only meaningful as a rank, because it is relative to other people's MMR.
|
On September 15 2012 18:38 paralleluniverse wrote:Show nested quote +On September 15 2012 18:29 skeldark wrote:On September 15 2012 17:01 paralleluniverse wrote:
Since there is no division tier now, after you've played enough games to usually get +/-12 points for a win/loss, then your MMR can simply be calculated by points - bonus pool. Why is this tool still necessary? when you get a game with 12/12 against someone that is the same league you are, and calculate adjusted points you know ... a number that is linear to your dmmr, still not your MMR. You will have such a game perhaps 1 out of 300. Also you still dont know anything about how people from different leagues. So you can only compare yourself to people from your same league when you find for them a 12 / 12 game in this league. So with observing 600 games you will be able to tell how 2 people stand to each other. I calculate mmr for every game you play and put it in comparison to the whole sc2 population. So you can compare yourself to everyone on your server over a number. And this for every game and see your MMR movement. (when everything works correct) On September 15 2012 17:19 enemy2010 wrote:I just got promoted to diamond with this graphs: BTW: First time in diamond league for me after over 1 year of platinum! YEAH!!!! Im sure about the diamond offsets. Not 100% about the platinum once. You are way under the line. I will check your games later and see if this are games vs diamond or platinum people and how many of this games are actuell good games. Can be that - A) only one of them is a good one and the graph was not stable yet and the errormargin and unluck push it to far down. B) you have only 1 or 2 good games and a new account? in this case the prediction maker is not accurate and you shoot up like hell - faster than i predict C) the games are all good and only against platinum opponents = i calculate platinum range to high PS: rightclick on the graph and click save as to get a easy screenshot. This one is not even necessary because when you give me your username i can see your whole graph on my back-end version. The fact that skill rating must converge to points in the long run, excluding bonus pool, means that after around 20 or 30 games have been played, your points must equal MMR up to a linear transformation. .
Thats what blizzard says... but thats not the truth.
. Special rules make sure points dont follow mmr if the mmr is for the league you are in, is in the down half. In this case a protection system jumps in to hide the fact that you are in the down half to let you feel better.
But that is still enough to compare with other people in the same league. For example, if 2 people in the same league have played over 30 games, and player A has higher (points - bonus pool) than player B, then it follows that player A has higher MMR than player B.
even if both are not caped like explained above, higher adjusted point in a league is an indicator not proof for higher dmmr.
If you dispute this, then under what situation would player B have higher MMR than player A?
Your adjusted did not follow the dmmr correct. You have different amount of games that follow special rules like the one in a ) and they get more "free" points. Also perhaps 30 games are not enough. Why 30 why not 20 or 40?
Since this allows for ranking, it is good enough as MMR is essentially only meaningful as a rank, because it is relative to other people's MMR.
This allows for a very inaccurate ranking and only for the top of the league and only inside the league (dmmr only league - MMR whole server).
Why do it bad if you can do it good?
|
@enemy2010
Update checked your data
You only have one good game but this one was against a platinum guy with low error margin.
Thats an indicator that my platinum range is really to high!
Will keep that in mind if i check the offset data next time when i have more games.
|
On September 15 2012 19:08 skeldark wrote:
@enemy2010
Update checked your data
You only have one good game but this one was against a platinum guy with low error margin.
Thats an indicator that my platinum range is really to high!
Will keep that in mind if i check the offset data next time when i have more games.
Well, I just understood half of what you said, but.... glad I could help :D
|
Pretty cool tool/method, I always wanted to be able to tell what my mmr in SC2 is (In WoW your mmr is displayed after each arena game ends, so I'm used to knowing) D:
|
I think there are some problems (at least for me) with the population gaussian graph. I'm marked correctly as silver/gold player in the MMR graph, but the population graph puts me way higher: a good chunk above gold. Since the two have always been quite similar, I feel there's something going wrong here.. I'll tell you if playing more games will fix this, but it would be better if this was't happening at all
|
Is it usual to take more than 20 games to find good game. it's on 0/22 for me now....
|
On September 15 2012 23:27 lInsta wrote:
Is it usual to take more than 20 games to find good game. it's on 0/22 for me now....
Sounds about right. I assume you don't have the ram scanner on?
|
On September 15 2012 11:14 skeldark wrote:
To get more games enable the Memory scanner ( 1 out of 20 ) is to low.
Out of 23 games, only 2 are "good" (ie. dot filled in), but I have the memory scanner enabled. Is there something else that could be effecting it?
|
A filled in dot signifies a win, not a good game.
If you click the little cog on the top of the window, go to Data, you can select "Mark assumptions / bad Games" to show the good/bad games.
Good is blue, bad is grey.
|
On September 15 2012 23:21 Malhavoc wrote:I think there are some problems (at least for me) with the population gaussian graph. I'm marked correctly as silver/gold player in the MMR graph, but the population graph puts me way higher: a good chunk above gold. Since the two have always been quite similar, I feel there's something going wrong here.. I'll tell you if playing more games will fix this, but it would be better if this was't happening at all
I think skeldark wrote some pages back that Blizzard has stopped spreading players like in a gaussian graph. If so, you can actually scrap it. Nice find still. I guess he forgot about it.
I wonder how people are spread nowadays though (more like 20% of players in every league from Bronze to Diamond; best 2% Masters, best 200 GM?).
|
|
|
The MMR graphs seem to show that diamond league now contains a narrower MMR range than it was in the previous season, in relation to other leagues. Probably because Blizzard moved the promotion lines around.
If diamond is smaller, is it because "low diamond" has shifted up, or because "high diamond" has shifted down?
|
About gauss / offsets / sc2ranks.
The offsets change to keep the leagues in shape. They dont try to do this very accurate they just put in numbers that kind of do the job. The playerbase shift away from a perfect gauss glock so offsets correction try to keep the leagues in it.
All of this dont affect the real skill function. Its only cosmetic for the fake frontend.
Sc2ranks dont detect active players like blizzard does, so their numbers are never equal to blizzard numbers.
At this point just ignore my gauss graph. I will check if i can find out gold-plat % in future but dont read to much into it.
About many bad games
With less tiers and smaller league sizes the change to be caped is higher now. Thats why there are so many bad game.
Only because they removed one hidding factor dont mean they gave up on hiding mmr...
|
|
|
|
|
|
EPT
