|
On July 27 2012 00:32 skeldark wrote:Show nested quote +On July 26 2012 10:31 SoleSteeler wrote:
Does this need to be updated again if SC2gears gets updated? my updates are indipendent from sc2gears. But i use the new features of sc2gears so you should allways be up to date. He would not do that if he dont have too. can you mark assumptions in the options and post the result of that. The graph looks like you dont have a good game for long time. Can be that you have bad games and only 20 assumptions and the frist good game again correct it. Just a guess lets see. You are on master line so you should have master enemy that produce 100% accurate data. oh and the result of the F analyer perhaps show the mistake
I'll get those when I get home from work. That particular game was a masters vs masters. It shot back up a couple games later.
|
On July 27 2012 01:40 InfCereal wrote:Show nested quote +On July 27 2012 00:32 skeldark wrote:On July 26 2012 10:31 SoleSteeler wrote:
Does this need to be updated again if SC2gears gets updated? my updates are indipendent from sc2gears. But i use the new features of sc2gears so you should allways be up to date. He would not do that if he dont have too. can you mark assumptions in the options and post the result of that. The graph looks like you dont have a good game for long time. Can be that you have bad games and only 20 assumptions and the frist good game again correct it. Just a guess lets see. You are on master line so you should have master enemy that produce 100% accurate data. oh and the result of the F analyer perhaps show the mistake I'll get those when I get home from work. That particular game was a masters vs masters. It shot back up a couple games later.
In this case he could not change the vs M game so he had to jump and the next v M jumped back.
= the game is broken.
When i see the amount of change i think he calculated the changepoints wrong by 1.
|
|
|
yeah obvious wrong data.
Change-point mistakes of +-1 can happen. I have many rules in place to prevent it in very few cases one of them slip throw.
Just delete the game in the dataeditor or open the game.dat and edit the game as bad ( the 1, 0 ) direct after your bnet-link
|
I get the following error in the log of sc2gears MMR tool:
ERROR: Cant find Memory-Profile in Active-Profiles! Perhaps its just a custom game...
Here's my recent log:
+ Show Spoiler +
18:41:11 --------------------------
18:41:11 Game Start detected
18:41:11 ERROR: Cant find Memory-Profile in Active-Profiles! Perhaps its just a custom game...
18:41:54 Webgrabber: Player true
18:41:54 No Profile, skip
18:41:54 Webgrabber: Opponent false
18:41:54 PRE-Trigger done
18:49:54 --------------------------
18:49:54 Game Start detected
18:49:54 ERROR: Cant find Memory-Profile in Active-Profiles! Perhaps its just a custom game...
18:50:38 Webgrabber: Player true
18:50:38 No Profile, skip
18:50:38 Webgrabber: Opponent false
18:50:38 PRE-Trigger done
19:0:39 Webgrabber: Player true
19:0:43 Webgrabber: Opponent true
19:0:43 DATA Alg.B: failed, No Bonus Win
19:0:43 Adjused Calculate Player:true
19:0:43 Adjused Calculate Opponent: true
19:0:43 Gamecheck: true
19:0:43 POST-Trigger done
19:0:43 Player-data-Qualitiy: 100%
19:0:43 Opponent-data-Qualitiy: 100%
19:0:44 OK Uploader: game uploaded
19:0:44 Ladder Game Done!
19:0:44 --------------------------
19:0:48 --------------------------
19:0:48 Game Start detected
19:0:48 ERROR: Cant find Memory-Profile in Active-Profiles! Perhaps its just a custom game...
19:1:31 Webgrabber: Player true
19:1:31 No Profile, skip
19:1:31 Webgrabber: Opponent false
19:1:31 PRE-Trigger done
19:10:51 Webgrabber: Player true
19:10:55 Webgrabber: Opponent true
19:10:55 Adjused Calculate Player:true
19:10:55 DATA Alg.B: failed, No Bonus Win
19:10:55 ERROR AFTERCHECK All algorithem failed
19:10:55 Adjused Calculate Opponent: false
19:10:55 Gamecheck: true
19:10:55 POST-Trigger done
19:10:55 Player-data-Qualitiy: 75%
19:10:55 Opponent-data-Qualitiy: 100%
19:10:56 OK Uploader: game uploaded
19:10:56 Ladder Game Done!
19:10:56 --------------------------
19:11:2 --------------------------
19:11:2 Game Start detected
19:11:2 ERROR: Cant find Memory-Profile in Active-Profiles! Perhaps its just a custom game...
19:11:45 Webgrabber: Player true
19:11:45 No Profile, skip
19:11:45 Webgrabber: Opponent false
19:11:45 PRE-Trigger done
19:24:23 Webgrabber: Player true
19:24:26 Webgrabber: Opponent true
19:24:26 DATA Alg.B: failed, No Bonus Win
19:24:26 Adjused Calculate Player:true
19:24:26 Adjused Calculate Opponent: true
19:24:26 Gamecheck: true
19:24:26 POST-Trigger done
19:24:26 Player-data-Qualitiy: 100%
19:24:26 Opponent-data-Qualitiy: 100%
19:24:27 OK Uploader: game uploaded
19:24:27 Ladder Game Done!
19:24:27 --------------------------
There's also a ERROR: Aftercheck All Algorithem [sic] failed in there too.
It seems to work fine, but what are these errors I get every time?
|
ERROR: Cant find Memory-Profile in Active-Profiles! Perhaps its just a custom game...
I means that the profile you entered in the program and the profile that is laddering 1v1 at the moment are not the same!
Check the profile you entered in the settings
|
^ I don't understand what you mean.
I mean, it seems like it's working just fine - my MMR is all plotted out just fine.
The profile I entered in the settings are:
http://us.battle.net/sc2/en/profile/1584798/1/Belial/
I'm just worried what this error is.
I only 1v1 ladder matchmaking, on that profile. I don't play customs, I don't have any other accounts.
|
How does the finetune algorithim work?
|
On July 29 2012 14:32 EPICWINZ wrote:
How does the finetune algorithim work?
I take data triples and check if the results are possible and how far they are away from my exception.
If i get impossible results on low margin ( e.g. lost 4 points after a WIN) i correct the game inside of the error margin of the functions that run before.
E,G,
Game 1: 1020 MMR WIN ERRORMARGIN +-16
Game 2: 1018 MMR WIN ERRORMARGIN +-24
Game 3: 1052 MMR LOOSE ERRORMARGIN +-24
The finetuner detects that you lost mmr in game 2 after winning game 1
So
1 - 2: -2 on win is wrong
2-3 +34 is very high for only 1 win
Because both sites of 2 show the same mistake i know i would correct game 2 in the right direction by adding points.
The Finetuner will add +18 to game 2.
Its inside of the errormargin of +-24 and get game 2 closer to the expectation to both sites!
If there is an obvious error outside of the errrormargin the finetuner will NOT correct it.
He only correct, if he can do it inside of the error-margin and only if correction would not affect the direction of the global graph.
So that he never produce follow up mistakes.
|
So im in plat now.... and on MMR stats im pretty close to the line where platinum is, does that means im stable in platinum or that im close too diamond?
|
On July 30 2012 19:17 Schalde wrote:
So im in plat now.... and on MMR stats im pretty close to the line where platinum is, does that means im stable in platinum or that im close too diamond?
If its close above the plat promotion line this means your mmr is a little bit over the 0 line of platinum. I think some people call it "low platinum".
If its close under diamond promotion line you are what people call "high paltinum", close above they call "low diamond".
Leagues are just random MMR values blizzard put in long time ago.
They dont mean anything, you just get a shiny icon if you pass this random lines. Thats it!
If you have 1000 MMR you are not low x or high y ... you are just 1000 MMR.
You see how good this is compare to other players on the server, when pressing the gauss button top left.
|
If I'm sitting at 2030ish MMR does the Blizzard system cause it to decay during inactivity?
Similar, if I have to go inactive, will I be placed back in Diamond should I cease playing? Or have I crossed the MMR line enough to be promoted?
|
On July 30 2012 23:58 etherealfall wrote:
If I'm sitting at 2030ish MMR does the Blizzard system cause it to decay during inactivity?
Similar, if I have to go inactive, will I be placed back in Diamond should I cease playing? Or have I crossed the MMR line enough to be promoted?
Your skill is your skill if you play or not.
Skill functions dont care how active you are. Only fake functions like the ladder points does.
You get promoted if you are stable in one tier inside a new division.
Once again. Promotion demotion dont say much. The number i show you does.
Leagues are a very inaccurate way to measure Skill.
I calculate the accurate skill number and you guys use it to to find out what your inaccurate skill number will be...
I think i just came to late and the hole league/icon idea is to established by now, even if it shows your skill total inaccurate.
To give you an example:
If a player is in gold this means: his mmr can be between 0 silver and max platinum = 60% of the players.
|
@skeldark
i am a big fan of your tool.
I would also love to have a feature that displayes where you stand in your league depending on your fellow players of your league. So you could basically have a low master MMR and still be at the (50%) quartil. i don't know how accurate the Gaussian curve is in the current ladder. So having a number % where you stand or showing it in the MMR graph would be really cool.
My statistics are a bid rusty. Your have a lot of data. Not sure if it is independed so as it seems to me more better players use the MMR tool. I am sure your knowledge is better. Can you conclude from your data to the distribution of the absolute numbers?
Edit: i know of the current graph of the mmr polulation you have build in. Is the ladder still a Gaussian curve?
|
On July 31 2012 00:15 OrbitalPlane wrote:
@skeldark
i am a big fan of your tool.
I would also love to have a feature that displayes where you stand in your league depending on your fellow players of your league. So you could basically have a low master MMR and still be at the (50%) quartil. i don't know how accurate the Gaussian curve is in the current ladder. So having a number % where you stand or showing it in the MMR graph would be really cool.
My statistics are a bid rusty. Your have a lot of data. Not sure if it is independed so as it seems to me more better players use the MMR tool. I am sure your knowledge is better. Can you conclude from your data to the distribution of the absolute numbers?
thx.
The gauss curve is not accurate. But the best i can do and accurate enough to tell low, high, middle.
I can not calculate the disruption with my data because my userbase ( tl user who care for their mmr) is biased.
The gauss curve you see is what it should be if gold-platinum is exactly 20%. Thats not what it is, but also not far away of it.
I could calculate your % out of gauss but i think you see it pretty easy and i have much real life stuff comming up at the moment.
If one day 50% are in masters, this means that the player skill distribution is far away from gauss.
In this case blizzards whole skill system failed.
|
On July 31 2012 00:14 skeldark wrote:Show nested quote +On July 30 2012 23:58 etherealfall wrote:
If I'm sitting at 2030ish MMR does the Blizzard system cause it to decay during inactivity?
Similarly, if I have to go inactive, will I be placed back in Diamond should I cease playing? Or have I crossed the MMR line enough to be promoted? Your skill is your skill if you play or not. Skill functions dont care how active you are. Only fake functions like the ladder points does. You get promoted if you are stable in one tier inside a new division. Once again. Promotion demotion dont say much. The number i show you does. Leagues are a very inaccurate way to measure Skill. I calculate the accurate skill number and you guys use it to to find out what your inaccurate skill number will be...
I see. I understand what you're saying.
But for the sake of a blunt answer, what would that be to my second question given a current MMR estimate of 2030ish with 108/168 good games. I'm asking this because I've been trying to get promoted for the sake of a promotion and things are going to get very busy soon so SC will have to be played less.
Thanks.
|
50 % quantil of master i meant. xD
This would indeed be sad days if 50% of the players would be in masters. :D
Thx for the answer that's what i also thought. Let's hope blizzard gives some nice statistics with HOTS.
|
On July 31 2012 00:27 etherealfall wrote:Show nested quote +On July 31 2012 00:14 skeldark wrote:On July 30 2012 23:58 etherealfall wrote:
If I'm sitting at 2030ish MMR does the Blizzard system cause it to decay during inactivity?
Similarly, if I have to go inactive, will I be placed back in Diamond should I cease playing? Or have I crossed the MMR line enough to be promoted? Your skill is your skill if you play or not. Skill functions dont care how active you are. Only fake functions like the ladder points does. You get promoted if you are stable in one tier inside a new division. Once again. Promotion demotion dont say much. The number i show you does. Leagues are a very inaccurate way to measure Skill. I calculate the accurate skill number and you guys use it to to find out what your inaccurate skill number will be... I see. I understand what you're saying. But for the sake of a blunt answer, what would that be to my second question given a current MMR estimate of 2030ish with 108/168 good games. I'm asking this because I've been trying to get promoted for the sake of a promotion and things are going to get very busy soon so SC will have to be played less. Thanks.
2030 is above master promotion line but only few points. If you stay stable above the line ( means you dont fall under it in next 5-10 games) you get promoted to master. This is the case because master only have one tier so: in master = stable.
If you are already in master you will only get demoted when you are stable in one diamond tier or fall down to platinum level.
If if a master player drop down to low diamond and raise a little again he dont get demoted because he is not stable.
On July 31 2012 00:29 OrbitalPlane wrote:
quartil of master i meant. xD
This would indeed be sad days if 50% of the players would be in masters. :D
Thx for the answer that's what i also thought. Let's hope blizzard gives some nice statistics with HOTS.
When we learned on thing than not to trust the statistic data blizzard publish. We used published data from blizzard in the beginning to calculate mmr but we found out very fast that this data is completely wrong.
I dont know if they did a mistake or try to troll us or just don't care and invented random numbers.
Also keep in mind there are 2 types of %, Points and Players.
e.g.
2300 master mmr is low masters in points .
But you are above 33% of master players so you are mid master in distribution.
|
Thanks skeldark! Hopefully i stabilize then eh.
|
First of all thanks for the work you put into this. It's nice to see a more detailed measure of skill than just a League symbol.
One thing that bothers me a little right now is that the opponent line in the tool always shows my opponent being exactly 10 MMR above me, regardless of me havine win- or lossstreaks. Is it that the tool just estimated my opponents MMR, because it didn't have proper data for him? Else I would expect my opponents MMRs to sometimes be lower than mine (after a lossstreak for example) or more higher (after a winstreak maybe), because Blizzard should give me opponents of different skill level.
On a sidenote I just have like 10 good games with MMR tool. So maybe this will change when I have played more games (?).
Also from your errormargin post above, your max errormargin is +-24 (usually I get +-16). That means that you expect the real Blizzard MMR to rise or fall 24 points at max per game, right? (That would be so sad, because Gold League would be so far away  . I know it doesn't mean anything, but a Gold icon looks so much more shiny than a silver one  ).
|
|
|
|
|
|
EPT
