|
On June 27 2012 23:37 arie3000 wrote:Show nested quote +On June 27 2012 09:24 skeldark wrote:On June 27 2012 07:29 maLaK1 wrote:
how can I set the dialog of the tool to not maximize every time I start Sc2gears?
I want a small dialog like it was before will add that in next version
Update:2) with last version, many people who play on different servers and a lot of uploaded gamedata, i can make a statistic over: Skill diffrence of different server! That will definitely be VERY interesting! Realize what sort-of shitstorm you're going to create with that...  I tried it yesterday on my mac, installed the latest SC2gears and dowleaded the plugin, but indeed it doesn't show up. Even if you tell it to 'show non-working (or words to that effect) plugins', it doesn't show up. Hopefully you can get to work for Mac, if you need help or screenshots, post it!
I think the problem is that part of my code throw an expectation on loading and sc2gears dont output this mistake.
No easy fix can be everything but most likely the libs for scanner and registry.
On June 27 2012 23:53 Mendelfist wrote:Show nested quote +On June 26 2012 19:43 skeldark wrote:
to be honest i dont trust A any more. A is the algo that finds your tier. Your tier is important for your opponent mmr. It dont care for your own. My latest version use a different version of A. instead of calculating all possible tiers for you he use the information of F. So i calculate the opponent after you. Looks way better but im not sure yet.
Is every MMR-stats-client calculating/guessing the player tiers based on locally collected data, or do you have some sort of global database with division tiers (based on division name)? I would think that if enough people use the client you would soon know the tier of most divisions, and there would be little or no guesswork involved. Until a new season starts at least. This tool is most excellent, and I look forward to every weekend when I have time to play and see what happens to my MMR, but it still makes some wrong guesses it seems.
I have both. I have a way to calculate the tiers global and the client is able to use this information. The db file the tool download includes an tier and offset list.
However i dont use it at the moment. If i use it, i have to be 100% sure about the tier. Because it will change everyone's graph and i need game rows to analyse. So basic i use "local" data to find them global. Its always a guess because there is no way to know a tier 100%.
The question is how good is the guess. I think im at 90% right now.
Also there are so many tiers, its not worth to do it every season. By the time i have enough divisions the season is already over.
When i look at your graph it look pretty good for me. What mistakes do you mean?
i see one jump that can not explained by deviation.
But my old version did not upload bad games so can be that i missing some games between.
4.6
-fixed bug that some bad games are not stored
-option to dont start fullscreen
-performance improved
|
On June 28 2012 00:04 skeldark wrote:
When i look at your graph it look pretty good for me. What mistakes do you mean?
i see one jump that can not explained by deviation.
But my old version did not upload bad games so can be that i missing some games between.
Yes, I have marked two games here that look especially suspect:
![[image loading]](http://www.lysator.liu.se/~john/mmrstats2.jpg)
and I think that there may be more games to the left of it that are not correct. Here I have made a graph of my opponents relative MMR (I forgot what the usual term for it is, "division" MMR or something?) covering approximately the same time period:
![[image loading]](http://www.lysator.liu.se/~john/opmmr2.jpg)
It is very noisy due to Blizzards new shitty (sorry, I mean "relaxed") matchmaking system, but the trend is still very visible. (Ignore the "Plot Area" text)
|
On June 28 2012 01:13 Mendelfist wrote:Show nested quote +On June 28 2012 00:04 skeldark wrote:
When i look at your graph it look pretty good for me. What mistakes do you mean?
i see one jump that can not explained by deviation.
But my old version did not upload bad games so can be that i missing some games between. Yes, I have marked two games here that look especially suspect: and I think that there may be more games to the left of it that are not correct. Here I have made a graph of my opponents MMR compared to my own (I forgot what the usual term for it is, "division" MMR or something?) covering approximately the same time period: It is very noisy due to Blizzards new shitty (sorry, I mean "relaxed") matchmaking system, but the trend is still very visible. (Ignore the "Plot Area" text)
that was the jump i was talking over.
there are obvious missing games between.
The wrong dot is a silver player he put on T0 what he is obvious not. when i look closer t1 should make less mistake even with the lost game. dont know why he take 0. the 3 predicted game before and the missing game after throws him off tho.
the graph on the right is at lower bond so total correct.
Dmmr say nothing for you. the graph would show chaos even if every opponent is = your mmr.
|
On June 28 2012 01:20 skeldark wrote:Show nested quote +On June 28 2012 01:13 Mendelfist wrote:On June 28 2012 00:04 skeldark wrote:
When i look at your graph it look pretty good for me. What mistakes do you mean?
i see one jump that can not explained by deviation.
But my old version did not upload bad games so can be that i missing some games between. Yes, I have marked two games here that look especially suspect: and I think that there may be more games to the left of it that are not correct. Here I have made a graph of my opponents MMR compared to my own (I forgot what the usual term for it is, "division" MMR or something?) covering approximately the same time period: It is very noisy due to Blizzards new shitty (sorry, I mean "relaxed") matchmaking system, but the trend is still very visible. (Ignore the "Plot Area" text) that was the jump i was talking over. there are obvious missing games between. The wrong dot is a silver player he put on T0 what he is obvious not. when i look closer t1 should make less mistake even with the lost game. dont know why he take 0. the 3 predicted game before and the missing game after throws him off tho. the graph on the right is at lower bond so total correct. Dmmr say nothing for you. the graph would show chaos even if every opponent is = your mmr.
Yes, I agree that the right part of the graph looks correct. It also looks approximately like the DMMR graph.
I want to point out that the DMMR graph is my *opponents* DMMR, not my own, and I think it is very useful, because there are no tier errors in it. My own tier doesn't change. The trend of my opponents DMMR should be similar to the graph your tool makes if it works correctly, I think.
|
On June 28 2012 01:36 Mendelfist wrote:Show nested quote +On June 28 2012 01:20 skeldark wrote:On June 28 2012 01:13 Mendelfist wrote:On June 28 2012 00:04 skeldark wrote:
When i look at your graph it look pretty good for me. What mistakes do you mean?
i see one jump that can not explained by deviation.
But my old version did not upload bad games so can be that i missing some games between. Yes, I have marked two games here that look especially suspect: and I think that there may be more games to the left of it that are not correct. Here I have made a graph of my opponents MMR compared to my own (I forgot what the usual term for it is, "division" MMR or something?) covering approximately the same time period: It is very noisy due to Blizzards new shitty (sorry, I mean "relaxed") matchmaking system, but the trend is still very visible. (Ignore the "Plot Area" text) that was the jump i was talking over. there are obvious missing games between. The wrong dot is a silver player he put on T0 what he is obvious not. when i look closer t1 should make less mistake even with the lost game. dont know why he take 0. the 3 predicted game before and the missing game after throws him off tho. the graph on the right is at lower bond so total correct. Dmmr say nothing for you. the graph would show chaos even if every opponent is = your mmr. Yes, I agree that the right part of the graph looks correct. It also looks approximately like the DMMR graph. I want to point out that the DMMR graph is my *opponents* DMMR, not my own, and I think it is very useful, because there are no tier errors in it. My own tier doesn't change. The trend of my opponents DMMR should be similar to the graph your tool makes if it works correctly, I think.
Yes and no. You just take the dmmr without the special rules. Most importand, you dont check if the mmr is caped so you work with wrong values.
If you look at my tool you see that most opponents are just 10 points above you. Thats because they are caped and the dmmr value for them is wrong. So i fill in this assumption of their mmr so you see a nice graph without holes. Caped dmmr is the biggest problem and makes over half of all data invalid!
I checked your graph and forced the game into t1. Result is : he is t1 BUT the missing game make the gap anyway.
The anaylser found out , there is a mistake in the data and there is no way to correct it because of missing games.
He try to minimize the mistake and the best way to do so, is to put the opponent of this game in T0 and move the gap 1 dot to the left. This is 10 points better than T1 and the gap on the right. Random mistakes because he have to judge on 1 game and the mistake is smaller than the deviation.
This mistake does not change any other point of the graph and stay local. That is the problem with the new analyser. He is more accurate but he really needs all games. Missing games confuse him but he will correct himself pretty fast again.
My latest patch fix the bug that some bad games are not stored what happend in your case.
|
On June 28 2012 01:40 skeldark wrote:
Yes and no. You just take the dmmr without the special rules. Most importand, you dont check if the mmr is caped so you get total wrong dmmr values.
If you look at my tool you see that most opponents are just 10 points above you.Thats because they are caped and the dmmr value for them is wrong.
That's only partly true, I think. I usually just quit when I see my opponent as favored, because I don't want to waste time on predictable games. These games won't show in your tool because the pretrigger is too slow to catch them. That's why you think my opponents are just 10 points above me. :-) I agree however that the cap rules (which I haven't implemented) will cause the DMMR values to be wrong, but I'm not interested in absolute values, only trends. To prove that the noise is caused by Blizzards shitt... I mean relaxed matchmaking and not my algorithm you can compare these graphs that I made from season 6 and season 7:
![[image loading]](http://www.lysator.liu.se/~john/opmmr.jpg)
|
On June 28 2012 02:04 Mendelfist wrote:Show nested quote +On June 28 2012 01:40 skeldark wrote:
Yes and no. You just take the dmmr without the special rules. Most importand, you dont check if the mmr is caped so you get total wrong dmmr values.
If you look at my tool you see that most opponents are just 10 points above you.Thats because they are caped and the dmmr value for them is wrong. That's only partly true, I think. I usually just quit when I see my opponent as favored, because I don't want to waste time on predictable games. These games won't show in your tool because the pretrigger is too slow to catch them. That's why you think my opponents are just 10 points above me. :-) I agree however that the cap rules (which I haven't implemented) will cause the DMMR values to be wrong, but I'm not interested in absolute values, only trends. To prove that the noise is caused by Blizzards shitt... I mean relaxed matchmaking and not my algorithm you can compare these graphs that I made from season 6 and season 7:
The important point is that caped dmmr is not independent wrong. Its create an depended error over average. Thats the main reason why you cant work with them.
About the 10 points: this is the filler value. He adds that whenever he dont have good data. When the pretrigger was to slow, webgrabber went wrong, calculation went wrong,. bnet error OR the game is caped! (and some other special rules)
You can look at the opponent graph in my tool and enablge "mark predictions". Whats left is the good data.
You graph show me that you dont have many caped opponents. This is luck and can change.
Everyone under 100 can be caped and way lower than this. So if you take the avg middle of the values you will judge yourself way to high! I did the same mistake in my old B algorithm and still have this mistake in part of A.
That is the reason why my newer generation of algorithm dont look at the enemy mmr to calculate your mmr any more.
Something else: Favourite is most likely40-60 - 30-70 chance. So you have still a 30-40% winchance.
Use your chances ^^
|
On June 28 2012 02:12 skeldark wrote:
Everyone under 100 can be caped and way lower than this. So if you take the avg middle of the values you will judge yourself way to high!
Yes, I think I understand this, but I'm not doing any averaging of my data. In season 7 the MMR cap is clearly visible around game 41. All outliers are above the graph and none are below, but don't you agree that it looks like my DMMR was slightly below 100 (edit: or rather at 100) at that time? Don't you agree that my DMMR around game 121 (where the MMR cap no longer is active) was about 300? What I'm saying is that I don't agree with you when you say that DMMR says nothing, I find it very accurate for judging my development despite the noise, at least as long as it doesn't go below 100.
Something else: Favourite is most likely40-60 - 30-70 chance. So you have still a 30-40% winchance.
Use your chances ^^
When they implemented the new sloppy matchmaking I tried and found that I won very few of them. Maybe it's psychological. In any case I very much prefer to play even matches.
|
On June 28 2012 02:45 Mendelfist wrote:Show nested quote +On June 28 2012 02:12 skeldark wrote:
Everyone under 100 can be caped and way lower than this. So if you take the avg middle of the values you will judge yourself way to high! Yes, I think I understand this, but I'm not doing any averaging of my data. In season 7 the MMR cap is clearly visible around game 41. All outliers are above the graph and none are below, but don't you agree that it looks like my DMMR was slightly below 100 (edit: or rather at 100) at that time? Don't you agree that my DMMR around game 121 (where the MMR cap no longer is active) was about 300? What I'm saying is that I don't agree with you when you say that DMMR says nothing, I find it very accurate for judging my development despite the noise, at least as long as it doesn't go below 100. Show nested quote +
Something else: Favourite is most likely40-60 - 30-70 chance. So you have still a 30-40% winchance.
Use your chances ^^
When they implemented the new sloppy matchmaking I tried and found that I won very few of them. Maybe it's psychological. In any case I very much prefer to play even matches.
i agree the dmmr of your opponent say something. But its very inaccurate and you can easy take wrong data with this method and it looks good. F is way way more accurate than the your mmr near to opponent mmr method.
about the fav. If you have a problem with it just put something on the monitor for the waiting time so you dont see it ^^
Sounds simple but this way you dont defeat yourself and see this games as chance to proof you improved.
They are nothing to loose but much to win games!
|
On June 28 2012 03:15 skeldark wrote:
i agree the dmmr of your opponent say something. But its very inaccurate and you can easy take wrong data with this method and it looks good. F is way way more accurate than the your mmr near to opponent mmr method.
Well, if we go back to the original graphs I dont think it is more accurate in this case. There are more problems there than just the single silver game that you pointed out. It doesn't matter if you remove one, two or three games, the jump up to 1160 is too large.
I think I can see the reason if I press "Mark assumtions". Of these 50 games 28 are gray. And no, It's not because I lost on purpose (only maybe 5 times). You maybe don't agree, but I trust the DMMR graph a lot more in this case. There is a lot of noise in it, but that is easy to remove mentally. Don't interpret me wrong, I am not saying that DMMR is a solution for anything, just that there is something wrong with the MMR stats graph in this particular case, and it's more than one game. I just used the DMMR graph as evidence.
|
On June 28 2012 04:35 Mendelfist wrote:Show nested quote +On June 28 2012 03:15 skeldark wrote:
i agree the dmmr of your opponent say something. But its very inaccurate and you can easy take wrong data with this method and it looks good. F is way way more accurate than the your mmr near to opponent mmr method. Well, if we go back to the original graphs I dont think it is more accurate in this case. There are more problems there than just the single silver game that you pointed out. It doesn't matter if you remove one, two or three games, the jump up to 1160 is too large. I think I can see the reason if I press "Mark assumtions". Of these 50 games 28 are gray. And no, It's not because I lost on purpose (only maybe 5 times). You maybe don't agree, but I trust the DMMR graph a lot more in this case. There is a lot of noise in it, but that is easy to remove mentally. Don't interpret me wrong, I am not saying that DMMR is a solution for anything, just that there is something wrong with the MMR stats graph in this particular case, and it's more than one game. I just used the DMMR graph as evidence.
I calculate dmmr. With many special cases. They are gray because you can not trust the DMMR data! I know when the data is correct and when not. You take all data and think because you can calculate the dmmr this is your dmmr. But its not.
What you calculate as dmmr is NOT your dmmr in this games.
That i mark them gray means i calculated that this is bad data and holds no information. You use this bad data because you dont realise its not the dmmr. So marking them gray and delete the data is not a mistake its a step ahead!
The "noise" is the deviation of the f function. you have this one too in your dmmr because you use the same function.
The jump is correct. There are missing games... and also the deviation follows the jump what makes him higher.
You look one one graph and judge per eye what looks better. I did this too. But i later found out, that most of the ones that look good are incorrect. I can produce wonderfull graphs that look total correct with the data but are total wrong.
Now i dont trust my feeling or my eye any-more. I test over the hole data and search for obvious mistakes and than recalculate the possibility,
I do this for several month now and i total understand where you come from. I know what you see in the dmmr of your opponent and how you think you can judge your mmr from this. I did this too! Analyser B and A did exact what you did on a more formal way.
I analysed 20 k games this method and tweaked the analyser for weeks until i found out, this way has a structural problem.
If you do this with the eyes by looking at graphs you take averages in your head if you want or not. You start to interpretate whats not there. Look at the dmmr graphs you posted. Dont you see the average of it when you look at it? you even say "look at the trend".
This is NOT the average dmmr of your opponent because many points in this graph is not the dmmr of your opponents.
You do exactly what i did 6 weeks ago at version 0.9.
When you want to discuss the theory's behind it you find me or notthat often on tl teamspeak.
|
To take two examples: My second last game was a win (+9). I had 167 adjusted points before the match. Thus I calculated my opponents DMMR as 167 - 89 = 78.
In my last game I had 176 adjusted points before the match. I won 13 points. I calculated my opponents DMMR as 176 + 29 = 205.
You claim that these are not the DMMRs of my opponents. Why? What's wrong with it?
|
On June 28 2012 06:02 Mendelfist wrote:
To take two examples: My second last game was a win (+9). I had 167 adjusted points before the match. Thus I calculated my opponents DMMR as 167 - 89 = 78.
In my last game I had 176 adjusted points before the match. I won 13 points. I calculated my opponents DMMR as 176 + 29 = 205.
You claim that these are not the DMMRs of my opponents. Why? What's wrong with it?
the first guy is caped his dmmr is to low he can be -infinity up to 73
under 100 is the "hand" rule. The real function to calculate if someone is a cap canidate is more complicated.
one of many special rules... Like i said, the hole thing is not so easy.
|
|
|
On June 28 2012 06:04 skeldark wrote:Show nested quote +On June 28 2012 06:02 Mendelfist wrote:
To take two examples: My second last game was a win (+9). I had 167 adjusted points before the match. Thus I calculated my opponents DMMR as 167 - 89 = 78.
In my last game I had 176 adjusted points before the match. I won 13 points. I calculated my opponents DMMR as 176 + 29 = 205.
You claim that these are not the DMMRs of my opponents. Why? What's wrong with it? the first guy is caped his dmmr is to low he can be -infinity up to 73 one of many special rules...
Hm, yes you are right. I had a look at these tables, which I didn't read too carefully last time. I didn't think the cap was that aggressive.
|
Yes, I just noticed that. Well, this will cause the low spikes in my opponents DMMR graph to be shifted upwards, and more the lower my ajusted points are.
I'm still trying to cling to my idea that my eyes can compensate for this if my adjusted points are not too low, even though skeldark tries to convince me otherwise. :-) Are there any other pitfalls in looking at DMMR graphs? skeldarks seems to claim they are useless, but I kind of like them, especially since I have data collected from all seasons.
|
On June 28 2012 06:43 Mendelfist wrote:Yes, I just noticed that. Well, this will cause the low spikes in my opponents DMMR graph to be shifted upwards, and more the lower my ajusted points are. I'm still trying to cling to my idea that my eyes can compensate for this if my adjusted points are not too low, even though skeldark tries to convince me otherwise. :-) Are there any other pitfalls in looking at DMMR graphs? skeldarks seems to claim they are useless, but I kind of like them, especially since I have data collected from all seasons.
They are not useless. They are an important part of calculating mmr. But they are just one part.
I give you an example: your caped opponents are at -9001 mmr . So under bronce. You are in middle of your opponents so 300 -9000
you at -4000 to gold thats like -2000 under bronce. possible with the data....
Sure you can guess thats not the case. But you have to back the guess up with calculation and you just guess by looking at the graph.
I try it other way. I do what you do i calculate all this dmrr and much more. So just look at my graph its way more accurate .
The hole dmmr calculation is in there, together with many other things. If you want to use your old data and know the exact change point and adjustedpoints you can translate your data into my gamedata. Open the games.dat, its human readable, you will understand the format i think.
|
On June 28 2012 07:36 skeldark wrote:If you want to use your old data and know the exact change point and adjustedpoints you can translate your data into my gamedata. Open the games.dat, its human readable, you will understand the format i think.
That's an idea, but aren't you dependent on knowing league and division offsets for your algorithm? Blizzard has changed this in the past.
Another thing: The opponent MMR graph in your tool seems to be somewhat misleading. If you discard a DMMR because it is capped, for example between -infinity to 73, shouldn't it be shown as at most 73? The graph seems to use some other higher value. My opponent in my next last match for example has MMR 1174 (gray) which is outside the possible range.
|
On June 28 2012 08:07 Mendelfist wrote:Show nested quote +On June 28 2012 07:36 skeldark wrote:If you want to use your old data and know the exact change point and adjustedpoints you can translate your data into my gamedata. Open the games.dat, its human readable, you will understand the format i think. That's an idea, but aren't you dependent on knowing league and division offsets for your algorithm? Blizzard has changed this in the past. Another thing: The opponent MMR graph in your tool seems to be somewhat misleading. If you discard a DMMR because it is capped, for example between -infinity to 73, shouldn't it be shown as at most 73? The graph seems to use some other higher value. My opponent in my next last match for example has MMR 1174 (gray) which is outside the possible range.
tiers only change a little bit never big, because any change is a correction to keep it in the gauss standard deviation. But that is a complete different topic...
Like i said the assumptions for opponents are filler points. they are not true at all. 73 would not be true also. Its just there so you have a graph and not only single points. This way its obvious they are not true and to be honest i did not care so much because its only for the opponent and he is not very interesting.
Also judging your own tier is way harder than opponent tiers, so he is less accurate.
I know it looks like own tier is easier because i have way more games of your tier and only one game for every opponent tier. But your mmr have something i can use: they played by the same guy! the opponent mmr have no rule they follow so even with more games its harder to judge. I can only judge on your mmr, and compare your and opponent mmr is problematic (what i try to explain for 2 pages now ) because the caped games will, filtered or not, change the diffrence.
I even thought about removing the opponent completely because i was afraid people could think they can judge them-self over the opponent. Instead i put him on your line when i dont know the value, so even if people do this mistake they are at-least not total wrong.
I hope this showes what i meant when i said: "I total understand where you come from". I know its easy mistake to judge on your opponent because i did the same in the beginning and this is my way to solve the problem in the program.
|
Dakota_Fanning
Hungary2349 Posts
On June 28 2012 00:04 skeldark wrote:
I think the problem is that part of my code throw an expectation on loading and sc2gears dont output this mistake.
No easy fix can be everything but most likely the libs for scanner and registry.
That cannot be the problem. Plugin's code is not executed until the plugin is manually started from the Plugin manager.
If a plugin is not enabled (this is the case with a fresh install of Sc2gears and the plugin), on startup Sc2gears only detects the plugins in the plugins folder, but does not load the jars. Plugin jars are only loaded if the plugin is started. So if the plugin does not appear in the plugin list, it cannot be due to failing of loading the jars or starting the plugin.
I still think it's due to file permission errors. For a plugin to appear in the plugin list, only the following things are required:
-Have a folder inside the "Sc2gears/Plugins" folder.
-Must have a "Sc2gears-plugin.xml" plugin descriptor file which must be syntactically valid (if it's not valid, proper error message is logged).
-In the descriptor the plugin API version must be valid (if not, error message will be logged).
-Its main class must be unique among the installed plugins (again, error message is logged if not).
-The plugin must have at least 1 jar file (error message is logged if not).
If all these conditions are met, the plugin will appear in the plugin manager. You can see, I log errors if any of the conditions are not met.
|
|
|
|
|
|
EPT
