Match Making Rating Tool - Page 137

korona

1098 Posts

April 02 2013 23:00 GMT

#2721

On April 03 2013 07:16 DoubleDice wrote:
Last game opponent have won but still same problem, what maybe happening?

LOG FILE:
+ Show Spoiler +

18:40:43 MMR-Stats 9.0
18:40:43 --------INIT START--------
18:40:43 OK File-Encoding UTF-8
18:40:43 Loading HOTS DB file...
18:40:43 OK Loading DB version 12.2 done
18:40:45 OK DB up to date!
18:40:45 Loading character profiles:
18:40:45 Profile: http://us.battle.net/sc2/en/profile/612917/2/DoubleDice/
18:40:45 OK Settings loaded!
18:40:45 OK Timeclient: Systemtime offset: -1 sec
18:40:45 OK Gamedata: 7 Games loaded.
18:40:45 Loading Archive...
18:40:45 OK Gamedata: 7 Games loaded.
18:40:45 Mac-OS detected!
18:40:45 Disable Memoryreader...
18:40:45 Disable Gameobserver....
18:40:45 Memory scanner is disabled
18:40:45 OK Triggers ready!
18:40:47 OK: Program is up to date
18:40:47 --------INIT END--------
19:14:29 ERROR: preplayer data is for wrong character. Setting pregame data invalid
19:15

Post-trigger: Reading web profiles
19:15:15 Webgrabber: Player true
19:15:20 Webgrabber: Opponent true
19:15:20 Post-trigger: Done reading web profiles
19:15:20 Calculating adjusted points for the players (to be successful either Alg. A or B needs to succceed for certain player
19:15:20 DATA Alg. A : failed, No pre-game data (RAM scanner possibly off) or pre-game data bad:
19:15:20 DATA Alg.B: failed, Unknown change points for player: http://us.battle.net/sc2/en/profile/612917/2/DoubleDice/
19:15:20 A -9999 B -9999

19:15:20
19:15:20 ERROR AFTERCHECK All algorithem failed
19:15:20 DATA Alg. A : failed, No pre-game data (RAM scanner possibly off) or pre-game data bad:
19:15:20 DATA Alg.B: failed, No Bonus Win: http://us.battle.net/sc2/en/profile/287860/1/Vinq/
19:15:20 A -9999 B -9999

19:15:20
19:15:20 ERROR AFTERCHECK All algorithem failed
19:15:20 ERROR LASTCHECK : Both players invalid
19:15:20 Gamecheck: false
19:15:20 GAME:
HotS
19:15:20 PREPLAYER 1:
Data: false
Reading: false
Adjusted: false
Dmmr: false
id:
playerlink:
game: Unknown
server:
league: ERROR
division_name: ERROR
Main race: unkown
Game race: unkown
gamedate: ERROR
mapname: ERROR
totalbonuspool: -9999
Wins: -9999
Looses: -9999
points: -9999
bonus: -9999
change_points: -9999
adjusted_points: -9999

19:15:20 PREPLAYER 2:
Data: false
Reading: false
Adjusted: false
Dmmr: false
id:
playerlink:
game: Unknown
server:
league: ERROR
division_name: ERROR
Main race: unkown
Game race: unkown
gamedate: ERROR
mapname: ERROR
totalbonuspool: -9999
Wins: -9999
Looses: -9999
points: -9999
bonus: -9999
change_points: -9999
adjusted_points: -9999

19:15:20 POSTPLAYER 1:
Data: false
Reading: false
Adjusted: false
Dmmr: false
id: 2-612917
playerlink: http://us.battle.net/sc2/en/profile/612917/2/DoubleDice/
game: HotS
server: US
league: diamond
division_name: banshee rho
Main race: z
Game race: z
gamedate: 4/2/2013
mapname: Neo Planet S LE
totalbonuspool: 338
Wins: 42
Looses: 42
points: 385
bonus: 1
change_points: -9999
adjusted_points: 48

19:15:20 POSTPLAYER 2:
Data: false
Reading: false
Adjusted: false
Dmmr: false
id: 1-287860
playerlink: http://us.battle.net/sc2/en/profile/287860/1/Vinq/
game: HotS
server: US
league: diamond
division_name: boros zeta
Main race: t
Game race: t
gamedate: 4/2/2013
mapname: Neo Planet S LE
totalbonuspool: 338
Wins: 69
Looses: 64
points: 395
bonus: 0
change_points: -9999
adjusted_points: 57

19:15:20 Dataset: (GAME: HotS)
--- GAMESTART ---
TIME: 1364940868426
VALID: false
Result for player 1: L
--- GAMEEND ---

19:15:20 POST-Trigger done
19:15:20 <FONT COLOR="DD0000">Player-data-Qualitiy: 50% </font>
19:15:20 <FONT COLOR="DD0000">Opponent-data-Qualitiy: 50% </font>
19:15:21 OK Uploader: game uploaded
19:15:21 Ladder Game Done!
19:15:21 --------------------------

The program is still functioning correct (at least regarding both old and new example game data) like I said in my previous post. It only gives that one false error message, which I have already fixed for the future version. But that is only an error message that did not affect how the program works at all. Blank pregame profiles had already been set bad earlier.

We don't have a memory scanner that supports Mac. Mac does not have Windows registry into which the SC2 client inputs APM values meant for Razer peripherals. By those values we detect when the game starts. On Mac we cannot detect it, so we don't have pre-game profiles for the players. I can't even investigate this further as I don't have Mac. And as there is not pre-game profile for the opponent, his change points can only be calculated if he wins and has bonus pool left. Before patch 2.0.4 the change points were available when opponent lost, but not anymore as Blizzard hid them.

And if you look at the example player you gave:
totalbonuspool: 338
bonus: 0

This means that maximum bonus pool was 338 at that moment, but he had 0 bonus points left after the match. --> We don't know how many points he got. If he had bonus points still left, the change points would have been the X points match history tells he got divided by 2. Now the change points he got could be anything between x to x/2 with max value of 24 (2 exceptions: if match history tells he got 47 or 48 points, his change points would be 24. Or if it tells he got 1 point, his change points would be 1). As his change points were unknown they were set to -9999, which is default value used by the program for unknown/bad values.

noeljones

Australia119 Posts

April 03 2013 06:40 GMT

#2722

How long do you expect it will take to find new values for the hots promotion line?
Thanks for all your work

TrickNachos

United States23 Posts

April 03 2013 08:11 GMT

#2723

I have 5/158 good games, but now it's only showing the latest 3 games for some reason? Where is the rest of the graph?

korona

1098 Posts

April 03 2013 08:18 GMT

#2724

On April 03 2013 15:40 noeljones wrote:
How long do you expect it will take to find new values for the hots promotion line?
Thanks for all your work

I will probably take a look in next week's week-end. During last season there was barely enough data after 2 weeks to be able to make educated guesses for offsets for the lower leagues (later rechecks confirmed the guesses were close enough). But for higher leagues there will likely be enough data.

I haven't yet updated the analysis tools, but expect that they only need minor changes (I presume there won't be complicated dependencies in them, but I may be wrong).

korona

1098 Posts

April 03 2013 08:21 GMT

#2725

On April 03 2013 17:11 TrickNachos wrote:
I have 5/158 good games, but now it's only showing the latest 3 games for some reason? Where is the rest of the graph?

Maybe you have set the filter to last 10 games or have wrong character chosen or have archive open, or have the filter set to last league & you were just promoted.

Edit: and that low amount of good games is due you have not enough rights to run the mem scanner. Thus the tool acts like the memory reader would be disabled.

korona

1098 Posts

April 03 2013 08:32 GMT

#2726

Unplanned/unofficial support for tracking unranked MMR

I realized yesterday that there is an 'unplanned' support in the MMR tool to track unranked MMR. As there is no feasible (or user friendly) way to fix it (prevent it), I'll explain it. And you can use it to track MMR for your unranked characters if you wish. Just do not mix unranked data with ranked data. Also note that if a reliable way to detect unranked games in the future appears (e.g. a field in a replay file is found) I will switch it on in some future version, but it won't 'fix' the old data.

By design the tool does not store the data for unranked matches. But as web profiles are unreliable, we cannot let false positives to cause data from the ranked matches to be dropped. Thus there is a rule that unranked detection needs that both pre-game and post-game profiles are partially valid for the player himself. If there is any problem regarding those parts of the profiles then that match is set to ranked. Two basic examples what would cause pre-profile of the player be invalid: 1) Blizzard has maintenance & the web profile is unavailable, 2) player is playing placement matches and he does not yet have web profile for 1v1.

The second reason actually made the unplanned unranked MMR tracking for characters that only play unranked possible. If you haven't finished your placements & only play unranked you will not have 1v1 web profile for that character. Thus unranked detection won't activate. Thus you can track unranked MMR for characters that have not finished their 1v1 placements. Yes the tool still marks those matches 'ranked', but that does not matter if you only play unranked with those characters. If you finish your placements, the unranked detection will switch on and data for all unranked matches from that point forward will be dropped (unless e.g. the web profiles are unavailable or match is less than 30 sec (real-time) long).

Also as Mac players don't have pre-game profiles, the unranked detection is always off for them. Don't let ranked & unranked games mix (manually delete unranked data if they are recorded for your ranked character).

TrickNachos

United States23 Posts

April 03 2013 09:52 GMT

#2727

Cool! I'll be doing this on my unranked characters then xD. So this should mean it might be a little easier than expected to track unranked mmr? That would be pretty crazy =D.

Also about the wrong character korona... I don't have the wrong character chosen, it isn't the archive that is open. Filter isn't set the last league and sadly I'm not promoted (even though I'm #1 dia with 8 consective wins atm -_-).

How do I let the memory scanner have enough rights?

Also, the graph says 157, 158 , 159 etc at the bottom so maybe i have too many games for the graph?

EDIT: Anyone who is looking to get into master I got my promotion at 1,777 on EU server which is weird since according to mmr stats my mmr has been much higher than that (1900 range). What weird is I even played a diamond to get into masters. I had a 9 win streak though at rank 1 dia so I hope that helps someone out there. Just don't lose! Lol...

korona

1098 Posts

April 03 2013 15:32 GMT

#2728

On April 03 2013 18:52 TrickNachos wrote:
Cool! I'll be doing this on my unranked characters then xD. So this should mean it might be a little easier than expected to track unranked mmr? That would be pretty crazy =D.

No. It is as hard as it has always been. This also clearly shows how web profile based unranked / ranked detection is unreliable as Blizzard decided to hide what mode you are playing. All it needs is that there is something wrong in the web profiles. For example after today's maintenance EU web profiles have been inaccessible (are now fixed, but were still broken ~one hour ago). You cannot do the unranked/ranked detection based on web profiles that are not available.

And like I said in my earlier post regarding the unranked MMR the counting is not the problem: http://www.teamliquid.net/forum/viewmessage.php?topic_id=334561&currentpage=132#2637 It is the detection & sorting correct games to correct profiles.

Here some problems regarding adding a support fo unranked profiles:

* Web profile based unranked / ranked detection is not 100% reliable --> the tool would store match data under wrong profiles

* Amount of character profiles would become double --> user would have hard time to find / remember to select the correct profile unless improvements to the UI would be made. Also automatic profile detection would become more complex

* If the tool would store match data under wrong profiles, there would need to be a feature that allows user to move matches easily between profiles --> But you cannot trust this to basic users. Some would mix all up and get corrupted profiles contained mixed match data from several profiles. Would need several warnings & somehow making sure that data cannot be accidently moved

* If the match data is mixed, you cannot rely on it at all. E.g. offset calculation would be based on chance - was that data mixed or was it correct . The MMR graphs themselves recover when next 'good game' is played, but it would lead to spikes if there is a difference between separate ranked & unranked MMRs. At the moment I can filter out the unreliable data that has a chance to be from unranked matches regarding e.g. offset analysis. But not necessarily could after the changes

* Adding unranked profiles would add lots of new breaking points to the code + as the data would be mixed it would become harder to detect if the emerging problems would be caused by a bug or just because of bad data

* Numerous other concerns

* Adding support for unranked profiles would get us in worse situation than we are currently. There is no 100% unranked/ranked detection at the moment.

* All of the above changes would need lots of work regarding implementation & much more regarding testing and I don't have that time at the moment. Had to budget much more time and effort even for the HotS conversion than I really afforded (many personal things were postponed due to the conversion that should have been started earlier). MMR tool is a legacy program (legacy in this case means it was not made by the current developers) that has tens of thousands lines of code (Yes skeldark made a large effort to create this tool in the first place). On Java side (nearly entirely my responsibility) the conversion required some 2500 to 3000 new / changed lines (+ other code that was removed during refactoring). That is not technically much, but the testing took multiple times more time than the actual implementation to make sure e.g. some sneaky dependencies did not break. And it seems we got a clean release: there has not been any reported bugs that would affect how the program works. After the release some bugs caused by the old code has been found & fixed. Also some new bugs that e.g. caused some wrong error messages have been found & fixed. But nothing yet has been found that would affect the main functionality of the tool. Also not rushing the release was a good decision. Even during the last day before the release some bugs that would have affected the functionality emerged & were fixed.

Current situation:

At the moment most of the unranked matches are dropped automatically (if web profiles are available). If some are recorded here and there the user can manually delete them (these are usually matches that last under 30 sec). The unranked detection does not give many false positives (e.g. I haven't heard any yet) which would cause ranked match data to be dropped (tie games are exception. Tie games are supposed to be dropped). Also at the moment there are clear rules how I can ignore the data that has chance to be unranked when calculating offsets etc. And now we have even unofficial support for tracking unranked MMR if the user does not play his placement matches with that character.

This seems a good place to be at compared to the potentially numerous problems that adding separate unranked profiles would causer at this point of time as detection is not 100%. At the moment it is not worth the effort do the unranked profile separation. The corner stones for future implementation have been already laid down (needed data structures and save file support has already been added). If a reliable detection method emerges in future & there is time to do the implementation & testing, unranked profiles may become reality in some point of the future but not today.

On April 03 2013 18:52 TrickNachos wrote:
Also about the wrong character korona... I don't have the wrong character chosen, it isn't the archive that is open. Filter isn't set the last league and sadly I'm not promoted (even though I'm #1 dia with 8 consective wins atm -_-).

And still the reason likely is some filter. Blizzard broke their EU web profiles this morning (they seem to be now back online. One hour ago they were still broken). When web profiles were unavailable your league was likely set 'unknown' & season potentially to be incorrect for those matches.

It might be possible to detect the league based on data stored by SC2 in the replay, but there are questions. That league data in the replay is always status before the match & Blizzard seems at least on the profiles show 'the best' league in that format (weather it be WoL or HotS). Not necessarily worth adding that functionality either...

I cannot at the moment check the user graphs based on uploaded data as I have not had time to update the analysis tools yet. Also if there was some bug regarding the public client it might not show up in the analysis client as some of its code is entirely separate. Would need the games_i.data to check (this also could be formed from the uploaded data, but tools not yet updated). But this problem is likely caused by some filter setting .

On April 03 2013 18:52 TrickNachos wrote:
How do I let the memory scanner have enough rights?

You probably just need to run the sc2gears as administrator. And based on your PMs it seems to have worked for you. Sorry we cannot provide basic IT support.

On April 03 2013 18:52 TrickNachos wrote:
Also, the graph says 157, 158 , 159 etc at the bottom so maybe i have too many games for the graph?

There is no limit (the more data you have the more it needs memory & may take longer to calculate the graph). Some graph filter is likely selected (make sure they are all 'all').

On April 03 2013 18:52 TrickNachos wrote:
EDIT: Anyone who is looking to get into master I got my promotion at 1,777 on EU server which is weird since according to mmr stats my mmr has been much higher than that (1900 range). What weird is I even played a diamond to get into masters. I had a 9 win streak though at rank 1 dia so I hope that helps someone out there. Just don't lose! Lol...

This info is not necessarily too useful at the moment as offsets are not correct. Indeed if the last measured MMR was against diamond league player the master promotion line would likely be somewhere little over 1700 on EU (e.g. my personal promotion was at ~1750 based on diam opponent). But if the last player was plat or master from which your MMR was calculated, it would give some totally different value (offsets are incorrect).

Scumi

Germany108 Posts

April 03 2013 16:49 GMT

#2729

I played a bunch of laddergames and got the following message every time.

+ Show Spoiler +

18:24:6 Game Start detected
18:24:6 Searching memory for character profiles for game HotS
18:24:6 Reading Memory ...OK
18:24:6 ERROR: Could not find memory profile for the opponent! Perhaps its just a custom game...
18:24:46 Pre-trigger: Reading web profiles
18:24:51 Webgrabber: Player true
18:24:51 No Profile, skip
18:24:51 Webgrabber: Opponent false
18:24:51 Pre-trigger: Done reading web profiles
18:24:51 Pre-Trigger done

korona

1098 Posts

April 03 2013 17:03 GMT

#2730

On April 04 2013 01:49 Scumi wrote:
I played a bunch of laddergames and got the following message every time.

+ Show Spoiler +

Yes. Just noticed it myself. It seems in 2.0.7 static memory addresses were slightly altered. At least the second address was found with one run with the dynamic version of the mem reader (the other was different player slot, but likely because custom games use different slots than 1v1 games). Sent request for InfCereal to update the static reader.

Depending if he has time today & if Skeldark is available to publish the new version, we might have an update later today (either partial (only mem reader), or full v 9.1 (there are some changes that have not yet been tested, even if they are unlikely to cause problems). If it does not make it today we will try to get it soon.

Edit: Waiting test results from release candidate 9.1. It includes updated memory reader & small updates to the tool like the the getting started screen. It was originally intended to be released during the weekend. If tests pass it will likely be released today. Please note I had intended that 9.1 changes would have been tested more thoroughly, but as changes were quite simple, it should be safe to release if basic tests pass.

TrickNachos

United States23 Posts

April 04 2013 02:07 GMT

#2731

Korona, first of all I'd like thank you for all your hard work, time, and effort that is put into this project. Everyone here on teamliquid appreciates what you're doing for the community and we love the fact that we have some estimations on where our hidden MMR is since blizzard has chosen to hide it from us =). Talking about unranked support when detection isn't at 100% is too much to ask at this time, I was just spouting out ideas, that's all. You are right about the web profiles as my graph seems to have been fixed automatically, I have idea why it was like that, but when I restarted my computer, my graph was back to normal, so nothing is wrong here. Thank you for your continued support with mmrstats.

skeldark

Germany2223 Posts

April 04 2013 03:27 GMT

#2732

9.1 up

ktimekiller

United States690 Posts

April 04 2013 09:06 GMT

#2733

My games are not being recognized before and after 9.1

1:35

Game Start detected 1:35:49 Pre-trigger: Reading web profiles 1:35:52 Webgrabber: Player true 1:35:52 No Profile, skip 1:35:52 Webgrabber: Opponent false 1:35:52 Pre-trigger: Done reading web profiles 1:35:52 Pre-Trigger done

korona

1098 Posts

April 04 2013 09:52 GMT

#2734

On April 04 2013 18:06 ktimekiller wrote:
My games are not being recognized before and after 9.1

1:35

Hi! As you see that part of the log tells about reading the web profiles in the start of the game. Please check if you have set correct web profile URL in the settings + I would need to see more of the log than just that part.

Yesterday EU web profiles were unavailable for several hours & after that failed more frequently than normal: Also US profiles failed yesterday more frequently than normal but were more stable than the EU profiles. With a quick look to Today's uploaded data, the others seem to be getting data ok...

Aelfric

Turkey1496 Posts

April 05 2013 06:49 GMT

#2735

Does master leagues have offsets? In wol i remember they are not having any offsets and all master leagues were equal. Is it still the same? So if you are mostly playing vs master league opponents can you rely on what the program shows about your mmr?

Hairy

United Kingdom1169 Posts

April 05 2013 07:05 GMT

#2736

On April 05 2013 15:49 Aelfric wrote:
Does master leagues have offsets? In wol i remember they are not having any offsets and all master leagues were equal. Is it still the same? So if you are mostly playing vs master league opponents can you rely on what the program shows about your mmr?

You are correct that all master leagues were created equal, but I think (and correct me if I'm wrong) that they actually removed league offsets across ALL leagues a little while ago.

From what I understand, the MMRs are actually accurate. What aren't accurate (yet) are the league promotion lines.

leopardb

Germany92 Posts

April 05 2013 08:21 GMT

#2737

Hi, i'm using MMRStats under linux and i tried a new experiment : SC2Gears/MMRStats with Wine and the latest windows JRE. The thing works, SC2Gears even sees when a games starts or ends (saying "game started"), MMRStats also works, but as with the native linux java machine, the memory scanner does not work. Now the 2 programms SC2.exe and java(SC2Gears) run on the same PREFIX (which means the share the memory and can "see" each other). The offset must naturally be different and i know that they're static right now but my question is : do you think the new memory scanner will be able to find the right address in my setup ? after all, the memory is the same when using SC2 under wine since the DLL are the windows ones (wine is not an emulator). Also one problem is : in both possibilities native and wine java machines, the registry cannot be read (because of the way wine exposes it to the processes). Is it a/the problem ?

InfCereal

Canada1759 Posts

April 05 2013 10:36 GMT

#2738

On April 05 2013 17:21 leopardb wrote:
Hi, i'm using MMRStats under linux and i tried a new experiment : SC2Gears/MMRStats with Wine and the latest windows JRE. The thing works, SC2Gears even sees when a games starts or ends (saying "game started"), MMRStats also works, but as with the native linux java machine, the memory scanner does not work. Now the 2 programms SC2.exe and java(SC2Gears) run on the same PREFIX (which means the share the memory and can "see" each other). The offset must naturally be different and i know that they're static right now but my question is : do you think the new memory scanner will be able to find the right address in my setup ? after all, the memory is the same when using SC2 under wine since the DLL are the windows ones (wine is not an emulator). Also one problem is : in both possibilities native and wine java machines, the registry cannot be read (because of the way wine exposes it to the processes). Is it a/the problem ?

We have a dynamic one that would. I have no idea how linux w/ wine works, but all the dynamic scanner does is read the memory from the module base (The portion of the sc2 process containing all the IDs) to it's end, and returning a value once it finds the long ID of any of your accounts.

We never touch the registry to my knowledge.

L33tmaster

United States13 Posts

April 05 2013 11:05 GMT

#2739

On March 31 2013 08:03 JazVM wrote:
FYI (dont know if it is valid info though) I got promoted from diamond to masters at around 1650-1700 according to the tool.

I got promoted at 2045. NA server.

korona

1098 Posts

April 05 2013 12:09 GMT

#2740

On April 05 2013 16:05 Hairy wrote:

Show nested quote +

Inside each league offsets do not matter. E.g. if you are in diamond and play a 'good game' against another diamond player, you get ok relational MMRs against him / other diamonds (with certain error marginals). But against people from other leagues there are offsets. And those offsets in many cases are incorrect at the moment.

For example:

On April 05 2013 20:05 L33tmaster wrote:

Show nested quote +

I got promoted at 2045. NA server.

L33tmaster's last 'good game' when he was in diamond was likely against master player (actually it might have been even when he was in plat against master). JazVM's last good game was likely against platinum player (or if it was against diam, there was several games after that good game). (note that I have not yet looked at general data and only base these 'guesses' only based on small sample of data). As the offsets are incorrect they had considerably different calculated MMRs when they were promoted.

On April 05 2013 19:36 InfCereal wrote:

Show nested quote +

Windows registry is used to detect the match start. SC2 client inputs APM values to the registry originally meant for Razer's peripheral. Based on those values we detect when the match starts. If this detection fails there are no pre-game profiles for either players & memory reader is never even attempted to be used.

Prev 1 135 136 137 138 139 207 Next

Please or register to reply.

Match Making Rating Tool - Page 137

Completed

Ongoing

Upcoming