EPTJust a thought.
Match Making Rating Tool - Page 187
| Forum Index > SC2 General |
|
ScorpionSC
United States21 Posts
Just a thought. | ||
|
korona
1098 Posts
On May 19 2014 06:39 ScorpionSC wrote: Getting some weird errors and MMR-stats hasn't worked right for me in a while. I deleted everything and reinstalled, here is the game log. + Show Spoiler + 16:03:35 Adjusted calculator ERROR: Invalid player data: http://us.battle.net/sc2/en/profile/2792558/1/Decy/ 16:03:35 DATA Alg. A : success: http://us.battle.net/sc2/en/profile/2898584/1/ShadowedRadi/ 16:03:35 DATA Alg.B: success:http://us.battle.net/sc2/en/profile/2898584/1/ShadowedRadi/ 16:03:35 Gamecheck: true 16:03:35 POST-Trigger done 16:03:35 Data quality needs to be 100% to be able to calculate MMR 16:03:35 Data quality for calculating Player's MMR: 75% 16:03:35 Data quality for calculating Opponent's MMR: 50% 16:03:37 OK Uploader: game uploaded 16:03:37 Ladder Game Done! 16:03:37 -------------------------- 16:03:37 -------------------------- 16:03:37 Game Start detected 16:03:37 Searching memory for character profiles for game HotS 16:03:37 Reading Memory ...OK 16:03:37 OK Found Memory-Profile in Active-Profiles! 16:04:17 Pre-trigger: Reading web profiles 16:04:20 FATALERROR WEBGRABBER could not fetch season number! EXPECTED: 18. Player: http://us.battle.net/sc2/en/profile/2792558/1/Decy/ 16:04:20 Webgrabber: Player false 16:04:23 Webgrabber: Opponent true 16:04:23 Pre-trigger: Done reading web profiles 16:04:23 Pre-Trigger done 16:27:49 -------------------------- 16:27:49 Game Start detected 16:27:49 Searching memory for character profiles for game HotS 16:27:49 Reading Memory ...OK 16:27:49 OK Found Memory-Profile in Active-Profiles! 16:28:29 Pre-trigger: Reading web profiles 16:28:32 FATALERROR WEBGRABBER could not fetch season number! EXPECTED: 18. Player: http://us.battle.net/sc2/en/profile/2792558/1/Decy/ 16:28:32 Webgrabber: Player false 16:28:36 Webgrabber: Opponent true 16:28:36 Pre-trigger: Done reading web profiles 16:28:36 Pre-Trigger done 16:43:51 Post-trigger: Start 16:44:51 Post-trigger: Reading web profiles 16:44:54 FATALERROR WEBGRABBER could not fetch season number! EXPECTED: 18. Player: http://us.battle.net/sc2/en/profile/2792558/1/Decy/ Any ideas what is going on? The tool tries to solve the season number primarily from the web profile, but that information disappeared from the profiles in the start of this season. If it cannot get the season number from the web profiles, it tries to fetch it from the SC2 API. The SC2 API returns an error to all those who use RoS CE icons like you (the API always breaks if new icons are used. Blizzard usually fixes that a week or two after the introduction of the icons, but for RoS icons they have not done so). When the season number is not available, the data for that player is set bad for that match. If the tool manages to fetch the season number for the other player then that number is used for the match. You can solve your problem by not using D3 RoS CE icons. (Thought there was an in-game bug that prevents the user from selecting the RoS CE icons again (they disappear), if he switches to some old icon. I don't know if they have solved that problem yet.) | ||
|
korona
1098 Posts
On May 19 2014 09:23 ScorpionSC wrote: Hey Korona, I just had a thought. I know that a lot of the times a game is not able to be used to determine MMR because there is one or more pieces of information that makes it impossible to calculate (like bonus pool trickery etc). What about having an option for after a match to be able to let the user input the unknown information? Like if you can't use my last match because the opponent had bonus pool, couldn't I go determine how much bonus pool he got for the last match, enter it and have the tool then calculate the data etc? Would obviously only be able to "adjust" the data for the last game played and only for a small set amount of time after the match, but seems like it might help get a few more "good" games rather than taking upwards of 50 matches before a calculation can happen. Just a thought. The user himself cannot determine the opponent's data any better than the tool as you need pre-profile info to be able to calculate the changes for the opponent. Opponent's data is used to calculate the player's MMR. But the user could check the pre-match and post-match data for himself. Also the user can check which part of the points he received was taken from the bonus pool and which part was 'new points' by hovering the mouse cursor over the points in the SC2 in-game-score screen (this cannot be done for the opponent). The user's data is used for calculating the opponent's MMR. But these kind of tasks cannot usually be trusted to the user. Yes some would carefully input the correct information, but many would make mistakes. Also some might purposely enter wrong data. And the user inputted invalid data would make it a mess e.g. for determining the offsets from the uploaded data. Also the current way the MMR tool handles the data collection sets limitations. It is complex with lots of checks and other moving parts. It would take lots of rework to make such feature. And such feature would not help regarding the opponent's data from which the user's MMR is calculated. | ||
|
korona
1098 Posts
Regarding the out-of-sync issue of the web profiles I have to see if I have time to make larger changes how data is fetched during the summer. Some kind of web profile polling based solution with more concurrency might do the trick. But in all logical paths that I have thought of there are problematic things and I am not sure if it is worth of the effort as it would require larger changes to the current implementation. Of course if Blizzard would fix their web-profile updates this problem would disappear. | ||
|
Danglars
United States12133 Posts
On May 19 2014 16:42 korona wrote: I'm so ho happy for the effort, and I continue to use the tool on my games to help the cause. Let's hope Blizzard has some solutions for their web profile updating procedures soon.2014-05-19: Update regarding the offsets. So considerable percentage of the matches are affected by the out-of-sync web profile issue that I got very low amount of offset calculation results when I ran the calculations yesterday. This means I have to solve the offsets by hand based on the user graphs and that is time consuming and error-prone process. I did the first pass yesterday but did not find values that I would be pleased with. I will continue improving the numbers in the near future. Regarding the out-of-sync issue of the web profiles I have to see if I have time to make larger changes how data is fetched during the summer. Some kind of web profile polling based solution with more concurrency might do the trick. But in all logical paths that I have thought of there are problematic things and I am not sure if it is worth of the effort as it would require larger changes to the current implementation. Of course if Blizzard would fix their web-profile updates this problem would disappear. | ||
|
YoloSwaggins87
United Kingdom1 Post
| ||
|
korona
1098 Posts
On June 08 2014 21:54 YoloSwaggins87 wrote: I Guys do you know when the new off sets will be calculated and the mmr stats will be updated? Hopefully in the coming week. Been busy, but I did look at the data for couple of hours earlier today. Seemed to confirm my hypothesis regarding some leagues that I made some weeks ago, but it was too time consuming to confirm it back then due data errors. Also the two lowest leagues would have been shots in dark back then, but now are hopefully solvable in reasonable time. But that is the next step. After that the new version of MMR tool needs to be tested and that will take time too (did the archiving some three weeks ago, but did not have time to test). I hope I have time to check the WoL offsets too, but as there is a lot less data (roughly 4 % as much as HotS data) it will be quick and dirty check. But I will be AFK for an extended weekend in the coming week, so we will have to see if I make it before that or will it take longer. -- But here is some data regarding offsets without giving the exact numbers: ma-gm has roughly doubled, di-ma has considerably increased, but the master threshold has been separately set to a lower point than the offset would suggest, pl-di and go-pl seem unchanged (or small increase for pl-di and small decrease for go-pl), si-go and br-si both have decreased but I have not yet solved them. This means that most players from mid gold to mid to high diam have relatively correct MMR with the old offsets. Also if you play only against players from your league, you have relatively correct MMR. But data errors still happen due to web-profiles that are often updating out-of-sync and lots of accounts are suffering from them. But if you have lots of 'good games', then it is usually straightforward to deduce which ones are good and which ones are incorrect. I also noticed that both offset changes that Blizzard did this year happened 21 days after the start of the both seasons. Thus it is likely that the next offset changes happen 3 weeks after the start of the next season. | ||
|
kalimann
9 Posts
| ||
|
korona
1098 Posts
On June 12 2014 21:15 kalimann wrote: I get you are updated some things etc., I just want to be sure you are aware some stats get as skewed as this: http://i.imgur.com/1qzK2PB.png Data errors happen more frequently as Blizzard's web profiles are updating quite often out-of-sync. In the web-profile match history updates usually immediately, but the ladder page may update immediately or may take several minutes before the update. If the ladder page has not updated when the tool fetches the data 1 minute after the match, it may result a data error if the tool does not recognize that something is wrong (there are plenty of different checks, but sometimes it is not possible detect if the profile was out-of-sync). In these cases it may actually calculate your opponent's previous opponent's MMR. The games from which your MMR is calculated are called 'good games' in MMR tool slang. To graphically show 'the good games' go to 'Confic' > 'Data' > 'Mark assumptions / bad games' and save the settings. Now you can graphically see from which games the MMR was calculated. If you have plenty of 'good games', you can easily regocnize which ones are inline with each others (likely correct calculations) and which are not (likely incorrect). You can set the likely incorrect games 'bad' from 'Config' > 'Data' > 'Refresh' > select correct game from the list > 'Mark as bad game' and save when done. But Blizzard changed offsets 1st of May. Thus games played after that are not directly comparable with games played before that. The current version of the MMR tool does not yet support the new offsets. But hopefully later today a new version of MMR tool is published that will archive the games that were played before the change and update the offsets. It will fix your graph if you did not have any data errors. But I suspect there are some. I should probably do some graphical guide in the future how to recognize bad values as they are now more frequent due to the out-of-sync issue of the web profiles. | ||
|
kalimann
9 Posts
 Does the tool update itself or is it done manually? | ||
|
korona
1098 Posts
On June 13 2014 00:11 kalimann wrote: Ok I see this, to be honest I still don't quite understand how the calculations are made, but that's not as important. Does the tool update itself or is it done manually? The DB file updates automatically when Sc2gears is launched if there is a new DB file available. The tool itself does not update automatically. A big "Download new Version" button will appear in the upper toolbar when a new version of the tool is available (it checks the version when you start Sc2gears). When you push the update button, it downloads and unpacks the new version. You need to restart Sc2gears after that. | ||
|
korona
1098 Posts
Please restart sc2gears to get 'Download new Version' button to appear. The change notes are available from the original post. Some further notes can be found below. -- Notes regarding the new offsets: Blizzard changed the offsets 1st of May. This was 21 days after the start of this season (S18). Last time they changed the offsets was 21 days after the start of last season in January. Thus one could expect that the next time Blizzard changes the offsets is 3 weeks after the start of next season. A log containing start dates for different seasons & offset change dates was added to the original post. -- Due to the out-of-sync issue of the web profiles the offsets were mostly solved based on the user graphs. This means there could be some inaccuracies (especially bronze-silver offset), but the higher leagues should be close. Here are the new offsets (from 2014-05-01): br-si 200, si-go 220, go-pl 280, pl-di 240, di-ma 580, ma-gm 160 The old offsets were (from 2014-01-24 to 2014-05-01): br-si 280, si-go 260, go-pl 290, pl-di 235, di-ma 450, ma-gm 85 As you see two lowest offsets were decreased, go-pl & pl-di remained pretty much the same (or potentially exact the same), di-ma increased and ma-gm roughly doubled. For other leagues than GM Blizzard usually has used league thresholds that reflect the offsets. This time they seem to have set master threshold much lower thought (offset suggests 1520 but the threshold seems to be at 1410). The new threshold estimates are (from 2014-05-01) : si 200, go 420, pl 700, di 940, ma 1410 Regarding WoL there is much less collected data and thus exact offset values are hard to find. But the user graphs looked mostly good with the old offsets and I did not see any clear discontinuities. Thus I will not change the WoL offsets. However I slightly dropped WoL platinum threshold (gold platinum offset may have slightly been dropped or stayed the same, some of the user profiles are inconsistent regarding this and there is not enough data to recognize which ones are correct). Following offset change times were used regarding the archiving: NA: 2014-05-01 21:25:00 GMT EU (KR, SEA and China): 2014-05-02 00:25:00 GMT Due to the out-of-sync issue of the web profiles the change times may not be exact (there was not enough 'good games' during those few hours to pinpoint the exact time). Thus I checked when the promotions started happening and used that data. The NA time is likely very close, but EU time could be earlier. For KR, Sea and China there was not enough data for that night, so I used the EU change time for them. The NA change time seemed to be some minutes earlier than Mintograde's data collection from nios.kr suggested (I hope he continues to collect data for the next season too, EU data would also be nice). -- The out-of-sync update issue of the web profiles continues. Thus many user accounts have some data errors that cause either upward or downward spikes. But if there are plenty of 'good games' then the incorrect values can usually be easily identified. I solved the offsets mainly based on EU data.The MMR graphs for EU profiles seemed to look generally better with the new offsets than graphs for NA profiles. In the past Blizzard has used same offsets for all servers. I hope they have not started using different values for different servers, but if they have the difference is still small. -- Regarding the tool: The archiving functionality was thoroughly tested, but I had no time for play testing. Most of the changes were archiving related, but few quite simple new data checks were added into the tool. They are for detecting if the web profiles are suffering from 'missing bonus pool issue' that occurred at the start of this season (they only catch few special cases). The code should be fine, but there is always a possibility as there was no play-testing that something went wrong. As the web profiles are not currently suffering from the 'missing bonus pool issue' (except right after placement / promotion it may take couple of minutes before the bonus pool appears in the web profile), the checks should usually not activate. If you are seeing following messages (not after placement / promotion) in the 'Main Log': "ERROR PRECHECK PRE-PLAYER: pre-player unused bonus pool is too low compared to post-player unused bonus pool!" or "ERROR PRECHECK PLAYER: post-player unused bonus pool is too low compared to pre-player unused bonus pool!", then please inform it in this thread and also add both pre and post match data from that match from the 'Game Log' inside spoiler tags. The out-of-sync issue of the web-profiles may rarely activate the checks too thought, but in those cases it is desired. -- I will be AFK from tomorrow for an extended weekend, so you may not hear from me until next week. | ||
|
grumpyone
48 Posts
| ||
|
kervv12
2 Posts
http://postimg.org/image/gtaw4rhrt/ thanks for making this and i hope you could help me. | ||
|
InfCereal
Canada1755 Posts
On June 16 2014 07:12 kervv12 wrote: okay so im having a little problem i think. i played 28 games with your tool, but it still says that these are pre-results and it needs better data. what do i have to do to give it better data? this is what i see http://postimg.org/image/gtaw4rhrt/ thanks for making this and i hope you could help me. You can turn on the memory reader in the settings to improve accuracy. Otherwise, keep playing. There's a set of rather strict requirements for accurate calculations, so perfect games are few and far between. However, once you get a good game, data will extrapolate upon it, giving you a fairly accurate view, even without good games. | ||
|
NeXaSC2
44 Posts
| ||
|
kervv12
2 Posts
| ||
|
InfCereal
Canada1755 Posts
On June 16 2014 23:39 kervv12 wrote: i already got memory reader on with all the games, and i feel that 24 of those games are all good games, if not. what would be considered a good game? It's nothing you're in control of. Just keep playing. | ||
|
korona
1098 Posts
It seems that the ladder lock period has started last night at local times. Thus 'good games' recorded after that are incorrect (you may optionally set them manually bad via 'Config' > 'Data' > 'Refresh' > select game from the list > 'mark as bad game'). Games played after downloading the DB file will be ok. Likely start times for the ladder lock for different servers are listed inside spoiler tags: + Show Spoiler + NA : Mon 2014-06-23 07:00:00 GMT (started ~8 hours before DB update) EU : Sun 2014-06-22 23:00:00 GMT (started ~15 hours before DB update) KR : Sun 2014-06-22 15:00:00 GMT (started ~24 hours before DB update) SEA: Sun 2014-06-22 16:00:00 GMT (started ~23 hours before DB update) CN : Sun 2014-06-22 16:00:00 GMT (started ~23 hours before DB update) Sorry for the mix-up. 4 days ago a blue poster told that the lock period was supposed to start at June 30: http://us.battle.net/sc2/en/forum/topic/13141104496?page=2#23 At the start of the season I had seen earlier date via game client but thought that Blizzard had changed it. I should not have trusted the blue poster and checked it myself. The ladder lock period seems to already have started. Thus the information given by Psione was either false or they had forgotten to change it also to the game servers. The final bonus pools for S18 for all servers are: master & gm: 1825 diamond & below: 1058 -- Edit: Blizzard just released a news post both on EU & NA that claims that the ladders are locked June 30. So it seems they forgot to update the game servers. We will have to wait and see how they will fix this - Will they release the lock or will it remain locked for two weeks. http://us.battle.net/sc2/en/blog/14502317/2014-season-2-lock-incoming-6-23-2014 Edit 2: Blizzard decided that the ladder will remain locked for 2 weeks: http://us.battle.net/sc2/en/forum/topic/13243434219 | ||
|
Malhavoc
Italy308 Posts
| ||
| ||
