|
On May 10 2015 06:07 ApBuLLet wrote:So I'm using v 22.1 of the DB file, running MMR Stats 10.3 in SC2Gears 14.3.3 and I haven't gotten a "good" game yet in 52 games (it says Needs a game with better data 0/52). I have it correctly linked to my battle.net profile which is here http://us.battle.net/sc2/en/profile/625953/1/ApBuLLet/ and it seems that everything is correct in my profile. Any idea why I haven't been able to get any games with good data yet? I initially started playing again about a month ago after a long break and I was stuck in gold league due to MMR decay (I was masters all throughout WoL). I figured this was why the data was bad before, plus the season lock came shortly after I started playing again. However, I have played a good number of games in the new season and have gotten back up to diamond which seems to be where I belong so I don't understand why the games still have bad data. Is there something else I need to do to get good data out of my games? Edit: I also have memory scanning enabled, just FYI
Check the link in the end of the previous message (and links inside it). From February 2014 web profiles have also been affected by out-of-sync issue that leads that data from many games has to be marked bad.
Also worth noting that considerable amount of your recorded games may have been from before db 22.1 was released (data was locked most of the ladder lock period and long time at start of season)
|
On May 10 2015 08:57 korona wrote:Show nested quote +On May 10 2015 06:07 ApBuLLet wrote:So I'm using v 22.1 of the DB file, running MMR Stats 10.3 in SC2Gears 14.3.3 and I haven't gotten a "good" game yet in 52 games (it says Needs a game with better data 0/52). I have it correctly linked to my battle.net profile which is here http://us.battle.net/sc2/en/profile/625953/1/ApBuLLet/ and it seems that everything is correct in my profile. Any idea why I haven't been able to get any games with good data yet? I initially started playing again about a month ago after a long break and I was stuck in gold league due to MMR decay (I was masters all throughout WoL). I figured this was why the data was bad before, plus the season lock came shortly after I started playing again. However, I have played a good number of games in the new season and have gotten back up to diamond which seems to be where I belong so I don't understand why the games still have bad data. Is there something else I need to do to get good data out of my games? Edit: I also have memory scanning enabled, just FYI Check the link in the end of the previous message (and links inside it). From February 2014 web profiles have also been affected by out-of-sync issue that leads that data from many games has to be marked bad. Also worth noting that considerable amount of your recorded games may have been from before db 22.1 was released (data was locked most of the ladder lock period and long time at start of season)
I think you're right that a good number of those 50 something games were from before the new db version, so instead of 50+ I would say it's been more like 25-30 games. I also think that I am in the lower to mid part of my league and it sounds like that makes it more difficult to get good data. On top of that, I wouldn't be surprised if there is an out of sync issue as well. Is that something that happens randomly or does it seem to happen more to certain people? I'm wondering if my extended period of inactivity may have caused an issue.
Either way, I suppose I will just keep playing and see what happens. Thanks for your the response, I'll post again in here if I get good data or if I play many more games without getting any.
|
Thanks for the info ! I was worried about the fact it had no pre-game data, I did not know this was related to the lack of Ram reading. I guess I'll just keep the tool in the background and eventually it will catch a good game at some point !
On May 10 2015 05:21 korona wrote:Show nested quote +On May 10 2015 04:04 Stephanus wrote:Hello, I have played several games and still nothing comes out. I have checked my battle net page, it seems correct. I am on mac. It keeps saying there is no info on me, and that data quality is 50%. Here is my log : + Show Spoiler +20:56:55 Post-trigger: Start 20:57:55 Post-trigger: Reading web profiles 20:57:59 Webgrabber: Player true 20:58:02 Webgrabber: Opponent true 20:58:02 Post-trigger: Done reading web profiles 20:58:02 Unranked detection starts (detection is disabled for lost matches) 20:58:02 Unranked detection ends 20:58:02 Bonus calc (Stephanus diam-): max end: 385, max fetch: 385 20:58:02 Bonus calc (%D0%90%D0%BB%D0%B5%D0%BA%D1%81%D0%B5%D0%B9 diam-): max end: 385, max fetch: 385 20:58:02 Calculating adjusted points for the players (either Alg. A or B needs to succceed for certain player) 20:58:02 DATA Alg. A : failed, No pre-game data (RAM scanner possibly off) or pre-game data bad: 20:58:02 DATA Alg.B: failed, No Bonus Win: http://eu.battle.net/sc2/en/profile/4318628/1/Stephanus/ 20:58:02 ERROR AFTERCHECK Both alg. A and B failed: http://eu.battle.net/sc2/en/profile/4318628/1/Stephanus/ 20:58:02 ERROR PRECHECK PLAYER: 0 Ladderpoints - No Data: http://eu.battle.net/sc2/en/profile/1137667/2/Алексей/ 20:58:02 ERROR LASTCHECK : Both players invalid 20:58:02 Gamecheck: false 20:58:02 POST-Trigger done 20:58:02 Data quality needs to be 100% to be able to calculate MMR 20:58:02 Data quality for calculating Player's MMR: 50% 20:58:02 Data quality for calculating Opponent's MMR: 50% 20:58:05 OK Uploader: game uploaded 20:58:05 Ladder Game Done! 20:58:05 -------------------------- First I presume you have the latest DB file installed (v 22.1). With the older 22.0 data was locked and no good results were possible. On Mac the biggest problem is that game launch is not detected (the technique used is based on Windows registry). This means the tool never gets pre-game data for any match on Mac. The primary method of calculating points change for each match requires both pre-game and post-game data. But it is possible to solve the change points without pre-game data too. But the chances are slimmer. There are 2 things that must happen. And if they happen then there are the usual requirements for data quality. (During WoL it was easier when point change for lost matches was always known, now you never know it based on web profiles for lost matches unless you have pre-game data too). The two things are 1) opponent must win, 2) opponent must have unused bonus pool. If these two things happen and the data fulfills numerous other requirements, then it is possible to calculate your MMR from post-game data. This actually works the other way too. To calculate opponent's MMR from only post-game data requires that you win and you have remaining bonus pool (and then the numerous other requirements). Check this link and links inside it for some simplified info on other requirements (Remember those are very old offsets / thresholds in those examples, but basic principle is still valid, Also remember that there is no pre-game data nor memory reader for Mac): http://www.teamliquid.net/forum/viewpost.php?post_id=22798150
|
I continually get an error saying either no updated happened or middle updated happened. Everything else seems to be working fine just when it goes to take data from bnet it messes up. Is this just bad luck or do I have an incorrect setting? BTW awesome job on this thank you!
|
I've updated the DB, still got 3 "VALID: false" games in a row. Is it expected?
|
On May 12 2015 00:15 NjalSC wrote:
I've updated the DB, still got 3 "VALID: false" games in a row. Is it expected?
Not sure what that means. I need more data from the MMR tool log regarding that ('main log' and 'game log').
On May 11 2015 05:24 aevanee wrote:
I continually get an error saying either no updated happened or middle updated happened. Everything else seems to be working fine just when it goes to take data from bnet it messes up. Is this just bad luck or do I have an incorrect setting? BTW awesome job on this thank you!
Yes. That out-of-sync issue of Blizzard's web profiles is quite annoying. It is up to 'luck' if the web profiles update as expected. In the future version I increase the post game data fetching delay a little bit that might help things little (not much thought as in worst cases it has taken more than 5 minutes for the division page to update).
|
I need few people to do some practical testing with the next version of the MMR tool. The tests involve playing some 1v1 ladder games with the new version of MMR tool (upload enabled and memory reader enabled) and copy pasting some log data for me. Setup of the tests includes some manual updates (taking backup copy of certain file, replacing it with contents from zip file and in the end restoring the backup copy). As always the data provided via the tests is kept private.
Suitable person should play either on EU or NA servers and preferably have played more than 25 games this season already. The tests should be done on Windows. Please PM me if you are interested and I will then provide instructions. I will update this message when I have enough testers so check end of this message before offering to help.
Edit: 2015-05-12 Testers are still needed
Edit 2015-05-13 Now I have likely enough testers. Thank you!
|
On May 12 2015 01:40 korona wrote:Show nested quote +On May 12 2015 00:15 NjalSC wrote:
I've updated the DB, still got 3 "VALID: false" games in a row. Is it expected? Not sure what that means. I need more data from the MMR tool log regarding that ('main log' and 'game log').
http://pastebin.com/F8a3x4ep here you go
|
On May 12 2015 06:48 NjalSC wrote:Show nested quote +On May 12 2015 01:40 korona wrote:On May 12 2015 00:15 NjalSC wrote:
I've updated the DB, still got 3 "VALID: false" games in a row. Is it expected? Not sure what that means. I need more data from the MMR tool log regarding that ('main log' and 'game log'). http://pastebin.com/F8a3x4ep here you go
Thanks! These seem to be because of the out-of-sync issue on Blizzard's web profiles
These error messages are related to that:
ERROR PRECHECK Player: WEBSERVER no update happened!
ERROR PRECHECK PRE-PLAYER: WEBSERVER middle update happened!
Meaning that the data either did not change at all between the start of the match and end of the match or the data changed partially during the data fetch (You can check that yourself too by comparing preplayer 1 to postplayer 1 and similarly data for the second player. For example if win and lose counts don't change means the division page did not update. Often only match history changes (change_points, map) when the division page (wins, loses, points) remains unchanged).
Just play more. It is up to 'luck' when the profiles update fully immediately.
-----
More testers for the next version of the MMR tool is still needed. Please see more details: http://www.teamliquid.net/forum/viewpost.php?post_id=24244196
|
On May 12 2015 06:57 korona wrote:Show nested quote +On May 12 2015 06:48 NjalSC wrote:On May 12 2015 01:40 korona wrote:On May 12 2015 00:15 NjalSC wrote:
I've updated the DB, still got 3 "VALID: false" games in a row. Is it expected? Not sure what that means. I need more data from the MMR tool log regarding that ('main log' and 'game log'). http://pastebin.com/F8a3x4ep here you go Thanks! These seem to be because of the out-of-sync issue on Blizzard's web profiles These error messages are related to that: ERROR PRECHECK Player: WEBSERVER no update happened! ERROR PRECHECK PRE-PLAYER: WEBSERVER middle update happened! Meaning the data either did not change at all between the start of the match and end of the match or the data changed partially during the data fetch. Just play more. It is up to 'luck' when the profiles update fully immediately. ----- More testers for the next version of the MMR tool is still needed. Please see more details: http://www.teamliquid.net/forum/viewpost.php?post_id=24244196
I see! Thanks.
|
If you are willing to speed up the release of the next version of the MMR tool (that is needed to be able to collect data to calculate the new offsets) please consider offering testing help (requires playing games and copy pasting some log data, the new version is mostly ready, but requires some practical testing to be sure that everything is ok).
See details http://www.teamliquid.net/forum/viewpost.php?post_id=24244196
|
About how many games are we speaking and until when?
|
On May 13 2015 03:09 Archiatrus wrote:
About how many games are we speaking and until when?
Not too many. Handful - both wins and loses to make sure everything is ok (depends also on data what server returns).
I don't personally have any stabilized accounts and have not enough time to stabilize one myself as requires that 25 to 30 games from the first placement match, especially when I need to practice first to get rustiness off to get decent MMR immediately. That is why I am preferably seeking those who have already played that much already earlier (25+ games from start of the season).
Regarding time: preferably soon. And even if somebody might be able to provide data faster to allow publishing (e.g, if you split your games to several days), the test data that is 'late' will be valuable too. Solving offsets requires lots and lots of data (will take weeks to collect after the new version has been published) and even small amount may contain 'key data' (only small percent of the data is usable for offset calculations).
|
Now enough volunteers have approached me regarding testing of the next version. So no need for others to offer help anymore.
Now we wait for them complete the tests (depending on their schedules it may take couple of days). But if everything goes ok, the tool can likely be released on weekend (I did quite extensive testing with staged special cases myself last weekend so I don't expect big problems if any).
|
BTW Korona I think they lowered the MMR requirements for some leagues, I was promoted to Plat with a MMR around 650.
|
On May 15 2015 16:51 Chanyman wrote:
BTW Korona I think they lowered the MMR requirements for some leagues, I was promoted to Plat with a MMR around 650.
They changed the offsets and thresholds at the start of the season. The numbers MMR tool currently gives are not comparable with the league thresholds shown in MMR tool. But the numbers are comparable with other good results against same opponents from same league (e.g. if good result was against platinum all following good results from platinum opponents are comparable, but not comparable with good results from gold opponents, good results from gold opponents are comparable with good results from other gold opponents and so on, this is of course if there is no data errors).
If no problems are found the next version of the MMR tool is published this weekend (so far at glance there has not been problems with the user tests, but I have not done deeper analysis yet + couple have not yet completed them). The new version enables uploading data to new server from which the offsets can be solved.
Note the following was meant only for Chanyman:
But if you have had good results after start of this season (and before new version is published), you may PM me link to your games_i.data (in a zip package, the unpacked file is located [PATH_TO]\Sc2gears\User content\Plugin file cache\mmr.plugin.MMRPlugin). That data may be useful when collecting data to solve the offsets. But it is likely going to take weeks to collect enough data to solve the offsets after the new version is released.
|
So just to clarify, does this mean that the MMR ratings are accurate but the league boundaries (eg plat diamond boundary at 940 MMR) might not be accurate?
|
On May 16 2015 02:52 Average Joe wrote:
So just to clarify, does this mean that the MMR ratings are accurate but the league boundaries (eg plat diamond boundary at 940 MMR) might not be accurate?
In short yes. League thresholds are likely not accurate (lower border is comparatively accurate, if the good games come from that league opponents). But as offsets are not known the MMRs are comparable based on your opponents league.
Good games against bronze are comparable to good games against bronze
Good games against silver are comparable to good games against silver
Good games against platinum are comparable to good games against platinum
Good games against diamond are comparable to good games against diamond
Good games against master are comparable to good games against master
Good games against gm are comparable to good games against gm
Good games from different league players might be comparable. It is likely that Blizzard kept at least some of the offsets about the same as the old offsets were quite optimal for many leagues. But there are offsets that have likely changed such as diamond-master offset (diamond range had become huge, more than twice of some lower leagues). But when the new offsets have been solved, the old results will be automatically fixed (data will be collected with the next version of MMR tool and then offsets will be solved).
A hypothetical situation old platinum-diamond offset was 260. Now if the new one is 300 (not known), the results from diamond players will be 40 of from the results from platinum player (if you are a platinum).
|
2015-05-17: Results from the practical tests of v. 11.0
During the past week 6 testers played games on both NA and EU using the new version of the MMR tool. Big thanks for all participants!
The main goals for the tests were:
1) Test how the new server handles traffic + are there problems with uploads or downloads
2) Collect MMR tool logs to identify possible issues
3) Collect MMR tool logs to see how big impact the out-of-sync issue on Bnet profiles is causing
Before these practical tests I had done quite extensive testing via staged special scenarios. Thus I did not expect any considerable issues or games stoppers to be found in the practical tests - None were found.
The 6 testers played total 77 games throughout the week (one continued with 12 games more, but I have no logs for those games (only uploaded match data)).
The test results are summed up below. They have been grouped based on the test goals:
1) There was no problems with the new server during the tests. Both uploads and downloads were ok.
2) No considerable issues regarding the tool were found. Section of the version history had to be rewritten to be more precise. I also flagged data for two matches for later check regarding possible minor issue (potentially incorrect data quality test had marked them bad, but the end result (marking them bad) was most likely correct).
But there was 1 potential issue that should be checked in the future. From some 10 to 15 games (mostly from EU server) for either of the players pre-game data was from before the previous match and the data from the end of the match was indeed from the end. This means that the division web profile had not yet been updated to correspond the pre-game situation when pre-game data was fetched (40 sec after the start of the game), but when the game ended it was 'swiftly' updated to correspond the new situation. In these cases the optimal solution would be to mark pre-game data bad as it is still possible in limited cases to solve MMR from only post-game data. It is possible that it is already done this way in the tool, but I have to recheck this for the future version (there is a possibility that both the pre-game and post-game data is marked bad for that player). This situation is caused by the out-of-sync issue of Bnet web profiles (only Blizzard can fix it).
This was a little bit surprising because the start of the game is registered at the end of the loading screen. If average game search (after pressing 'play' button on SC2 client) takes roughly 30 seconds and the loading screen another 30 seconds (start of the game would be registered ~25 seconds in). it means that there was already at least ~1 minute time for the web profiles to update. 40 seconds delay is added on top of that before the data is fetched. Thus the web profiles have much more time to be updated regarding the pre-game data than post-game data (there is 80 seconds delay before the data is fetched after the replay file has been saved).
3) The situation regarding the out-of-sync issue that is happening on Blizzard's web profiles seems more dire than I expected. Since February 2014 the web profiles have been suffering from it. Typically after the match the match history for both players is updated immediately. But the ladder division page either updates immediately or it may take several minutes before it is updated (some cases even more than 5 minutes). And as result the match history and the ladder division page are not always in sync.
I went through all the match data the testers provided. I marked A) how many games contained ok data for both players, B) ok data for one player and C) invalid data for both players. These results do not include handful of games where the web profiles had not been available.
For EU server the results were: A) 4 ok for both (7.7%), B) 16 ok for 1 player (34.6%) and C) 25 invalid for both (57.7%). Considerable portion of the 'invalid for both' came from a certain day and there were comparatively less 'invalid for both' results in later days. Thus it may depend on the day (or in our perspective on 'luck').
There were 2 'good games' amongst the EU data: - 1 good result for the player himself (calculated from opponent's data) and 1 for the opponent (calculated from the player's data).
For NA server the results were: A) 4 ok for both (21%), B) 6 ok for 1 player (31.5%) and C) 9 invalid for both (47.3%).
There were no 'good games' amongst the NA data. For 12 NA games I don't have MMR tool logs, but at least 4 of them likely had ok data for at least 1 player (2 likely for both).
Based on this small sample it seems that web profile data is usually more reliable on NA server than on EU server. The figures for negative cases were higher than I have seen in previous seasons. It is possible that data reliability differs from day to day e.g. based on server load (or like it shows to us by 'luck').
Not all of the previous 'invalid for both' or 'valid for 1' results were caused by the out-of-sync issue. Some were caused by the opponent playing unranked match. When playing unranked matches the ranked division page does not update and for the tool it looks mostly the same as the out-of-sync issue. I did not calculate how many opponents were likely playing unranked mode, but there were several.
But what did we learn? Even if considerable portion of the games are affected by the out-of-sync issue, for roughly half or more of the matches the data for at least one player is ok. This means that if other requirements are also met, it is possible to calculate MMR at least for the other player (from matches where data is at least for other ok). Please also remember that player's MMR is calculated from the opponent's data & the opponent's MMR is calculated from the player's data.
There was only 1 'good game' for the player himself & 1 for the opponents' amongst the 77 matches. But this was not mainly because there was so much invalid data. It was because all testers had fairly low adjusted points, which usually translates that they are low regarding their league's MMR range (1 was closing the range where MMRs can be calculated more often). Thus to get more 'good games' they either need to increase their MMR to get matched against higher opponents or get matched against a opponent from the top of the lower league (who has high range MMR regarding her lower league). Blizzard has also claimed that they have tightened up search range for the matches. It could also affect the frequency of 'good games' for low MMR range players in each league.
Edit: Interesting detail. Data for the first match played after the release of the new version likely resulted 'good game' for both the player and the opponent (did not check via client, only based on raw data, it might also be that only good game for 1, but is close for both).
|
2015-05-17: New version of MMR tool (v. 11.0) was released that requires manual installation - new server taken into use
As the old server for the MMR tool is not available anymore a new server had to be taken into use. Thus the new version of the tool requires manual installation.
Blizzard changed the league offsets at the start of the S22 (S2 / 2015). The new offsets and thresholds are not yet known. To be able to solve the offsets lots of data from the users is needed (usually takes weeks to collect enough). It is important that as many MMR tool users as possible would adopt the new version quickly (more users --> more data --> offsets solved faster). If you have friends who play SC2 and have used MMR tool in the past please remind them about the new version that requires manual installation and direct them to this thread. I suspect that considerable portion of the MMR tool users do not follow this thread actively.
Download link, installation instructions, change notes and related info can be found from the original post under same topic as this.
|
|
|
|
|
|
EPT
