|
On September 01 2013 05:26 Nelvalhil wrote:Hey, I was curious about when I will get placed into a higher league since I'm facing gold/plat and former dia players while I'm still in silver. So I've downloaded MMR tools and have read the instructions, I have the latest version of MMR stats installed and it recognizes my account. But now one thing is; I still says needs a game with better data, I've played atleast 15/20 games on ladder 1v1 with MMR stats running. Also memory is enabled. Am I doing something wrong? I really want to know what my MMR is and when I get promoted into the higher leagues. I've posted some screenshots taken of MMR stats and I hope someone can tell me what I'm doing wrong Cheers! Screenshots: http://i41.tinypic.com/110yclz.pnghttp://i43.tinypic.com/20r7904.pngEDIT: My SC2gears also says I'm using 60ish mb ram of 250ish, is that normal? I've got 4x4 gb ram installed
The MMR tool has not registered any games. Make sure the SC2gears auto-save is enabled and auto-save settings are ok (make sure the replays are actually auto-saved by sc2gears). If that's ok then make sure you have correct account in MMR tool settings.
|
On September 02 2013 01:40 IMScientist wrote:So i finally got back to plat after 43 games 28 wins 15 losses win rate of 65% vs all, was promoted at 862 mmr points
As a small update I searched for accounts that had 3 weeks or longer inactivity period during last season. found some 900+ such accounts from last season's data (total characters ~9950, ~270k matches). And at least more than 30 had symptoms of sudden MMR drop (I did not check all 900+. And many had no good games after inactivity period). And all that I checked (10+ of those 30+) had inactivity period where I expected it to be based on graph. But have not had time to check their base data. But it seems it does not affect all --> That MMR decay might not be a feature but a bug. Of course if it is some kind of 'reset' then activity in other game modes may prevent it.
Not sure when I have time to continue analysis. At least next weekend I will most likely be AFK.
Edit: And I quickly also checked the offsets. Diamond-master seems same as last season. From others there is not enough data yet, but that small amount suggested that also the other offsets are likely the same.
Edit2: I seem to have used older files (likely counted that user amount from even older file). That data was before season lock. Final count of users was 11472 and 292k games. Many players activated during the season lock period, so likely more matching cases will be found when I use the full data.
Edit 3 (2013-09-04): I am adding more developer side tools that make it easier to analyse the data. At the moment it is easy to spot the affected accounts as added inactivity detection already earlier. But going through the base data would take lots and lots of time. I am adding tools to more easily rule out cases where inactivity periods are actually caused, e.g. because the user did not use MMR tool for a time. Or used older version of the tool or db file that caused incorrect results. Also as the Blizzard's MMR drop bug/feature does not affect every account that was inactive I have to try to find rule what triggers it. Excalibur_z's recent post regarding remaining bonus pools reminded that I should also check the bonus pools before and after inactivity. Maybe inactivity in this case is not based on time, but if unused bonus pool rises too high (totally speculation. I have not checked the data). But will likely happen at earliest next week as I will be AFK for an extended weekend.
|
Blizzard seems to have implemented some kind of MMR decay mechanism that has been in place at least from second season of HotS (S13, First season was S12. Current season is 15). It is unknown whether it is intentional or some kind of bug.
This would explain a lot. I don't have the 1v1 sc2gears data (hard drive issues) to back it up, but I had gotten into masters 1v1 two seasons ago (when there were more masters people than there should have been), before subsequently placing diamond last season (understandable when Blizzard restricted masters again - my division was filled with former masters players) - so a high diamond MMR. My unranked 1v1s last season were similar, against masters/top diamonds. Last season I then decided to get my 2v2, 3v3, and 4v4 random ladders to masters - which I did, then when I switched back to 1v1 unranked after a month or so of inactivity I was playing gold and platinum players (lots of plats, which suggested an MMR reset), which was baffling.
I just did my placements for 2v2, 3v3 and 4v4, and placed platinum in each of them - I had gotten into masters with a lot of games (hundreds) and hadn't played since being promoted so my MMR should not have dropped. My initial 1v1 ranked placement did put me into diamond this season, but my initial games are suggesting a low MMR.
|
Do we have data on the decay that would indicate the rate of decay and/or if it also increases uncertainy?
|
On September 02 2013 03:48 vaderseven wrote:
Do we have data on the decay that would indicate the rate of decay and/or if it also increases uncertainy?
Can't comment yet as have not gone through the data well enough. May also be constant amount for each inactivity. But there are plenty of accounts _not_ showing this behavior at all after inactivity period. Also it has crossed my mind could the uncertainty increase and MMR just drop faster via loses? But then again there are accounts that had won both last matches before inactivity and first matches after inactivity. And both gave calculated results that were showing the drop. Also the MMR seems behaving normally after that. So it is unlikely that uncertainty is affected. But not sure when I have time to investigate further.
|
hey  can u please share such information - how much mmr was required to promote to gml this season (i.e. today)
|
On September 02 2013 11:49 Val_ wrote:hey can u please share such information - how much mmr was required to promote to gml this season (i.e. today)
Not enough data yet. There are only 2 gm values for EU at moment (there was 7 EU matches that contained gm:s). Last value for lower of those two is 2131. But as there are plenty of accounts that have nearly that, it is likely near what the low border was at start.
|
thats more then enough for me, thank you
(have 2119 btw)
edit:
2175 was enough for me to promote to gml a minute ago
friend of mine was promoted on 2200 so maybe 1950 gml line should be corrected a bit 
+ Show Spoiler +
|
On September 02 2013 02:34 korona wrote:Show nested quote +On September 01 2013 05:26 Nelvalhil wrote:Hey, I was curious about when I will get placed into a higher league since I'm facing gold/plat and former dia players while I'm still in silver. So I've downloaded MMR tools and have read the instructions, I have the latest version of MMR stats installed and it recognizes my account. But now one thing is; I still says needs a game with better data, I've played atleast 15/20 games on ladder 1v1 with MMR stats running. Also memory is enabled. Am I doing something wrong? I really want to know what my MMR is and when I get promoted into the higher leagues. I've posted some screenshots taken of MMR stats and I hope someone can tell me what I'm doing wrong Cheers! Screenshots: http://i41.tinypic.com/110yclz.pnghttp://i43.tinypic.com/20r7904.pngEDIT: My SC2gears also says I'm using 60ish mb ram of 250ish, is that normal? I've got 4x4 gb ram installed The MMR tool has not registered any games. Make sure the SC2gears auto-save is enabled and auto-save settings are ok (make sure the replays are actually auto-saved by sc2gears). If that's ok then make sure you have correct account in MMR tool settings.
First of all thanks for replying,
Second; My autosafe is enabled, atleast I think it is, where do I find all those settings?
http://i44.tinypic.com/ibcbhs.pngg
|
Here are current entry estimations for GM border when it opened for S15 (as now more than half of spots have been used entry barrier is likely much higher):
EU 2130
US 1800
Last season was first time when I set separate values for EU & US as they were for first time clearly separated (last season's US value may have been little too high and EU value little too low). Last season's values were:
EU 1950
US 1900
Highest recorded values during this season (S15) are:
EU 2503
US 2151
And highest values from last season (S14):
EU 2754 (from middle of August. He finished very high at top-10. Also there are close ones from few other top names. There are lots of top readings as apparently certain shady figure used MMR tool in August for a period, but since at least his most known account has been banned by Blizzard if recent news are to believed (not for using MMR tool. And no good MMR readings from his own accounts as he never activated memory reader))
US 2346 (from July. There are close values from beginning of August from top-15 finishers)
Based on these I will set top-gm estimates to 2750 for EU and 2500 for US (old values EU 2400 and US 2300. If I remember correctly S14 EU high value was based on same person as S15 high value).
There are quite low amount of GM level users and most are at low to mid GM range & many of them do not keep MMR tool always on. They record values for their opponents too, but are usually not matched against very top (usually in full season data there are records for the top too but those records are not necessary fresh). Big thanks to all GMs who use the tool! I hope more would consider using the tool in the future.
For first time there was one Korean GM using MMR tool last season during July (month, not the player). Highest recorded value was some 2500 from a barcode who ended finishing in top-30. Thus it is likely ok to use new EU values for Korea too, even if true values might be lower. Please note these are just the estimate borders that do not affect calculations. True values for GM are dynamic. No SEA GM data yet for this season. Likely remains about the same as last season's due low population.
For China there is one GM reading in last season's data (it was first HotS season for China and everyone started from blank MMR). It was 1569 recorded at 2013-08-25. He was not the lowest and he made it to GM this season too. Based on him I put starting value for Chinese GM-border to 1550. As SEA border is the same, I will use SEA top value for China too. There are so small number of Chinese users that there is no data regarding Chinese offsets. I just presume they have the same offsets as other servers.
Old values (start of S14):
## US, EU, KR, SEA, CN
1900,2300
1950,2400
1950,2400
1550,1900
1950,2400 // HotS was launched in China at the middle of season. Values were just copy-paste from EU
New values (start of S15):
## US, EU, KR, SEA, CN
1800 1830,2500
2130,2750
2130,2750
1550,1900
1550,1900
I will later (not necessary today) update these to the MMR tool too. If I finetune the numbers I will edit the changes into this message too.
Edit 1: now 28 hours later some US masters have increased their MMR, but not have reached GM. Pretty much all over 1860 are in GM, but there are plenty of masters between 1800 and 1860. Even if the data suggests that the initial MMR barrier to GM was around 1800, I will set it to 1830. Also there are some new Chinese data regarding top masters. Highest has 1460 and is not in GM. Thus the 1550 I decided earlier based on that one week-old last season value is likely very close (likely something between 1460 to 1550).
Edit 2: These numbers have now been published via new DB file. After checking graphs I may drop EU border to 2100 in some later release as it still inside error boundaries. But will let it stay in 2130 for some time.
|
On September 03 2013 00:17 Nelvalhil wrote:Show nested quote +On September 02 2013 02:34 korona wrote:On September 01 2013 05:26 Nelvalhil wrote:Hey, I was curious about when I will get placed into a higher league since I'm facing gold/plat and former dia players while I'm still in silver. So I've downloaded MMR tools and have read the instructions, I have the latest version of MMR stats installed and it recognizes my account. But now one thing is; I still says needs a game with better data, I've played atleast 15/20 games on ladder 1v1 with MMR stats running. Also memory is enabled. Am I doing something wrong? I really want to know what my MMR is and when I get promoted into the higher leagues. I've posted some screenshots taken of MMR stats and I hope someone can tell me what I'm doing wrong Cheers! Screenshots: http://i41.tinypic.com/110yclz.pnghttp://i43.tinypic.com/20r7904.pngEDIT: My SC2gears also says I'm using 60ish mb ram of 250ish, is that normal? I've got 4x4 gb ram installed The MMR tool has not registered any games. Make sure the SC2gears auto-save is enabled and auto-save settings are ok (make sure the replays are actually auto-saved by sc2gears). If that's ok then make sure you have correct account in MMR tool settings. First of all thanks for replying, Second; My autosafe is enabled, atleast I think it is, where do I find all those settings? http://i44.tinypic.com/ibcbhs.pngg
S2gears auto-save settings can be found from 'Settings' > 'Misc settings'. It is mainly divided into two tabs: 'Replay auto-save' and 'Folders'. Especially make sure the folders are correct. I would recommend that you enable 'Play "Replay saved" when replays are saved' so you hear when a replay file is saved. Also 'Auto-open new replays' is handy as it opens them automatically in sc2gears analyzer.
|
On August 29 2013 02:08 korona wrote:
Blizzard seems to have implemented some kind of MMR decay mechanism that has been in place at least from second season of HotS (S13, First season was S12. Current season is 15). It is unknown whether it is intentional or some kind of bug.
I was inactive for the past 4 weeks on KR server. I was in diamond last season with a low number of games played (and was diamond two seasons ago as well). I just played my placement match, and got placed in Gold. I'll be playing some more today to try to get some good games on the graph for this season.
My profile is http://kr.battle.net/sc2/ko/profile/913627/2/qdlbp/.
Edit: I seem to be getting matched against players who are actually Gold, and have been Gold in their previous seasons. I got some good games in the new season, here's what the graph looks like so far:
+ Show Spoiler +
Games 14, 15, 16, and 17 are in the new season - the previous games were played 4+ weeks ago.
I looked through the profiles of my opponents from 4+ weeks ago, and most of them are in diamond again this season.
|
On September 03 2013 02:51 korona wrote:Show nested quote +On September 03 2013 00:17 Nelvalhil wrote:On September 02 2013 02:34 korona wrote:On September 01 2013 05:26 Nelvalhil wrote:Hey, I was curious about when I will get placed into a higher league since I'm facing gold/plat and former dia players while I'm still in silver. So I've downloaded MMR tools and have read the instructions, I have the latest version of MMR stats installed and it recognizes my account. But now one thing is; I still says needs a game with better data, I've played atleast 15/20 games on ladder 1v1 with MMR stats running. Also memory is enabled. Am I doing something wrong? I really want to know what my MMR is and when I get promoted into the higher leagues. I've posted some screenshots taken of MMR stats and I hope someone can tell me what I'm doing wrong Cheers! Screenshots: http://i41.tinypic.com/110yclz.pnghttp://i43.tinypic.com/20r7904.pngEDIT: My SC2gears also says I'm using 60ish mb ram of 250ish, is that normal? I've got 4x4 gb ram installed The MMR tool has not registered any games. Make sure the SC2gears auto-save is enabled and auto-save settings are ok (make sure the replays are actually auto-saved by sc2gears). If that's ok then make sure you have correct account in MMR tool settings. First of all thanks for replying, Second; My autosafe is enabled, atleast I think it is, where do I find all those settings? http://i44.tinypic.com/ibcbhs.pngg S2gears auto-save settings can be found from 'Settings' > 'Misc settings'. It is mainly divided into two tabs: 'Replay auto-save' and 'Folders'. Especially make sure the folders are correct. I would recommend that you enable 'Play "Replay saved" when replays are saved' so you hear when a replay file is saved. Also 'Auto-open new replays' is handy as it opens them automatically in sc2gears analyzer.
Woo-Hoo! Thank you very very very much!!! :D
|
I know i will sound silly and impacient, but I have 50 games played and still the program doesn't throw the mmr =(
My friend told me that he got it after 15 games so I just want to know if there is something im doing wrong or I only need to wait more.
Thanks!
|
On September 04 2013 10:43 lokes wrote:
I know i will sound silly and impacient, but I have 50 games played and still the program doesn't throw the mmr =(
My friend told me that he got it after 15 games so I just want to know if there is something im doing wrong or I only need to wait more.
Thanks!
Easy answer if I guessed your account name correct. You use old version of the MMR tool (you can check version number from the log). Update it to the latest version (9.6). Download button should have appeared in the top toolbar. If not, then restart the MMR tool. If it still does not appear, then download it via link given in the original post and install manually.
(Blizzard changed slightly the structure of web profile pages. Thus the old version cannot read all the info it needs from them)
|
Oh thanks you veru much. I knew it would be an easy solution that would make me look stupid =P
I will comment when I get the mmr.
|
Is there a way to manually import a game?
In this case, I forgot to open Sc2Gears before playing so it couldn't run the trigger...etc.
|
On September 05 2013 00:48 mythandier wrote:
Is there a way to manually import a game?
In this case, I forgot to open Sc2Gears before playing so it couldn't run the trigger...etc.
No, there's no way. It relies on the real time data after a game, so retroactively calculating mmr on old games would provide inaccurate results.
|
Update regarding the MMR decay feature/bug analysis.
Yesterday I wrote some code that made it easier to filter out irrelevant or bad data (for example many do not keep MMR tool always open and nearly half of the accounts seemingly being inactive are due to this) & validate the data faster (it is still quite slow as has to be done manually). I checked 110 profiles that had valid inactivity period of 42 days or more during last season (note during the season, not during season switch). Most of such accounts either did not have any 'good games' or had only 'good games' on either side of the inactivity period and thus those accounts were ignored. Thought at least some of these had dropped a league in the beginning of this season, even if their last season graph had been in higher league range (I did not check all).
But amongst those were 9 accounts that had good data on both sides of the inactivity period. 6 of these accounts had a MMR drop exactly where the inactivity period was. 3 had no signs of MMR changes due inactivity (though one was mid-bronze and would have dropped to negative range if there was a similar drop as in others). As an interesting note 3 of these 9 accounts were from users that play at least two servers and had been active on another server during the inactivity period.
I quickly checked the in-game profiles of 2 of the unaffected players. Person A who had longest inactivity seemed to be quite active in arranged teams, but his match history contained only max 1 week old matches (match history is nowadays capped to 200). Based on sc2ranks A had played team games during July that is in middle of his 1v1 inactivity period. Second longest unaffected person B also had his match history full, but based on it he is active unranked 1v1 player. According to sc2ranks B had arranged team games during his ranked 1v1 inactivity during July & August. For the bronze player I did not check his history. I should similarly check some of the positive profiles, but as there are lots, I will do it later. But as Blizzard has limited match history to only 200 matches it may take some time to find profile that has match history from the inactivity period & before it.
Here are 3 example pictures of what MMR graph from drop cases looks like (actually have similar also from non-drop cases. Also note that MMR tool always calculates the estimate matches backwards until a previous 'good game'. The bonus pool 'after' is the pool in the start of the match that ended the inactivity. Fine-tune is also off so it does not fine-tune results inside error margins): example A (this person has mem reader off, otherwise he would have lots of good games as is so high), example B (typo. matches after is 39-41) and example C
I will be AFK from Friday morning for an extended weekend and will continue analysis sometime next week (there are plenty of accounts having sudden MMR drop with less than 42 days of inactivity).
Edit: example picture B has typo. The w-l after the match that ended inactivity is 39-41.
Edit 2: Quick look to offsets. Still low amount of data from lower leagues, but results look mostly the same as last season. It is unlikely that there were changes or the changes were small.
|
United States12269 Posts
Those are some interesting examples. How did you come by 42 days/6 weeks as the inactivity period, and are there similar results for 35 days, 28 days, 21 days, 14 days? I kind of have my doubts about them demoting players above X bonus pool forcibly, but they've done some dramatic changes with HotS so this may not be too far-fetched after all.
|
|
|
|
|
|
EPT![[image loading]](http://i.imgur.com/VbeL5Lo.png)
![[image loading]](http://i.imgur.com/lJUcU9H.png)
