|
United States12235 Posts
It's long been established that the SC2 league populations center around the concept of "active players". If you're wondering why that target distribution of 2/18/20/32/20/8 is never what's reported on Nios or SC2Ranks, well, that's why. Those sites track all accounts and not solely active players. One of the main reasons for this is that we don't know what constitutes an "active player".
So, then, what is an "active player"?
Blizzard uses the bonus pool as their primary activity metric. The bonus pool accumulates at a fixed rate, so if you let it pile up, then you haven't been playing games and therefore aren't active. A couple of years ago I emailed the designer asking why our different forms of activity filters never matched up with Blizzard's distribution, and I received the response "if you were to look in terms of 'bonus weeks behind' you may find closer results." Now, I'm not sure about whether there is a secondary measurement being considered (perhaps games played), I'm just evaluating all possibilities.
+ Show Spoiler +
Side note: If Blizzard looks at league distributions on a continual basis--that is, in the middle of a season--then surely there is some additional filter in place like number of games played. It couldn't be one game or five games because just placing in a league shouldn't qualify you as an active player, even though early in a season your unspent bonus pool will be quite low.
Shadowed from SC2Ranks provided me with a data snapshot from this afternoon (Nios, your site is great but it doesn't have as many data points as SC2R, sorry!). Now that Shadowed has incorporated bonus pool tracking, I can finally look more closely at how Blizzard might be filtering for active users.
I created a spreadsheet based on that snapshot using the AM region, 1v1 bracket (150K 1v1 users, largest of all regions).
Link to Spreadsheet on Google Drive
Note: Even though I use column headers based on multiples of 180, the sub-Master leagues have already been adjusted for the 0.58 bonus pool accumulation rate. It was just easier to express "bonus weeks behind" in terms of the more round Master value (180 per week).
Each cell represents a league population percentage compared with the other leagues following the same criteria. That is, the B4 cell (Master, >=1 game, <=180 bonus pool) means that based on the snapshot from July 31, 2013, 6% of all users with >=1 game played and who have <=180 bonus pool are in Master league. This value is notably inflated over the target 2% value even though Nios and SC2Ranks find that 3% of the entire account population is in Master. You can see in the spreadsheet that a lot of the other leagues are closer to their targets, there are a lot of 18%s in the <=180 Diamond region, a lot of 20%s all over Platinum, a lot of 32%s all over Gold. What isn't apparent is one fixed coordinate that fits all the leagues. This leads me to believe that the league adjustments may in fact be seasonal.
Season Locks
The bonus pool stops accumulating during a season lock, which lasts for one week. Blizzard's stated reason for this is to allow players to spend their remaining bonus pool and establish final divisional rankings. However, the main reason for season locks may actually be far more practical. If Blizzard takes a snapshot of the remaining leagues at the time of the league lock, then that gives them a full week to analyze the current league populations. This would allow them plenty of time to run solves on whether they need to adjust the league thresholds for next season and what the new values should be.
Here's a look at a chart from the S4 Season Lock on August 20, 2013 which compares the Bonus Pool Population with the Actual Population and the Target Distribution:
![[image loading]](http://i.imgur.com/O12IoEi.png)
What you can see in this chart is that the Actual Population of all players is way off, but when you look in terms of remaining bonus pool, the numbers suddenly look a lot closer to the Target Distribution.
Here's how the data points break down:
1 bonus week: Master - 5.4%, Diamond - 15.27%, Platinum - 18.95%, Gold - 28.84%, Silver - 20.26%, Bronze - 11.26%
2 bonus weeks: Master - 5.56%, Diamond - 15.34%, Platinum - 19.37%, Gold - 29.58%, Silver - 20.07%, Bronze - 10.05%
3 bonus weeks: Master - 5.62%, Diamond - 15.06%, Platinum - 19.53%, Gold - 30.31%, Silver - 20.18%, Bronze - 9.26%
4 bonus weeks: Master - 5.64%, Diamond - 14.57%, Platinum - 19.46%, Gold - 31.27%, Silver - 20.46%, Bronze - 8.57%
5 bonus weeks: Master - 5.53%, Diamond - 13.87%, Platinum - 19.21%, Gold - 31.94%, Silver - 21.17%, Bronze - 8.25%
6 bonus weeks: Master - 5.09%, Diamond - 12.60%, Platinum - 18.35%, Gold - 31.96%, Silver - 22.93%, Bronze - 9.03%
What this tells us is that at the time of the S4 Season Lock, of the players who had less than 1080*0.58=626 bonus pool, 12.6% were in Diamond, far short of the 18% target. However, of the players who had less than 180*0.58=104 bonus pool, 15.27% were in Diamond. Master is always pretty steady at around 5%, above the 2% target. This means that out of all the active players, Master is inflated by the end of the season. This could mean that Blizzard might want to raise the requirements to get into Master league in order to make it more exclusive in the following season, which would also have an effect in making Diamond a little bigger.
What we don't know is which line reflects the actual activity metric, and that's why I have 6 different "bonus weeks behind" lines. In most of them, Diamond, Gold and Platinum are slightly underrepresented while Master, Silver and Bronze are slightly overrepresented.
|
Cool, thanks Excalibur_Z. Love these articles.
Blizzard behind the scene stuff is so awesome 
|
United States12235 Posts
Updated with a chart of the populations according to different bonus pool thresholds from the 2013 Season 4 league lock.
|
Wait what? Only 1 guy commented on this for an entire month? :o
|
On August 30 2013 04:32 Koshi wrote:
Wait what? Only 1 guy commented on this for an entire month? :o
:O
such an amazing research getting lost... 
|
On August 30 2013 04:32 Koshi wrote:
Wait what? Only 1 guy commented on this for an entire month? :o
Yeah, the bonus pool on this thread must be really high right now. Well perhaps since there is no useful information but a neat fact though.
|
On August 30 2013 04:32 Koshi wrote:
Wait what? Only 1 guy commented on this for an entire month? :o
Seriously, I missed this the first time around, but it is really interesting. Glad this got bumped!
Basically it appears that there are a lot of bronze/silver players who only play their placement games (plus maybe a few more) each season and then nothing else. It's amazing how all of 1 -> 6 bonus weeks are so similar to each other.
Thanks for the insight Excalibur_Z!
|
How did I miss this thread? Really interesting read
|
|
|
I ll give it 1 more bump. Cuz stats are for cool kids.
|
Very interesting, it's unfortunate that this thread got buried.
|
Missed the thread the two other times it got bumped, so thx Koshi.
What I read Blizzard sets the threshold manually? Can this also mean thats the reason why the WOL ladder is a bit f*ed up since a few seasons, since noone (except players like me) cares about WOL anymore? Maybe its also bc they removed GM-league without updating some of their other code.
[WOL league distribution NA: Masters 0.05% (5 Users) Diamond 1.20% (131 Users) Platinum 5.43% (592 Users) Gold 26.26% (2,864 Users) Silver 51.57% (5,625 Users) Bronze 15.49%(1,690 Users)]
|
On August 01 2013 09:37 Excalibur_Z wrote:
What this tells us is that at the time of the S4 Season Lock, of the players who had less than 1080*0.58=626 bonus pool, 12.6% were in Diamond, far short of the 18% target. However, of the players who had less than 180*0.58=104 bonus pool, 15.27% were in Diamond. Master is always pretty steady at around 5%, above the 2% target. This means that out of all the active players, Master is inflated by the end of the season. This could mean that Blizzard might want to raise the requirements to get into Master league in order to make it more exclusive in the following season, which would also have an effect in making Diamond a little bigger.
I have a feeling if you take a look right now, you'll see that the bolded part is true. I know of a number of people that were Master last season in the middle of their league that are now Diamond. After playing their placement match, it put them into Diamond instead of back into Masters, which it used to typically do.
It's interesting to think that Blizzard might actually be altering the criteria it takes to get into certain leagues based on the number of "active" players in each league from the previous season. This would of course make a lot of sense if you think about it. If they have set criteria to base whether or not a player should be in a certain league at the start of a season, it would make sense for the percentage of players in each league to be very close to what they are looking for. Then as the season goes on other players will meet that criteria and be promoted inflating the league once more and forcing Blizzard to set their criteria differently. Of course you should take into account players that do get demoted during the season, which might offset the inflation.
It would be interesting if Blizzard were to set an absolute set percentage for each league forcing people out of the league either by promotion of demotion.
That is of course, what I think is going on.
|
Personally I feel division should carry over to the start of next season, but have a simple promotion/demotion policy. Way too many people are being demoted 1-2 leagues at the start of every season after performing reasonably well in their former league, which sends a bad message to the players.
There is no reason demotions should be disable during the course of a season, particularly for the higher leagues. Win-rate, activity, and maybe even some 3rd criteria like a mixture of spending quotient/supply cap time/apm/etc could be used to assess whether a player should remain in their current league, receive a promotion, or a demotion.
Currently the ladder is a complete joke and does not in any way even resemble the term "ladder". It's simply Blizzard throwing you wherever they feel like at the start of the season with absolutely no explanation or clear reasoning. I can understand that the dynamics of the population shifts and league parameters need to be adjusted but that should be done during the course of a season based on player performance.
Is Diamond League too large? Fine. Demote the players occupying a seat in diamond that are either
a.) Not playing enough
b.) Not performing well
It is just such a terrible experience right now and impossible to maintain any sense of direction or to assess progression.
Edit: Or simply find a way to put this into the hands of the players or let us have a little control over our own fates.
Example - if a particular league has become too large and needs to be whittled down, maybe impose a requirement that all players outside of the Top 50 of their division will be demoted, or cut the players who simply aren't putting any time in, or maybe even have an informal tournament setup.
The way it is handled now just doesn't work and I can't think of anything that makes me less motivated to play than a random, undeserved demotion after a meaningless placement match. That system needs to be removed as well. I can understand the purpose of playing 5 placement matches, but that single match placement format proves absolutely nothing
|
On September 03 2013 07:23 NubainMuscle wrote:+ Show Spoiler +
Personally I feel division should carry over to the start of next season, but have a simple promotion/demotion policy. Way too many people are being demoted 1-2 leagues at the start of every season after performing reasonably well in their former league, which sends a bad message to the players.
There is no reason demotions should be disable during the course of a season, particularly for the higher leagues. Win-rate, activity, and maybe even some 3rd criteria like a mixture of spending quotient/supply cap time/apm/etc could be used to assess whether a player should remain in their current league, receive a promotion, or a demotion.
Currently the ladder is a complete joke and does not in any way even resemble the term "ladder". It's simply Blizzard throwing you wherever they feel like at the start of the season with absolutely no explanation or clear reasoning. I can understand that the dynamics of the population shifts and league parameters need to be adjusted but that should be done during the course of a season based on player performance.
Is Diamond League too large? Fine. Demote the players occupying a seat in diamond that are either a.) Not playing enough b.) Not performing well It is just such a terrible experience right now and impossible to maintain any sense of direction or to assess progression. + Show Spoiler +Edit: Or simply find a way to put this into the hands of the players or let us have a little control over our own fates.
Example - if a particular league has become too large and needs to be whittled down, maybe impose a requirement that all players outside of the Top 50 of their division will be demoted, or cut the players who simply aren't putting any time in, or maybe even have an informal tournament setup.
The way it is handled now just doesn't work and I can't think of anything that makes me less motivated to play than a random, undeserved demotion after a meaningless placement match. That system needs to be removed as well. I can understand the purpose of playing 5 placement matches, but that single match placement format proves absolutely nothing
This is what Blizzard does periodically. They do it at the end/start of ladder seasons. That's why you see the people being placed into different leagues instead of carrying over.
I think for a majority of players it's motivating to try to get back into *X* league each season (that's what we really want right? Encouraging people to play). If you were placed into *X* league but were constantly scared that you would be demoted all season you probably would not play until the season lock anyway (and then if your MMR dropped you'd still get placed into a lower league the following season).
I wouldn't mind if seasons were a bit shorter (we'll see how it goes with mirroring WCS seasons) or if there was a "mid season MMR/League/inactivity check" to promote/demote/kick players as needed.
|
Probably explains why I got bumped into plat after a crappy on and off season playing random :D
I'm guessing Diamond got too large because of how many people were demoted from masters the season before, and therefor people not playing well got kicked down.
Or am I misunderstanding?
|
Excalibur, I feasted upon your excellent OP.
What I would really love to see would be what the above graph looks like for WOL 1v1. I know much fewer people play WOL 1v1, but I feel quite confident that even the active player distribution for WOL would still look pretty screwed up (far off from blizzard's stated population targets). If the above graph IS from WOL then you will have successfully shocked me. I checked the distributions on SC2 ranks fairly regularly last season for WOL and HOTS, and they don't seem to have changed much, which strengthens my opinion that your graph is of HOTS 1v1 and that WOL 1v1 would still look pretty bad even after adjustment for active players.
I wonder if blizzard is adjusting MMR league boundaries JOINTLY for WOL and HOTS? It would make sense that fewer skilled players are left playing WOL... Still though, a grand total of 20 1v1 Masters in Americas region right now seems awfully low even if the boundaries vary jointly for WOL and HOTS. If blizzard IS doing this, then of course they would base the distribution off of HOTS and have that one look mostly correct (as it does in your graph) and leave WOL distribution naked to suffer the wrath of an eroded playerbase.
P.S. Bonus pool max at time of snapshot = 1337. That can't be a coincidence, right 
P.P.S. If you do make a WOL 1v1 graph, you might want to exclude China. It could be something else going on, but it looks like China is pretty different (unrepresentative) of the other regions so including them would skew the data massively.
|
On August 30 2013 04:52 Big J wrote:Show nested quote +On August 30 2013 04:32 Koshi wrote:
Wait what? Only 1 guy commented on this for an entire month? :o such an amazing research getting lost...
such amazing..
so research...
much statistic..
wow!
|
|
|
Why have seasons at all? Why have leagues? It's very silly. Just display an ELO rating.
|
because reality makes people feel even more insignificant
thanks for the data though!
|
Leagues motivate me, being no. 65376 rather than 83244 on the EU ladder do not.
The competition within the league is a little stupid, but ppl only care about promotions anyway.
|
Now, after their situation report, I think we can safely say that their active threshold is 2 weeks time.
Do you have any new data from sc2ranks with the situation of players with bonus pool <= 2 weeks time from the last season lock?
|
United States12235 Posts
Unfortunately I can't get any more SC2Ranks data parses which contain bonus pool information, and Nios can't give me any either.
|
damn wrong thread and I have nothing interesting to say.
|
Thank you for this post!!! Have you thought about linking it to the more popular one: http://www.teamliquid.net/forum/starcraft-2/195273-comprehensive-sc2-league-and-ladder-guide
I haven't seen anything else online that actually shows a league distribution sorted by bonus pool remaining, and it is very helpful in showing that the ladder distribution is not as incorrect as it would seem on SC2Ranks or nios.
Also, I wonder about this - would there be anyway to somehow calculate the "league distribution" of non-laddering players who are active in playing unranked?
|
I grabbed all the remaining bonus pool values for 102,158 players (using the NA player list from nios.kr - http://nios.kr/sc2/us/1v1/hots/).
The program to do this was written in .NET, source code here: http://19hz.info/Starcraft/SC2RanksParser/SC2RanksParser/SC2RanksParser.vb
Using the method of bonus pool weeks remaining, these are the results:
![[image loading]](http://19hz.info/Starcraft/2014-06-14Chart.jpg)
![[image loading]](http://19hz.info/Starcraft/2014-06-14-RawCountTable.jpg)
The smaller the "Weeks Remaining," the more "active" a player can be considered.
There are about 10,000 "super-active" players - people that are probably playing 7+ games a week. (estimating this based on your bonus pool changing by about 20 every time you win or lose).
There might be about 25,000 "semi-active" players - people that play 3-4 games a week.
A huge part of the "playerbase" (say, half) have probably played less than 15 games total.
Overall, one could argue that for "active" players, Bronze and Silver leagues are actually undersized, and that Gold to Diamond leagues are oversized. Diamond league especially seems inflated.
Raw data is here: http://19hz.info/Starcraft/US Battle Net Bonus Pool.xlsm
Thanks Excalibur_Z for the background info. Please let me know if you see any issues with the data/analysis.
|
United States12235 Posts
This is great stuff, thanks a lot for doing the scraping. You've crossposted to Reddit as well I trust?
I think they might have changed the activity metric. No parses had been made since they started to adjust the boundaries based on decayed players around the beginning of 2014. If you compare your chart to mine from last year -- and both use the same method -- the lines now seem to be kind of all over the place.
The last piece of the puzzle would appear to be the amount of real time elapsed in between games played. There are many ways to spend, say, 100 bonus pool: 1) play 8 games in a day, 2) play 1 game per day everyday for 8 days, 3) play 4 games today and 4 games a month from today, and so on. And the same is true for letting 100 bonus pool accumulate, where you take some time off but it doesn't have to be all at once. What if decayed players are flagged in some way? I don't really see how we would get this information for every single game, but we could at least get it from the most recent game, though that might not tell us everything.
Here's how I'm thinking they might categorize decayed players now and factor them in:
1) If you played 0 games in any 2-week period within a season, you are flagged as "decayed" and removed from the raw counts for the rest of the season.
2) If you are currently "decayed" or were previously "decayed" but you play a game, is your "decayed" flag removed for that season?
The reason I say we may not be able to get this data is it would require full histories of every player for a season. It's easy enough to see "last game played: a month ago". It's entirely different to find "the gap between the 5th and 6th game played this season: 3 weeks".
It's also possible that I'm wrong and they didn't change the metric, but instead they're making manual changes. In any event this is a fascinating and much-needed update to the topic.
|
Wow, impressive work.
I don't think Blizzard is analyzing inside the period, doesn't seem to be worth the effort for 10,000 players. It's much more credible that they do manual adjustments in order to compensate the decay.
|
On June 16 2014 01:55 Excalibur_Z wrote:
This is great stuff, thanks a lot for doing the scraping. You've crossposted to Reddit as well I trust?
I think they might have changed the activity metric. No parses had been made since they started to adjust the boundaries based on decayed players around the beginning of 2014. If you compare your chart to mine from last year -- and both use the same method -- the lines now seem to be kind of all over the place.
The last piece of the puzzle would appear to be the amount of real time elapsed in between games played. There are many ways to spend, say, 100 bonus pool: 1) play 8 games in a day, 2) play 1 game per day everyday for 8 days, 3) play 4 games today and 4 games a month from today, and so on. And the same is true for letting 100 bonus pool accumulate, where you take some time off but it doesn't have to be all at once. What if decayed players are flagged in some way? I don't really see how we would get this information for every single game, but we could at least get it from the most recent game, though that might not tell us everything.
Here's how I'm thinking they might categorize decayed players now and factor them in:
1) If you played 0 games in any 2-week period within a season, you are flagged as "decayed" and removed from the raw counts for the rest of the season.
2) If you are currently "decayed" or were previously "decayed" but you play a game, is your "decayed" flag removed for that season?
The reason I say we may not be able to get this data is it would require full histories of every player for a season. It's easy enough to see "last game played: a month ago". It's entirely different to find "the gap between the 5th and 6th game played this season: 3 weeks".
It's also possible that I'm wrong and they didn't change the metric, but instead they're making manual changes. In any event this is a fascinating and much-needed update to the topic.
Yes I did post it on reddit.
http://www.reddit.com/r/starcraft/comments/286ugt/na_battlenet_player_activity_analysis_based_on/
I've also thought about tracking "last game(s) played" as well but to do so in the script would at least double the time required to run the program, which already takes 8-9 hours for 100,000 players, and I would have run this program constantly (every single day) to make sure I grab every single player's match history. I suppose we could run this for a smaller set of players but that's getting into statistics I'm not too familiar with anymore
That being said I'm doing a similar run for the European servers, and I'll be doing Korea next.
Thanks for your guidance!
|
It's only for NA right ? So there is at least 30k active players only on NA, that's a good number I think ! Especially since it counts only 1v1 games. Doesn't count customs, multi and arcades, which represent the major part of sc2 players.
#sc2nowaydead
|
Not sure I understand all the fine details, but a very interesting read nonetheless! And I certainly get the drift of it. Thanks for all the hard work! Talk of the ladder is usually uninteresting (for the same reason talk about balance is uninteresting), but a technical analysis with actual information sheds a lot of light.
I always wondered how Blizzard handled the problem of league distribution, since distribution of players varies a lot depending on definition of "players". Drilling into the meaning of "activity" helps reveal how Blizzard solves this puzzle.
Thanks!
|
On June 16 2014 20:51 Faust852 wrote:
It's only for NA right ? So there is at least 30k active players only on NA, that's a good number I think ! Especially since it counts only 1v1 games. Doesn't count customs, multi and arcades, which represent the major part of sc2 players.
#sc2nowaydead
Yes, this is just for NA 1v1 ladder.
I'm running the same analysis for EU right now - but this time it's taking over 24 hours. Eventually the data will be there though. It's running off my desktop so I can leave it on indefinitely.
I plan on doing the same thing with the KR server as well.
Someone on Reddit suggested I do this every week to keep up with activity statistics, so I might end up doing that too
|
On June 17 2014 21:47 frajen86 wrote:Show nested quote +On June 16 2014 20:51 Faust852 wrote:
It's only for NA right ? So there is at least 30k active players only on NA, that's a good number I think ! Especially since it counts only 1v1 games. Doesn't count customs, multi and arcades, which represent the major part of sc2 players.
#sc2nowaydead Yes, this is just for NA 1v1 ladder. I'm running the same analysis for EU right now - but this time it's taking over 24 hours. Eventually the data will be there though. It's running off my desktop so I can leave it on indefinitely. I plan on doing the same thing with the KR server as well. Someone on Reddit suggested I do this every week to keep up with activity statistics, so I might end up doing that too
Would be awesome if you could do that !
|
One thing about team ladder I've noticed is that I can't always tell how much bonus pool someone has in a team setting because some people are on two teams that end up in the same division, and I think Battle.net only shows bonus pool for the lower ranked team in that case.
So I think I'm just going to give up on the detailed team ladder stats for now
The best I could do is go through the team list and get summary stats on when the last game the team played was, and how many total wins/losses they have.
|
EU analysis done. Raw data is here: http://19hz.info/Starcraft/EU Battle Net Bonus Pool.xlsm
Analysis:
![[image loading]](http://19hz.info/Starcraft/2014-06-17-EU_Chart.jpg)
![[image loading]](http://19hz.info/Starcraft/2014-06-17-EU_RawCountTable.jpg)
There are about 12,000 "super-active" players - people that are probably playing 7+ games a week. (estimating this based on your bonus pool changing by about 20 every time you win or lose).
There might be about 30,000 "semi-active" players - people that play 3+ games a week.
About half the total player count has played less than 15 games total.
For "active" players, Bronze and Silver leagues are actually undersized, Gold and Platinum leagues are relatively close to their target sizes, and Diamond and Master leagues are inflated.
KR server analysis should be done tomorrow.
Reddit crosspost: http://www.reddit.com/r/starcraft/comments/28faj7/eu_battlenet_player_activity_analysis_1v1_ladder/
|
United States12235 Posts
haha. I remember you told one of the guys you would change your topic name to "ladder activity" rather than player activity, but I guess you forgot to do that. Oh well!
EU appears to be basically the same, which is good, because it means the data collection and analysis aren't flawed.
|
On June 18 2014 12:19 Excalibur_Z wrote:
haha. I remember you told one of the guys you would change your topic name to "ladder activity" rather than player activity, but I guess you forgot to do that. Oh well!
EU appears to be basically the same, which is good, because it means the data collection and analysis aren't flawed.
It's in there, it says 1v1 Ladder now!
And yes it looks the same. Diamond league seems big.
|
Is it possible to extract the race distribution per league as well? I always wondered if there is a difference between the "super-active" and the "I-play-one-game-per-season" player in terms of races actively played (and secretly hoped it is nearer the 1/3-1/3-1/3 distribution).
|
On June 15 2014 17:16 frajen86 wrote:I grabbed all the remaining bonus pool values for 102,158 players (using the NA player list from nios.kr - http://nios.kr/sc2/us/1v1/hots/).
Interesting data. It also highlights the problem with the current 'harsh' MMR decay. As so considerable portion of players are inactive most of the time, the decay causes general MMR level to go down.
If you are going to run the data collection (semi-) regularly, I would suggest to collect some more individual data, so it would be possible follow up the number of players who have longer than 2 weeks breaks during the 1v1 season (or who have several such breaks) (decay starts after 2 weeks and reaches its max value after 4 weeks). This would give better estimate how many players face direct MMR decay during the season (Of course those numbers would be only pointers as the unranked 1v1 games reset the decay counter too and they cannot be followed).
But what extra data would be needed: 1) Timestamp for each player data (fetch time) e.g. to be able to calculate max bonus pool (and potentially calculated max pool value based on it), 2) recording player-id:s (IDs cannot change like nick, for example id for http://eu.battle.net/sc2/en/profile/123456/1/example/ would be 1-123456 or long ID would be 2-S2-1-123456, where first number is the server 1 US, 2 EU, 3 KR, 5 CN, 6 SEA, 'S2' is static for all. Of course the URL itself contains that info too), 3) league, 4) division name (or id, but division name is ok as names are unique), win and loss counts and ladder points. By comparing the earlier data one can see if the player has played ranked 1v1 games during the period.
Bonus pool accumulation starting times for this season S18:
+ Show Spoiler +## US, EU, KR, SEA, CN
Sun 13 Apr 2014 07:00:08 GMT
Sat 12 Apr 2014 23:00:08 GMT
Sat 12 Apr 2014 15:00:08 GMT
Sat 12 Apr 2014 16:00:08 GMT
Sat 12 Apr 2014 16:00:08 GMT
or as long in milliseconds (Unix epoch time):
1397372408000
1397343608000
1397314808000
1397318408000
1397318408000
Edit 1: You have some problem regarding EU master players who have 9 weeks worth of bonus pool (should be 230 instead of 11).
|
On June 18 2014 18:14 korona wrote:Interesting data. It also highlights the problem with the current 'harsh' MMR decay. As so considerable portion of players are inactive most of the time, the decay causes general MMR level to go down. If you are going to run the data collection (semi-) regularly, I would suggest to collect some more individual data, so it would be possible follow up the number of players who have longer than 2 weeks breaks during the 1v1 season (or who have several such breaks) (decay starts after 2 weeks and reaches its max value after 4 weeks). This would give better estimate how many players face direct MMR decay during the season (Of course those numbers would be only pointers as the unranked 1v1 games reset the decay counter too and they cannot be followed). But what extra data would be needed: 1) Timestamp for each player data (fetch time) e.g. to be able to calculate max bonus pool (and potentially calculated max pool value based on it), 2) recording player-id:s (IDs cannot change like nick, for example id for http://eu.battle.net/sc2/en/profile/123456/1/example/ would be 1-123456 or long ID would be 2-S2-1-123456, where first number is the server 1 US, 2 EU, 3 KR, 5 CN, 6 SEA, 'S2' is static for all. Of course the URL itself contains that info too), 3) league, 4) division name (or id, but division name is ok as names are unique), win and loss counts and ladder points. By comparing the earlier data one can see if the player has played ranked 1v1 games during the period. Bonus pool accumulation starting times for this season S18: + Show Spoiler +## US, EU, KR, SEA, CN
Sun 13 Apr 2014 07:00:08 GMT
Sat 12 Apr 2014 23:00:08 GMT
Sat 12 Apr 2014 15:00:08 GMT
Sat 12 Apr 2014 16:00:08 GMT
Sat 12 Apr 2014 16:00:08 GMT
or as long in milliseconds (Unix epoch time):
1397372408000
1397343608000
1397314808000
1397318408000
1397318408000
Edit 1: You have some problem regarding EU master players who have 9 weeks worth of bonus pool (should be 230 instead of 11).
Good catch, fixed for next run.
Not exactly sure how to exactly use all that data right now, but it's possible to collect. The only thing that I think might be an issue is tracking the "last ranked or 1v1 game played" data, because I would have to open up an extra page for every single player, and at this rate doubling the program time would cause the whole program to take maybe 7-8 days to run, which means to keep up weekly updates I would have to be running this thing constantly.
For the next runthrough I'll try to add some of the metadata to the raw sheet and we can go from there (as far as analysis goes)
Thank you!
|
On June 18 2014 15:39 Archiatrus wrote:
Is it possible to extract the race distribution per league as well? I always wondered if there is a difference between the "super-active" and the "I-play-one-game-per-season" player in terms of races actively played (and secretly hoped it is nearer the 1/3-1/3-1/3 distribution).
It's do-able though and I'll try it for the next analysis set (probably released next week).
Thanks!
EDIT: actually nios.kr already grabs race stats for me so this should be pretty easy.
|
On June 18 2014 20:12 frajen86 wrote:Show nested quote +On June 18 2014 18:14 korona wrote:On June 15 2014 17:16 frajen86 wrote:I grabbed all the remaining bonus pool values for 102,158 players (using the NA player list from nios.kr - http://nios.kr/sc2/us/1v1/hots/). Interesting data. It also highlights the problem with the current 'harsh' MMR decay. As so considerable portion of players are inactive most of the time, the decay causes general MMR level to go down. If you are going to run the data collection (semi-) regularly, I would suggest to collect some more individual data, so it would be possible follow up the number of players who have longer than 2 weeks breaks during the 1v1 season (or who have several such breaks) (decay starts after 2 weeks and reaches its max value after 4 weeks). This would give better estimate how many players face direct MMR decay during the season (Of course those numbers would be only pointers as the unranked 1v1 games reset the decay counter too and they cannot be followed). But what extra data would be needed: 1) Timestamp for each player data (fetch time) e.g. to be able to calculate max bonus pool (and potentially calculated max pool value based on it), 2) recording player-id:s (IDs cannot change like nick, for example id for http://eu.battle.net/sc2/en/profile/123456/1/example/ would be 1-123456 or long ID would be 2-S2-1-123456, where first number is the server 1 US, 2 EU, 3 KR, 5 CN, 6 SEA, 'S2' is static for all. Of course the URL itself contains that info too), 3) league, 4) division name (or id, but division name is ok as names are unique), win and loss counts and ladder points. By comparing the earlier data one can see if the player has played ranked 1v1 games during the period. Bonus pool accumulation starting times for this season S18: + Show Spoiler +## US, EU, KR, SEA, CN
Sun 13 Apr 2014 07:00:08 GMT
Sat 12 Apr 2014 23:00:08 GMT
Sat 12 Apr 2014 15:00:08 GMT
Sat 12 Apr 2014 16:00:08 GMT
Sat 12 Apr 2014 16:00:08 GMT
or as long in milliseconds (Unix epoch time):
1397372408000
1397343608000
1397314808000
1397318408000
1397318408000
Edit 1: You have some problem regarding EU master players who have 9 weeks worth of bonus pool (should be 230 instead of 11). Good catch, fixed for next run. Not exactly sure how to exactly use all that data right now, but it's possible to collect. The only thing that I think might be an issue is tracking the "last ranked or 1v1 game played" data, because I would have to open up an extra page for every single player, and at this rate doubling the program time would cause the whole program to take maybe 7-8 days to run, which means to keep up weekly updates I would have to be running this thing constantly.
Everything I mentioned is available from the ladder page and from its URL from which you fetch the remaining bonus pool, thus no need to open several pages for each player. (the last ranked game can be solved after the data fetches by comparing records from different weeks. That does not give exact times, but gives data if each player has played ranked games during the period between the fetches).
Actually the division page also contains date when the player joined the division (+ the other info mentioned before). That also could be used for recognizing if the division has changed or 'leave league' feature been used + when there has been activity at least.
On June 18 2014 20:12 frajen86 wrote:For the next runthrough I'll try to add some of the metadata to the raw sheet and we can go from there (as far as analysis goes)
That would be nice.
Also regarding parsing the division page. Blizzard usually does not make changes to it. They may add data, but the old data can still be found by using old key elements. Also if they remove some data or data goes missing, the elements remain. For example this season they removed season numbers. The elements are still there, but the content is missing. They may add the missing content again later and usually do not change anything regarding the elements. (in other words using regexps like you are using is the fastest way to do it and the solution will likely continue to work even if Blizzard makes additions / removals in the future)
|
|
|
Very interesting statistics. Korea has the highest retention rate of very players, but EU has the highest number of very active master players, much higher than KR or NA. perhaps it is the large number of small tourneys in EU that created this effect.
|
Sorry for the bump, but the content here would be especially valuable given the ladder's current state. Is there any possible way to redo this analysis in the current season? The analysis performed by frajen86 in particular.
I've seen mention by Excalibur_Z and Korona regarding other threads (i.e. MMR plugin for SC2gears and Blizzard's response for League Distribution), but this would be helpful.
|
On March 10 2015 05:48 HaloLegend98 wrote:Sorry for the bump, but the content here would be especially valuable given the ladder's current state. Is there any possible way to redo this analysis in the current season? The analysis performed by frajen86 in particular. I've seen mention by Excalibur_Z and Korona regarding other threads (i.e. MMR plugin for SC2gears and Blizzard's response for League Distribution), but this would be helpful.
See this thread: http://www.teamliquid.net/forum/starcraft-2/459913-ranked-ladder-activity-analysis
Heads-up, that thread is a bit more than just bonus pool metric.
Although I already put up some numbers from the American region relating just to bonus pool: http://www.teamliquid.net/forum/viewpost.php?post_id=23869343
I don't have every region up/full analysis because the whole thing takes a while to run. Part of the program goes through every single page in nios.kr, loads every nios.kr player's profile page on battle.net, and grabs the bonus pool. Loading every single player's battle.net profile has taken up to a few days in the past. AFAIK remaining bonus pool is not in the Starcraft 2 API so I can't get it as easily as nios.kr gets a player's points, or total wins, etc.
|
OK thanks. I should have found this on my own...
1. Is this posted on Reddit? I have a feeling that it would help clear out a lot of unrest. I don't think that ladder distributions are going to be seeing major promotions. It's like many are waiting for a handout from Blizzard. Seeing this might calm down the tension over the league issue.
edit: found this from December http://www.reddit.com/r/starcraft/comments/2q8lg1/ranked_ladder_activity_analysis_update_from_dec/
So it is out there, but the community is still convinced that promotions are necessary. Strange.
2. I understand that long term issues regarding the tail ends of the distribution are a problem, but can this analysis help shed light about people complaining about Masters players playing Plats, i.e. MMR being a little wonky? I personally have not had matchmaking issues. I've used the MMR plugin for SC2gears and it has worked very effectively reflecting my experience. But others maintain consistent matchmaking problems.
edit: I found this
http://www.teamliquid.net/forum/starcraft-2/479763-ladder-league-distribution-update-march-5?page=5
|
|
|
|
|
|
EPT![[image loading]](http://19hz.info/Starcraft/2014-06-18-KR_Chart.jpg)
![[image loading]](http://19hz.info/Starcraft/2014-06-18-KR_RawCountTable.jpg)
