|
|
|
I've been "ploughing" fast through the pages here, and many of the findings are very interesting, but people need to understand that if Blizzard uses a "standard" rating system, you cant just infer some random formula out of data. Yes, it may be fruitfull in some cases, but it cant replicate the real system in most cases. As mentioned, most rating systems of these sorts is based on statistical modelling, for example Microsoft uses Bayesian inference in its TrueSkill system to estimate the players' skill. Another example is the ELO rating system used in Chess.
To make a successful model one needs tons of data and slowly test model after model untill one finds one that may be useable. Most likely one needs to start of with some hierarchical bayesian model where one does some empirical Bayes style to update parameters as matches are played.
|
United States12245 Posts
On February 11 2011 05:04 SDream wrote:This guy has just been promoted: http://www.sc2ranks.com/la/211989/FelipeESDLAFrom this division: http://www.sc2ranks.com/div/66992/division-ursa-pi(probably rank S platinum, he had 2,974 more or less) To this division: http://www.sc2ranks.com/div/45047/division-drone-whiskey(confirmed rank D diamond, he is active players with zero bonus pool as of now) I'd like to emphasize how he was an ACTIVE platinum player but the system still took the time to finally move him to a D rank diamond, even though we have confirmed F and E rank diamond on this LA server. Why is that? Moving average? That's the answer, but then I have this other question: how to make your moving average smaller in range? Being more stable yourself?
A moving average would stabilize faster with enough losses to offset wins, so probably.
|
|
|
You can't ignore that you don't know that offset from the division he came, nor the offset of the division he went.
What if It's a A Rank Gold and he goes to S rank platinum, or S Gold and then goes to A platinum? There will be a difference of 2 division tiers, you can't ignore that, can you?
As I said before, most people go to E Rank diamond, some goes straight to D Rank. It is possible to have a MMR of platinum and still be active and in gold, cause your moving average is not there yet.
|
|
|
|
|
|
|
United States12245 Posts
You don't need to be an active player to be promoted, you only need to play against active players. I've spelled this out many many times already.
|
On February 12 2011 17:39 Excalibur_Z wrote:
You don't need to be an active player to be promoted, you only need to play against active players. I've spelled this out many many times already.
I posted a pretty long rant but believe the following three sentences encapsulate my view:
Thanks for this vague and unhelpful statement. I guess you've been right all along and the system is impossible to crack, all the front-end data is irrelevant and everything I've been doing, despite being spot-on, is all some hocus pocus.
You can go fuck yourself, and I'll post my work somewhere else.
User was temp banned for this post.
|
o.O
Anyway, F Rank can't have this objective, it would be silly. It ought to be a consequence, not an intended objective.
The reason I keep believing we have only 2 tiers on gold, platinum and silver is that:
1) I could only confirm 2 tiers with the divisions "filling process".
2) If you look at "max points" on average, it seems that diamond and bronze have the same amount of tiers and silver, gold and platinum have 5 less divisions (as they have 315 less points on average, that would be 5 less divisions). That actually makes sense with leagues offsets being 150 and divisions offsets 63 a little more or a little less than that.
Then I call it S rank for the one that is only 150 points far away from the lowest tier of the other division and then A for the second one.
Considering what I now think I understand better, the moving average, I think it's somewhat safe to assume the same phenomenum as diamond: A rank on silver, gold an platinum are almost as rare as F Rank diamond, S rank gold, silver and platinum are almost as common as E rank diamond.
The reason for that is cause:
1) The system will prefer to put you on a worst league first ->
2) Most players are most likely to get better little by little, but not slowly enough for the moving average to be able to put you on a new division without skipping entirally division tiers, that's because the 79th placed guy can't really beat the 78th placed guy 100% of the time, then a high enough moving average that requires 63+ equivalent on points of MMR will most often than not be present ->
3) As it will promote you only once, you'll be stuck on that division tier and will just ever move again if you are promoted again, where everything I said will happen again.
Master league then would promote people faster, cause it won't require your moving average to be confortable on 1 division tier, it will be enough to just be confortable on master.
I think that it is it.
Even though we can't crack the system 100%, I feel like I understand it enough to know where I stand among the active players, I think that the active, inactive differences are the mistery I am most curious about now.
Thanks for your contributions michael, see you.
|
Just tossing this out there.
You figured out that:
Post promotion points = unspent bonus pool +73
Points offset between diamond and masters is 150
Points offset between known diamond divisions is 63
It is assumed that:
Points offset between all divisions is 150
If tiers in lower divisions exist they are at 63 point offset
Then there is the data point that just got left behind once the promotion was solved at unspent bonus pool +73. There was still the fact that promotions were happening and the point loss was a multiple of 63 (the original assumption)
Why did the original assumption work out so neatly, when the actual method was unspent bonus pool +73?
Do the three data points we know plus the old method plus the assumptions possibly give a way to determine additional tiers for the continuous ladder?
|
Heey, I just got promoted, to:
Diamond: 2169 pts, Division Dragoon Echo
Umm, I honestly forget what division I was in and how many points I had, but I think you can find out from sc2 ranks?? If anyone wants to, my username is ffadicted.200
bnet profile: http://us.battle.net/sc2/en/profile/2143320/1/ffadicted/
It's worth noting that at the time of promotion, there were NOT 100 in my division, and the division is not in the OP, so it leads me to beleive that it's a new divison?
The highest amount of points seems a bit low, so maybe it's a low ranked division, which kind of upsets me cuz I wanted to be in a higher ranked one idk why I didn't 
|
United States12245 Posts
On February 13 2011 01:44 Eeryck wrote:
Just tossing this out there.
You figured out that:
Post promotion points = unspent bonus pool +73
Points offset between diamond and masters is 150
Points offset between known diamond divisions is 63
It is assumed that:
Points offset between all divisions is 150
If tiers in lower divisions exist they are at 63 point offset
Then there is the data point that just got left behind once the promotion was solved at unspent bonus pool +73. There was still the fact that promotions were happening and the point loss was a multiple of 63 (the original assumption)
Why did the original assumption work out so neatly, when the actual method was unspent bonus pool +73?
Do the three data points we know plus the old method plus the assumptions possibly give a way to determine additional tiers for the continuous ladder?
The point loss wasn't 63 in all cases, it just was for me and a couple of other people. For the others who had exact match histories we had those jotted off to the side as unexplainable anomalies. It wasn't until Master league where we noticed the pattern that we were able to go back and see if it fit, which it did (which was very disappointing because it clouds all of the offsets). Because of that, we may not be able to find the offsets of the other leagues and their division tiers (we had to rely on the top 200 to get the ones we know).
|
One thing that I don't like about this system of new rating = 73 + unspent bonus pool is that it can lead to abuse: the latter you start (so the more bonus pool you have), and the faster you get promoted (so the less bonus point you spent), the better your new rating will be. This is troublesome because some time to get promoted faster you need to lose game, so losing games can be helpfull for the rating! Of course you will always converge to the same rating, but if there are tourmanents with a rating cut-off at a precise time, it could lead to abuses.
|
United States12245 Posts
On February 13 2011 04:12 gondolin wrote:
One thing that I don't like about this system of new rating = 73 + unspent bonus pool is that it can lead to abuse: the latter you start (so the more bonus pool you have), and the faster you get promoted (so the less bonus point you spent), the better your new rating will be. This is troublesome because some time to get promoted faster you need to lose game, so losing games can be helpfull for the rating! Of course you will always converge to the same rating, but if there are tourmanents with a rating cut-off at a precise time, it could lead to abuses.
I can kind of understand that. If you were to just say "starting tournament, 2k Master players only" then that would discriminate against the Master player who had 500 points but a record of like 10-2 who may be playing the top 100 players in every game. I don't think this information will change anything with regard to tournament invites, but it's good to be aware anyway.
|
On February 13 2011 02:38 ffadicted wrote:Heey, I just got promoted, to: Diamond: 2169 pts, Division Dragoon Echo Umm, I honestly forget what division I was in and how many points I had, but I think you can find out from sc2 ranks?? If anyone wants to, my username is ffadicted.200 bnet profile: http://us.battle.net/sc2/en/profile/2143320/1/ffadicted/It's worth noting that at the time of promotion, there were NOT 100 in my division, and the division is not in the OP, so it leads me to beleive that it's a new divison? The highest amount of points seems a bit low, so maybe it's a low ranked division, which kind of upsets me cuz I wanted to be in a higher ranked one idk why I didn't
Thanks to Master league, we have a huge number of diamond divisions not filled, some are pretty old actually.
|
This thread to forever to find, but I finally managed to find it.
I was Promoted to Diamond (from Plat) about 30 minutes ago
My sc2ranks profile is: http://sc2ranks.com/us/341638/Gofarman
My record at time of promotion was 310/297 (6 game played since sc2ranks last updated)
-12/-10/+21/+17/+11/+11 that was the last 6 games played *(was promoted after the bold game)
My points at time of promotion was 2931 (in platinum) after promotion was 2750 a loss of 181 during promotion.
I went from Rank 1 of Division Baneling X-Ray to Rank 9 Tyrador Chi.
Idk if this will help you guys out at all but that's what I have got.
|
On February 12 2011 19:44 michaelhasanalias wrote:Show nested quote +On February 12 2011 17:39 Excalibur_Z wrote:
You don't need to be an active player to be promoted, you only need to play against active players. I've spelled this out many many times already. I posted a pretty long rant but believe the following three sentences encapsulate my view: Thanks for this vague and unhelpful statement. I guess you've been right all along and the system is impossible to crack, all the front-end data is irrelevant and everything I've been doing, despite being spot-on, is all some hocus pocus. You can go fuck yourself, and I'll post my work somewhere else. User was temp banned for this post.
I was supposed to answer your "im offended" post, which I havent had the chance to before now since i've been away for some time, but it seems like you have deleted that one. Never the less I intend to answer it any way. The reason for writing what I did is because you dont seem to understand how these systems are built, namely on Bayesian analysis. It's not my intention to "stfu noob, Im a statistician, I know how this works", but rather to point out that you are missing a vital point. For example how the standard deviation (also known as the "sigma", allthough it could be any greek letter) is used to decided if a player is sufficiently enough inside a determined interval which would qualify you for promotion. You might hit some nails/catch the essence with your formulae, but your method doesnt describe accurately how the ranking system works (assumed that when people says it's confirmed that they use approx. the WoW system, its actually confirmed). If you've ever tried to publish something in a serious journal, you would know the endless back and forth between you and the refs before they accept your paper, so I don't understand the sudden "f*ck off" attitude.
Now, I just started getting into the system, and I hope to do some testing with Bayesian models shortly since Im quite familiar with models of such types. If we could join forces in finding a accurate description of the system Im more than willing to contribute.
|
Nice post, it's interesting that he doesn't understand his formula cannot be accurate when the most complicated aspects of it are an addition and multiplication symbol.
|
|
|
|
|
|
EPT
