EPT1916 + 1430 (3346)
-
2554 + 409 (2963)
=
383 - 150 = 233(S) - 63 = 170(A) = 107(B) = 44(C) = -19(D)
Right now I'd guess D, but I wouldn't bet anything cause I am not the one that collected the data
Edit: fixed
SC2 Ladder Analysis: Division Tiers - Page 48
| Forum Index > SC2 General |
|
SDream
Brazil896 Posts
1916 + 1430 (3346) - 2554 + 409 (2963) = 383 - 150 = 233(S) - 63 = 170(A) = 107(B) = 44(C) = -19(D) Right now I'd guess D, but I wouldn't bet anything cause I am not the one that collected the data Edit: fixed | ||
|
turdburgler
England6749 Posts
On March 11 2011 14:23 Excalibur_Z wrote: There are no "better" Master divisions, they're all the same. Now, if you were talking about Diamond divisions, it's definitely possible to start out in a D-rank Diamond division, get promoted to Master, then get demoted to an S-rank Diamond division. We've certainly seen that happen. Didn't you read the OP fully?  aparently not : < but thanks for the reply | ||
|
Grildrak
Sweden44 Posts
 should have checked everything before posting it (was a little too happy finally finding it :D )My calculation: The player in my division do now have 2029 point and since this game (also including it) he gained 113 point so he had 1916 before this game was played. He is right now on 1361 points in his bonus pool and since that game (also including it) he used 80 (counting 50% of the points from his wins as he have a so big bonus pool we know 50% is bonus) and the game was played 23 hours ago so he got 11 point added to his pool so bonus was 1430 The opponent have do now have 2627 point and since the game (also including it) he gained 99 point so he had 2528 before this game was played. The opponent is right now on 342 points in his bonus pool and since that game (also including it) he used 78 (counting 50% of the points from his wins as he have a so big bonus pool we know 50% is bonus) and the game was played 23 hours ago so he got 11 point added to his pool so bonus was 409 | ||
|
SDream
Brazil896 Posts
| ||
|
Kyamo
Canada129 Posts
Opponent: adjusted points before game = -33 points change for game = -11 Plat Reaver Sigma Me: adjusted points before game = 156 points change for game = +12 Gold Meinhoff Whiskey So this time, I get offset of 156 - (-33) = 189. This is very close to my last result, so those two guys are probably in the same tier of plat. Also, hmm, 189 = 63*3. | ||
Excalibur_Z
United States12235 Posts
On March 12 2011 08:21 Kyamo wrote: Ok, I got another 12/11, and this time I got all the info directly after the game. Opponent: adjusted points before game = -33 points change for game = -11 Plat Reaver Sigma Me: adjusted points before game = 156 points change for game = +12 Gold Meinhoff Whiskey So this time, I get offset of 156 - (-33) = 189. This is very close to my last result, so those two guys are probably in the same tier of plat. Also, hmm, 189 = 63*3. Could also be proof of a 150 league offset, since it's an -11/+12 game. This is good stuff, please keep it coming. | ||
|
DiDigital
75 Posts
| ||
|
Excalibur_Z
United States12235 Posts
On March 12 2011 10:15 DiDigital wrote: I imagine you guys are already considering this but I thought I'd mention it. When looking at these +12/-12 or close games and comparing the point difference you need to remember that there is also the difference in the players MMR affecting the points. In other words consider a struggling 100 adjpnt plat player with a 100 adjpnt gold MMR playing against a 100ap gold player with a 100ap plat mmr. Both players will be looking at a 12/12 result despite the match being very uneven. This result would imply a 100 point offset that is completely irrelevant. The variance between your points and the opponent's MMR would have to be smallest at a +12 or -12 game because that's precisely in the middle of the 0-24 range. If both players win and lose 12 adjusted points, then each player's MMR is close to each other's points which means their points are close to each other and their MMRs are also close to each other. There certainly would have to be some variance, yes, but it probably won't be very large. If there's a very large discrepancy then that could be evidence of a different division tier or league offset. At least, that's what we're banking on. | ||
|
DiDigital
75 Posts
For example, just looking at players adjusted point totals from the spreadsheet you linked the average adjusted point total is 122 (max 404 silver, minimum -86 plat) and a standard deviation of 82. Logically we can assume that most of these players are in the league that their mmr dictates, and that at some point players will be promoted or demoted before their points reach too far into the next league. Another assumption is that the average score will be lower than the league modifier. Because of the division modifiers we can also assume that most players scores will be inflated closer to the top of the division rather than the middle. In other words, I haven't ran the numbers for all the bnet, but based on this sample I would assume that the league offsets have to be greater than 130. Since this data is mostly silver/gold I think it also may imply some uniformity to the league point offsets, since silver/gold would be smaller offsets if the offsets were not uniform. | ||
|
Excalibur_Z
United States12235 Posts
| ||
|
DiDigital
75 Posts
On March 12 2011 11:58 Excalibur_Z wrote: you play against a Master player and that player loses 12 points, then you know your MMR is pretty close to whatever his adjusted points were before that game. If your opponents points are not indicative of their mmr then you will become even more confused when looking at the results. When players first start, or go on winning or losing streaks their points will not be indicative of their mmr. Like if your 25-0 player played another 25-0 player. I suppose this is getting away from the main point, that looking at individual games is dangerous. There needs to be a lot of measurable consistency between a lot of results. | ||
|
Excalibur_Z
United States12235 Posts
| ||
|
SDream
Brazil896 Posts
Edit: It has been pretty much confirmed, but it's hard to do a 100% confirmation. Rank S Zerus Eta Hanson Beta Nova Folsom Alfa Rank A Akilae Rho Ursa Pi Koronis Whiskey Norad Hotel Colosso Yankee Aiur Jaca Reagan Zero Porcos de Guerra Tango Boros Chi Azimar Psi Duran Pedra Aparição Sapo Entregador União Taarim Omicron Aldaris Ostra Halcyon Zeta Fole Lima Fênix Kappa Rank B Khalis Marte Nargil Foca Mutano Épsilon Mohandar Tau Fanáticus Ilha Endiabrado Bravo Urun Xingu Char Dado Swann Miu Rebocador Kilo Sargas Novembro Mar Sara Rato -------------------------- Names in Portuguese (so you won't find it like this on sc2ranks) | ||
|
SDream
Brazil896 Posts
Rank S Hacylon Novembro Pretor Ilha Rank A Infestador Pedra Rank B Argo Eco Rank C Destruidora Tau Supremacia Gama Rank D Zangão Whiskey Nave Batedora Kilo Samiku Pi Rank E Médico Miu Preservador Dado Corvo Chi Furtivo Sigma Viking Rato Raynor Delta Ehlna Bravo Azimar casa Shiloh Upsilon Adun Beta Vigia Marte Roxara Psi Turaxix Ômega Augustogrado Xingu Aparição Zero Lenassa Phi Rank F Ambunave Água --------------- This might help me figuring out the offset from platinum <-> diamond I just have to win more... I might stop playing random if I have to. Embrace the zerg, let's see. | ||
|
Kyamo
Canada129 Posts
On March 12 2011 11:58 Excalibur_Z wrote: Hm, what would be an example of a game where the MMRs are drastically different? If you're like 25-0 for example and you play against a Master player and that player loses 12 points, then you know your MMR is pretty close to whatever his adjusted points were before that game. I think I see what he is saying. I will try to explain. If the game is 12/12 then we know that player A's adj pts are approximately equal to player B's MMR. Call this value X. So Apts = Bmmr = X. We also know that player A's MMR is approximately equal to player B's adj pts. Call this value Y. So Bpts = Ammr = Y. But there is no guarantee that X = Y. Usually they will be close, because the system will try to match people with similar MMRs. So usually, Ammr is approximately equal to Bmmr, and X would be approximately equal to Y. Also, usually Apts will be close to Ammr and Bpts close to Bmmr. However, it is possible that they could be not very close, if a closer match was not found and both players pts are not close to their MMRs. For example, maybe Player A has 200 pts and a MMR of 400, and Player B has a MMR of 200 and 400 pts. This would be a 12/12 game, despite the fact these players are not equal. I think this scenario would be fairly rare though (at least for large differences), since 3 things must be true for this to happen. (both players with inaccurate points in opposite directions, and disparate MMR match) So I think enough data should get rid of this problem. We just have to keep in mind that any one game could be an outlier. | ||
|
Excalibur_Z
United States12235 Posts
On March 12 2011 23:55 Kyamo wrote: I think I see what he is saying. I will try to explain. If the game is 12/12 then we know that player A's adj pts are approximately equal to player B's MMR. Call this value X. So Apts = Bmmr = X. We also know that player A's MMR is approximately equal to player B's adj pts. Call this value Y. So Bpts = Ammr = Y. But there is no guarantee that X = Y. Usually they will be close, because the system will try to match people with similar MMRs. So usually, Ammr is approximately equal to Bmmr, and X would be approximately equal to Y. Also, usually Apts will be close to Ammr and Bpts close to Bmmr. However, it is possible that they could be not very close, if a closer match was not found and both players pts are not close to their MMRs. For example, maybe Player A has 200 pts and a MMR of 400, and Player B has a MMR of 200 and 400 pts. This would be a 12/12 game, despite the fact these players are not equal. I think this scenario would be fairly rare though (at least for large differences), since 3 things must be true for this to happen. (both players with inaccurate points in opposite directions, and disparate MMR match) So I think enough data should get rid of this problem. We just have to keep in mind that any one game could be an outlier. Oh no... you're right =( I forgot about that scenario. That could definitely impact our data and affect our conclusions. So ideally what we would need is a trend of games, maybe 2 or 3 in a row, where both players earn or lose around 11-13 points. I wonder just how often those 12/12 outliers will appear... | ||
|
SDream
Brazil896 Posts
On March 13 2011 01:43 Excalibur_Z wrote: Oh no... you're right =( I forgot about that scenario. That could definitely impact our data and affect our conclusions. So ideally what we would need is a trend of games, maybe 2 or 3 in a row, where both players earn or lose around 11-13 points. I wonder just how often those 12/12 outliers will appear... The one we saw that the difference was of 40 points, I think it was a 12/12 game and I said it ought to be an anomaly or something. | ||
|
michaelhasanalias
Korea (South)1231 Posts
I think there is a cap on uncertainty, and once you hit that cap, the floor of your maximum uncertainty, when it passes the cutoff for the next league, you will be promoted. Here is the data if anyone wants to look: https://spreadsheets.google.com/ccc?key=0Aj5e5-iDDBbddGJ4dTE4cFBNY0t6Rnp6cmYzaDhPc2c&hl=en#gid=0 Bronze placement Silver after 13 games Gold after 4 games Plat after 5 games Dia after 5 games | ||
|
SDream
Brazil896 Posts
And I found one, so I'll share: Diamond guy won 12, -34 adjusted points (before match) on a known Rank E diamond division. Platinum guy lost 12, 130 adjusted points (before match) on a probably Rank A platinum division. That's a 164 difference. Considering that platinum S and diamond F divisions exist between these guys, then we should have at least a 3 "tiers" difference. 164/3 = 54+/- That doesn't give enough room for a league offset greater than 60+/-. Rank A plat with 189 = rank S plat with 126 = rank F diamond with 63 = rank E diamond with 0 Feasible, I'd need to confirm if these platinum rank A are right first and then look for more data to confirm, it's difficult to confirm with my weird matchmaking, so I'll have to stalk more random guys >.> Edit: the rank A platinum is pretty much confirmed. --------- Edit: Another One: Diamond guy lost 12, -26 adj points (before) in a known Rank E diamond division. Platinum guy won 12, 100 adj points (before) in a probably S(B?) platinum division. That's a difference of 126 points, which is 63*2. IMO that confirm it's a rank S platinum indeed and that the jump to platinum <-> diamond is of only another division! Edit: Rank S platinum pretty much confirmed. --------- Edit2: Another One: Platinum guy won 12, 112 adj points (before) in a probably S platinum division. Platinum guy lost 11, 160 adj points (before) in a probably A platinum division. That's a difference of 48 points, 1 division, being the first >second. Confirming the S and A status I gave them. woot | ||
|
DiDigital
75 Posts
When the system is calculating your points to determine how many you should earn from the match up it is likely doing this: Your Points + League Offset - Division Modifier. In the first two examples you gave it should look like this: Diamond Offset = Plat Offset + Offset in addition; thus (Po+O)-34-(6*63) = Po+130-(1*63) (Po+O)-412 = Po+67 O = 479 and the second game, (Po+O)-26-(6*63) = Po+100-(0*63) (Po+O)-404 = Po+100 O = 504 If the difference between S ranked Plat divisions and F ranked diamond divisions is 63 points that would imply a diamond offset of 504 points. If you could find the same pattern by looking at higher ranked known diamond divisions and plat divisions you could probably prove this to be true, and at the same time imply that similar structure exists in the lower leagues. | ||
| ||
