
SC2 Ladder Analysis: Division Tiers - Page 51
Forum Index > SC2 General |
Meldrath
United States620 Posts
![]() | ||
SDream
Brazil896 Posts
Edit: ExcZ! I am afraid the "bug" we saw on PTR will indeed happen at live o.o http://sea.battle.net/sc2/en/profile/105510/1/ash/matches Edit2: Or maybe he was just paired with diamonds? It does happens more frequently on low population realms... so it's just a coincidence. Edit3: China also has it and sc2ranks is already ready for it: http://www.sc2ranks.com/stats/league/all/1/all | ||
![]()
Excalibur_Z
United States12235 Posts
On April 12 2011 07:33 SDream wrote: Points are still reset to spent bonus pool + 73, so there would be no big surprise if there is another offset separeting master league from GM league... Edit: ExcZ! I am afraid the "bug" we saw on PTR will indeed happen at live o.o http://sea.battle.net/sc2/en/profile/105510/1/ash/matches Edit2: Or maybe he was just paired with diamonds? It does happens more frequently on low population realms... so it's just a coincidence. Edit3: China also has it and sc2ranks is already ready for it: http://www.sc2ranks.com/stats/league/all/1/all Oh wow, that's amazing. That's a huge dropoff in points. I'm not seeing that for some of the other players in that division. | ||
SDream
Brazil896 Posts
http://sea.battle.net/sc2/en/profile/308360/1/GnSStarcym/matches This is quite terrible for LA server, cause maybe half or more of the GMs there could be "bugged" till they drop, then get their points back and go up again, crazy stuff. wait and see... Edit: Another one: http://www.battlenet.com.cn/sc2/zh/profile/415723/1/DhcBajiao/matches I'd guess right now that GM offset is like 450 points from Master. 300-450 I think... (maybe more) | ||
InfSalvation
1 Post
On March 30 2011 02:05 Torte de Lini wrote: Math isn't strong suit, is it? There's a decimal point man lolol Apparently, neither is yours. That's called scientific notation, notice the e^? | ||
Grildrak
Sweden44 Posts
Opponent: points before game: 136 points change for game: +24 (12 bonus pool) bonus pool before game: 27 Master division Aldaris Gravity Me: points before game: 327 points change for game = -12 bonus pool before game: 10 Diamond division Tosh Alamo | ||
akaMadMike
Norway93 Posts
In short what I have understodd is the win/loss ratio determends where your MMR will be - winning alot it wil gradually go up untill you start loosing and eventually finds an equilibrium for a little while (correct me if I am wrong). But how far back does the games affect your MMR? Say you bought the game and started out as a noob/semi noob. You loose mor than you win and are placed in bronze, where you probably belong. After 3-4 months of just fooling arround in bronze just building random stuff (not executing BO's etc) and not having a clue about the mechanics etc, you have played 60-70 games with a 30% win ratio. Then you join a local LAN where a guy points out the importance of BO's and introduce Day[9] daily for you. all of a sudden you start to understand the mechanics and your game develop rapidly. Then to my real question: Since you now have lost more or less 40-50 games and your win ratio is about 30%. Will this mean you have to win alot of games to make it to about 50% win ratio before the MMR cosider you good enough to promote? Will it be better to practice alot of KOTH or custom rather than destroing your win/loss ratio? | ||
Lysenko
Iceland2128 Posts
On April 14 2011 22:35 akaMadMike wrote: In short what I have understodd is the win/loss ratio determends where your MMR will be - winning alot it wil gradually go up untill you start loosing and eventually finds an equilibrium for a little while (correct me if I am wrong). But how far back does the games affect your MMR? Edit: What I'm about to tell you is a summary of much of what's in this thread, as it relates to your question: http://www.teamliquid.net/forum/viewmessage.php?topic_id=142211 Win/loss ratio doesn't directly determine anything. The system looks at the existing MMR of your opponent and upon winning or losing moves your MMR up or down more or less depending on their rating vs. your rating. So, for example, if you were bronze and happened to be matched with a much higher rated diamond player because the two of you were the only two on the server, a win would GREATLY increase your MMR while a loss would have little effect. Now, that's what happens to your hidden "matchmaking rating," or MMR. How many points you get or don't get on the ladder ranking screen is based on a comparison between your *points* (adjusted based on your division and subtracting bonus points you've earned) and your opponent's hidden matchmaking rating. This works this way because the system wants your adjusted points to converge on your MMR (which should be pretty close to your opponent's MMR) as fast as possible, so that ladder rankings for active players reflect their underlying MMR score. To answer your question directly: if you're starting a game, only your current MMR, the current MMR of your opponent, and whether you win or lose affects what your MMR will be after the game. Your win/loss history is already entirely taken into account by your current MMR and has no effect beyond that. As for promotions, because single games can make the MMR swing around a bunch pretty fast, promotions are based on a moving average of MMR that smooths out the swings. I don't think anyone outside Blizzard knows over how many games the MMR is averaged, but it's probably in the 10-30 range. | ||
Zog
57 Posts
On April 14 2011 22:35 akaMadMike wrote: Then to my real question: Since you now have lost more or less 40-50 games and your win ratio is about 30%. Will this mean you have to win alot of games to make it to about 50% win ratio before the MMR cosider you good enough to promote? Will it be better to practice alot of KOTH or custom rather than destroing your win/loss ratio? Yeah, probably you will have to win as many games as you already lost before going out of bronze ( or before the system matches with silver players). Your MMR must be really low with 30% ratio. Destroying your win / loss ratio is not a good idea, you can lower it till you go down one league if you feel the need, but shouldn't more. I free lost 60 games in a row to go from diamond to bronze (race switch), I had to win at least 30 more games than defeats before going out of bronze. That was a bit of overkill. | ||
akaMadMike
Norway93 Posts
On April 15 2011 01:33 Zog wrote: Yeah, probably you will have to win as many games as you already lost before going out of bronze ( or before the system matches with silver players). Your MMR must be really low with 30% ratio. Destroying your win / loss ratio is not a good idea, you can lower it till you go down one league if you feel the need, but shouldn't more. I free lost 60 games in a row to go from diamond to bronze (race switch), I had to win at least 30 more games than defeats before going out of bronze. That was a bit of overkill. Thanks! So I guess if you got a friend which recently bought the game just to play team with you and a few others, one will come better off to start prac in custom games rather than trowing away the first 15-20 matches or so..? | ||
Lysenko
Iceland2128 Posts
On April 15 2011 01:47 akaMadMike wrote: Thanks! So I guess if you got a friend which recently bought the game just to play team with you and a few others, one will come better off to start prac in custom games rather than trowing away the first 15-20 matches or so..? In principle it should not make a difference, but I'd say yes, only because it appears that 1.3 did not entirely fix the bug that made it difficult for Bronze players with negative MMR ratings and zero points to advance. | ||
Kyamo
Canada129 Posts
Edited to remove the data I added, since I realized it was probably the promotion game, so useless data. | ||
![]()
Excalibur_Z
United States12235 Posts
On April 16 2011 07:46 Kyamo wrote: So has anyone found a 12/12 game between a grandmasters and masters player, to try to figure out the offset yet? Edited to remove the data I added, since I realized it was probably the promotion game, so useless data. I'll keep an eye out, but I'm only one guy! | ||
SDream
Brazil896 Posts
On April 16 2011 13:21 Excalibur_Z wrote: I'll keep an eye out, but I'm only one guy! All I can find is -23 +2 games between GMs and diamonds, sorry guys, it might take some months ;_; I recomend stalking people like rootcatz, cause he's master and will probably fight GM pretty often (I'd hope...). Edit: right now 11 players on LA region GM league have less than 30 points, which means negative adjusted points and a pretty high offset on GM league. | ||
![]()
Excalibur_Z
United States12235 Posts
12 GM 253 -7 Master 546 12 GM 241 -11 Master 666 -11 GM 216 12-15? Master 588 -12 GM 228 15 Master 558 -12 GM 206 12 Master 637 12 GM 198 -11 Master 598 -12 GM 166 13 Master 560 -12 GM 178 13 Master 573 12 GM 134 -9 Master 480 12 GM 68 -12 Master 484 12 GM 56 -12 Master 496 So looks like it's anywhere from 350-450 from Master league. We also don't know the variance of a +12/-12 pairing... maybe 40-50 points? I'll keep looking. | ||
SDream
Brazil896 Posts
On April 17 2011 02:22 Excalibur_Z wrote: Rating Change and Adjusted Points before the game for both players: 12 GM 253 -7 Master 546 12 GM 241 -11 Master 666 -11 GM 216 12-15? Master 588 -12 GM 228 15 Master 558 -12 GM 206 12 Master 637 12 GM 198 -11 Master 598 -12 GM 166 13 Master 560 -12 GM 178 13 Master 573 12 GM 134 -9 Master 480 12 GM 68 -12 Master 484 12 GM 56 -12 Master 496 So looks like it's anywhere from 350-450 from Master league. We also don't know the variance of a +12/-12 pairing... maybe 40-50 points? I'll keep looking. It seems to be something around 430 (390-460) probably 410-440, so we are getting closer, I'd say chances are it's 420-436 with 99% confidence. Also there is a wrong information about season 2 in the OP. Divisions names won't be used twice, so the division tier won't change (but it won't be usefull for the season 2 so... yeah). Variance for 12/12 was usually 5 or less, but it could go up to very high levels as a rare occasion. | ||
DiDigital
75 Posts
Is this line correct? I put these games into excel and ran solver to calculate league offset and point variance based on points earned. Since these are all at least 12 point games (I threw out the one with the ?) for one player and there are no division tiers in these leagues excel should spit out the answer pretty easily. Solver concluded that point variance is roughly 28 for every point away from 12 with a league offset of 430. The problem though is that it could not meet all of the constraints I laid out for it, specifically that the amount of points left unexplained by the result be less than half of the estimate point variance. In this case the line in question is off by 23 points from the estimated total. If each point you win represents a roughly 28 point differential between players then the variance should never be more than 14 points, otherwise the number of points won/loss should be rounded up or down. When you do this same test on lower leagues you have the same problem, but this is caused by the division tiers messing with points won/loss. Obviously this assumes that these calculations are linear, which is unlikely, but that is probably not the issue here as the players should have similar mmrs and this is a big outlier. Do you have any more 12 point gm/m games? | ||
Mendelfist
Sweden356 Posts
On April 29 2011 10:17 DiDigital wrote: Solver concluded that point variance is roughly 28 for every point away from 12 I came up with the number 27 for gold league earlier: http://www.teamliquid.net/forum/viewmessage.php?topic_id=169830¤tpage=41#801 I'm glad our numbers agree. :-) | ||
DiDigital
75 Posts
| ||
Kyamo
Canada129 Posts
| ||
| ||